1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2: %       Kottos-Smilansky

3: %       Long paper on chaotic scattering on Graphs

4: %       June 2002

5: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

6: %-----------------------LaTex------------------------------

7: \documentclass[12pt]{iopart}

8: \usepackage{iopams}

9: \usepackage{epsf}

10: \usepackage{euscript}

11: %\usepackage{graphicx}

12: %%% Various definitions of bold Z, Q, R, N. %%%

13: \def\bbbone{{\mathchoice {\rm 1\mskip-4mu l} {\rm 1\mskip-4mu l}

14: {\rm 1\mskip-4.5mu l} {\rm 1\mskip-5mu l}}}

15: \def\bbbn{{\rm I\!N}}

16: \def\bbbr{{\rm I\!R}}

17: \def\bbbq{{\mathchoice {\setbox0=\hbox{$\displaystyle\rm Q$}\hbox{\raise

18: 0.15\ht0\hbox to0pt{\kern0.4\wd0\vrule height0.8\ht0\hss}\box0}}

19: {\setbox0=\hbox{$\textstyle\rm Q$}\hbox{\raise

20: 0.15\ht0\hbox to0pt{\kern0.4\wd0\vrule height0.8\ht0\hss}\box0}}

21: {\setbox0=\hbox{$\scriptstyle\rm Q$}\hbox{\raise

22: 0.15\ht0\hbox to0pt{\kern0.4\wd0\vrule height0.7\ht0\hss}\box0}}

23: {\setbox0=\hbox{$\scriptscriptstyle\rm Q$}\hbox{\raise

24: 0.15\ht0\hbox to0pt{\kern0.4\wd0\vrule height0.7\ht0\hss}\box0}}}}

25: \def\bbbz{{\mathchoice {\hbox{$\sf\textstyle Z\kern-0.4em Z$}}

26: {\hbox{$\sf\textstyle Z\kern-0.4em Z$}}

27: {\hbox{$\sf\scriptstyle Z\kern-0.3em Z$}}

28: {\hbox{$\sf\scriptscriptstyle Z\kern-0.2em Z$}}}}

29: 

30: \begin{document}

31: \renewcommand{\baselinestretch}{1.5}

32: \title{ Quantum Graphs: A simple model for Chaotic Scattering}

33: \author{Tsampikos Kottos$^1${\footnote {corresponding author:

34: tsamp@chaos.gwdg.de}}

35: and Uzy Smilansky$^2$ \\

36: $^1$ Max-Planck-Institut f\"ur Str\"omungsforschung, 37073 G\"ottingen,

37: Germany,\\

38: $^2$ Department of  Physics of Complex Systems,

39: The Weizmann Institute of Science, 76100 Rehovot, Israel}

40: \date{\today }

41: 

42: \begin{abstract}

43: We connect quantum graphs with infinite leads, and turn them to scattering systems.

44: We show that they display all the features which characterize quantum scattering

45: systems with an underlying classical chaotic dynamics: typical poles, delay time and

46: conductance distributions, Ericson fluctuations, and when considered statistically,

47: the ensemble of scattering matrices reproduce quite well the predictions of 

48: appropriately defined Random Matrix ensembles. The underlying classical dynamics

49: can be defined, and it provides important parameters which are needed for the quantum

50: theory. In particular, we derive exact expressions for the scattering matrix, and an

51: exact trace formula for the density of resonances, in terms of classical orbits,

52: analogous to the semiclassical theory of chaotic scattering. We use this in order to

53: investigate the origin of the connection between Random Matrix Theory and the

54: underlying classical chaotic dynamics. Being an exact theory, and due to its relative

55: simplicity, it offers new insights into this problem which is at the fore-front of

56: the research in chaotic scattering and related fields.\\

57: 

58: \hspace {-0.5cm} submitted to J. Phys. A Special Issue -- Random Matrix Theory

59: \end{abstract}

60: 

61: 

62: %-------------------------------------------------------------------------

63: 

64: \section{\bf Introduction}

65: \label{sec:introduction}

66: 

67: Quantum graphs of one-dimensional wires connected at nodes were introduced already

68: more than half a century ago to model physical systems. Depending on the envisaged

69: application the precise formulation of the models can be quite diverse and ranges

70: from solid-state applications to mathematical physics 

71: \cite{A85,FJK87,ML77,A81,CC88,NYO94,Ibook,MT01,Z98,V98,E95,R83}. Lately, quantum 

72: graphs attracted also the interest of the quantum chaos community because they can 

73: be viewed as typical and yet relatively simple examples for the large class of 

74: systems in which classically chaotic dynamics implies universal correlations in the 

75: semiclassical limit \cite{KS97,KS99,A99,BK99,SS00,T00,BSW02,BG00,K01}. Up to now we 

76: have only a limited understanding of the reasons for this universality, and quantum 

77: graph models provide a valuable opportunity for mathematically rigorous investigations 

78: of the phenomenon. In particular, for quantum graphs an exact trace formula exists 

79: \cite{R83,KS97,KS99} which is based on the periodic orbits of a mixing classical 

80: dynamical system. Moreover, it is possible to express two-point spectral correlation 

81: functions in terms of purely combinatorial problems \cite{BK99,SS00,T00,BSW02}.

82: 

83: By attaching infinite leads at the vertices, we get non-compact graphs, for which a 

84: scattering theory can be constructed \cite{KS00,BG01,TM01}. They display many of the 

85: feature that characterize scattering systems with an underlying chaotic classical 

86: dynamics \cite{M73,S89,GR89,RBUS90,EVP99}, and they are the subject of the present 

87: paper. The {\it quantum} scattering matrix for such problems can be written explicitly, 

88: together with a trace formula for the density of its resonances. These expressions 

89: are the analogues of the corresponding semiclassical approximations available in the 

90: theory of chaotic scattering \cite{M73,S89,GR89}, albeit here they are exact. With 

91: these tools we analyze the distribution of resonances, partial delay times, and the 

92: statistics of the fluctuating scattering amplitudes and cross sections. Moreover, 

93: we address issues like the statistical properties of the ensemble of the scattering 

94: matrices, and conductance distribution. Finally we analyze the effect of non-uniform  

95: connectivity on the statistical properties of the scattering matrix. A main part of 

96: our analysis will be focused on the comparison of the statistical properties of the 

97: above quantities with the Random Matrix Theory (RMT) predictions 

98: \cite{M75,MPS85,JSA92,HILSS92,BB94,FS96b,FS96,FS97,BFB97,GM98,MB99}.

99: 

100: The paper is structured in the following way. In section (\ref {sec:definitions}), 

101: the mathematical model is  introduced and the main definitions are given. Section 

102: (\ref{sec:s-matrix}) is devoted to the derivation of the  scattering matrix for

103: graphs. The trace formula for the  density of resonances of quantum graphs is

104: presented in section (\ref {sec:classical}) which includes also the analysis of 

105: the underlying classical  system. The next section, is dedicated to the analysis 

106: of various statistical  properties of the $S$-matrix. Our numerical data are

107: compared both with the predictions of RMT and with the semiclassical expectations. 

108: Finally, in section (\ref {sec:star}), we analyze the class of star-graphs for which 

109: several results can be analytically derived. Our conclusions are summarized in 

110: section (\ref {sec:conclusions}).

111: 

112: %-------------------------------------------------------------------------

113: 

114: \section{\bf Quantum Graphs: Definitions}

115: \label{sec:definitions}

116: 

117: We start by considering a {\it compact} graph ${\cal G}$. It consists of $V$ {\it 

118: vertices} connected by $B$ {\it bonds}. The number of bonds which emanate from each 

119: vertex $i$ defines the valency $v_i$ of the corresponding vertex (for simplicity we 

120: will allow only a single bond between any two vertices). The graph is called $v-regular$ 

121: if all the vertices have the same valency $v$. The total number of bonds is $B={1\over 

122: 2} \sum_{i=1} ^V v_{i}$. Associated to every graph is its connectivity matrix $C$. It 

123: is a square matrix of size $V$ whose matrix elements $C_{i,j}$ take the values $1$ if 

124: the vertices $i,j$ are connected with a bond, or $0$ otherwise. The bond connecting 

125: the vertices $i$ and $j$ is denoted by $b \equiv (i,j)$, and we use the convention 

126: that $i<j$. It will be sometimes convenient to use the ``time reversed" notation, where 

127: the first index is the larger, and $\hat b \equiv (j,i)$ with $j>i$. We shall also use 

128: the directed bonds representation, in which $b$ and $\hat b$  are distinguished as two 

129: directed bonds conjugated by time-reversal. We associate the  natural metric to the 

130: bonds, so that $x_{i,j}\  (x_{j,i})$  measures the distance from the vertex $i\  (j)$ 

131: along the bond.  The length of the bonds are denoted by $L_{b}$ and we shall henceforth 

132: assume that they are {\it rationally independent}. The mean length is defined by 

133: $\left<L \right>\equiv (1/B) \sum_{b=1}^B L_b$ and in  all numerical calculations 

134: bellow it will be taken to be $1$. In the {\it directed- bond} notation $L_{b} = 

135: L_{\hat b}$.

136: 

137: The {\it scattering} graph ${\tilde {\cal G}}$ is obtained by adding leads which

138: extend from  $M (\leq V)$ vertices to infinity.  For simplicity we connect at most

139: one lead to any vertex. The valency of these  vertices increases to ${\tilde v}_i

140: =v_i+1$. The $M$ leads are denoted by the index $i$ of the vertex to which they are

141: attached while $x_i$ now measures the distance from the vertex along the lead $i$.

142: 

143: The Schr\"odinger operator (with $\hbar=2m=1$) is defined on the graph ${\tilde {\cal

144: G}}$ in the following way: On the bonds $b$, the components $\Psi_b$ of the total wave

145: function $\Psi$ are solutions of the one - dimensional equation

146: \begin{equation}

147: \label{schrodinger}

148: \left( -i{\frac{{\rm d\ \ }}{{\rm d}x}}-A_b\right) ^2\Psi _b(x)=k^2\Psi

149: _b(x), \,\,\,\,\,\,\,\,\,\,\,\, \bigskip\ b=(i,j)

150: \end{equation}

151: where $A_b$ (with $\Re e(A_{b})\ne 0$ and $A_{b}= -A_{\hat b}$) is a ``magnetic

152: vector potential" which breaks the time reversal symmetry. In most applications

153: we shall assume that all the $A_{b}$'s are equal and the bond index will be dropped.

154: The components of the wave functions on the leads, $\Psi_i(x)$, are solutions of

155: \begin{equation}

156: \label{schrodinger1}

157: -{\frac{{\rm d^2\ \ }}{{\rm d}x^2}}\Psi _i(x)=k^2\Psi_i(x),

158: \,\,\,\,\,\,\,\,\,\,\,\, \bigskip\ i=1,...,M.

159: \end{equation}

160: At the vertices, the wavefunction satisfies boundary conditions which ensure current 

161: conservation. To implement the boundary conditions, the components of the wave function 

162: on each of the bonds $b$ and the leads $i$ are expressed in terms of counter propagating  

163: waves with a wave-vector $k$:

164: \begin {eqnarray}

165: &&{\rm On\ the \  bonds:\ } \Psi_{b} = a_{b} {\rm e}^{i(k+A_{b})x_{b}}

166: +c_{ b} {\rm e}^{i(-k+A_{b})x_{b}} \nonumber  \\

167: &&{\rm On\  the \ leads\ :\ } \Psi_i =  I_{i} {\rm e}^{-ikx_{i}} + O _{i}

168: {\rm e}^{ ikx_{i}} \ .

169: \label{wfun1}

170: \end{eqnarray}

171: The amplitudes $a_b,c_b$ on the bonds and $I_i,O_i$ on the lead are related  by

172: \begin{eqnarray}

173: \label{vertexcond}

174: {\hspace {-20mm}}

175: {\left (

176: \begin {array} {c}

177:    O_i  \\  a_{i,j_1}\\ \cdot  \\   a_{i,j_{v_i}}

178:   \end {array}

179: \right )} =

180: \Sigma^{(i)}

181: {\left ( \begin{array} {c}

182:    I_i  \\ c_{j_1,i} \\ \cdot  \\ c_{j_{v_i},i}

183:   \end {array} \right ) }, \quad\quad \quad\quad

184: \Sigma^{(i)}=

185: {\left (

186: \begin{array}{cccc}

187: \rho^{(i)}  & \tau^{(i)}_{j_1} & \cdot   & \tau^{(i)}_{j_{v_i}} \\

188: \tau^{(i)}_{j_1} & \tilde \sigma^{(i)}_{j_1,j_1} & \cdot  &

189: \tilde \sigma^{(i)}_{j_1,j_{v_i}} \\

190: \cdot  & \cdot &\cdot &\cdot \\

191: \tau^{(i)}_{j_{v_i}} & \tilde \sigma^{(i)}_{{j_{v_i}},j_1} &

192: \cdot & \tilde \sigma^{(i)}_{j_{v_i},j_{v_i}} \\

193: \end{array}

194: \right)}.

195: \end{eqnarray}

196: These equalities impose the  boundary conditions at the vertices. The vertex scattering 

197: matrices $\Sigma^{(i)}_{j,j'}$, are $\tilde v_i \times \tilde v_i$ unitary symmetric 

198: matrices, and $j,j'$ go over all the $v_i$ bonds and the lead which emanate from $i$. 

199: The unitarity of $\Sigma^{(i)}$   guarantees current conservation at each vertex.

200: 

201: On the right hand side of (\ref{vertexcond}), the vertex scattering matrix $\Sigma^{(i)}$

202: was written explicitly in terms of the vertex reflection amplitude $\rho^{(i)}$, the

203: lead-bond transmission amplitudes $\{ \tau^{(i)}_{j}\}$, and the $v_i\times v_i$ bond-bond

204: transition matrix $\tilde \sigma^{(i)}_{j,j'}$, which is {\it sub unitary} ($|\det \tilde

205: \sigma^{(i)}|<1 $), due to the coupling to the leads.

206: 

207: Graphs  for which there are no further requirements on the $\Sigma^{(i)}$ shall be 

208: referred  to as {\it  generic}. It is often convenient to compute the vertex scattering

209: matrix  from a requirement that the wave function is continuous and satisfies $Neumann$

210: boundary conditions at that vertex. These graphs shall be referred to as Neumann graphs, 

211: and the resulting $\tilde \Sigma^{(i)}$ matrices read:

212: \begin {equation}

213: \label{Neumann}

214: \tilde \sigma^{(i)} _{j,j'}= {2 \over {\tilde v}}  -\delta _{j,j'};\quad

215: \tau^{(i)}_j =   {2 \over {\tilde v}};\quad

216: \rho^{(i)} ={2 \over {\tilde v}}-1\ .

217: \end{equation}

218: 

219: Vertices which are not coupled to leads have $\rho ^{i}=1 ,\ \tau^{(i)}_j =0$, while

220: the bond-bond transition matrix $\tilde \sigma^{(i)}_{j.j'}={2 \over v}-\delta _{j,j'}$ 

221: is unitary.

222: 

223: %-------------------------------------------------------------------------

224: 

225: \section{\bf The S-matrix for Quantum Graphs}

226: \label{sec:s-matrix}

227: 

228: It is  convenient to discuss first graphs with leads connected to all the vertices

229: $M=V$. The generalization to an arbitrarily number $M\leq V$ of leads (channels) is

230: straightforward and will be presented at the end of this section.

231: 

232: To derive the scattering matrix, we first write the  bond wave functions using the 

233: two  representations which are conjugated by   ``time reversal":

234: \begin{eqnarray}

235: \label{wfun2}

236: \Psi_{b}(x_b) &=& a_{b} {\rm e}^{i(k+A_{b})x_{b}}

237: +c_{ b} {\rm e}^{i(-k+A_{b})x_{b}}\  = \nonumber \\

238: \Psi _{\hat b}(x_{\hat b}) &=&a_{\hat b}{\rm e}^{i(k+A_{\hat b})x_{\hat b}}+

239: c_{\hat b}{\rm e}^  {i(-k+A_{\hat b})x_{\hat b}}\  = \nonumber  \\

240: &=&a_{\hat b}{\rm e}^  {i(-k-A_{\hat b})x_b}{\rm e}^  {i(k+A_{\hat b})L_b} +

241: c_{\hat b}{\rm e}^  {i(-k+A_{\hat b})L_b}{\rm e}^  {i(k-A_{\hat b})x_b} \ .

242: \end{eqnarray}

243: Hence,

244: \begin{equation}

245: c_b=a_{\hat b}{\rm e}^  {i(k+A_{\hat b})L_b},\,\,\,\,

246: a_b=c_{\hat b}{\rm e}^  {i (-k+A_{\hat b})L_b }.\label{wfun3}

247: \end{equation}

248: In other words, but for a phase factor, the outgoing wave from the vertex $i$ in the 

249: direction $j$ is identical to the incoming wave at $j$ coming from $i$.

250: 

251: Substituting $a_b$ from Eq.~(\ref{wfun3}) in Eq.~(\ref{vertexcond}), and solving for 

252: $c_{i,j}$ we get

253: \begin{eqnarray}

254: \label{wfun4}

255: c_{i,j'}&=&\sum_{r,s} \left({\bf 1}-\tilde S_B(k;A) \right

256: )^{-1}_{(i,r),(s,j)} D_{(s,j)}

257: \tau _s^{(j)} I_j \nonumber \\

258: O_i&=&\rho^{(i)} I_i + \sum_{j'}\tau_{j'}^{(i)} c_{ij'}

259: \end{eqnarray}

260: where ${\bf 1}$ is the $2B\times 2B$ unit matrix. Here, the ``bond scattering matrix'' 

261: ${\tilde S}_b$ is a sub-unitary matrix in the $2B$ dimensional space of directed bonds 

262: which propagates the wavefunctions. It is defined as $\tilde S_B(k,A)=D(k;A)\tilde R $, 

263: with

264: \begin{eqnarray}

265: \label{DandT}

266: D_{ij,i^{\prime }j^{\prime }}(k,A) &=&\delta _{i,i^{\prime }}\delta

267: _{j,j^{\prime }}{\rm e}^{ikL_{ij}+iA_{i,j}L_{ij}} \\

268: \ \tilde R_{ji,nm} &=&\delta _{n,i}C_{j,i}C_{i,m}{\tilde \sigma}

269: _{ji,im}^{(i)}.

270: \nonumber

271: \end{eqnarray}

272: $D(k,A)$ is a  diagonal unitary matrix which depends only on the metric properties of 

273: the graph, and provides a phase which is due to free propagation  on the bonds. The 

274: sub-unitary matrix $\tilde R$ depends on the connectivity and on the bond-bond transition 

275: matrices ${\tilde \sigma}$. It  assigns a scattering amplitude for transitions between 

276: connected directed bonds. $\tilde R$ is sub-unitary, since

277: \begin{equation}

278: \label{subR}

279: |\det \tilde R| = \prod_{i=1}^V |\det \tilde \sigma ^{(i)}|<1.

280: \end{equation}

281: 

282: Replacing $c_{i,j'}$ in the second of Eqs.~(\ref{wfun4}) we get the following relation 

283: between the outgoing and incoming amplitudes $O_i$ and $I_j$ on the leads:

284: \begin{equation}

285: \label{inout}

286: O_i = \rho^{(i)} I_i + \sum_{j'j r s}\tau_{j'}^{(i)}

287: \left({\bf 1}-\tilde S_B(k;A) \right )^{-1}_{(i,r),(s,j)} D_{(s,j)} \tau

288: _s^{(j)} I_j

289: \end{equation}

290: Combining (\ref{inout}) for all leads $i=1,\dots ,V$, we obtain the unitary $V\times V$ 

291: scattering matrix $S^{(V)}$,

292: \begin{equation}

293: \label{scatmat}

294: S^{(V)}_{i,j}  = \delta_{i,j} \rho^{(i)}

295:   + \sum_{r,s} \tau^{(i)}_r

296: \left ({\bf 1}-\tilde S_B(k;A) \right )^{-1}_{(i,r),(s,j)} D_{(s,j)}

297: \tau _s^{(j)}.

298: \end{equation}

299: From Eq.~(\ref{scatmat}), we see that the scattering matrix may be decomposed into 

300: two parts $S^{(V)}(k)=S^{D}+S^{fl}(k)$ which are associated with two well separated 

301: time scales of the scattering process. $S^{D}(=\delta_{i,j} \rho^{i})$ is the prompt 

302: reflection at the entrance vertex and induces a ``direct" component. In general, it

303: varies very slowly with energy, and is system dependent. On the other hand the

304: ``chaotic" component of the $S$ matrix, $S^{fl}(k)$, starts by a transmission from

305: the incoming lead $i$ to the bonds $(i,r)$ with transmission amplitudes 

306: $\tau^{(i)}_{r}$. The wave gains a phase ${\rm e}^{i(k+A_{b})L_{b}}$ for each bond it 

307: traverses and a scattering amplitude $\tilde \sigma^{(i)}_{r,s}$ at each vertex. This 

308: multiple scattering inside the interaction region becomes apparent when the expansion

309: \begin{equation}

310: \label{multi}

311: ({\bf 1}-\tilde S_B(k;A))^{-1}= \sum_{n=0}^{\infty} \tilde S_B ^n(k;A)

312: \,\,\,\,\,\, 

313: \end{equation}

314: is substituted in    (\ref{scatmat}). Eventually the wave is transmitted

315: from the bond $(s,j)$

316: to the lead $j$ with an  amplitude $\tau _s^{(j)}$. Explicitly,

317: \begin{equation}

318: S^{(V)}_{i,j}  = \delta_{i,j} \rho^{(i)} +  \sum_{t \in {\cal

319: T}_{i\rightarrow j}}

320: {\cal B}_{t} {\rm e}^{i (k l_t +  \Theta_t)}

321: \label {sexplicit}

322: \end{equation}

323: where ${\cal T}_{i\rightarrow j}$ is the set of the trajectories on $\tilde

324: {\cal G}$ which lead from $i$ to $j$.  ${\cal B}_{t}$ is the amplitude

325: corresponding to a path $t$  whose length and directed length are $l_t=

326: \sum_{b\in t}L_b$ and $\Theta_t=\sum_{b\in t} L_bA_b$  respectively. 

327: Thus the

328: scattering amplitude $S^{(V)}_{i,j}$ is  a sum of a large number of partial

329: amplitudes, whose complex  interference brings about the typical irregular

330: fluctuations of $|S^{(V)}_{i,j}|^2$ as a function of $k$.

331: 

332: One of the basic concepts in the quantum theory of scattering are the 

333: resonances.

334: They represent long-lived intermediate states to which bound states of a 

335: closed

336: system are converted due to coupling to continua. On a formal level, 

337: resonances

338: show up as poles of the scattering matrix $S^{(M)}$ occurring at complex 

339: wave-numbers

340: $\kappa_n = k_n - \frac i2 \Gamma_n$, where $k_n$ and $\Gamma_n$ are the 

341: position

342: and the width of the resonances, respectively. From (\ref{scatmat}) it 

343: follows that

344: the resonances are the complex zeros of

345: \begin{equation}

346: \label {resonancecond}

347: \zeta_{\tilde {\cal G}}(\kappa) = \det \left({\bf 1} -\tilde S_B(\kappa

348: ;A)\right)=0 \ .

349: \end{equation}

350: The eigenvalues of $\tilde S_B$ are in the unit circle, and therefore the

351: resonances appear in the lower half of the complex $\kappa$ plane. Moreover

352: from

353: Eq.~(\ref{resonancecond}) it is clear that their formation is closely

354: related to the

355: internal dynamics inside the scattering region which is governed by $\tilde

356: S_B$.

357: 

358: There exists an intimate link  between the scattering matrix and the

359: spectrum of the

360: corresponding closed graph. It manifests the exterior -interior duality

361: \cite{DS92a} for  graphs. The spectrum of the closed graph is the set of

362: wave-numbers for which

363: $S^{(V)}$ has $+1$ as an eigenvalue. This corresponds to a  solution where

364: no currents flow in

365: the leads so that the conservation of current is  satisfied on the internal

366: bonds.

367: $1$ is in the spectrum of

368: $S^{(V)}$ if

369: \begin{equation}

370: \label{phsca4}

371: \zeta_{\cal G}(k)=\det\left[{\bf 1} -S^{(V)}(k)\right] = 0 \quad .

372: \end{equation}

373:   Eq.~(\ref{phsca4}) can be transformed in an alternative  form

374: \begin{equation}

375: \label{phscaa} {\hspace {-15mm}}

376: \zeta_{\cal G}(k)= \det [ {\bf 1} -\rho ]

377: {\det[{\bf 1}-D(k)R] \over \det[{\bf 1}-D(k) {\tilde R}]} = 0 \quad ; \quad

378: R_{i,r;s,j} = \tilde R_{i,r;s,j} +\delta_{r,s}

379: {\tau^{(r)}_{i}\tau^{(r)}_{j}\over 1-\rho^{(r)} }.

380: \end{equation}

381: which is satisfied once

382: \begin{equation}

383: \label{phscaa1}

384: \det[{\bf 1}-D(k)R] = 0 .

385: \end{equation}

386: In contrast to ${\tilde R}$, $R$ is a unitary matrix in the space of

387: directed bonds, and

388: therefore the spectrum is real. (\ref {phscaa1}) is the secular equation

389: for the

390: spectrum of the compact part of the graph, and it was derived in a

391: different way in \cite

392: {KS97}.

393: 

394: The difference $\delta R = R - \tilde R$ gets smaller as larger graphs are considered 

395: (for graphs with Neumann  boundary conditions it is easy to see that the difference 

396: is of order $\frac{1}{v}$). That  is, the leads are weekly coupled to the compact part 

397: of the graph, and one can use  perturbation theory for the computation of the resonance 

398: parameters. To lowest order, $(\delta R =0)$, the resonances coincide with the spectrum 

399: of the compact graph. Let $k_n$ be in the spectrum.  Hence, there exists a vector 

400: $|n \rangle$ which satisfies the equation

401: \begin{equation}

402: D(k_n) R\  |n \rangle = 1\ |n \rangle \ .

403: \end{equation}

404: To first order in $\delta R$, the resonances acquire a width

405: \begin{equation}

406: \label{pertu3}

407: \delta \kappa_n = -i  {\langle n|D(k_n)\delta R|n\rangle \over \langle n|L|n\rangle} \,.

408: \end{equation}

409: To check the usefulness of this result, we searched numerically for the 

410: true poles

411: for a few scattering graphs and compared them with the approximation 

412: ~(\ref{pertu3}).

413: In figure~1 we show the comparison for fully connected Neumann graphs

414: with $V=5,15$ and $A=0$.  As expected the agreement between the exact 

415: poles and the

416: perturbative results improves as $v$ increases.

417: 

418: \begin{figure}

419: \begin{center}

420: \epsfxsize.5\textwidth%

421: \epsfbox{sfig1.eps}

422: \caption { Poles of the $S^{(V)}$-matrix for regular Neumann graphs. The exact 

423: evaluated poles are indicated with $(\circ )$ while $(\star )$ are the results 

424: of the perturbation theory (\ref{pertu3}): (a) $V=5$ and $v=4$ and (b) $V=15$ 

425: and $v=14$.}

426: \end{center}

427: \label{fig:fig2}

428: \end{figure}

429: 

430: We finally comment that the formalism above can be easily modified for graphs where

431: not all the vertices are attached to leads. If the vertex $l$ is not attached, one

432: has to set $\rho^{(l)}=1, \tau^{(l)}_j =0$  in the definition of $\Sigma^{(l)}$.

433: The dimension of the scattering matrix is then changed accordingly.

434: 

435: For the sake of completeness we quote here an alternative expression for the $S$ 

436: matrix which applies for Neumann graphs, exclusively  \cite{KS99}. For this purpose 

437: we define the $V\times V$ matrix

438: \begin{equation}

439: \label{secu1}

440: h_{i,j}(k,A) =\left\{

441: \begin{array}{cc}

442: -\sum_{m\neq i}C_{i,m}\cot(kL_{i,m})  , &

443: i=j \\ \\

444: C_{i,j}{\rm e}^  {-iA_{i,j}L_{i,j}}(\sin (kL_{i,j}))^{-1}, & i\neq j \

445: \end{array}

446: \right.   \nonumber

447: \end{equation}

448: in terms of which,

449: \begin{equation}

450: S^{(V)} = (i {\bf 1} + h(k))^{-1} (i {\bf 1}-h(k))

451: \label{phsca3}

452: \end{equation}

453: where ${\bf 1}$ is the $V\times V$ unit matrix. $S^{(V)}$ is unitary 

454: since $h(k)$ is

455: hermitian.

456: 

457:   In the case of graphs connected to leads at an arbitrary  set of $M<V$

458: vertices with indices

459: $\left \{i_l \right\}, \ \ l=1,\cdots,M$, the $M \times M$ scattering

460: matrix $S^{(M)}$

461: has to be modified in the following way

462: \begin{equation}

463: S^{(M)} = 2i W \left ( h(k) + i W^{T} W \right ) ^{-1} W^{T} -{\bf 1} \ .

464: \label{phsca5}

465: \end{equation}

466: Here $W_{i_l,j}=\delta_{i_l,j}$ is the $M\times V$ leads - vertices coupling

467: matrix, ( $W={\bf 1}$ when $M=V$). This form of the $S$ matrix is

468: reminiscent of the

469: expression  which was introduced by Weidenm\"uller to generalize the

470: Breit-Wigner theory

471: for  many channels and internal states. However, (\ref {phsca5}) is an

472: exact expression

473: which involves no truncations.

474: 

475:   It follows from (\ref {phsca5}) that in the present case, one can identify

476: the poles of the

477: $S$ matrix with the zeros of

478: \begin{equation}

479: \label{resonN}

480: \zeta_{\tilde {\cal G}}(\kappa) = \det \left(h(\kappa) + i W^{T} W

481: \right)=0 \ .

482: \end{equation}

483: Its main advantage over (\ref{resonancecond}) is that it involves a determinant of a 

484: matrix of much lower dimension.

485: 

486: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

487: \section{\bf The trace formula for resonances and  classical scattering on

488: graphs}

489: \label{sec:classical}

490: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

491: 

492: In the spectral theory of bounded hamiltonian systems, the most fundamental object of 

493: study is the spectral density which consists of a sum of $\delta$ functions at the 

494: spectral points. For open systems, it is replaced by the resonance density, which

495: is defined on the real $k$ line and consists of an infinite sum of Lorentzians which

496: are centered at $k_n=\Re e \kappa_n$ and have width $\Gamma _n =-2\Im m \kappa_n$,

497: where $\kappa_n$ are the complex poles of the scattering matrix. In the present section 

498: we shall express the resonance density for scattering graphs in terms of periodic 

499: orbits of their compact part. This is the analogue of the trace formula for bounded 

500: graphs.

501: 

502: 

503: %-----------------------------------------------------------------------------------

504: \subsection{\bf The Trace Formula}

505: 

506: The resonance density $d_R(k)$ can be deduced from the total phase $\Phi(k) = \frac{1}

507: {i} \ln \det S^{(M)}(k) $ of the scattering matrix \cite {Krein}:

508: \begin{equation}

509: \label{resden}

510: d_R(k)\equiv{1\over 2\pi} {d\Phi\over dk} = -i{1\over 2\pi}{\partial\over

511: \partial k}\ln \det S^{(M)}(k).

512: \end{equation}

513: It is a smooth function for real $k$, and can be interpreted as the mean length of

514: the {\it delay} associated with the scattering at wave-number $k$ 

515: \cite{S89}.

516: 

517: Using Eq.~(\ref{scatmat}) and performing standard manipulations \cite{MW69}, we obtain 

518: the following expression for the phase $\Phi(k)$

519: \begin{equation}

520: \Phi(k)- \Phi(0) = -2 \Im m \ln \det (I- \tilde S_B(k;A)) + {\cal L}k \, .

521: \label{tra}

522: \end{equation}

523: where ${\cal L} =2\sum_{b=1}^B L_b $ is twice the total length of the 

524: bonds of

525: $\tilde {\cal G}$.

526: 

527: Using the expansion

528: \begin{equation}

529: \label{expan}

530: \ln \det (I- \tilde S_B(k;A)) = -\sum_{n=1}^{\infty} {1\over n} {\rm tr}

531: {\tilde

532: S_B}^n(k;A)

533: \end{equation}

534: we rewrite Eq.~(\ref{tra}) as

535: \begin{equation}

536: \label{tra1}

537: \Phi(k)= \Phi(0) + {\cal L}k +

538: 2 \Im m \sum_{n=1}^{\infty} {1\over n} {\rm tr} {\tilde S_B}^n (k;A).

539: \end{equation}

540: 

541: On the other hand, using Eq.~(\ref{DandT}) we can write ${\rm tr} {\tilde

542: S_B}^n(k;A)$

543: as sums over $n-$periodic orbits on the graph

544: \begin{equation}

545: \label{ppo}

546: {\rm tr} ({\tilde S_B}^n(k;A)) = \sum_{p\in {\cal P}_{n}} n_p {\tilde {\cal

547: A}_p}^r

548: {\rm e}^  {i(l_p k+\Theta_p)r} \quad ,

549: \end{equation}

550: where the sum is over the set ${\cal P}_n$ of primitive periodic orbits 

551: whose

552: period $n_p$ is a divisor of $n$, with $r=n/n_p$ (primitive periodic orbits

553: are those which

554: cannot be written as a repetition of a shorter periodic orbit). The 

555: amplitudes

556: ${\tilde {\cal A}}_p$ are the products of the bond-bond scattering

557: amplitudes $\tilde

558: \sigma^{(i)}_{b,b'}$ along the primitive loops i.e.

559: \begin{equation}

560: \label{ampli}

561: {\tilde {\cal A}_p} = \prod_{i=1}^{n_p} {\tilde \sigma^{(i)}_{b,b'}} \ .

562: \end{equation}

563: Substituting Eqs.~(\ref{tra1},\ref{ppo}) in Eq.~(\ref{resden}) one gets the resonance

564: density

565: \begin{equation}

566: \label{resdens}

567: d_R(k)  = {1\over 2 \pi}{\cal L}+{1\over \pi}\Re e\sum_{n=1}^{\infty}\sum_{p\in 

568: {\cal P}_{n}} n_p l_p r  {\tilde {\cal A}_p}^r {\rm e}^  {i(l_p k+\Theta_p)r}

569: \end{equation}

570: Eq.~(\ref{resdens}) is an {\it exact} trace formula for the resonance density. The

571: first term on the right hand side of Eq.~(\ref{resdens}) corresponds to the smooth

572: resonance density, while the second provides the fluctuating part. We notice that

573: the mean resonance spacing is given by

574: \begin{equation}

575: \label{mrd}

576: \Delta= {2\pi\over {\cal L}} \simeq {2 \pi\over 2 B \langle L\rangle}

577: \end{equation}

578: and it is the same as the mean level spacing obtained for the corresponding bounded 

579: graph ${\mathcal G }$ \cite{KS97}.

580: 

581: %------------------------------

582: \subsection{\bf Classical Dynamics}

583: 

584: We conclude this section with the discussion of the classical dynamics on the graph

585: ${\tilde {\cal G}}$. A classical particle moves freely as long as it is on a bond.

586: The vertices are singular points, and it is not possible to write down the analogue

587: of the Newton equations there.  In \cite {KS97,KS99} it was shown that it is possible

588: to define a classical evolution on the graph: A {\it Poincar\'{e} section} on the

589: graph consists of the discrete set of directed bonds. The phase-space density at a

590: (topological) time $n$ is the set of occupation probabilities $\rho_b(n)$ of the

591: directed bonds, and the classical evolution is governed by a Markovian master equation.

592: Applied to the compact part of a scattering graph it reads,

593: \begin{equation}

594: \label{master}

595: \tilde \rho _b(n+1)=\sum_{b^{\prime }} \tilde  U_{b,b^{\prime }}

596: \tilde \rho _{b^{\prime }}(n)

597: \end{equation}

598: where the transition matrix $ \tilde U_{b,b^{\prime }}$ is given by the 

599: corresponding

600: {\it quantum} transition probability

601: \begin{equation}

602: \tilde U_{ij,nm}= \left|\tilde R_{ij,nm}\right|^2=\delta _{j,n} |

603: \sigma^{(j)}_{ij,jm}|^2.

604:   \label{cl3}

605: \end{equation}

606:   Notice that $\tilde U$ does not involve any metric information on the 

607: graph.

608: 

609:   Due to loss of flux to the leads $\sum_{b'}{\tilde U}_{bb'}<1$, and the

610: phase-space measure is not preserved, but rather, decays in time. The

611: probability to remain on $\tilde {\cal G}$ is

612: \begin{equation}

613: {\tilde P}(n) \equiv \sum_{b=1}^{2B}{\rho}_b(n) = \sum_{b,b'}{\tilde

614: U}_{bb'}{\rho}_{b'}(n-1)

615: \simeq  {\rm e}^  {-\Gamma_{cl}n}{\tilde P(0)}

616: \end{equation}

617: where $\exp (-\Gamma_{cl})$ is the largest eigenvalue of the ``leaky" 

618: evolution

619: operator ${\tilde U}_{bb'}$.

620: 

621: For the $v$-regular graph with $\tilde \sigma^{(i)}$ given by (\ref 

622: {Neumann})

623: the spectrum of $\tilde U$ is restricted to the interior of a circle with

624: radius given by the maximum eigenvalue $\nu_1=(v-1)\tau^2+\rho^2$ with

625: corresponding eigenvector $|1>=(1/2B)(1,1,\cdots,1)^T$. Hence the decay rate

626: $\Gamma_{cl}=-\ln \nu_1$ for regular Neumann graphs take the simple form

627: \begin{equation}

628: \label{Ndrate}

629: \Gamma_{cl}=-\ln \left(1-\tau^2\right) \approx (2/(1+v))^2.

630: \end{equation}

631: We notice that removing the leads from the vertices and turning ${\tilde

632: {\cal G}}$ into a compact graph ${\cal G}$ we get $\Gamma_{cl}= 0$ since in

633: this case $\tau^{(i)}=0$ (and $\rho^{(i)}=1$) and the phase-space measure

634: is preserved as expected.

635: 

636: The inverse decay rate $T_{cl} = \Gamma_{cl}^{-1}$, gives the average classical delay 

637: time that the particle spends within the interaction region. Injecting a particle from 

638: the leads to the scattering domain, its probability to be on any bond randomizes, 

639: because at each vertex a Markovian choice of one out of $v$ directions is made. The 

640: longer a particle remains within the interaction regime, the more scattering events it 

641: experiences. The set of trapped trajectories whose occupancy decays exponentially in 

642: time is the analogue of the strange repeller in generic Hamiltonian systems displaying 

643: ``chaotic scattering".

644: 

645: 

646: %-------------------------------------------------------------------------

647: 

648: \section{\bf Statistical Analysis of the $S-$matrix}

649: \label{sec:statistics}

650: 

651: %---------------------------------------------------------------------------

652: 

653: So far we developed the scattering theory of graphs, pointing out their similarity

654: with scattering systems which display chaotic scattering in the classical limit. 

655: Due to the interference of a large number of amplitudes, the $S-$matrix fluctuates 

656: as a function of $k$, and its further analysis calls for a statistical approach 

657: which will be the subject of the present section. We shall show that quantum graphs 

658: possess typical poles, delay time and conductance distributions, Ericson fluctuations 

659: of the scattering amplitudes, and when considered statistically, the ensemble of 

660: scattering matrices are very well reproduced by the predictions of RMT. At the same 

661: time deviations from the universal RMT results, which are related to the system-

662: specific properties of some graphs, will be  pointed out. The study  of these 

663: deviations is especially convenient for graphs because of the transparent and simple 

664: scattering theory developed in terms of scattering trajectories.

665: 

666: An important parameter which is associated with the statistical properties of the 

667: $S$-matrix is the Ericson parameter defined through the scaled mean resonance width 

668: as:

669: \begin{equation}

670: \label {eripar}

671: \left<\gamma\right>_k \equiv  {\left<\Gamma_n\right>_k  \over \Delta}

672: \end{equation}

673: where $\langle \cdot \rangle _k$ denotes spectral averaging and $\Delta$ is the

674: mean spacing between resonances. The Ericson parameter determines whether the

675: resonances overlap $(\langle \gamma\rangle_k >1)$ or are isolated $(\langle

676: \gamma\rangle_k <1)$. Typical example for the two extreme situations are shown

677: in figure~2.

678: 

679: \begin{figure}

680: \begin{center}

681: \epsfxsize.5\textwidth%

682: \epsfbox{sfig2.eps}

683: \end{center}

684: \caption{ Representative examples of scattering cross sections 

685: $|S_{i,j}^{(V)}|^2$

686: for Neumann graphs in a regime of  (a) isolated resonances ($V=15$, 

687: $v=14$ with

688: corresponding $\left<\gamma\right>_k=0.59$) and (b) overlapping 

689: resonances ($V=49$,

690: $v=14$ with corresponding $\left<\gamma\right>_k\simeq 2$). The real 

691: parts of the

692: resonances in this energy interval are indicated by the vertical lines.}

693: \end{figure}

694: 

695: The degree of resonance overlap determines the statistical properties of the

696: $S$-matrix.  We shall show in the sequel that the mean width  can be

697: approximated  by the classical decay rate $\langle \gamma\rangle_k=

698: \gamma_{cl}$. For the $v$

699: regular graphs discussed above, we have

700: \begin{equation}

701: \label{clmr}

702: \gamma_{cl}\equiv {\Gamma_{cl}\over \Delta} \approx {4\over 2\pi} 

703: {v\over 1+v}

704: {V\over 1+v}.

705: \end{equation}

706: where we made use of Eqs.~(\ref{mrd},\ref{Ndrate}).

707: Thus changing $v$ and $V$ we can control the degree of overlap allowing to

708: test various phenomena.

709: 

710: In what follows, unless explicitly specified, we shall consider regular

711: graphs with

712: one lead attached to each vertex, i.e. $M=V$. Finally, the widths are

713: always scaled

714: by the mean spacing $\Delta$ i.e. $\gamma_n\equiv {\Gamma_n \over \Delta}$.

715: 

716: %---------------------------------------------------------------------------

717: \subsection{\bf The resonance width distribution }

718: 

719: The resonance width distribution can be computed for a given graph in the following 

720: way. Consider the complex $\kappa=x+iy$ plane, where the zeros of the secular function

721: \begin{equation}

722: \label {resonancecond1}

723: \zeta_{\tilde {\cal G}}(\kappa) = \det \left({\bf 1} -\tilde S_B(\kappa ;A)\right) = f(x,y) +i

724: g(x,y)

725: \ .

726: \end{equation}

727: are the poles of the $S$ matrix (resonances). The variables $(x,y)$

728: are expressed in units of $\Delta$. On average, the number of resonances

729: with real part, $x \in

730: [0,X]$ is $X$. The density of resonance widths reads

731: \begin{equation}

732: \hspace{-10mm}

733: {\mathcal P}(\gamma) =\lim_{ X \rightarrow \infty} \frac{1}{ X } \int_0^X

734: \delta\left (f(x,y= - \gamma)\right ) \delta\left (g(x,y= -  \gamma)\right )

735: \left |\frac{{\rm d}\zeta_{\tilde {\cal G}}(\kappa)}{ {\rm d} \kappa}\right |^2_{y= -

736: \gamma}   {\rm d}x

737: \ .

738: \label{eq:disty}

739: \end{equation}

740: We used the Cauchy-Riemann theorem for the evaluation of the Jacobian $(f_xg_y - f_y 

741: g_x)$ which multiplies the two $\delta$ functions that locate the complex zeros of 

742: $\zeta_{\tilde {\cal G}}$. We now recall that the $\kappa$ dependence of $\tilde S_B

743: (\kappa ;A)$ comes from the factors ${\rm e}^{i\kappa L_b}$ in (\ref {DandT}). This

744: implies that for a given $y$, $\zeta_{\tilde {\cal G}}$ is a quasi periodic function 

745: of $x$. Moreover, expanding the determinant (\ref {resonancecond1}) it is not difficult 

746: to show that the frequencies involved can be written as linear combinations of the bond 

747: lengths $\lambda =\sum  q_b L_b $ with integer coefficients $q_b =0, 1, {\rm or}\ 2$. 

748: Since the bond lengths are rationally independent, we find that $\zeta_{\tilde {\cal G}}$

749: depends on a {\it finite}  number of incommensurate frequencies. The expression (\ref

750: {eq:disty}) can be regarded as a ``time"  integral over a trajectory on a multidimensional

751: incommensurate torus, which covers the torus ergodically. Hence, the integral can be

752: replaced by a phase space average,

753: 

754: \begin{equation}

755: \hspace{-10mm}

756: {\mathcal P}(\gamma) =

757: \int_0^{2\pi}\frac {{\rm d} \psi_1}{2\pi} \cdots \int_0^{2\pi}  \frac 

758: {{\rm d}

759: \psi_J}{2\pi} \

760: \delta\left (f({\vec \psi} , - \gamma)\right )\ \delta\left (g({\vec \psi} ,

761: -\gamma)\right ) \

762: \left |\frac{{\rm d}\zeta_{\tilde {\cal G}}({\vec \psi,-\gamma})}{ {\rm d}

763: \kappa}\right |^2_ .

764: \label{eq:distyergod}

765: \end{equation}

766: Here ${\vec \psi}$ denotes  the vector of independent angles on the $J$ dimensional

767: torus. Although the above formula provides a general framework, its application to

768: actual graphs is a formidable task.

769: 

770: An important feature of the distribution of the resonances in the complex plane can

771: be deduced by studying the secular function $\zeta_{\tilde {\cal G}}(\kappa)$. Consider

772: $\zeta_{\tilde {\cal G}}(\kappa=0)$. If one of the eigenvalues of the matrix $\tilde R$

773: (\ref {DandT}) takes the value $1$, $\zeta_{\tilde {\cal G}}(k=0) =0$ and because of

774: the quasi-periodicity of $\zeta_{\tilde {\cal G}}$, its zeros reach any vicinity of

775: the real axis infinitely many times. The largest eigenvalue of the $\tilde R$ matrix

776: for $v$-regular Neumann graphs is $1$, and therefore the distribution of resonance 

777: widths is  finite in the vicinity of $\gamma=0$. For generic graphs, the spectrum of 

778: $\tilde R$ is inside  a circle of radius $\lambda_{max} <1$. This implies that the 

779: poles are excluded from a strip just under the real axis, whose width can be estimated 

780: by

781: \begin{equation}

782: \label{gapKS}

783: \Gamma_{gap}=-2 \ln( |\lambda_{max}|)/L_{max}.

784: \end{equation}

785: where $L_{max}$ is the  maximum bond length. The existence of a gap is 

786: an important

787: feature of the resonance width distribution ${\cal P}(\gamma)$ for 

788: chaotic scattering

789: systems.

790: 

791: A similar argument was used recently in \cite{BG01} in order to obtain an {\it upper}

792: bound for the resonance widths.  It is

793: \begin{equation}

794: \label{upl}

795: \Gamma_{max}\sim -2 \ln( |\lambda_{min}|)/L_{min}\ ,

796: \end{equation}

797: where $\lambda_{min}$ and $L_{min}$ are the minimum eigenvalue and bond length,

798: respectively.

799: 

800: The distribution of the complex poles  for a generic fully connected graph with $V=5$

801: is shown in figure~3a. The vertical line which marks the region from which resonances

802: are excluded was computed using (\ref {gapKS}).

803: \begin{figure}

804: \begin{center}

805: \epsfxsize.5\textwidth%

806: \epsfbox{sfig3.eps}

807: \caption{ (a) The 5000 resonances of a single realization of a complete $V=5$ graph 

808: with $A\ne 0$ and $M=V$. The solid line marks the position of the gap $\gamma_{\it gap}$. 

809: (b) The distribution of resonance widths ${\cal P}(\gamma)$. The solid line is the RMT 

810: prediction (\ref{FSpole}). The difference ${\cal P}(\gamma)-{\cal P}_{CUE} (\gamma)$ 

811: is shown in the inset.}

812: \end{center}

813: \end{figure}

814: 

815: Random matrix theory can provide an expression for the distribution of resonances. In the

816: case of non-overlapping resonances,  perturbation theory  shows that the resonance widths 

817: are distributed according to the so-called $\chi^2$ distribution

818: \begin{equation}

819: \label{chi2}

820: {\cal P}(\gamma) = \frac {(\beta M/2)^{\beta M/2}}{\langle \gamma\rangle_k{\it \Gamma}

821: (\beta M/2)} \left({\gamma\over\langle \gamma\rangle_k}\right)^{\beta M/2 -1} 

822: \exp(-\gamma \beta M/2\langle \gamma\rangle_k ),

823: \end{equation}

824: where $\beta=1(2)$ for systems which respect (break) time-reversal symmetry, and 

825: ${\it \Gamma}(x)$ is the gamma-function. Once $\langle \gamma\rangle_k$ becomes large 

826: enough the resonances start to  overlap, and (\ref{chi2}) does not hold. In the general 

827: case, Fyodorov and Sommers \cite{FS96,FS97} proved that the distribution of scaled 

828: resonance widths for the unitary random matrix ensemble, is given by

829: \begin{equation}

830: \label{FSpole}

831: {\cal P}(\gamma) = \frac {(-1)^M}{\Gamma(M)} \gamma^{M-1} {d^M\over 

832: d\gamma^M}

833: \left({\rm e}^  {-\gamma \pi g} {\sinh(\gamma\pi)\over(\gamma\pi)}\right)

834: \end{equation}

835: where the parameter $g={2\over (1-\langle S^D\rangle_k^2)} -1$ controls the degree of

836: coupling with the channels (and is assumed that $g_i=g \,\,\forall i=1,...M$). For

837: $g\gg 1$ (i.e. weak coupling regime) Eq.~(\ref{FSpole}) reduces to

838: (\ref{chi2}).

839: 

840: In the limit of $M\gg 1 $, Eq.~(\ref{FSpole}) reduces to the following expression

841: \cite{FS97}

842: \begin{equation}

843: {\cal P}(\gamma)=

844: \left\{

845: \begin{array}{lll}

846: {M\over 2\pi \gamma^2}&\ {\rm for}\ &{M\over \pi(g+1)}<

847: \gamma<{M\over\pi(g-1)} \\\nonumber

848: \ 0&\ \ \ \ & \ \ \ \ \ \  {\rm otherwise} \nonumber

849: \end{array}

850: \right.  \ .

851: \label{largeM}

852: \end{equation}

853: It shows that in the limit of large number of channels there exist a  strip in the 

854: complex $\kappa-$ plane which is free of resonances. This is in agreement with 

855: previous findings \cite{GR89,M75,SZ88}. In the case of maximal coupling i.e. $g=1$, 

856: the power law (\ref{largeM}) extends to  infinity, leading to divergencies of the 

857: various moments of $\gamma$'s. Using  (\ref{FSpole}) we recover the well known 

858: Moldauer-Simonius relation \cite{M75} for the mean resonance width \cite{FS97}

859: \begin{equation}

860: \label{SMrw}

861: \langle \gamma\rangle _k = -\frac {\sum_{i=1}^{V} \ln (|\langle S^{D}\rangle _k|^2)}{2\pi}.

862: \end{equation}

863: 

864: The resonance width distribution for a $V=5$ regular and generic graph is shown in

865: figure~3b together with the RMT prediction, which reproduces the numerical distribution

866: quite well. Figure~4 shows a similar comparison for a Neumann graph. The relatively

867: high abundance of resonances in the vicinity of the real axis conforms with the

868: expectations.

869: 

870: \begin{figure}

871: \begin{center}

872: \epsfxsize.5\textwidth%

873: \epsfbox{sfig4.eps}

874: \end{center}

875: \label{figure:n-resdist}

876: \caption {  Resonance width distribution ${\cal P}(\gamma)$ for a complete

877: Neumann graph with $M=V=5$ and $A\ne 0$. The solid line is the RMT prediction

878: (\ref{FSpole}). }

879: \end{figure}

880: 

881: 

882: %--------------------------------------------------------------------

883: 

884: \subsection{\bf The Form Factor}

885: 

886: To investigate further the dynamical origin of the resonance fluctuations, we study

887: the resonance two-point form factor $K(t)$. The main advantage of $K(t)$ is that it

888: allows us to study the resonance fluctuations in terms of classical orbits. It is

889: defined as

890: \begin{equation}

891: \label{rescor}

892: K_R(t)\equiv  \int {\rm d}\chi \ {\rm e}^{i 2\pi \chi {\cal L} t}

893: R_2(\chi).

894: \end{equation}

895: where $t$ measures lengths in units of the Heisenberg length $l_H={\cal L}$

896: and $

897: R_2(\chi)$ is the excess probability density of finding two resonances at a

898: distance

899: $\chi$,

900: \begin{equation}

901: \label{corr1}

902: \hspace {-15mm}

903: R_2(\chi;k_0) =\langle \tilde d_R  (k+ {\chi \over 2})\ \tilde d_R(k-

904: {\chi\over 2})\rangle_k = \langle {\Delta\over 2k_0}\int_{-k_0} ^{k_0}

905: {\tilde d}_R  (k+ {\chi \over 2})\ {\tilde d}_R(k- {\chi\over 2})

906: dk\rangle_{k_0}

907: \ .

908: \end{equation}

909: Above $\langle \ldots \rangle _{k_0}$ indicate averaging oven a number of spectral 

910: intervals of size $\Delta_k=2k_0$, centered around $k_0$. Here $\tilde d_R(k)$ is 

911: the oscillatory part of $d_R(k)$ (see Eq.~(\ref{resdens})). Substituting the latter 

912: in Eqs.~(\ref {rescor},\ref{corr1}) we obtain $K_R(t)$ in terms of periodic orbits 

913: and their repetitions

914: \begin{equation}

915: \label{ff1}

916: K_R(t)= {2{\cal N}\over{\cal L}^2}\left|\sum_p\sum_r l_p {\tilde {\cal 

917: A}}_p^r

918: {\rm e}^{i (kl_a+ \Theta_a)} \delta_{\cal N} (t-rl_p/{\cal L})\right|^2 .

919: \end{equation}

920: where $\delta_{\cal N}(x) = (\sin({\cal N}x/2))/({\cal N}x/2)$ and ${\cal N}={\Delta_k

921: \over\Delta}$. A similar sum contributes to the spectral form factor of the compact

922: graph \cite{KS97,KS99}. However, the corresponding amplitudes are different

923: due to the fact that ${\tilde {\cal A}}_p$ contains also the information about the

924: escape of flux to the leads.

925: 

926: \begin{figure}

927: \begin{center}

928: \epsfxsize.5\textwidth%

929: \epsfbox{sfig5.eps}

930: \caption{The form factors $K_R(t)$ for a complete $V=5$ graph, with either generic or

931: Neumann  boundary conditions, $A\neq 0$ and $M=V$. The data were averaged over $5000$ 

932: spectral intervals and smoothed on small $t$ intervals. Upper inset: $K_R(t)$ for small 

933: times.  Solid line correspond to the numerical data for the pentagon with generic boundary 

934: conditions while dashed line is the approximant $K_R(t)\approx K(t) exp(-\gamma_{cl} 

935: t)$. Lower inset: $K_R(t)$ calculated with high resolution. The lengths of  periodic 

936: orbits of the close graph are indicated by arrows.}

937: \end{center}

938: \end{figure}

939: 

940: Assuming that all periodic orbits decay at the same rate, one would 

941: substitute

942: ${\tilde {\cal A}}_p$ with ${\cal A}_p \exp(-n_p \gamma_{cl}/2)$ where 

943: ${\cal

944: A}_p$ is the weight assign to the periodic orbit of the corresponding 

945: close system.

946: Then Eq.~(\ref{ff1}) takes the following simple form

947: \begin{equation}

948: \label{corr6}

949: K_R(t) \approx K(t) {\rm e}^  {-\gamma_{cl}t}\,\,\,,

950: \end{equation}

951: where $K(t)$ is the form factor of the compact system \cite{EVP99}. 

952: Notice that

953: for $t\ll \gamma_{cl}^{-1}$ the resonance form factor $K_R(t)$ is equal 

954: to $K(t)$.

955: This is  reasonable since an open system cannot be distinguished from a 

956: closed

957: one during short times. This simple approximation is checked in the 

958: inset of

959: figure~5 (see dashed line) and it is shown to reproduce the numerical 

960: data rather

961: well in the domain $t\le 5$. The asymptotic decay is dominated by the 

962: resonances

963: which are nearest to the gap, and it cannot be captured by the crude 

964: argument

965: presented above. For generic graph, $K_R$ decays exponentially but with 

966: a rate

967: given by $\gamma_{as} =\gamma_{gap}$ (the best fit, indicated in 

968: figure~5 by the

969: dashed line, give $\gamma_{as}$ which agrees with  $\gamma_{gap}$ within 

970: $30\%$).

971: For the graph with Neumann boundary conditions, $\gamma_{gap}=0$ and one 

972: expects

973: an asymptotic power-law decay. The corresponding best fit (see dashed 

974: line in

975: figure~5) shows $t^{-2}$.

976: 

977: In Hamiltonian systems in more than one dimension, the size of the 

978: spectral interval

979: $\Delta_k$ where the spectral average is performed is limited by the 

980: requirement

981: that the smooth part of the spectral density is approximately constant. 

982: Here instead

983: we can take arbitrarily large spectral intervals since the smooth 

984: spectral density

985: is constant \cite{KS97,KS99}. This way, one can reach the domain where 

986: the function

987: $K_R(t)$ is composed of arbitrarily sharp spikes which resolve 

988: completely the length

989: spectrum for lengths which are both smaller and larger than $l_H$. In 

990: the inset of

991: figure~5 we show the numerical $K(t)$ calculated with high enough 

992: resolution. In the

993: same inset we mark with arrows the location of the lengths of short 

994: periodic orbits.

995: We notice that as long as $t{\cal L}$ is shorter than the length of the 

996: shortest

997: periodic orbit, $K(t)=0$. With increasing $t$, the periodic orbits 

998: become exponentially

999: dense and therefore the peaks start to overlap, giving rise to a quasi 

1000: continuum

1001: described approximately by Eq.~(\ref{corr6}).

1002: 

1003: 

1004: %----------------------------------------------------------------------------

1005: \subsection{\bf Ericson fluctuations}

1006: 

1007: As was mentioned above, the scattering cross sections are dominated by 

1008: either isolated

1009: resonances, or by overlapping resonances whose fluctuations follow a 

1010: typical pattern.

1011: These patterns were first discussed by Ericson \cite {E60} in the frame 

1012: of nuclear

1013: physics and were shown to be one of the main attributes of chaotic 

1014: scattering \cite

1015: {S89}. The transition between the two regimes is controlled by the 

1016: Ericson parameter

1017: (\ref {eripar}). Typical fluctuating cross sections are shown in figure~2.

1018: 

1019: \begin{figure}

1020: \begin{center}

1021: \epsfxsize.5\textwidth%

1022: \epsfbox{sfig6.eps}

1023: \end{center}

1024: \label{figu:autocorr}

1025: \caption{ The autocorrelation function $C(\chi,\nu=1)$ for regular 

1026: graphs with

1027: Neumann boundary conditions. ($\circ$) correspond to a graph with

1028: $\gamma=0.59$

1029: (isolated resonance regime), ($\star$) to a graph with $\gamma=1.36$

1030: (intermediate

1031: regime) while ($\diamond$) to a graph with $\gamma \simeq 2$ (overlapping

1032: resonances

1033: regime). The solid lines correspond to the theoretical expression

1034: (\ref{ericscl}):

1035: (a) The real part of $C(\chi,\nu=1)$ ; (b) The imaginary part of

1036: $C(\chi,\nu=1)$.}

1037: \end{figure}

1038: 

1039: 

1040: A convenient measure for Ericson fluctuations is the autocorrelation 

1041: function

1042: \begin{equation}

1043: \label{ericcor}

1044: C(\chi; \nu) =  {1\over\Delta j} \sum _{ j=j_{min}}^{j_{max}}\left 

1045: \langle  \

1046: S^{(M)}_{j,j+\nu}(k+{\chi\over 2}) \ S^{(M)\star}_{j,j+\nu}

1047: (k-{\chi \over 2}) \ \right  \rangle_{k}

1048: \end{equation}

1049: where $\Delta j = j_{max}-j_{min}+1$. To evaluate Eq.~(\ref{ericcor}) we

1050: substitute

1051: the expression of the $S$-matrix from Eq.~(\ref{sexplicit}) and  split the

1052: sum over

1053: trajectories into two distinct parts: the contributions of short

1054: trajectories are

1055: computed explicitly by following the multiple scattering expansion up to

1056: trajectories

1057: of length $l_{max}$. The contribution of longer orbits are approximated by

1058: using the

1059: diagonal approximation, which results in a Lorentzian with a width

1060: $\gamma_{Er}$.

1061: Including explicitly up to $n=3$ scattering events we get,

1062: \begin{eqnarray}

1063: \label{ericscl}

1064: C(\chi;\nu ) &\approx & G {\rm e}^  {il_{max}\chi} \frac {

1065: \gamma_{Er}}{\gamma_{Er}-i\chi}

1066: +\frac1{\Delta j}\sum_{j=j_{min}}^{j_{max}} [ \tau^4 {\rm e}^  {i\chi

1067: L_{j,j+\nu}}

1068: +\tau^4 \rho^4 {\rm e}^  {3i\chi L_{j,j+\nu} } \nonumber \\

1069: &+& \tau^6 \sum_{m\neq j,j+\nu} {\rm e}^  {i\chi(L_{j,m}+L_{m,j+\nu})} ]

1070: \end{eqnarray}

1071: where $G$ is determined by the condition $C(\chi=0;\nu)=1$. The 

1072: interplay between the

1073: contributions of long and short periodic orbits is shown in figure~6. 

1074: For overlapping

1075: resonances, the autocorrelation function is well reproduced by a 

1076: Lorentzian while for

1077: the case of isolated resonances one can clearly see the contributions of 

1078: short paths.

1079: 

1080: \begin{figure}

1081: \begin{center}

1082: \epsfxsize.5\textwidth%

1083: \epsfbox{sfig7.eps}

1084: \caption{ The mean resonance width $\left<\gamma\right>_k$, autocorrelation

1085: width $\gamma_{\it Er}$, the classical expectation $\gamma_{cl}$, and 

1086: the RMT

1087: prediction (\ref{SMrw}) vs. $V$ for various graphs with Neumann boundary

1088: conditions and constant valency $v=14$.}

1089: \end{center}

1090: \end{figure}

1091: 

1092: In figure~7 we report the mean resonance width $\langle \gamma\rangle_k$ calculated

1093: numerically for  various graphs, the parameter $\gamma_{\rm Er}$ extracted from the

1094: best fit of the  numerical $C(\chi)$ with Eq.~(\ref{ericscl}), together with the RMT

1095: prediction Eq.~(\ref{SMrw}), and the classical expectation given by Eq.~(\ref{clmr}).

1096: The results justify the use  of the classical estimate especially in the limit $V

1097: \rightarrow \infty$ for fixed $v/V$ (which is the analogue of the semiclassical limit).

1098: In this limit, the RMT and the classical estimate coincide.

1099: 

1100: %----------------------------------------------------------------------------

1101: 

1102: \subsection{\bf $S-$matrix statistics}

1103: 

1104: One can check further the applicability of RMT by studying the entire $S^{(M)}-$matrix 

1105: distribution function. The probability density of $M\times M$ unitary $S^{(M)}$-matrices 

1106: is defined in a $M+\beta M(M-1)/2$ parameter space and was found first in \cite{MPS85} 

1107: to be given by the Poisson Kernel

1108: \begin{equation}

1109: \label{sRMT}

1110: d{\cal P}_{\bar S}(S^{(M)})=p_{\bar S}(S)

1111: d\mu(S)=C_{\beta}\frac{[\det(1-{\bar S}

1112: {\bar S^{\dagger}})]^{(\beta M +2-\beta)/2}}{|\det(1-{\bar S}{\bar

1113: S^{\dagger}})|^{(\beta M +2-\beta)}} d\mu_{\beta}(S)

1114: \end{equation}

1115: where $C_{\beta}$ is a normalization constant which depend on the symmetry class. All

1116: system-specific relevant informations are included in the ensemble average $S^{(M)}-$

1117: matrix, defined as ${\bar S}_{i,j}= \langle S^D\rangle _k$. For regular graphs, ${\bar 

1118: S}$ is proportional to the unit matrix i.e. ${\bar S}_{i,j}= \frac{1-v}{1+v}\delta_{i,j}$,

1119: while the eigenvectors are distributed uniformly and independent from the eigenphases.

1120: The invariant measure $d\mu_{\beta}(S)$ is given as

1121: \begin{equation}

1122: \label{sRMT1}

1123: d\mu_{\beta}(S^{(M)}) = \prod_{i<j} |{\rm e}^  {i\phi_i}-{\rm e}^ 

1124: {i\phi_j}|^{\beta}

1125: \prod_{i=1}^Md\phi_i d\Omega

1126: \end{equation}

1127: where $d\Omega$ is the solid angle on the $M$ dimensional unit hyper-sphere.

1128: 

1129: \begin{figure}

1130: \begin{center}

1131: \epsfxsize.5\textwidth%

1132: \epsfbox{sfig8.eps}

1133: \caption{ The ${\cal P}(\psi)$ and ${\cal P}(\omega)$ distributions for a $2

1134: \times 2$ scattering matrix $S$. The solid lines correspond to the

1135: predictions of

1136: RMT (\ref{s2}). The upper panel correspond to $A=0$ while the lower panel to

1137: $A\neq 0$.}

1138: \end{center}

1139: \end{figure}

1140: For large $M$ values, a numerical check of this probability distribution is

1141: prohibitive. However, for $M=2$ one can find easily the exact form of

1142: (\ref{sRMT})

1143: and compare with the numerical results. The resulting distribution of the

1144: eigenphases $\phi_1,\phi_2$ of the $2\times 2$ scattering matrices of 

1145: regular

1146: graphs is given by

1147: %\widetext

1148: \begin{eqnarray}

1149: \label{s2}

1150: \hspace{-20mm}

1151: & &d{\cal P}_{\bar S} (S^{(2)}) = \\

1152: \hspace{-20mm}

1153: &=& \frac {C_{\beta} \left(4v\over (v+1)^2\right)^{\beta+2}

1154: \sin^{\beta}\left(\omega\over 2\right)}{\left[1-4{\bar 

1155: S}\cos(\frac{\omega}2)

1156: \cos\psi + 2{\bar S}^2 \left( 2\cos^2(\frac{\omega} 2)+\cos(2\psi)\right)-

1157: 4{\bar S}^3 \cos\psi \cos(\frac{\omega} 2)+{\bar 

1158: S}^4\right]^{\frac{\beta+2}2}}

1159: d\omega d\psi \nonumber

1160: \end{eqnarray}

1161: where we used the notation $\psi = \frac {\phi_1 +\phi_2}2$ and $\omega 

1162: = \phi_1

1163: - \phi_2$. Integrating Eq.~(\ref{s2}) over $\omega$, $\psi$ we get the 

1164: corresponding

1165: distribution functions ${\cal P}_{\bar S}(\omega)$ and ${\cal P}_{\bar 

1166: S}(\psi)$.

1167: Our numerical results for an ensample of $S$-matrices calculated for 

1168: different

1169: realizations of the lengths of the graphs are reported in figure~8 

1170: together with

1171: the RMT  predictions (\ref{s2}) for a regular graph with two channels. 

1172: An overall

1173: good agreement is seen both for $A=0$ and $A\neq 0$.

1174: 

1175: %-------------------------------------------------------------------------------

1176: \subsection{\bf Partial delay times statistics}

1177: 

1178: The Wigner delay time  captures the time-dependent aspects of quantum 

1179: scattering. It

1180: can be interpreted  as the typical time an almost monochromatic wave 

1181: packet remains

1182: in the interaction  region. It is defined as

1183: \begin{equation}

1184: \label{wignersmith1}

1185: T_W \ =\ {1\over 2 i M}{\rm tr} \left [  S^{(M)\,\,\dagger}

1186: \frac{dS^{(M)}(k)}{dk} \right ]\ =

1187: \ {1\over 2 M}

1188: \sum_{i=1}^M \frac{\partial\phi_i(k)}{\partial k}\ ,

1189: \end{equation}

1190: where $\phi_i$ are the eigen-phases of the $S^{(M)}$-matrix.

1191: The partial derivatives ${\partial \phi_i(k)\over \partial k }$ are the

1192: {\it partial delay times} and their statistical properties were studied

1193: extensively within the RMT \cite{FS96b,FS97,GM98}. For the one-channel 

1194: case it

1195: was found \cite{FS97,GM98} that the probability distribution of the scaled

1196: (with the mean level spacing $\Delta$) partial delay times $T_i={\Delta 

1197: \over

1198: 2\pi}{\partial \phi_i(k)\over \partial k }$ is

1199: \begin{equation}

1200: \label{wignersmith3}

1201: {\cal P}_{\bar S}^{(\beta)}(T) =\frac{({\beta\over 2})^{\beta/2}}

1202: {{\it \Gamma}({\beta\over 2})T^{2+\beta/2}}

1203: \int_0 ^{2\pi} {\cal P}(\phi)^{1+\beta/2} {\rm e}^  {-{\beta\over 2T}{\cal

1204: P}(\phi)}d\phi

1205: \end{equation}

1206: where ${\cal P}(\phi)$ is the Poisson Kernel Eq.~(\ref{sRMT}). The general

1207: case of

1208: $M$-equivalent open channels was studied in \cite{FS96b,FS97} where it was

1209: found that

1210: for broken time reversal symmetry the probability distribution of partial

1211: delay

1212: times is

1213: \begin{equation}

1214: \label{wignersmith4}

1215: {\cal P}(T)= \frac{(-1)^M}{M! T^{M+2}}\frac {\partial^M}{\partial(T^{-1})^M}

1216: \left[ {\rm e}^  {-g/T}I_0(T^{-1}{\sqrt(g^2-1)}) \right]

1217: \end{equation}

1218: where $I_0(x)$ stands for the modified Bessel function.

1219: 

1220: \begin{figure}

1221: \begin{center}

1222: \epsfxsize.5\textwidth%

1223: \epsfbox{sfig9.eps}

1224: \caption{

1225: The distribution of the scaled partial delay times $T$ for various graphs with

1226: Neumann boundary conditions. The dashed lines correspond to the RMT expectation

1227: (\ref{wignersmith3},\ref{wignersmith4}): (a) One channel and $A=0$; (b) $M=V$

1228: channels and $A\ne 0$.}

1229: \end{center}

1230: \end{figure}

1231: 

1232: To investigate the statistical properties of the partial delay times for our system we 

1233: had calculated ${\cal P}(T)$ for various graphs. The resulting distributions are shown 

1234: in figure~9  together with the RMT predictions (\ref{wignersmith3},\ref{wignersmith4}).

1235: An overall agreement is evident.  Deviations appear at the short time regime (i.e. 

1236: short  orbits), during which the ``chaotic" component due to multiple scattering is

1237: not yet fully developed \cite {RBUS90}.

1238: 

1239: %----------------------------------------------------------------------------

1240: 

1241: \subsection{\bf Conductance Distribution }

1242: 

1243: Due to the recent experimental investigation of electronic transport through mesoscopic

1244: devices  \cite{MRWHG92}, the study  of the statistical properties of the conductance 

1245: gained some importance. For a device connected to reservoirs by leads, the Landauer-

1246: B\"uttiker formula relates its conductance $G$ to the transmission coefficient $T_G$ 

1247: by the expression $G=(2{\rm e}^  2/\hbar) T_G$. When each lead supports one channel, 

1248: the transmission coefficient can be written in terms of the $S^{(M)}-$matrix as

1249: \begin{equation}

1250: \label{landauer}

1251: T_G= \sum_{i\neq j}^M |S_{i,j}^{(M)}|^2 = 1-|S_{j,j}^{(M)}|^2

1252: \end{equation}

1253: where $j$ is the input channel.

1254: \begin{figure}

1255: \begin{center}

1256: \epsfxsize.5\textwidth%

1257: \epsfbox{sfig10.eps}

1258: \caption{ Conductance distribution ${\cal P}(T_G)$ for a graph with $M=2$

1259: channels: (a) Time reversal symmetry is preserved ($A=0)$. The solid line is

1260: the RMT results Eq.~(\ref{ptrans}) where we had used Eq.~(\ref{s2}) for the

1261: Poisson Kernel while the dashed line is the approximate expression (\ref{Slarge1});

1262: (b) Broken time reversal symmetry ($A\ne 0$). The solid line is the RMT result

1263: Eq.~(\ref{Slarge2}).}

1264: \end{center}

1265: \end{figure}

1266: 

1267: In the absence of direct processes, the distribution of conductance ${\cal P}(T_G)$ 

1268: for arbitrarily number of channels was worked out within RMT, and the results 

1269: describe in a satisfactory way both the numerical calculations and the experimental 

1270: data (for a review see \cite{MB99} and references therein). However in cases where 

1271: direct processes appear significantly, one must use the Poisson's kernel (\ref{sRMT}) 

1272: in its full  generality. This is exactly the case with Neumann graphs since ${\bar S} 

1273: \neq 0$. The  probability distribution of the transmission $T_G$ is then

1274: \begin{equation}

1275: \label{ptrans}

1276: {\cal P} (T_G) = \int \delta(T_G-\sum_{i\neq j} |S_{i,j}|^2) {\cal P}_{\bar

1277: S}(S)

1278: d\mu(S).

1279: \end{equation}

1280: 

1281: For the case $M=2$ and for a diagonal ${\bar S}-$matrix (only direct

1282: reflection) with

1283: equivalent channels ${\bar S}_{11}={\bar S}_{22}={\bar S}$,

1284: Eq.~(\ref{ptrans}) can be

1285: worked out analytically \cite{BB94} in the limit of strong reflection and

1286: $T_G\ll 1$.

1287: The resulting expression is:

1288: \begin{equation}

1289: \label{Slarge1}

1290: {\cal P}_{\beta=1}(T_G)=

1291: {8\over \pi^2(1-{\bar S}^2)} T_G^{-1/2}\quad , \quad  T_G\ll (1-{\bar 

1292: S}^2)^2

1293: \end{equation}

1294: For $\beta=2$ one can  compute in a close form the whole distribution

1295: \cite{MB99}

1296: \begin{equation}

1297: \label{Slarge2}

1298: {\cal P}_{\beta=2}(T_G)= (1-{\bar S}^2) \frac{(1-{\bar S}^4)^2+2{\bar 

1299: S}^2 (1+

1300: {\bar S}^4)T_G+ 4{\bar S}^4T_G^2}{((1-{\bar S}^2)^2 + 4{\bar S}^2 

1301: T_G)^{5/2}}.

1302: \end{equation}

1303: 

1304: Our numerical results for a regular graph with two leads are plotted in 

1305: figure~10

1306: together with the RMT results 

1307: (\ref{ptrans},\ref{Slarge1},\ref{Slarge2}). Notice

1308: that,  the small conductances are emphasized, because of the presence of 

1309: direct

1310: reflection  and no direct transmission. At the same time, the most 

1311: pronounced

1312: differences between  the two symmetry classes are for $T_G\ll 1$, where for

1313: $\beta=1$ the conductance distribution diverges while it take a finite 

1314: value for

1315: $\beta = 2$.

1316: 

1317: %----------------------------------------------------------------------------

1318: 

1319: \section{\bf Scattering from star-graphs.}

1320: \label{sec:star}

1321: 

1322: So far, we have studied scattering from well connected graphs, and we have shown that

1323: many statistical properties of the scattering matrix are described by RMT. We shall

1324: dedicate this section to demonstrate the statistical properties of the $S-$matrix for

1325: graphs which have non-uniform connectivity.

1326: 

1327: A representative example of this category are the ``star" (or ``Hydra") graphs 

1328: \cite{KS99,BK99}. They consist of $v_0$ bonds, all of which emanate from a single 

1329: common vertex labeled with the index $i=0$. The vertex at $i=0$, will be referred

1330: to in the sequel as the  {\it head}. The total number of vertices for such a graph 

1331: is $V=v_0+1$, and the vertices at the end of the bonds will be labeled by $i=1

1332: \cdots v_0$. The star is a bipartite graph, a property which implies e.g., that there

1333: exists no periodic orbits of odd period \cite{KS99}. To turn a star graph  into a 

1334: scattering system we add a lead to its head.

1335: 

1336: It is a simple matter to derive the scattering matrix $S=S^{(M=1)}$ for a Neumann star. 

1337: It reduces to a phase factor, 

1338: \begin{equation}

1339: \label{contH}

1340: S(k)\equiv {\rm e}^  {i\phi(k)} = {-\sum_{i=1}^{v_0} \tan(kL_i) +

1341: i \over \sum_{i=1}^{v_0} \tan(kL_i) + i}

1342: \end{equation}

1343: 

1344: The spectrum $\{k_n\}$ of the close system can be identified as the set of wave-vectors

1345: for which $S(k)$ equals  1, which implies that no current flows in the lead. The 

1346: resulting quantization condition is

1347: \begin{equation}

1348: \label{eq:starb}

1349: 1-S(k)=0\longleftrightarrow {2\sum_{i=1}^{v_0} \tan(k_nL_i) \over

1350: \sum_{i=1}^{v_0}

1351: \tan(kL_i) + i}=0\quad,

1352: \end{equation}

1353: which is satisfied once $\sum_{i=1}^{v_0} \tan(k_nL_i)=0$. This is 

1354: identical with

1355: the condition derived in \cite{KS99}.

1356: 

1357: The poles $\{\kappa_n\}$ are the complex zeros of

1358: \begin{equation}

1359: \label{openH}

1360: \sum_{i=1}^{v_0}\tan(\kappa_nL_i) + i =0.

1361: \end{equation}

1362: To first order in $\frac {1}{v_0}$, we get

1363: \begin{equation}

1364: \label{pertH}

1365: \Gamma_n^{(1)} = {1\over\sum_{i=1}^{v_0}{L_i\over \cos(2k_nL_i) +1}}\

1366: \end{equation}

1367: which can be used as a starting point for the exact evaluation of the 

1368: poles. For the

1369: latter one has to perform a self consistent search for the complex zeros 

1370: of the

1371: secular equation (\ref{openH}). This is a time consuming process and the 

1372: correct

1373: choice of the initial conditions is very important.

1374: 

1375: In figure~11  we present our numerical results for the distribution of 

1376: rescaled

1377: resonance widths ${\cal P}(\gamma)$ for a star with $v_0=20$. The data 

1378: are in

1379: excellent agreement with the RMT expectation given in Eq.~(\ref{chi2}). 

1380: We point

1381: that in this case the coupling  to the continuum is weak since $g\simeq 

1382: 10 \gg 1$

1383: and therefore the $\chi^2$-distribution  with $M=1$ is applicable.

1384: 

1385: \begin{figure}

1386: \begin{center}

1387: \epsfxsize.5\textwidth%

1388: \epsfbox{sfig11.eps}

1389: \caption{ The rescaled resonance width distribution ${\cal P}(\gamma)$ for

1390: a star graph

1391: with $v_0=20$. The solid line is the RMT prediction Eq.~(\ref{chi2}).}

1392: \end{center}

1393: \end{figure}

1394: 

1395: Using Eq.~(\ref{contH}) we get the following relation for the scaled 

1396: delay times

1397: \begin{equation}

1398: \label{delayH}

1399: T(k) ={\Delta\over 2\pi}{2\sum_{i=1}^{v_0} {L_i\over \cos^2(kL_i)}\over 1+

1400: \left(\sum_{i=1}^{v_0}\tan k L_i\right)^2}

1401: \end{equation}

1402: which can be used to generate ${\cal P}(T)$. The latter is reported in figure~12

1403: together with the RMT prediction ~(\ref{wignersmith4}). We notice that although

1404: the tail of the distribution agrees reasonably well with the RMT prediction, there

1405: are considerable deviations at the origin.

1406: 

1407: A peculiarity of the star graph is that the mean time delay is twice larger than the

1408: expected one from semiclassical considerations. To be more specific, for a generic

1409: chaotic system coupled to $M$ channels, one has \cite{DS92a}

1410: \begin{equation}

1411: \label{taumH}

1412: \langle T\rangle _k \simeq {M+1\over M} {\Delta\over 2\pi} \langle l_p\rangle _k

1413: \end{equation}

1414: where $\langle l_p\rangle _k\simeq \Gamma_{cl}^{-1}$ is the average length of the 

1415: classical paths inside the interaction area. Thus for the star with $M=1$, we would 

1416: expect based on Eq.~(\ref{taumH}) that $\langle T\rangle _k \simeq 2 \Gamma_{cl}$. 

1417: However, due to the fact that all the periodic orbits on the  star graph are self 

1418: tracing we get an additional factor of $2$ and thus

1419: \begin{equation}

1420: \label{mtHc}

1421: \langle T\rangle _k= {\Delta\over 2\pi} 4 \Gamma_{cl}^{-1}.

1422: \end{equation}

1423: 

1424: \begin{figure}

1425: \begin{center}

1426: \epsfxsize.5\textwidth%

1427: \epsfbox{sfig12.eps}

1428: \caption {The distribution of the scaled partial delay times ${\cal P}(T)$ for a star 

1429: graph with $v_0=20$. The dashed line is the RMT prediction Eq.~(\ref{wignersmith3}).

1430: In the upper inset we show the same data in a double logarithmic plot. In the lower 

1431: inset we plot the numerical results for $<T>_k$ ($\circ$). The solid line is the 

1432: asymptotic value $1$ expected from semiclassical considerations.}

1433: \end{center}

1434: \end{figure}

1435: 

1436: The corresponding classical decay rate $\Gamma_{cl}$ can be found exactly by a direct

1437: evaluation of the eigenvalues of the classical evolution operator ${\tilde U}$. One 

1438: can easily show that

1439: \begin{equation}

1440: \label{UH1}

1441: {\tilde U} =\left( \begin{array}{ll}

1442:   0 & {\bf 1}  \\

1443: ({\tilde \sigma^{(0)}}) ^2 & 0

1444: \end{array}  \right)

1445: \end{equation}

1446: where ${\tilde \sigma}^{(0)}$ is the $v_0\times v_0$ vertex scattering matrix at

1447: the star head as defined in Eq.~(\ref{Neumann}), while ${\bf 1}$ denotes the

1448: $v_0\times v_0$ identity matrix. The square of the classical evolution operator

1449: ${\tilde U}^2$ has a block diagonal form

1450: \begin{equation}

1451: \label{UH2}

1452: {\tilde U}^2 =\left( \begin{array}{ll}

1453: ({\tilde \sigma^{(0)}}) ^2 & 0 \\

1454: 0 & ({\tilde \sigma^{(0)}}) ^2

1455: \end{array}  \right).

1456: \end{equation}

1457: and its spectrum consists of the values $1-{4\over (1+v_0)^2}$ with multiplicity

1458: 2 and $1-{4\over (1+v_0)}$ with multiplicity $2v_0-2$. Therefore the spectrum of

1459: ${\tilde U}$ is

1460: \begin{eqnarray}

1461: \label{UH3}

1462: \lambda _u &=& \pm \sqrt {1-{4\over(1+v_0)^2}}\\

1463:             &=& \pm \sqrt {1-{4\over 1+v_0}} \quad \quad {\rm with}\,\, {\rm

1464: multiplicity}

1465: \,\, v_0-1\nonumber.

1466: \end{eqnarray}

1467: For short times where the classical evolution is applicable, the dominant eigenvalue

1468: is $\lambda_u=\sqrt{1-{4\over 1+v_0}}$ leading to a classical decay rate 

1469: \begin{equation}

1470: \label{Hgammac}

1471: \Gamma_{cl} \simeq {2\over 1+v_0}.

1472: \end{equation}

1473: Substituting Eq.~(\ref{Hgammac}) we get eventually that $\langle T\rangle _k = {1+

1474: v_0\over v_0} \ \overrightarrow{\scriptstyle v_0{\rightarrow \infty }}\ 1$ which 

1475: indicate that the mean time a particle spend inside the interaction regime is

1476: proportional to the Heisenberg time. Our numerical calculations reported in the lower 

1477: inset of Fig.~12 agrees nicely with the semiclassical prediction ~(\ref{mtHc}).

1478: 

1479: Finally, we analyze the distribution of $S$ when generated over different realizations 

1480: of the lengths of the bond. For the one channel case this is equivalent with the 

1481: distribution ${\cal P} (\phi)$ of the phase of the $S$-matrix. To derive the latter, 

1482: it is convenient to rewrite the $S$-matrix in the following form

1483: \begin{equation}

1484: \label{hydraPs}

1485: S\equiv \exp(i\phi) =(-K + i)/(K + i),

1486: \end{equation}

1487: with $ K = \sum_{i=1}^v \tan(kL_i) $. The probability distribution of

1488: $K$ is

1489: \begin{eqnarray}

1490: \label{hydra2}

1491: {\cal P}(K)& = &\left \langle \delta\left(K-\sum_{i=1}^{v}

1492: \tan (kL_i) \right)\right \rangle_{L}\nonumber \\

1493: & = & \frac 1{2\pi} \int {\rm e}^  {iKx} dx \left( \frac1 {\Delta L}

1494: \int_{L_{min}}^{L_{max}} dL {\rm e}^  {-ix\tan(kL)}\right)^v \nonumber \\

1495: &=& \frac 1{\pi}

1496: \frac {v}{v^2+K^2} \ .

1497: \end{eqnarray}

1498: Thus, with $\bar {S} \equiv \langle S\rangle =\frac{1-v}{1+v}$, we get

1499: \begin{equation}

1500: \label{hydra3}

1501: {\cal P}(\phi) = {\cal P}(K) \left|\frac{dK}{d\phi}\right| =

1502: \frac 1{2\pi} \frac{1-{\bar S}^2}{1+{\bar S}^2-2{\bar S} \cos\phi} \ .

1503: \end{equation}

1504: Equation ~(\ref{hydra3}) reproduces the Poisson's kernel for a one-channel

1505: scattering

1506: matrix, derived in the framework of RMT \cite{FS97}. The conditions under

1507: which this

1508: result is derived in \cite{FS97} are fulfilled {\it exactly} in the present

1509: case. Our numerical results are reported in figure~13 and are in

1510: excellent agreement  with the theoretical prediction Eq.~(\ref{hydra3}).

1511: 

1512: \begin{figure}

1513: \begin{center}

1514: \epsfxsize.5\textwidth%

1515: \epsfbox{sfig13.eps}

1516: \end{center}

1517: \label{fig:starphase}

1518: \caption{The distribution of phases ${\cal P}(\phi)$ of the $S_H$-matrix

1519: for two stars with $v_0=15$ and $v_0=30$. The solid lines are the 

1520: corresponding

1521: theoretical predictions (\ref{hydra3}).}

1522: \end{figure}

1523: 

1524: %-------------------------------------------------------------------------

1525: 

1526: \section{\bf Conclusions}

1527: \label{sec:conclusions}

1528: 

1529: In this paper, we turned quantum graphs into scattering systems. We show that they

1530: combine the desirable features of both behaving ``typically" and being mathematically

1531: simple. Thus, we propose them as a convenient tool to study generic behavior of

1532: chaotic scattering systems.

1533: 

1534: The classical dynamics on an open graph was defined, and the classical staying

1535: probability was shown to decay in an exponential way. The resulting classical

1536: escape rate was calculated and used to describe the properties of the corresponding

1537: quantum system. The scattering matrix was written in terms of classical orbits and

1538: an exact trace formula for the resonance density was found. A gap for the resonance

1539: widths has been obtained for ``generic" graphs and its absence was explained for

1540: Neumann graphs. An analysis of the cross section autocorrelation function was 

1541: performed and its non-universal characteristics were explained in terms of the short

1542: classical scattering trajectories. Finally, due to the relative ease by which a large

1543: number of numerical data can be computed for the graph models, we had performed a

1544: detail statistical analysis of delay times, resonance widths and distribution of the

1545: $S$-matrix. Our results compares well with the predictions of RMT indicating that our

1546: model can be used in order to understand the origin of the connection between RMT and

1547: the underlying classical chaotic dynamics.

1548: 

1549: The results reported here, complete our previous investigations on graphs. We conclude

1550: that quantum graphs may serve as a convenient paradigm in the area of quantum chaos,

1551: both for spectral and scattering studies.

1552: 

1553: 

1554: \section{\bf Acknowledgments}

1555: 

1556: We acknowledge many useful discussions with Y. Fyodorov and H. Schanz. This work was

1557: supported by the Minerva Center for Nonlinear Physics, and an Israel Science Foundation

1558: Grant. (T.~K.) acknowledges a postdoctoral fellowship from the Feinberg School, The 

1559: Weizmann Institute of Science.

1560: 

1561: %-----------------------------------------------------------------------------

1562: 

1563: \begin{thebibliography}{99}

1564: 

1565: \bibitem{A85}  S. Alexander, Phys. Rev. B {\bf 27}, 1541 (1985).

1566: 

1567: \bibitem{FJK87}  C.~Flesia, R.~Johnston and H.~Kunz, Europhys. Lett. {\bf 3}

1568: , 497 (1987).

1569: 

1570: \bibitem{ML77}  R. Mitra and S. W. Lee, {\it Analytical techniques in the

1571: theory of guided waves,} Macmillan, New York.

1572: 

1573: \bibitem{A81}  P. W. Anderson, Phys. Rev. B {\bf 23}, 4828 (1981); B. 

1574: Shapiro,

1575: Phys. Rev. Lett. {\bf 48} 823 (1982).

1576: 

1577: \bibitem{CC88} J. T. Chalker and P. D. Coddington, J. Phys. C {\bf 21}, 2665

1578: (1988); Rochus Klesse and Marcus Metzler, Phys. Rev. Lett. {\bf 79}, 721

1579: (1997); R. Klesse, Ph. D. Thesis, Universitat zu K\"oln, AWOS-Verlag, 

1580: Erfurt,

1581:   1996.

1582: 

1583: \bibitem{NYO94}  Y.~Avishai, J. M. Luck, Phys. Rev. B {\bf 45}, 1074 (1992);

1584: T.~Nakayama, K.~Yakubo, R.~L.~Orbach, Rev. Mod. Phys. {\bf 66}, 381 (1994).

1585: 

1586: \bibitem{Ibook}  Y. Imry, {\it Introduction to Mesoscopic Systems}, Oxford,

1587: New York (1996); D. Kowal, U. Sivan, O. Entin-Wohlman, Y. Imry, Phys. Rev. B

1588: {\bf 42}, 9009 (1990).

1589: 

1590: \bibitem{MT01} J. Vidal, G. Montambaux, D. Doucot, Phys. Rev. B {\bf 62},

1591: R16294 (2000).

1592: 

1593: \bibitem{Z98} Zhang et al. Phys.~Rev.~Lett. {\bf 81}, 5540 (1998).

1594: 

1595: \bibitem{V98} Colin de Verdi\`ere. {\em Spectres de Graphes}. Soci\'et\'e

1596: Math\'ematique \phantom{$[$19$]$} de France, Marseille (1998).

1597: 

1598: \bibitem{E95} P.~Exner, Phys. Rev. Lett. {\bf 74}, 3503 (1995); P.~Exner,

1599: P.~Seba, Rep. Math. Phys. {\bf 28}, 7 (1989); J.~E.~Avron, P.~Exner, Y. 

1600: Last,

1601: Phys. Rev. Lett. {\bf 72}, 896 (1994); R. Carlson, Electronic Journal of

1602: Differential Equations {\bf 6}, 1 (1998).

1603: 

1604: \bibitem{R83} Jean-Pierre Roth, in: {\it Lectures Notes in Mathematics: 

1605: Theorie

1606: du Potentiel}, A. Dold and B. Eckmann, eds. (Springer--Verlag) 521-539.

1607: 

1608: 

1609: \bibitem{KS97} Tsampikos Kottos and Uzy Smilansky. Phys.~Rev.~Lett. {\bf 

1610: 79},

1611: 4794 (1997).

1612: 

1613: \bibitem{KS99}Tsampikos Kottos and Uzy Smilansky, Annals of Physics {\bf 

1614: 274},

1615: 76 (1999).

1616: 

1617: \bibitem{A99}Akkermans E, Comtet A, Desbois J, Montambaux G, Texier C,

1618: Annals of Physics {\bf 284}, 10 (2000).

1619: 

1620: \bibitem{BK99}Berkolaiko G. and Keating J.~Phys.~A {\bf 32} 7827 (1999);

1621: Berkolaiko G., Bogomolny E.B., and Keating J.~Phys.~A {\bf 34} 335 (2001).

1622: 

1623: \bibitem{SS00}H. Schanz and U. Smilansky. Phys.~Rev.~Lett. {\bf 84}, 

1624: 1427 (2000);

1625: Proceedings of the Australian Summer School on Quantum Chaos and 

1626: Mesoscopics,

1627: Canberra, Australia, 1999. [chao-dyn/9904007]; Tsampikos Kottos and Holger

1628: Schanz, Physica E {\bf 9}, 523 (2001).

1629: 

1630: \bibitem{T00} G. Tanner, J.~Phys.~A {\bf 33}, 3567 (2001).

1631: 

1632: \bibitem{BSW02} G. Berkolaiko, H. Schanz, R. S. Whitney, Phys. Rev. Lett.

1633: {\bf 88}, 104101 (2002).

1634: 

1635: \bibitem{BG00} F. Barra, P. Gaspard, J. Stat. Phys. {\bf 101}, 283 (2000);

1636: F. Barra, P. Gaspard, Phys. Rev. E {\bf 63}, 066215 (2001).

1637: 

1638: \bibitem{K01}L. Kaplan, Phys. Rev. E {\bf 64}, 036225 (2001); Tsampikos 

1639: Kottos

1640: in preparation (2002).

1641: 

1642: \bibitem{KS00} Tsampikos Kottos and Uzy Smilansky, Phys. Rev. Lett. {\bf85},

1643: 968 (2000).

1644: 

1645: \bibitem{BG01} F. Barra and P. Gaspard, Phys. Rev. E, {\bf 65}, 016205 

1646: (2002).

1647: 

1648: \bibitem{TM01} C. Texier and G. Montambaux, J.~Phys.~A {\bf 34} 10307 

1649: (2001);

1650: C. Texier, J.~Phys.~A {\bf 35} 3389 (2002).

1651: 

1652: \bibitem{M73} W. H. Miller, Adv. Chem. Phys. {\bf 25} 69 (1974).

1653: 

1654: \bibitem{S89}  U. Smilansky, in {\it Les Houches Summer School on Chaos

1655: and Quantum Physics}, M.-J. Giannoni et.al., eds. (North-Holland) 371-441

1656: (1989) (see also, {\it Les Houches Summer School on mesoscopic quantum 

1657: physics},

1658: E. Akkermans et. al., eds. (North-Holland) 373-433 (1994)).

1659: 

1660: \bibitem{GR89} P. Gaspard, in {\it ``Quantum Chaos", Proceedings of

1661: E. Fermi Summer School 1991}, G. Casati et. al., eds. (North-Holland)

1662: 307; P. Gaspard and S. A. Rice, J. Chem Phys {\bf 90}, 2225,2242,2255(1989);

1663: {\bf 91} E3279, (1989).

1664: 

1665: \bibitem{RBUS90}  R. Bl\"umel and U. Smilansky , Phys. Rev. Lett. {\bf 64},

1666: 241 (1990).

1667: 

1668: \bibitem {EVP99} B. Eckhardt, I. Varga, P. Pollner, Physica E {\bf 9}, 535

1669: (2001).

1670: 

1671: \bibitem{M75} P. A. Moldauer, Phys. Rev. C {\bf 11}, 426 (1975); P. A.

1672: Moldauer,

1673: Phys. Rep. {\bf 157}, 907 (1967); M. Simonius, Phys. Lett. {\bf 52B}, 279

1674: (1974).

1675: 

1676: \bibitem{MPS85} P. Mello, P. Pereyra, and T. H. Seligman, Ann. Phys. (NY)

1677: {\bf 161}, 254 (1985).

1678: 

1679: \bibitem{JSA92} R. A. Jalabert, A. D. Stone and Y. Alhassid, Phys. Rev. 

1680: Lett.

1681: {\bf 68}, 3468 (1992).

1682: 

1683: \bibitem{HILSS92} F. Haake, F. Izrailev, N. Lehmann, D. Saher, H-J Sommers,

1684: Z. Phys. B {\bf 88}, 359 (1992); N. Lehmann, D. Saher, V. Sokolov,  H-J

1685: Sommers,

1686: Nucl. Phys. A {\bf 582}, 223 (1995).

1687: 

1688: \bibitem{BB94} P. W. Brouwer and C. W. J. Beenakker, Phys. Rev. B {\bf 50},

1689: 11263 (1994).

1690: 

1691: \bibitem{FS96b}Y. V. Fyodorov and H.-J. Sommers, Phys. Rev. Lett. {\bf 76},

1692: 4709 (1996); Y. V. Fyodorov, D. V. Savin, and H.-J. Sommers, Phys. Rev. E

1693: {\bf 55} R4857 (1997).

1694: 

1695: \bibitem{FS96} H-J Sommers, Y. V. Fyodorov, and M. Titov, J. Phys. A:

1696: Math. Gen. {\bf 32} L77 (1999); Y. Fyodorov, H-J Sommers, Pis'ma Zh. Eksp.

1697: Teor.

1698: Fiz. {\bf 63} 970 (1996) [JETP Lett. {\bf 63},1026 (1996)]

1699: 

1700: \bibitem{FS97} Y. Fyodorov, H-J Sommers, J. Math. Phys. {\bf 38}

1701: 1918 (1997); P. Mello in {\it Les Houches Summer School on Chaos and

1702: Quantum Physics}, E. Akkermans et.al., eds (North-Holland) 437-491 (1994).

1703: 

1704: \bibitem{BFB97} P. W. Brouwer, K. M. Frahm, C. W. Beenakker, Phys. Rev. 

1705: Lett.

1706: {\bf 78}, 4737 (1997).

1707: 

1708: \bibitem{GM98} V. A. Gopar and P. Mello, Europhys. Lett. {\bf 42}, 131 

1709: (1998).

1710: 

1711: \bibitem{MB99} P. A. Mello and H. U. Baranger, Waves in Random Media 

1712: {\bf 9},

1713: 105 (1999).

1714: 

1715: \bibitem{DS92a} E. Doron, and U. Smilansky, Nonlinearity {\bf 5}, 1055 

1716: (1992);

1717: E. Doron, and U. Smilansky, Phys. Rev. Lett. {\bf 68}, 1255 (1992).

1718: 

1719: \bibitem {MW69}C. Mahaux and H. A. Weidenmuller, {\it ``Shell Model Approach

1720: in Nuclear Reactions''}, (North-Holland, Amsterdam), (1969).

1721: 

1722: \bibitem{Krein} M.~Sh. Birman and D.~R. Yafaev.

1723: \newblock The spectral shift function. {T}he work of {M}. {G}. {K}rein and

1724:    its further development.

1725: \newblock {\em St. Petersburg Math. J.}, 4:833--870, 1993.

1726: 

1727: \bibitem{SZ88} V. Sokolov and G. Zelevinsky, Phys. Lett. B {\bf202} 10 (1988); 

1728: E. Sobeslavsky, F. M. Dittes, and I. Rotter, J. Phys. A: Math. Gen. {\bf 28},

1729: 2963 (1995).

1730: 

1731: \bibitem{E60}T. Ericson, Phys. Rev. Lett. {\bf 5}, 430 (1960); T. Ericson,

1732: Ann. Phys. (NY) {\bf 23} 390 (1963).

1733: 

1734: \bibitem{MRWHG92}C. M. Marcus, A. J. Rimberg, R. M. Westervelt, P. F. 

1735: Hopkins,

1736: