1: \def\brr#1{\left(#1\right)}
2: \def\brf#1{\left\{#1\right\}}
3: \def\z{{\bf z}}
4: \def\u{{\bf u}}
5: \def\sign{\mathop{\rm sign}\nolimits}
6: \def\bbar#1{\overline{#1}}
7:
8:
9: \documentclass[12pt]{article}
10:
11: %%% Packages
12: \usepackage{amsfonts}
13: \usepackage{amsmath}
14: \usepackage{amsthm}
15: \usepackage[dvips]{color}
16: \usepackage[dvips]{graphicx}
17: %%%
18: \newtheorem{prop}{Proposition}
19: %\renewcommand{\theprop}{\!\!}
20:
21: \begin{document}
22:
23: \begin{titlepage}
24: \title{Star graphs and \v Seba billiards}
25: \author{G.~Berkolaiko$^1$, E.B.~Bogomolny,$^2$ and
26: J.P.~Keating$^{3,4}$\\
27: \\
28: \small\it
29: $^1$Dept. of Physics of Complex Systems, Weizmann Institute of
30: Science,\\
31: \small\it Rehovot 76100, Israel\\
32: \small\it
33: $^2$Laboratoire de Physique Th\'eorique et Mod\`eles Statistiques,
34: Universit\'e Paris-Sud,\\
35: \small\it 91405 Orsay Cedex, France\\
36: \small\it
37: $^3$ School of Mathematics, University of Bristol, Bristol BS8 1TW,
38: UK\\
39: \small\it
40: $^4$BRIMS, Hewlett-Packard Laboratories, Filton Road, Stoke
41: Gifford,\\
42: \small\it Bristol BS34 8QZ, UK}
43: \end{titlepage}
44:
45: \maketitle
46: \begin{abstract}
47: We derive an exact expression for the two-point correlation function
48: for quantum star graphs in the limit as the number of bonds tends to
49: infinity. This turns out to be identical to the corresponding result
50: for certain \v Seba billiards in the semiclassical limit. Reasons for this are
51: discussed. The formula we derive is also shown to be equivalent to a
52: series expansion for the form factor --- the Fourier transform of the
53: two-point correlation function --- previously calculated using periodic
54: orbit theory.
55: \end{abstract}
56: \thispagestyle{empty}
57: \newpage
58:
59: \section{Introduction}
60: The statistical distribution of quantum energy levels is a much
61: studied topic. It has been conjectured that generic, classically
62: integrable systems give rise to uncorrelated quantum spectra
63: \cite{BT}, while the energy levels of generic classically chaotic
64: systems have the same statistical properties as the eigenvalues of
65: random matrices \cite{BohGS}. This has been confirmed by
66: semiclassical theory \cite{Berry,Bog-Kea}, and in a large number of
67: numerical studies, but classes of systems have also been found for
68: which it is not true; these include geodesic motion on surfaces of
69: constant negative curvature \cite{BGGS}, and the cat maps \cite{Kea}.
70:
71: Quantum graphs \cite{KS1,KS2} are mathematical models introduced in order
72: to explore the connection between the periodic orbits of a system and
73: the statistical properties of its energy levels. The trace formula,
74: in which the level density is connected to
75: a sum over periodic orbits, is exact for graphs, rather than a semiclassical
76: approximation, and the orbits can be classified straightforwardly.
77: However, despite the fact that numerical computations have
78: revealed good conformance of the spectral statistics of many quantum graphs to
79: the predictions of Random Matrix Theory (RMT), few conclusive
80: analytical results have been obtained so far. This is due to
81: the fact that although some individual finite graphs can be shown to
82: reproduce certain features of RMT behaviour \cite{SS,Tan,Gas}, the
83: full RMT results can only be recovered in a limit in which one is
84: forced to consider larger and larger graphs, and this necessitates
85: finding general, combinatorial asymptotic techniques for dealing with
86: the (non-trivial) length degeneracies of the periodic orbits.
87:
88: One family of graphs in which this goal has been achieved are the star
89: graphs \cite{BK} (defined below and shown in Fig.~\ref{fig:star}),
90: but in this case the resulting spectral statistics are neither RMT nor
91: Poissonian (i.e.~those of random numbers).
92: It turns out, however, that it is not the first time that such
93: statistics have arisen in the connection with the study of quantum
94: chaos.
95: Our purpose here is to demonstrate that the star graphs have
96: exactly the same two point spectral correlations as a large class of
97: quantum systems, which we will refer to as {\em \v Seba billiards}.
98:
99: The original \v Seba billiard, a rectangular quantum billiard
100: perturbed by a point singularity (also illustrated in
101: Fig.~\ref{fig:star}), was introduced in \cite{S} as an example of a
102: system whose classical counterpart is integrable (the singularity
103: affects only a set of measure zero of the orbits) but which nonetheless
104: exhibits properties of quantum chaos. This construction was later
105: generalized to all integrable systems \cite{AS} perturbed in the same
106: way. We will refer to any
107: system in this class as a \v Seba billiard.
108:
109: The energy levels of a \v Seba billiard can be found by solving an
110: explicit equation which depends on the levels of the
111: original unperturbed system and on the boundary conditions imposed
112: at the singularity. This equation takes the general form
113: \begin{equation}
114: \label{eq:Sebacond}
115: \lambda\xi(z)=1,
116: \end{equation}
117: where $\xi(z)$ is the meromorphic function
118: \begin{equation}
119: \label{eq:SE_Seba}
120: \xi(z) = \sum_n \frac{|\psi_n({\bf x_0})|^2}{E_n-z},
121: \end{equation}
122: the sum being suitably regularized to ensure convergence. Here
123: $\brf{E_i}$ are the eigenvalues of the unperturbed system, $\psi_n({\bf x_0})$
124: is the value of the $n$th unperturbed eigenfunction at the position
125: ${\bf x_0}$ of the singularity, and the coupling constant
126: $\lambda$ parametrizes the boundary conditions \cite{S,AS}.
127: Assuming that $\brf{E_i}$ are given
128: by a Poisson process, one can then calculate the associated
129: spectral statistics, such as
130: the joint level distribution,
131: asymptotics of the level spacing distribution \cite{AS}, and the
132: two-point spectral correlation function \cite{BGS}. The results
133: show the presence of spectral correlations but are
134: substantially different from the RMT forms.
135:
136: Here we apply the methods developed for \v Seba billiards in \cite{BGS}
137: to calculate the two-point
138: spectral correlation function for star graphs, starting from an expression
139: which is analogous to (\ref{eq:SE_Seba}). The formula obtained will
140: be shown to be a resummation of the expansion computed from the
141: periodic orbit sum in \cite{BK}. Our main result will be that
142: this correlation function is the same as that already found for \v Seba
143: billiards in the case when $|\psi_n({\bf x_0})|^2={\rm constant}$ (e.g.~when
144: the billiard is rectangular with periodic boundary conditions) and
145: $\lambda \rightarrow \infty$. We finish with a discussion of reasons
146: for this coincidence.
147:
148:
149: \begin{figure}
150: \label{fig:star}
151: \vskip 7cm
152: \special{psfile="star_seba.eps" voffset=0 hoffset=0 hscale=80
153: vscale=80}
154: \caption{A star graph with $v$ edges (a) and a \v Seba billiard (b).}
155: \end{figure}
156:
157:
158:
159: %%%%%%%%%%%%%%%%%%%%%%%% New Section %%%%%%%%%%%%%%%%%%%%%%%%%%%%
160:
161: \section{Quantum star graphs}
162: Star graphs are metric graphs of the
163: type shown on Fig.~\ref{fig:star} with a Schr\"odinger equation
164: \begin{equation}
165: \label{eq:schrod}
166: -\frac{d^2}{d x_j^2} \Psi_j = k^2 \Psi_j, \quad x_j\in[0,L_j],
167: \end{equation}
168: defined on the bonds and boundary conditions, for example
169: \begin{gather}
170: \Psi_j(0) = \Psi_k(0), \label{eq:BC1}\\
171: \sum_j \frac{\partial}{\partial x_j} \Psi_j(0) = 0, \label{eq:BC2}\\
172: \frac{\partial}{\partial x_j} \Psi_j(L_j) = 0,
173: \label{eq:BC3}
174: \end{gather}
175: specified on the vertices. Here $L_j$ is the length of
176: the $j$-th bond, $j=1\ldots v$, and the real variable $x_j$
177: varies from 0 to $L_j$,
178: with 0 corresponding to the central vertex and $L_j$
179: to the outer vertex. The lengths $L_j$ are assumed to be
180: incommensurate; see \cite{BK} for further details.
181: We refer to positive values of the parameter $k$
182: for which the system~(\ref{eq:schrod})-(\ref{eq:BC3}) is solvable as
183: eigenvalues of the quantum star graph.
184:
185: Denoting the ordered sequence of eigenvalues by
186: $\brf{k_i}_{i=1}^\infty$, we define the spectral density by
187: \begin{equation}
188: \label{eq:def_dens}
189: d(k)=\sum_{i=1}^\infty \delta(k-k_i).
190: \end{equation}
191: The statistic we shall mainly be concerned with is the two-point
192: correlation function
193: \begin{equation}
194: \label{eq:def_r2}
195: R_2(x) = \frac1{\bbar{d}^2}\left\langle
196: d(k)d\brr{k+\frac{x}{\bbar{d}}} \right\rangle-\delta(x),
197: \end{equation}
198: where $\bbar{d} = \langle d(k) \rangle$ is the mean density, $\delta(x)$ is
199: the Dirac $\delta$-function, and the
200: average $\langle\ \cdot\ \rangle$ is either over $k$, or over the bond
201: lengths $L_j$ (we shall specify which in each particular context).
202: $R_2(x)$ is an even function and hence so is its Fourier transform,
203: \begin{equation}
204: \label{eq:def_Ktau}
205: K(\tau) = 1+2\Re \int_0^\infty (R_2(x)-1)e^{2\pi ix\tau} d\tau,
206: \end{equation}
207: which is called the form factor.
208:
209: A complete series expansion of the $v \rightarrow \infty$ limit of
210: $K(\tau)$ in powers of $\tau$ around $\tau=0$ was derived
211: for the star graphs in \cite{BK} using the trace formula and a
212: classification of the periodic orbits:
213: \begin{equation}
214: \label{Ktau}
215: K(\tau) = \exp(-4\tau) + \sum_{j=2}^\infty\sum_{M=0}^\infty
216: \frac{4^j}{j!}C_{j,M} \tau^{M+j+1},
217: \end{equation}
218: where
219: \begin{equation}
220: C_{j,M} = (-2)^M \sum_{K=0}^M \frac{(K+j-1)!(M-K+j-1)!}{(M+j-1)!}
221: F_j(K,M-K),
222: \label{eq:CM1}
223: \end{equation}
224: with
225: \begin{equation}
226: \label{eq:notation_F}
227: F_1(K,N) = \frac{\binom{K+N}{N}}{(N+1)!(K+1)!},
228: \end{equation}
229: and
230: \begin{equation}
231: \label{eq:FKN}
232: F_j(K,N) = \sum_{k=0}^K\sum_{n=0}^N F_1(k,n)F_{j-1}(K-k,N-n).
233: \end{equation}
234: Explicitly,
235: \begin{equation}
236: \label{eq:Ktau_firstfew}
237: K(\tau) = e^{-4\tau} + 8\tau^3 - \frac{32}{3}\tau^4 +
238: \frac{16}{3}\tau^5 - \frac{128}{15}\tau^6 + \frac{16}{9}\tau^7 +
239: \frac{64}{63}\tau^8 + o(\tau^8).
240: \end{equation}
241: In this calculation, the average in (\ref{eq:def_r2}) was over $k$.
242: The result is in excellent agreement with the
243: numerical data (see Fig.~\ref{fig:pade}) but is limited by the fact that
244: the radius of
245: convergence of the series is finite, being approximately $0.63$ (found
246: by applying Cauchy's test to the coefficients in the series, but see also
247: Fig.~\ref{fig:pade}). The
248: range of convergence can be extended using Pad\'e
249: approximation (again, see Fig.~\ref{fig:pade}), which suggests that the
250: singularity causing the divergence is not on the positive real line
251: \cite{thesis}.
252:
253: \begin{figure}[t]
254: \vskip 8.5cm
255: \special{psfile="pade_res_bbk.eps" voffset=-25 hoffset=-10
256: hscale=90 vscale=80}
257: \caption{The sum of the first 30 terms in the expansion~(\ref{Ktau})
258: (dashed line), which converges in the range
259: $\tau\le\tau^*\approx0.63$, compared to the results of a numerical
260: computation \cite{KS2} of $K(\tau)$ (circles). Also shown
261: are the Pad\'e approximations to the series of order
262: $[21/20]$ (thin
263: solid line) and $[23/23]$ (thick solid line).}
264: \label{fig:pade}
265: \end{figure}
266:
267:
268: Here we approach the problem from a different direction:
269: it is possible to solve equations (\ref{eq:schrod})-(\ref{eq:BC3}) to
270: derive an explicit condition on $k$ to be an eigenvalue.
271: Indeed, the general solution of (\ref{eq:schrod}) on a star graph
272: can be written in the form $\Psi_j(x) = A_j\cos(k(x+\phi_j))$,
273: $j=1,\ldots,v$. Applying condition (\ref{eq:BC3}), we obtain
274: $\phi_j=-L_j$ while condition~(\ref{eq:BC1}) on the central vertex
275: implies $A_j\cos(L_jk) = \mbox{const}$. Finally, applying
276: condition~(\ref{eq:BC2}) and dividing by $A_j\cos(L_jk)$ we obtain
277: \begin{equation}
278: \label{eq:SE}
279: \sum_{j=1}^v \tan{L_jk} = 0.
280: \end{equation}
281: Similar expressions can easily be found when different boundary
282: conditions are applied at the central vertex. The general equation
283: reads
284: \begin{equation}
285: \label{eq:SE_gen}
286: \sum_{j=1}^v \tan{L_jk} = \frac{1}{\lambda},
287: \end{equation}
288: where $\lambda$ is arbitrary parameter. However, in the limit as
289: $v\to \infty$, $\lambda$ fixed, the two-point correlation function turns
290: out to be independent of $\lambda$ (see the comment following equation
291: (\ref{eq:ren_r2})). Our calculations
292: will therefore be performed for $\lambda^{-1}=0$.
293:
294: Note the similarity between (\ref{eq:SE_gen}) and the quantization
295: condition (\ref{eq:Sebacond}) for \v Seba billiards when
296: $|\psi_n({\bf x_0})|^2={\rm constant}$.
297:
298: Condition (\ref{eq:SE}) means that $k$ is an eigenvalue if and only if
299: it is a zero of the function $F(k) = \sum_{j=1}^v \tan L_jk$, and so
300: we can express the density $d(k)$ as
301: \begin{equation}
302: \label{eq:dens_int}
303: d(k) = \frac1{2\pi}\int |F'(k)| e^{izF(k)} dz = \frac1{2\pi}\int
304: \sum_{s=1}^v\frac{L_s}{\cos^2{L_sk}} e^{iz\sum_{j=1}^v \tan{L_jk}}
305: dz.
306: \end{equation}
307: Our analysis of the spectral correlations will be based on this representation.
308:
309: %%%%%%%%%%%%%%%%%%%%%%%%% New Section %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
310: \section{Mean density.}
311:
312: As an example of the techniques to be employed later, we begin by
313: calculating the mean density $\bbar{d}$
314: defined as
315: \begin{equation}
316: \label{eq:dens_def}
317: \bbar{d} = \lim_{\Delta L \to 0, {k\to \infty}} \Big\langle d(k)
318: \Big\rangle_{\brf{L_j}}
319: \end{equation}
320: where now the average is with
321: respect to the individual lengths of the bonds, rather than over $k$:
322: \begin{equation}
323: \label{eq:aver}
324: \langle\ \cdot\ \rangle_{\brf{L_j}} = \int_{L_0}^{L_0+\Delta L}
325: \hspace{-5mm}\cdots \int_{L_0}^{L_0+\Delta L}\cdot\hspace{5mm}
326: \frac{dL_1}{\Delta L}\cdots\frac{dL_v}{\Delta L}.
327: \end{equation}
328: That is, we assume that the lengths are independent random variables
329: distributed uniformly on the interval $[L_0, L_0+\Delta L]$. We also
330: assume that $\Delta L$ and $k$ tend to their respective limits in such
331: a way that $\Delta L k\to\infty$.
332:
333: Applying this averaging to (\ref{eq:dens_int}) we obtain
334: \begin{eqnarray}
335: \Big\langle d(k) \Big\rangle_{\brf{L_j}}\hspace{-3mm}
336: &=& \frac1{2\pi} \int_{-\infty}^\infty dz \sum_{s=1}^v \int
337: \cdots \int_{L_0}^{L_0+\Delta L} L_s\frac{e^{iz\sum_{j=1}^v
338: \tan{kL_j}}}{\cos^2{kL_s}} \frac{dL_1}{\Delta L} \cdots
339: \frac{dL_v}{\Delta L} \nonumber\\
340: &=& \frac{v}{2\pi} \int_{-\infty}^\infty dz
341: \brr{\int_{L_0}^{L_0+\Delta L}\hspace{-0.3cm}e^{iz\tan{kL}}
342: \frac{dL}{\Delta L}}^{v-1}
343: \brr{\int_{L_0}^{L_0+\Delta L} \hspace{-0.3cm}
344: L\frac{e^{iz\tan{kL}}}{\cos^2{kL}} \frac{dL}{\Delta L}} \nonumber\\
345: \label{eq:calc_av}
346: &\equiv& \frac{v}{2\pi} \int_{-\infty}^\infty {\tilde f}^{v-1}(z)
347: {\tilde g}(z) \ dz.
348: \end{eqnarray}
349: Here
350: \begin{equation}
351: \label{eq:g_av}
352: {\tilde g}(z) = \int_{L_0}^{L_0+\Delta L}L\frac{e^{iz\tan{kL}}}{\cos^2{kL}}
353: \frac{dL}{\Delta L} \approx \frac{L_0}{\Delta L k}
354: \int_{\tan{kL_0}}^{\tan{k(L_0+\Delta L)}} e^{iz\tan{kL}}\ d\tan{kL},
355: \end{equation}
356: where we were able to approximate $L$ by $L_0$ because it is slowly
357: varying (compared with $\tan{kL}$) and ultimately we will take the
358: limit $\Delta L \to 0$. Now, since
359: $\tan{kL}$ is a periodic function with the period of $\pi/k$, and the
360: integration is performed over the interval containing approximately
361: $\Delta L k/\pi$ periods, we can further approximate
362: \begin{eqnarray}
363: \label{eq:g_av2}
364: {\tilde g}(z) &=& \frac{L_0}{\Delta L k}
365: \brr{\frac{\Delta L k}{\pi}\int_{-\infty}^{\infty}
366: e^{iz\tan{kL}}\ d\tan{kL} + O(1)} \nonumber\\
367: &\approx& 2L_0\delta(z),
368: \end{eqnarray}
369: where $O(1)$ is a quantity which is bounded as $k\Delta L \to\infty$.
370: Similarly,
371: \begin{eqnarray}
372: \label{eq:f_av}
373: {\tilde f}(z) &=& \int_{L_0}^{L_0+\Delta L} e^{iz\tan{kL}} \frac{dL}{\Delta
374: L} = \frac{L_0}{\Delta L k} \int_{\tan{kL_0}}^{\tan{k(L_0+\Delta L)}}
375: e^{iz\tan{kL}} \frac{d\tan{kL}}{1+\tan^2{kL}} \nonumber\\
376: &\approx& \frac1{\pi} \int_{-\infty}^\infty
377: \frac{e^{iz\alpha}}{1+\alpha^2} d\alpha = e^{-|z|},
378: \end{eqnarray}
379: where the last integral was evaluated by closing the contour in either the
380: upper ($z>0$) or lower ($z<0$) half-plane.
381:
382: Substituting the results into (\ref{eq:calc_av}) we obtain for the
383: average density
384: \begin{equation}
385: \label{eq:av_d_res}
386: \bbar{d} = \frac{v}{2\pi}2L_0\int_{-\infty}^\infty
387: e^{-(v-1)|z|}\delta(z)dz = \frac{L_0v}{\pi},
388: \end{equation}
389: which coincides with the result of averaging over $k$ with the
390: bond-lengths fixed \cite{KS1,KS2,BK}.
391:
392:
393: %%%%%%%%%%%%%%%%%%%%%%%%% New Section %%%%%%%%%%%%%%%%%%%%%%%%%%%%%
394: \section{Two-point correlation function}
395: The two-point correlation function is
396: given by
397: \begin{equation}
398: \label{eq:twopntnorm}
399: R_2(x) = \lim_{\Delta L \to 0, k\to\infty} \frac1{{\bbar{d}}^2}
400: R\brr{k,k+\frac{x}{\bbar{d}}},
401: \end{equation}
402: where $\bbar{d}$ is the mean density, the limit is taken in such a way
403: that $k\Delta L \to\infty$, and we take
404: \begin{eqnarray}
405: \label{eq:twopnt}
406: R(k_1,k_2) &=& \langle d(k_1) d(k_2) \rangle_{\brf{L_j}} \\
407: \nonumber
408: &=& \left\langle \int_{-\infty}^\infty \sum_{r,s=1}^v
409: \frac{L_rL_s e^{i\sum_{j=1}^v(z_1\tan{k_1L_j}+z_2\tan{k_2L_j})}}
410: {\cos^2k_1L_r\cos^2k_2L_s}
411: \frac{d\z}{4\pi^2} \right\rangle_{\brf{L_j}},
412: \end{eqnarray}
413: with $\z = (z_1,z_2)$.
414:
415: In this case, the analogue of (\ref{eq:calc_av}) is that
416: \begin{eqnarray}
417: \label{eq:twopntred}
418: R(k_1,k_2) = \int_{-\infty}^\infty
419: \left\{vg(\z)f^{v-1}(\z)
420: + v(v-1)\phi_1(\z)\phi_2(\z)f^{v-2}(\z)\right\} \frac{d\z}{4\pi^2},
421: \end{eqnarray}
422: where
423: \begin{eqnarray}
424: \label{eq:intf}
425: f(\z) &=& \frac1{\Delta L}\int_{L_0}^{L_0+\Delta L}
426: e^{i(z_1\tan(k_1L)+z_2\tan(k_2L))} dL,\\
427: \label{eq:intg}
428: g(\z) &=& \frac1{\Delta L}\int_{L_0}^{L_0+\Delta L}
429: \frac{L^2}{\cos^2k_1L\cos^2k_2L}
430: e^{i(z_1\tan(k_1L)+z_2\tan(k_2L))} dL,\\
431: \label{eq:intphi1}
432: \phi_1(\z) &=& \frac1{\Delta L}\int_{L_0}^{L_0+\Delta L}
433: \frac{L}{\cos^2k_1L} e^{i(z_1\tan(k_1L)+z_2\tan(k_2L))} dL,\\
434: \label{eq:intphi2}
435: \phi_2(\z) &=& \frac1{\Delta L}\int_{L_0}^{L_0+\Delta L}
436: \frac{L}{\cos^2k_2L} e^{i(z_1\tan(k_1L)+z_2\tan(k_2L))} dL.
437: \end{eqnarray}
438:
439: Substituting $k_1=k$, $k_2=k+\pi x/(vL_0)$, where $x$
440: is fixed, and taking
441: the limits
442: $k\to \infty$, $\Delta L\to 0$ (while $k\Delta L \to \infty$), we
443: obtain for the first integral
444: \begin{eqnarray}
445: f(\z) &=& \frac1{\Delta L}\int_{L_0}^{L_0+\Delta L}
446: e^{i\brr{z_1\tan(kL)+z_2\tan\brr{kL + \frac{\pi xL}{vL_0}}}} dL
447: \nonumber\\
448: \label{eq:f_step1}
449: &\approx&
450: \frac1\pi \int_{-\pi/2}^{\pi/2} e^{i\brr{z_1\tan\phi+z_2\tan\brr{\phi +
451: \frac{\pi x}{v}}}} d\phi,
452: \end{eqnarray}
453: where we have again used $L/L_0\approx 1$ and, as in the
454: transition from (\ref{eq:g_av}) to (\ref{eq:g_av2}), we have
455: approximated $f$ by the integral over one period. We now write
456: \begin{equation}
457: \label{eq:tans}
458: \tan\brr{\phi+\frac{\pi x}{v}} = \frac{\tan\phi + \tan\brr{\frac{\pi
459: x}{v}}} {1-\tan\phi\tan\brr{\frac{\pi x}{v}}} = -\beta +
460: \frac{1+\beta^2}{\beta-\tan\phi},
461: \end{equation}
462: where $\beta = (\tan(\pi x/v))^{-1} \propto v/(\pi x)$ (we are
463: interested in the $v\to\infty$ limit). Performing the change of variables
464: $\alpha=\tan\phi-\beta$, we arrive at
465: \begin{equation}
466: \label{eq:bessel}
467: f(\z) \approx \frac{e^{i\beta(z_1-z_2)}}{\pi}
468: \int_{-\infty}^\infty e^{iz_1\alpha - iz_2\frac{\beta^2+1}{\alpha}}
469: \frac{d\alpha}{(\alpha+\beta)^2+1}.
470: \end{equation}
471: Note that $f(\z)$ is invariant under the exchange $z_1\leftrightarrow
472: z_2$ and $\beta\to-\beta$, which can be verified by the change of
473: variables $\alpha=(\beta^2+1)/y$ in (\ref{eq:bessel}).
474:
475: To evaluate the integral in (\ref{eq:bessel}) we
476: differentiate it with respect to $z_1$ and $z_2$ to get
477: \begin{eqnarray}
478: \frac{\partial f}{\partial z_1} - \frac{\partial f}{\partial z_2}
479: \!&=&\! \frac{ie^{i\beta(z_1-z_2)}}{\pi}
480: \int_{-\infty}^\infty e^{iz_1\alpha - iz_2\frac{\beta^2+1}{\alpha}}
481: \brr{2\beta+\alpha+\frac{\beta^2+1}{\alpha}}\frac{d\alpha}{(\alpha+\beta)^2+1}\nonumber\\
482: \label{eq:diff}
483: &=&\! \frac{ie^{i\beta(z_1-z_2)}}{\pi}
484: \int_{-\infty}^\infty e^{iz_1\alpha - iz_2\frac{\beta^2+1}{\alpha}}
485: \frac{d\alpha}{\alpha} = -e^{i\beta(z_1-z_2)} \Phi(z_1,z_2),
486: \end{eqnarray}
487: where
488: \begin{eqnarray}
489: \label{eq:Phi_def}
490: \Phi(z_1,z_2) &\equiv& -\frac{i}{\pi}\int_{-\infty}^\infty
491: e^{iz_1\alpha - iz_2\frac{\beta^2+1}{\alpha}} \frac{d\alpha}{\alpha}\\
492: &=& 2\sign(z_1)H(-z_1z_2)
493: J_0\brr{2\sqrt{-(\beta^2+1)z_1z_2}}, \nonumber
494: \end{eqnarray}
495: $J_0(x)$ being the Bessel function of the first kind and $H(x)$
496: the Heaviside function (characteristic function of the half axis
497: $[0,\infty)$).
498:
499: Applying the method of characteristics to the PDE
500: \begin{equation}
501: \label{eq:pde}
502: \frac{\partial f}{\partial z_1} - \frac{\partial f}{\partial z_2} =
503: -e^{i\beta(z_1-z_2)} \Phi(z_1,z_2),
504: \end{equation}
505: we obtain the solution
506: \begin{equation}
507: \label{eq:ans_f}
508: f(\z) = e^{-|z_1+z_2|} - \int_0^{z_1}
509: e^{i\beta(2y-z_1-z_2)}\Phi\brr{y,z_1+z_2-y}dy.
510: \end{equation}
511:
512: \medskip
513: Treating the integral for $g(\z)$ (see (\ref{eq:intg})) in a
514: fashion similar to the one used to obtain (\ref{eq:bessel}) leads us to
515: \begin{eqnarray}
516: \label{eq:bessel_g}
517: g(z_1,z_2)
518: &\approx& \frac{L_0^2}{\pi}\int_{-\pi/2}^{\pi/2}
519: \frac{e^{i(z_1\tan(\phi)+z_2\tan(\phi+\pi x/v))}}
520: {\cos^2(\phi)\cos^2(\phi+\pi x/v)} d\phi \\
521: &=&
522: L_0^2\frac{e^{i\beta(z_1-z_2)}}{\pi}
523: \int_{-\infty}^\infty e^{iz_1\alpha - iz_2\frac{\beta^2+1}{\alpha}}
524: \brr{1+\brr{\frac{1+\beta^2}{\alpha}+\beta}^2} d\alpha.\nonumber
525: \end{eqnarray}
526: Comparing this integral to the one in
527: (\ref{eq:Phi_def}), and noting that
528: \begin{equation}
529: \label{eq:rubbish}
530: 1+\brr{\frac{1+\beta^2}{\alpha}+\beta}^2 =
531: \frac{\beta^2+1}{\alpha}\brr{\alpha+\beta+\frac{\beta^2+1}{\alpha}+\beta},
532: \end{equation}
533: we have that
534: \begin{equation}
535: \label{eq:rep_g}
536: g(\z) = L_0^2(\beta^2+1)\brr{\frac{\partial}{\partial z_1} -
537: \frac{\partial}{\partial z_2}}\left[ e^{i\beta(z_1-z_2)}
538: \Phi(z_1,z_2) \right].
539: \end{equation}
540:
541: One can derive a similar expression for the functions $\phi_1(\z)$,
542: \begin{equation}
543: \label{eq:rep_phi1}
544: \phi_1(\z) \approx L_0 \frac{e^{i\beta(z_1-z_2)}}{\pi}
545: \int_{-\infty}^\infty e^{iz_1\alpha - iz_2\frac{\beta^2+1}{\alpha}} d\alpha = L_0
546: e^{i\beta(z_1-z_2)} \frac{\partial}{\partial z_1} \Phi(z_1,z_2),
547: \end{equation}
548: and $\phi_2(\z)$,
549: \begin{eqnarray}
550: \label{eq:rep_phi2}
551: \phi_2(\z) &\approx& L_0 \frac{e^{i\beta(z_1-z_2)}}{\pi}
552: \int_{-\infty}^\infty e^{iz_1\alpha - iz_2\frac{\beta^2+1}{\alpha}}
553: \frac{(\beta^2+1)d\alpha}{\alpha^2}\nonumber\\
554: &=&
555: - L_0 e^{i\beta(z_1-z_2)} \frac{\partial}{\partial z_2} \Phi(z_1,z_2) .
556: \end{eqnarray}
557:
558: Now we have all the ingredients necessary for evaluating the
559: integral in (\ref{eq:twopntred}).
560: Substituting the expression for $g(\z)$, (\ref{eq:rep_g}),
561: into the first half of the integral and integrating it by parts we obtain
562: \begin{eqnarray}
563: \label{eq:int_bp}
564: \int\frac{d\z}{4\pi^2} v f^{v-1}g
565: &=&vL_0^2\int \frac{d\z}{4\pi^2} f^{v-1}(\beta^2+1)
566: \brr{\frac{\partial}{\partial z_1}
567: - \frac{\partial}{\partial z_2}}\left[ e^{i\beta(z_1-z_2)}
568: \Phi \right]\nonumber\\
569: &=& -v L_0^2\int \frac{d\z}{4\pi^2} (\beta^2+1) e^{i\beta(z_1-z_2)}
570: \Phi \brr{\frac{\partial}{\partial z_1} -
571: \frac{\partial}{\partial z_2}}\left[ f^{v-1}(\z)\right] \nonumber\\
572: &=& v(v-1) L_0^2\int \frac{d\z}{4\pi^2} (\beta^2+1) f^{v-2}
573: e^{2i\beta(z_1-z_2)} \Phi^2.
574: \end{eqnarray}
575: Thus
576: \begin{equation}
577: \label{eq:r2_good}
578: R_2(x) = \frac{v(v-1) L_0^2}{\bbar{d}^2}\int \frac{d\z}{4\pi^2} f^{v-2}
579: e^{2i\beta(z_1-z_2)} \left[ (\beta^2+1) \Phi^2 -
580: \frac{\partial \Phi}{\partial z_1} \frac{\partial \Phi}{\partial
581: z_2}\right].
582: \end{equation}
583: Now we need to take the limit $v\to\infty$. To do so
584: we write $f^{v-2}(\z) = e^{(v-2)\ln f}$ and rescale $f(\z)$
585: \begin{equation}
586: \label{eq:f_rescaled}
587: f(\u/\beta) = e^{-\frac{|u_1+u_2|}{\beta}} - \frac1{\beta}
588: \int_0^{u_1} e^{i(2y-u_1-u_2)} \Psi(y,u_1+u_2-y)dy,
589: \end{equation}
590: and hence, to the leading order in $1/\beta = \pi x/v$, we have
591: \begin{eqnarray}
592: \label{eq:lnf}
593: (v-2)\ln f(\u) &\approx&
594: -\pi x\brr{|u_1+u_2|+\int_0^{u_1}e^{i(2y-u_1-u_2)}
595: \Psi(y,u_1+u_2-y)dy}\nonumber\\
596: &\equiv& -\pi xQ,
597: \end{eqnarray}
598: where $\Psi$ is the rescaled function $\Phi$,
599: \begin{equation}
600: \label{eq:Psi}
601: \Psi(\u) = \Phi\brr{\frac{\u}\beta} =
602: 2\sign(u_1)H(-u_1u_2) J_0\brr{2\sqrt{-u_1u_2}},
603: \end{equation}
604: and we have taken the limit $v\to\infty$ ($\beta\to\infty$).
605:
606: Renormalizing the rest of (\ref{eq:r2_good}) and taking the
607: limit $v\to\infty$ we obtain
608: \begin{equation}
609: \label{eq:ren_r2}
610: R_2(x) = \frac14\int d\u e^{-\pi x Q}
611: e^{2i(u_1-u_2)} \left[ \Psi^2 -
612: \frac{\partial \Psi}{\partial u_1} \frac{\partial
613: \Psi}{\partial u_2}\right].
614: \end{equation}
615: The only change when the above calculation is generalized to other
616: boundary conditions at the central vertex (i.e.~to nonzero values of
617: $\lambda^{-1}$ in (\ref{eq:SE_gen})) is the appearance of a factor
618: $e^{-\lambda^{-1} (z_1 + z_2)}$ next to every occurrence of $d{\bf z}$ in the
619: above integrals. For $\lambda$ fixed, this factor disappears after rescaling
620: $\z=\u/\beta$ and taking the limit $\beta\to\infty$. Hence
621: equation (\ref{eq:ren_r2}) is then independent of $\lambda$. In the case
622: when $\lambda^{-1}={\tilde \lambda}^{-1}v$, the dependence of the spectral
623: statistics on the
624: boundary conditions at the central vertex persists. The above expressions then
625: coincide with those for those for \v Seba billiards with a renormalized
626: coupling consant, given in \cite{BGS}.
627:
628: For the derivatives of the function $\Psi$ one has
629: \begin{eqnarray}
630: \label{eq:Psi_der}
631: \frac{\partial \Psi}{\partial u_1} &=&
632: 2\brr{J_0(0)\delta(u_1) +
633: \sign(u_1)H(-u_1u_2)
634: \frac{u_2J'_0\brr{2\sqrt{-u_1u_2}}}{\sqrt{-u_1u_2}}},\\
635: \frac{\partial \Psi}{\partial u_2} &=&
636: 2\brr{-J_0(0)\delta(u_2) +
637: \sign(u_1)H(-u_1u_2)
638: \frac{u_1J'_0\brr{2\sqrt{-u_1u_2}}}{\sqrt{-u_1u_2}}},
639: \end{eqnarray}
640: therefore, using $J_0(0) = 1$ and $J'_0(x)=-J_1(x)$,
641: \begin{equation}
642: \label{eq:Psi_prod_der}
643: \frac{\partial \Psi}{\partial u_1} \frac{\partial \Psi}{\partial
644: u_2} = -4\brr{\delta(u_1)\delta(u_2) +
645: H(-u_1u_2) J_1^2\brr{2\sqrt{-u_1u_2}}}.
646: \end{equation}
647: Thus
648: \begin{equation}
649: \label{eq:r2_almost_fin}
650: R_2(x) = 1 + \!\!\int\!\! e^{-\pi xQ+2i(u_1-u_2)}
651: \left[J_0^2\brr{2\sqrt{-u_1u_2}}+J_1^2\brr{2\sqrt{-u_1u_2}}\right]
652: \!H(-u_1u_2) d\u.
653: \end{equation}
654: Now we perform the change of variables $u_2 \mapsto -u_2$ arriving at
655: the following integral representation of the two-point correlation function,
656: \begin{equation}
657: \label{eq:r2_fin}
658: R_2(x) = 1 + \int_D e^{-\pi
659: xM(\u)+2i(u_1+u_2)}
660: \left[J_0^2\brr{2\sqrt{u_1u_2}}+J_1^2\brr{2\sqrt{u_1u_2}}\right]
661: d\u.
662: \end{equation}
663: Here the domain of integration $D$ includes first and third quadrants
664: of the $u_1u_2$-plane and $M(\u)$ is given by
665: \begin{eqnarray}
666: \nonumber
667: M(\u) &\equiv& M(u_1,u_2) = |u_1-u_2| +\int_0^{u_1}e^{i(2y-u_1+u_2)}
668: \Psi(y,u_1-u_2-y)dy\\
669: \label{M_def}
670: &=& |u_1|+|u_2| - 2i\sign(u_1)\sum_{r,s=1}^\infty
671: \frac{(iu_1)^r (iu_2)^s(r+s-2)!}{r!s!(r-1)!(s-1)!}.
672: \end{eqnarray}
673:
674:
675: Equation (\ref{eq:r2_fin}) constitutes an exact formula for $R_2(x)$ for star
676: graphs in the limit $v \rightarrow \infty$. It is our main result. The point
677: we seek to draw attention to is that it is exactly the same as the one
678: obtained in \cite{BGS} for \v Seba billiards
679: when $|\psi_n({\bf x_0})|^2={\rm constant}$ in (\ref{eq:SE_Seba}) and
680: $\lambda \rightarrow \infty$.
681: We will expand on this
682: observation later. First, we consider some of the properties of the
683: two-point correlation function and the form factor in more detail.
684:
685:
686: %%%%%%%%%%%%%%%%%%%%%%%%%%%% New Section %%%%%%%%%%%%%%%%%%%%%%%%%%
687: \section{Expansion for large $x$}
688:
689: To derive an expansion of the two point correlation function $R_2(x)$
690: for large $x$ we notice that since $M(-\u) = \bbar{M(\u)}$, the
691: integral over the third quadrant in (\ref{eq:r2_fin})
692: is equal to the complex conjugate of the integral over second
693: quarter-plane, i.e.
694: \begin{equation}
695: \label{eq:r2_x_large}
696: R_2(x) = 1 + 2\Re \int\!\!\int_0^\infty e^{-\pi
697: xM(\u)+2i(u_1+u_2)} J(\u) d\u,
698: \end{equation}
699: where
700: \begin{equation}
701: \label{eq:expans_bessel}
702: J(\u) = J_0^2(2\sqrt{u_1u_2})+J_1^2(2\sqrt{u_1u_2}) =
703: \sum_{n=0}^\infty \frac{(-1)^nu_1^nu_2^n(2n)!}{(n+1)!(n!)^3}.
704: \end{equation}
705: Now we can use the expansion of $M(\u)$, (\ref{M_def}), to expand
706: $R_2(x)$ in the powers of $1/x$. We substitute $u_i =
707: \gamma_i/(x\pi)$ and obtain
708: \begin{eqnarray}
709: \nonumber
710: R_2(x) &=& 1 + 2\Re \frac1{x^2\pi^2}\int\!\!\int_0^\infty
711: d\gamma_1d\gamma_2 e^{-\gamma_1-\gamma_2} \left[1+
712: \frac{2i\brr{\gamma_1+\gamma_2-\gamma_1\gamma_2}}{x\pi}\right.\\
713: &\hphantom{= 1}& - \left.
714: \frac{\brr{5\gamma_1\gamma_2 +
715: 2\gamma_1^2 + 2\gamma_2^2 - 5\gamma_1\gamma_2^2 -
716: 5\gamma_1^2\gamma_2 + 2\gamma_1^2\gamma_2^2}}{x^2\pi^2} +
717: O\brr{\frac1{x^3}} \right]
718: \nonumber\\
719: &=& 1 + 2\Re \left[\frac1{x^2\pi^2} + \frac{2i}{x^3\pi^3} -
720: \frac1{x^4\pi^4} + \ldots \right].
721: \label{eq:r2_exp}
722: \end{eqnarray}
723: To compare this to the expansion (\ref{eq:Ktau_firstfew}) of $K(\tau)$
724: we note that if $K(\tau) = 1 + \sum_{k=1}^\infty a_k \tau^k$ for
725: $\tau>0$ then, inverting the Fourier transform in~(\ref{eq:def_Ktau}),
726: \begin{eqnarray}
727: \label{eq:FT_form}
728: R_2(x) - 1 &=& 2 \Re \lim_{\epsilon\to0} {\int_0^\infty
729: (K(\tau)-1) e^{-2\pi i(x-i\epsilon)\tau}d\tau}\\
730: &=& 2 \Re{\sum_{k=1}^\infty
731: \brr{\frac{-i}{2\pi}}^{k+1} \frac{a_k k!}{x^{k+1}} }.
732: \end{eqnarray}
733: Applying this to
734: \begin{equation}
735: \label{eq:K_tau_exp}
736: K(\tau) = 1-4\tau+8\tau^2-\frac{8}{3}\tau^3 + O(\tau^4),
737: \end{equation}
738: we see that the first few coefficients of the two expansions
739: agree. The proof that it is so for all coefficients is given by the
740: following proposition.
741: \begin{prop}
742: The asymptotic expansion (\ref{eq:r2_exp}) of the two-point correlation
743: function and the expansion~(\ref{Ktau}) of the
744: form factor coincide under the Fourier transformation
745: \begin{equation}
746: \label{eq:agreement_theorem}
747: \int\!\!\int_0^\infty e^{-\pi xM(\u)+2i(u_1+u_2)} J(\u) d\u
748: = \int_0^\infty\brr{K(\tau')-1}e^{-2\pi ix\tau'}d\tau'.
749: \end{equation}
750: \end{prop}
751:
752: \begin{proof}
753: The Fourier transform in (\ref{eq:agreement_theorem})
754: establishes the correspondence between the terms in the asymptotic
755: expansion of
756: \begin{equation}
757: \label{eq:def_tildeR2}
758: \widetilde{R_2}(x) = \int\!\!\int_0^\infty e^{-\pi
759: xM(\u)+2i(u_1+u_2)} J(\u) d\u
760: \end{equation}
761: and the terms of the small $\tau$ expansion of $K(\tau)$. This
762: correspondence is
763: \begin{equation}
764: \label{eq:FT_rule}
765: \frac1{(2\pi ix)^k} \longleftrightarrow
766: \frac{\tau^{k-1}}{(k-1)!}.
767: \end{equation}
768:
769: Our plan is to modify the integrand in the definition of
770: $\widetilde{R_2}(x)$, getting rid of the factor $e^{2i(u_1+u_2)}J(\u)$,
771: expand the integral in inverse powers of $x$ and apply the
772: correspondence rule~(\ref{eq:FT_rule}) to recover~(\ref{Ktau}).
773:
774: First of all, as one can verify by direct substitution of the series
775: for $M(u_1,u_2)$,
776: \begin{multline}
777: \label{eq:dalphas_xM}
778: \brr{\frac{\partial}{\partial \alpha_1} + \frac{\partial}{\partial
779: \alpha_2}} \brr{xM\brr{\frac{\alpha_1}{x}, \frac{\alpha_2}{x}}}
780: = \sum_{r,s=0}^\infty i^{r+s} \binom{r+s}{r}
781: \frac{(\alpha_1/x)^r(\alpha_2/x)^s}{r!s!}\\
782: = 2 e^{i(\alpha_1+\alpha_2)/x}
783: J_0\brr{\frac{2\sqrt{\alpha_1\alpha_2}}{x}},
784: \end{multline}
785: and
786: \begin{multline}
787: \label{eq:dxm_u/x}
788: \frac{\partial}{\partial x} \brr{xM\brr{\frac{\alpha_1}{x},
789: \frac{\alpha_2}{x}}}
790: = \sum_{r,s=1}^\infty 2i^{r+s+1}
791: \frac{(r+s-1)!(\alpha_1/x)^r(\alpha_2/x)^s}{r!s!(r-1)!(s-1)!}\\
792: = -\frac{2i\sqrt{\alpha_1\alpha_2}}{x}
793: J_1\brr{\frac{2\sqrt{\alpha_1\alpha_2}}{x}}
794: e^{i(\alpha_1+\alpha_2)/x}.
795: \end{multline}
796: Applying (\ref{eq:dxm_u/x}),
797: \begin{multline}
798: \label{eq:big_id1}
799: \frac{\partial^2}{\partial x^2} e^{-\pi xM\brr{\frac{\alpha_1}{x},
800: \frac{\alpha_2}{x}}}
801: \\
802: = e^{-\pi xM} \brr{-4\pi^2\frac{\alpha_1\alpha_2}{x^2}J_1^2e^{2\phi}
803: - \frac{2\pi i}{x^3}\brr{2J_0e^\phi \alpha_1\alpha_2 +
804: iJ_1e^\phi \sqrt{\alpha_1\alpha_2}(\alpha_1+\alpha_2)} },
805: \end{multline}
806: where $\phi=i(\alpha_1+\alpha_2)/x$ and for simplicity we have omitted the
807: argument \newline $(\alpha_1/x,\alpha_2/x)$ of the functions $M$,
808: $J_0$ and $J_1$.
809:
810: Similarly, using~(\ref{eq:dalphas_xM}), we have
811: \begin{multline}
812: \label{eq:big_id2}
813: \brr{\frac{\partial}{\partial \alpha_1} + \frac{\partial}{\partial
814: \alpha_2}}^2
815: e^{-\pi xM\brr{\frac{\alpha_1}{x}, \frac{\alpha_2}{x}}}
816: \\
817: = e^{-\pi xM} \brr{4\pi^2J_0^2e^{2\phi}
818: - \frac{2\pi i}{\alpha_1\alpha_2x}\brr{2J_0e^\phi \alpha_1\alpha_2
819: + iJ_1e^\phi \sqrt{\alpha_1\alpha_2}(\alpha_1+\alpha_2)} }.
820: \end{multline}
821: Noticing the similarity between (\ref{eq:big_id1}) and
822: (\ref{eq:big_id2}), we subtract the first from the second, with the
823: appropriate factors, to obtain
824: \begin{multline}
825: \label{eq:Bogom_id}
826: \frac1{4\pi^2}
827: \left[\frac{1}{x^2}\brr{\frac{\partial}{\partial\alpha_1}
828: + \frac{\partial}{\partial\alpha_2}}^2 -
829: \frac{1}{\alpha_1\alpha_2} \frac{\partial^2}{\partial
830: x^2}\right]
831: e^{-\pi xM\brr{\frac{\alpha_1}{x},\frac{\alpha_2}{x}}}\\
832: = \frac{1}{x^2} \left[J_0^2+J_1^2\right] e^{2\phi} e^{-xM},
833: \end{multline}
834: where, as before, the argument $(\alpha_1/x,\alpha_2/x)$ of $M$,
835: $J_0$ and $J_1$ has been omitted. The right hand side of
836: (\ref{eq:Bogom_id}) is exactly the
837: integrand of (\ref{eq:r2_x_large}) if we perform the change of
838: variables $u_i=\alpha_i/x$ and, therefore,
839: \begin{equation}
840: \label{eq:Bogom_rep}
841: \widetilde{R_2}(x)
842: = \iint_0^\infty \frac{d\alpha_1 d\alpha_2}{4\pi^2}
843: \left[\frac{1}{x^2}\brr{\frac{\partial}{\partial\alpha_1}
844: + \frac{\partial}{\partial\alpha_2}}^2 -
845: \frac{1}{\alpha_1\alpha_2} \frac{\partial^2}{\partial
846: x^2}\right]
847: e^{-\pi xM\brr{\frac{\alpha_1}{x},\frac{\alpha_2}{x}}}.
848: \end{equation}
849:
850: The first term in the integral can be evaluated as follows,
851: \begin{multline}
852: \iint_0^\infty \frac{d\alpha_1 d\alpha_2}{4\pi^2x^2}
853: \brr{\frac{\partial}{\partial\alpha_1}
854: + \frac{\partial}{\partial\alpha_2}}^2
855: e^{-\pi xM\brr{\frac{\alpha_1}{x},\frac{\alpha_2}{x}}}\\
856: = \left( -\int_0^\infty \frac{d\alpha_2}{4\pi x^2}
857: \left[\Theta \right]_{\alpha_1=0}^\infty
858: - \int_0^\infty \frac{d\alpha_1}{2\pi x^2}
859: \left[\Theta \right]_{\alpha_2=0}^\infty\right),
860: \end{multline}
861: where
862: \begin{equation}
863: \label{eq:Theta}
864: \Theta = \brr{\frac{\partial}{\partial\alpha_1}
865: + \frac{\partial}{\partial\alpha_2}} e^{-\pi
866: xM\brr{\frac{\alpha_1}{x},\frac{\alpha_2}{x}}}
867: = 2 e^{i(\alpha_1+\alpha_2)/x}
868: J_0\brr{\frac{2\sqrt{\alpha_1\alpha_2}}{x}} e^{-\pi xM}.
869: \end{equation}
870: Since
871: \begin{equation}
872: \label{eq:isaev_shtirlitz}
873: \left[ \Theta \right]_{\alpha_1=0}^\infty = -2e^{i\alpha_2/x}
874: e^{-\pi \alpha_2}, \qquad
875: \left[ \Theta \right]_{\alpha_2=0}^\infty = -2e^{i\alpha_1/x}
876: e^{-\pi \alpha_1},
877: \end{equation}
878: we obtain
879: \begin{equation}
880: \iint_0^\infty \frac{d\alpha_1 d\alpha_2}{4\pi^2x^2}
881: \brr{\frac{\partial}{\partial\alpha_1}
882: + \frac{\partial}{\partial\alpha_2}}^2
883: e^{-\pi xM\brr{\frac{\alpha_1}{x},\frac{\alpha_2}{x}}}
884: = \frac{1}{2\pi x^2} \frac{2}{\pi-i/x}.
885: \end{equation}
886: Now we can expand the result in inverse powers of $x$ and apply
887: the correspondence rule (\ref{eq:FT_rule}). We obtain
888: \begin{equation}
889: \frac{1}{\pi x} \frac{1}{\pi x-i}
890: = -\sum_{k=0}^\infty \brr{\frac{i}{\pi x}}^{k+2}
891: \longleftrightarrow 2 \sum_{k=0}^\infty
892: \frac{(-2\tau)^{k+1}}{(k-1)!}
893: = 2\brr{e^{-2\tau}-1}.
894: \end{equation}
895:
896: Next we need to expand the second part of the integrand in
897: (\ref{eq:Bogom_rep}),
898: \begin{multline}
899: \label{eq:second_part}
900: \frac{\partial^2}{\partial x^2}
901: e^{-\pi xM} =
902: \frac{\partial^2}{\partial x^2} e^{-\pi(\alpha_1+\alpha_2)}
903: \exp\brr{2\pi i\sum_{r,s=0}^\infty
904: \frac{(i\alpha_1)^{r+1} (i\alpha_2)^{s+1}(r+s)!}
905: {x^{r+s+1}r!s!(r+1)!(s+1)!} }\\
906: = e^{-\pi(\alpha_1+\alpha_2)}\frac{\partial^2}{\partial x^2}
907: \left[ \sum_{j=0}^\infty \frac{(2\pi i)^j}{j!}
908: \brr{\sum_{r,s=0}^\infty
909: \frac{(i\alpha_1)^{r+1} (i\alpha_2)^{s+1}(r+s)!}
910: {x^{r+s+1}r!s!(r+1)!(s+1)!}}^j
911: \right].
912: \end{multline}
913: Using the same notation as in (\ref{eq:notation_F}),
914: \begin{multline}
915: \brr{\sum_{r,s=0}^\infty
916: \frac{(i\alpha_1)^{r+1}
917: (i\alpha_2)^{s+1}(r+s)!}{x^{r+s+1}r!s!(r+1)!(s+1)!}}^j
918: = \brr{\sum_{r,s=0}^\infty
919: \frac{(i\alpha_1)^{r+1} (i\alpha_2)^{s+1}}{x^{r+s+1}}F_1(r,s)}^j\\
920: = \sum_{R,S=0}^\infty \frac{(i\alpha_1)^{R+j}
921: (i\alpha_2)^{S+j}}{x^{R+S+j}} F_j(R,S),
922: \end{multline}
923: where, as before, $F_j(R,S)$ is the $j$th convolution of $F_1(R,S)$
924: with itself. Thus
925: \begin{multline}
926: \label{eq:second_part_cont}
927: \frac{\partial^2}{\partial x^2}
928: e^{-\pi xM\brr{\frac{\alpha_1}{x},\frac{\alpha_2}{x}}}
929: = e^{-\pi(\alpha_1+\alpha_2)} \sum_{j=1}^\infty
930: \frac{(2\pi i)^j}{j!}\\
931: \times \sum_{R,S=0}^\infty \frac{(R+S+j-1)!(i\alpha_1)^{R+j}
932: (i\alpha_2)^{S+j}}{(R+S+j+1)!x^{R+S+j+2}} F_j(R,S).
933: \end{multline}
934: Finally we integrate against
935: $d\alpha_1d\alpha_2/(4\pi^2\alpha_1\alpha_2)$ to arrive at
936: \begin{multline}
937: - \iint_0^\infty \frac{d\alpha_1
938: d\alpha_2}{4\pi^2\alpha_1\alpha_2}
939: \frac{\partial^2}{\partial x^2}
940: e^{-\pi xM\brr{\frac{\alpha_1}{x},\frac{\alpha_2}{x}}}\\
941: = -\sum_{j=1}^\infty \frac{(2\pi i)^j}{4\pi^2j!}
942: \sum_{R,S=0}^\infty \frac{(R+S+j+1)!(R+j-1)!(S+j-1)!}
943: {(R+S+j-1)!(-i\pi)^{R+S+2j}x^{R+S+j+2}} F_j(R,S) \\
944: % = \sum_{j=1}^\infty \frac{(-2)^j}{4j!}
945: % \sum_{R,S=0}^\infty \frac{(R+S+j+1)!(R+j-1)!(S+j-1)!}
946: % {(R+S+j-1)!(-i\pi x)^{R+S+j+2}} F_j(R,S)\\
947: \longleftrightarrow
948: % \sum_{j=1}^\infty \frac{(-2)^j}{4j!}
949: % \sum_{R,S=0}^\infty \frac{(-2\tau)^{R+S+j+2}(R+j-1)!(S+j-1)!}
950: % {\tau(R+S+j-1)!} F_j(R,S)\\
951: \tau \sum_{j=1}^\infty \frac{(4\tau)^j}{j!}
952: \sum_{R,S=0}^\infty \frac{(-2\tau)^{R+S}(R+j-1)!(S+j-1)!}
953: {(R+S+j-1)!} F_j(R,S).
954: \end{multline}
955: This is exactly the same as the $j$ sum in (\ref{Ktau}) with the
956: exception of the extra $j=1$ term in the summation above. For $j=1$
957: we have
958: \begin{multline}
959: 4\tau^2 \sum_{R,S=0}^\infty \frac{(-2\tau)^{R+S}R!S!}
960: {(R+S)!} F_j(R,S)
961: = \sum_{R,S=0}^\infty \frac{(-2\tau)^{R+S+2}}{(R+1)!(S+1)!}\\
962: = \brr{\sum_{R=0}^\infty \frac{(-2\tau)^{R+1}}{(R+1)!}}
963: \brr{\sum_{S=0}^\infty \frac{(-2\tau)^{S+1}}{(S+1)!}} =
964: (1-e^{-2\tau})^2\\
965: = 1-2e^{-2\tau} + e^{-4\tau},
966: \end{multline}
967: which, together with the terms $1$ and $2(e^{-2\tau}-1)$,
968: gives the correct contribution $e^{-4\tau}$.
969: \end{proof}
970:
971:
972: %%%%%%%%%%%%%%%%%%%%% New Section %%%%%%%%%%%%%%%%%%%%%%%%%%%
973: \section{Singularities of the form factor}
974:
975: One can also obtain some information about the
976: singularities of $K(\tau)$ by Fourier transforming the integral representation
977: (\ref{eq:r2_x_large}). There is, however, a subtle
978: problem associated with this approach. The form factor is by definition an
979: even function defined on the real line. What we want to get from
980: transforming (\ref{eq:r2_x_large}) is an analytic function which
981: coincides with the form factor for real $\tau>0$, so as to be able to
982: study its complex singularities.
983:
984: As we saw above,
985: \begin{equation}
986: \label{eq:R2tilde}
987: \widetilde{R_2}(x) = \int\!\!\int_0^\infty e^{-\pi
988: xM(\u)+2i(u_1+u_2)} J(\u) d\u = \int_0^\infty (K(\tau')-1)
989: e^{-2\pi ix\tau'}.
990: \end{equation}
991: Integrating (\ref{eq:R2tilde}) against $e^{2\pi ix\tau}$ on the real
992: line we obtain
993: \begin{equation}
994: \label{eq:invR2tilde}
995: \int_{-\infty}^\infty \widetilde{R_2}(x) e^{2\pi ix\tau} dx =
996: K(\tau) - 1, \qquad \tau>0.
997: \end{equation}
998: One can check that this leads to the correct power series expansion
999: of the form factor: give $x$ a small negative imaginary part,
1000: $x\mapsto x-i\epsilon$, in $\widetilde{R_2}(x)$ (this is
1001: consistent with (\ref{eq:R2tilde})), substitute in the asymptotic
1002: expansion (56), and integrate term-by-term.
1003:
1004: We now use $\widetilde{R_2}(-x)=\bbar{\widetilde{R_2}(x)}$ to
1005: write
1006: \begin{equation}
1007: \int_{-\infty}^\infty e^{2\pi ix\tau} \widetilde{R_2}(x) dx
1008: = \int_0^\infty \brr{e^{2\pi ix\tau}\widetilde{R_2}(x) + e^{-2\pi
1009: ix\tau}\bbar{\widetilde{R_2}(x)}} dx.
1010: \label{eq:pFT}
1011: \end{equation}
1012: The only factor $\widetilde{R_2}(x)$ which depends on $x$ is $e^{-\pi
1013: xM(\u)}$ and
1014: \begin{equation}
1015: \label{eq:FT_exM}
1016: \int_0^\infty e^{2\pi ix\tau} e^{-\pi xM(\u)} dx =
1017: \frac{1}{\pi(M(\u)-2i\tau)},
1018: \end{equation}
1019: thus we have for the form factor
1020: \begin{equation}
1021: \label{eq:int_rep_ff}
1022: K(\tau) = 1 + \frac1\pi \int\!\!\int_0^\infty \left[
1023: \frac{e^{2i(u_1+u_2)}}{M(\u)-2i\tau} +
1024: \frac{e^{-2i(u_1+u_2)}}{\bbar{M(\u)}+2i\tau}\right] J(\u) d\u.
1025: \end{equation}
1026:
1027: The representation~(\ref{eq:int_rep_ff}) presents us with a way to
1028: find the singularities of $K(\tau)$. These are given
1029: by the condition $\tau=M(\u_s)/(2i)$ and $\tau=\bbar{M(\u_s)/(2i)}$,
1030: where
1031: the point $\u_s$ is such that
1032: \begin{equation}
1033: \label{eq:sing_cond}
1034: \frac{\partial M}{\partial u_1}(\u_s) =
1035: \frac{\partial M}{\partial u_2}(\u_s) = 0.
1036: \end{equation}
1037: The derivative with respect to $u_2$ is
1038: \begin{equation}
1039: \label{eq:deriv_u2}
1040: \frac{\partial M}{\partial u_2} = 1 - 2\int_0^{u_1} \left[
1041: e^{i(y+z)} J_1\brr{2\sqrt{yz}} \sqrt{y/z} - i e^{i(y+z)}
1042: J_0\brr{2\sqrt{yz}} \right] dy,
1043: \end{equation}
1044: where $z=y-u_1+u_2$ and we have assumed that $u_1>u_2>0$. It is obvious
1045: from the expansion~(\ref{M_def}), however, that the function $M(\u)$
1046: is continuously differentiable if
1047: $u_1u_2>0$ and hence that the expression~(\ref{eq:deriv_u2}) is valid for
1048: all $u_1>0$ and $u_2>0$.
1049: The integral in (\ref{eq:deriv_u2}) is not easy to analyse and
1050: to simplify it we reduce our search to the line $u_2=u_1$, where
1051: \begin{equation}
1052: \frac{\partial M}{\partial u_2}(u_2=u_1) = 1 -
1053: 2\int_0^{u_1}e^{2iy}J_1(2y)dy + 2i\int_0^{u_1} e^{2iy}J_0(2y)dy.
1054: \end{equation}
1055: Performing the second integration by parts,
1056: \begin{equation}
1057: \label{eq:int_parts}
1058: \int_0^{u_1} e^{2iy}J_0(2y)dy =
1059: \left.\frac{e^{2iy}J_0(2y)}{2i}\right|_0^{u_1} +
1060: \frac{2}{2i}\int_0^{u_1}e^{2iy}J_1(2y)dy,
1061: \end{equation}
1062: we obtain, after simplification,
1063: \begin{equation}
1064: \label{eq:deriv_res}
1065: \frac{\partial M}{\partial u_2}(u_2=u_1) = e^{2iu_1}J_0(2u_1).
1066: \end{equation}
1067: Since
1068: $ \frac{\partial M}{\partial u_1}(u_2=u_1) = \frac{\partial
1069: M}{\partial u_2}(u_2=u_1)$, we see that
1070: the zeros of the derivatives of $M(\u)$
1071: on the line $u_2=u_1$ are given by the zeros of the Bessel function
1072: $J_0$. The nearest zero is at $u_s\approx 1.202$.
1073: Thus one of the singularities of $K(\tau)$ lies at
1074: $\tau_s = M(1.202,1.202)/(2i) = 0.462-0.420 i$. We note that
1075: $|\tau_s|=0.624$, which coincides with our previous numerical estimate
1076: of the radius of convergence of
1077: the series expansion of $K(\tau)$ in powers of $\tau$ around $\tau=0$.
1078: This strongly suggests that this singularity is the closest to the
1079: origin. To this end, we can prove the following.
1080:
1081:
1082: \begin{prop}
1083: \label{prop:nearest_sing}
1084: Among the singularities arising from stationary points of\linebreak
1085: $M(u_1, u_2)$ along the line $u_2=u_1$,
1086: the singularity at $\tau_s = M(1.202,1.202)/(2i)$ is the
1087: nearest to the origin.
1088: \end{prop}
1089: \begin{proof}
1090: To show that the statement is true we need to prove that the
1091: function $|M(u,u)|^2$ is a nowhere decreasing function of
1092: $u$. On the line
1093: $u_1=u_2=u$ we have
1094: \begin{equation}
1095: \label{eq:M_on_line}
1096: M(u,u) = \int_0^{2u} e^{iy}J_0(y) dy =
1097: 2e^{2iu}u\brr{J_0(2u)-iJ_1(2u)}.
1098: \end{equation}
1099: Thus $|M(x/2,x/2)|^2 = x^2\brr{J_0^2(x)+J_1^2(x)}$ and its
1100: derivative is, after simplification, $\frac{d}{dx}|M(x/2,x/2)|^2 =
1101: 2xJ_0^2(x) \geq 0$.
1102: \end{proof}
1103:
1104: \begin{figure}[t]
1105: \vskip 8.5cm
1106: \special{psfile="cosfit.eps" voffset=-25
1107: hoffset=-10 hscale=90 vscale=80}
1108: \caption{The coefficients of the power series expansion of $K(\tau)$
1109: normalized by $\rho^n$ (crosses), compared to
1110: (\ref{eq:leading_contrib_ser_K}). As expected, the agreement
1111: improves as $n$
1112: increases.}
1113: \label{fig:cosfit}
1114: \end{figure}
1115:
1116: It is straightforward to approximate the behaviour of $K(\tau)$ near
1117: these singularities. We expand
1118: \begin{eqnarray}
1119: \nonumber
1120: M(\u) &\approx& M(\u_s) + \frac12 \frac{\partial^2 M}{\partial
1121: u_1^2}(\u_s) (u_1-u_2)^2 + \frac12 \frac{\partial^2 M}{\partial
1122: u_2^2}(\u_s) (u_2-u_s)^2\\
1123: \nonumber
1124: &&\qquad + \frac{\partial^2 M}{\partial u_1\partial u_2}(\u_s)
1125: (u_1-u_s)(u_2-u_s)\\
1126: &=& M(\u_s) + \alpha_s\brr{(u_1-u_s)^2+(u_2-u_s)^2}.
1127: \label{eq:M_expand_sing}
1128: \end{eqnarray}
1129: For the singularity associated with the first Bessel zero,
1130: $\alpha_s\approx 0.385-0.349 i$. Then, when $\tau$ is real,
1131: \begin{equation}
1132: \label{eq:ff_around_sing}
1133: K(\tau)\! \approx \!\frac1{\pi\alpha_s} \iint_{0}^\infty
1134: \frac{J(\u) e^{2i(u_1+u_2)}d\u}{(u_1-u_s)^2 + (u_2-u_s)^2 +
1135: (M(\u_s)-2i\tau)/\alpha_s} + \mbox{c.c.}
1136: \end{equation}
1137: The main
1138: contribution to the integral around these singularities is
1139: \begin{equation}
1140: \label{eq:ln_sing}
1141: K(\tau) \propto -C\ln\brr{1-\frac{2i\tau}{M(\u_s)}} -
1142: \bbar{C}\ln\brr{1+\frac{2i\tau}{\bbar{M(\u_s)}}},
1143: \end{equation}
1144: where $C=J(\u_s)e^{4iu_s}/\alpha_s$. Expanding
1145: (\ref{eq:ln_sing}) into a series around $\tau=0$ we get
1146: \begin{equation}
1147: \label{eq:leading_contrib_ser_K}
1148: K(\tau) \propto 2\Re\brr{C\sum_{n=1}^\infty
1149: \rho^n\frac{e^in\phi}{n}\tau^n} = 2A\sum_{n=1}^\infty\cos(\phi n + \psi)
1150: \frac{\rho^n}{n} \tau^n,
1151: \end{equation}
1152: where, for the singularity analysed above,
1153: $A = \left|J(\u_s)e^{4iu_s}/\alpha_s\right| \approx 0.519$,
1154: $\psi = \arg\brr{J(\u_s)e^{4iu_s}/\alpha_s}\approx -0.737$, $\rho =
1155: |2i/M(\u_s)|\approx 1.602$ and $\phi = \arg\brr{2i/M(\u_s)}\approx
1156: 0.737$. By Darboux's Principle, the coefficients of the
1157: expansion (\ref{eq:leading_contrib_ser_K}) should
1158: comprise the leading contribution to large-order asymptotics of
1159: the exact coefficients given by
1160: (\ref{Ktau}) and (\ref{eq:CM1}). To compare them we plot the
1161: exact coefficients $na_n/\rho^n$ against the approximate coefficients
1162: $2A\cos(\phi n + \psi)$. The result is shown in Fig.~\ref{fig:cosfit}.
1163:
1164:
1165: %%%%%%%%%%%%%%%%%%%%%%%%%%% New Section %%%%%%%%%%%%%%%%%%%%%%%%%
1166:
1167: \section{Small $x$ limit of $R_2(x)$}
1168: \label{sec:int_smallx}
1169:
1170: Returning to (\ref{eq:ren_r2}), one can check that
1171: the function $\Psi$, defined by (\ref{eq:Psi}), satisfies the
1172: equation
1173: \begin{multline}
1174: \label{eq:PDPhi}
1175: \left[\frac{\partial^2}{2\partial u_1 \partial u_2} +
1176: i\brr{\frac{\partial}{\partial u_1} -
1177: \frac{\partial}{\partial u_2}}\right]
1178: \brr{e^{2i(u_1-u_2)}\Psi^2}\\
1179: = e^{2i(u_1-u_2)} \brr{\frac{\partial\Psi}{\partial z_1}
1180: \frac{\partial\Psi}{\partial z_2} - \Psi^2}.
1181: \end{multline}
1182: Substituting it into (\ref{eq:ren_r2}) and integrating by parts we
1183: obtain
1184: \begin{multline}
1185: \label{eq:r2_better}
1186: R_2(x) = - \frac14\int d\u e^{-\pi x Q}
1187: \left[\frac{\partial^2}{2\partial u_1 \partial u_2} +
1188: i\brr{\frac{\partial}{\partial u_1} -
1189: \frac{\partial}{\partial u_2}}\right]
1190: \brr{e^{2i(u_1-u_2)}\Psi^2} \\
1191: = \int \frac{d\u}{4} e^{2i(u_1-u_2)}\Psi^2
1192: \left[i\brr{\frac{\partial}{\partial u_1} -
1193: \frac{\partial}{\partial u_2}}-\frac{\partial^2}{2\partial
1194: u_1 \partial u_2}\right] \brr{e^{-\pi x Q}}.
1195: \end{multline}
1196: Now, using the identities
1197: \begin{equation}
1198: \label{eq:iden_Q}
1199: \frac{\partial Q}{\partial u_1} - \frac{\partial Q}{\partial u_2} =
1200: e^{i(u_1-u_2)}\Psi, \qquad
1201: \frac{\partial^2 Q}{2\partial u_1 \partial u_2} =
1202: -i e^{i(u_1-u_2)}\Psi,
1203: \end{equation}
1204: which one can derive using the series expansion of
1205: $Q(u_1,u_2)=M(u_1,-u_2)$, we write
1206: \begin{multline}
1207: \label{eq:diff_op_eQ}
1208: \left[i\brr{\frac{\partial}{\partial u_1} -
1209: \frac{\partial}{\partial u_2}}-\frac{\partial^2}{2\partial
1210: u_1 \partial u_2}\right] \brr{e^{-\pi x Q}}\\
1211: = e^{-\pi x Q} \brr{ -i\pi x\brr{\frac{\partial Q}{\partial u_1} -
1212: \frac{\partial Q}{\partial u_2}} + \frac{\pi x}2 \frac{\partial^2
1213: Q}{\partial u_1 \partial u_2} -
1214: \frac{(\pi x)^2}2 \frac{\partial Q}{\partial u_1}
1215: \frac{\partial Q}{\partial u_2}}\\
1216: = -e^{-\pi x Q} \brr{\frac{3i\pi x}{2}e^{i(u_1-u_2)}\Psi +
1217: \frac{(\pi x)^2}2 \frac{\partial Q}{\partial u_1}
1218: \frac{\partial Q}{\partial u_2} }.
1219: \end{multline}
1220: Thus we obtain, finally,
1221: \begin{equation}
1222: \label{eq:r2_smallx}
1223: R_2(x) = -\int \frac{d\u}{8}
1224: e^{2i(u_1-u_2)-\pi x Q}\Psi^2 \left[ \pi^2x^2\frac{\partial
1225: Q}{\partial u_1}\frac{\partial Q}{\partial u_2} +
1226: 3i\pi x \Psi e^{i(u_1-u_2)}\right].
1227: \end{equation}
1228: From (\ref{eq:r2_smallx}) one can see that the two-point
1229: correlation function $R_2(x)$ is linear in $x$ for small $x$. The
1230: slope was computed in \cite{BGS}:
1231: \begin{equation}
1232: \label{eq:R_2_small_x}
1233: R_2(x) = \frac{\pi\sqrt{3}}{2}x + O(x^2).
1234: \end{equation}
1235:
1236:
1237: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1238: \section{Discussion}
1239:
1240: The derivation presented above provides a proof that two-point spectral
1241: correlations for certain \v Seba billiards and quantum star graphs
1242: are the same, in the appropriate limits. This
1243: initially surprising fact has its explanation in the following
1244: observations. First, the dynamics in both systems is centered around a
1245: single point scatterer; in star graphs it is the central vertex,
1246: and in \v Seba billiards the singularity. Furthermore, in between
1247: scatterings the dynamics is integrable in both cases.
1248:
1249: Second, applying the Mittag-Leffler
1250: theorem to the meromorphic function $\tan z$, we have that
1251: \begin{equation}
1252: \label{eq:mitt_leff}
1253: \tan z = \sum_{n=-\infty}^\infty \brr{\frac1{n\pi+\pi/2-z} -
1254: \frac1{n\pi+\pi/2}}.
1255: \end{equation}
1256: We can therefore rewrite (\ref{eq:SE_gen}) in a form similar to
1257: (\ref{eq:Sebacond}) when $|\psi_n({\bf x_0})|^2={\rm constant}$. It
1258: thus becomes less surprising that the
1259: two point correlation functions of the two systems are the same,
1260: because in the limit
1261: $v\to\infty$ the poles in (\ref{eq:SE}) have properties
1262: similar to those of a Poisson sequence.
1263: %? sequence'
1264:
1265: Third, from the mathematical point of view star graphs and \v Seba billiards
1266: are similar in that in both cases the scattering centre corresponds
1267: quantum mechanically to a perturbation of rank one.
1268:
1269: Finally, we remark that our results demonstrate that, at least
1270: as regards the
1271: special case considered here, graphs are able to reproduce
1272: features of other, experimentally realizable, quantum systems, and also that
1273: they provide further confirmation that spectral statistics can be computed
1274: exactly using the trace formula when the periodic orbit statistics are
1275: known \cite{BK}.
1276:
1277: \section*{Acknowledgments}
1278: One of us (G.B.) would like to thank the
1279: Laboratoire de Physique Th\'eorique et Mod\`eles Statistiques,
1280: Universit\'e Paris-Sud,
1281: and BRIMS, Hewlett-Packard Laboratories Bristol,
1282: for their hospitality.
1283:
1284: \begin{thebibliography}{99}
1285: \bibitem{BT} M.V.~Berry and M.~Tabor, {\it Proc. R. Soc. London A} {\bf
1286: 356} 375 (1977).
1287: \bibitem{BohGS} O.~Bohigas, M.-J.~Giannoni, and C.~Schmit, {\it
1288: Phys. Rev. Lett.} {\bf 52} 1 (1984).
1289: \bibitem{Berry} M.V.~Berry, {\it Proc. R. Soc. London A} {\bf 400} 229 (1985).
1290: \bibitem{Bog-Kea} E.B.~Bogomolny and J.P.~Keating, {\it Phys. Rev. Lett.}
1291: {\bf 77} 1472 (1996).
1292: \bibitem{BGGS} E.B.~Bogomlny, B.~Georgeot, M.J.~Giannoni, and
1293: C.~Schmit, {\it Phys. Rep.} {\bf 291} 220 (1997).
1294: \bibitem{Kea} J.P.~Keating, {\em Nonlinearity} {\bf 4} 309 (1991).
1295: \bibitem{KS1} T.~Kottos and U.~Smilansky, {\it Phys. Rev. Lett.} {\bf
1296: 79} 4794 (1997).
1297: \bibitem{KS2} T.~Kottos and U.~Smilansky, {\it Ann. Phys.} {\bf
1298: 274} 76 (1999).
1299: \bibitem{SS} H.~Schanz and U.~Smilansky, { Proceedings of the
1300: Australian Summer School in Quantum Chaos and Mesoscopics},
1301: Canberra (1999).
1302: \bibitem{Tan} G.~Tanner, {\it J. Phys A} {\bf 33} 3567 (2000).
1303: \bibitem{Gas} F.~Barra and P.~Gaspard, On the level spacing
1304: distribution in quantum graphs. Preprint, 1999.
1305: \bibitem{BK} G.~Berkolaiko and J.P.~Keating, {\it J. Phys. A} {\bf
1306: 32} 7827 (1999).
1307: \bibitem{S} P.~\v Seba, {\it Phys. Rev. Lett.} {\bf 64} 1855 (1990).
1308: \bibitem{AS} S.~Albeverio and P.~\v Seba, {\it J. Stat. Phys.} {\bf
1309: 64} 369 (1991).
1310: \bibitem{BGS} E.~Bogomolny, U.~Gerland, and C.~Schmidt, Singular
1311: Statistics. Preprint, 1997 (submitted to {\it Phys. Rev.} E, 2000).
1312: \bibitem{thesis} G.~Berkolaiko, Quantum star graphs, PhD thesis,
1313: University of Bristol (2000).
1314: \end{thebibliography}
1315:
1316:
1317: \end{document}
1318:
1319: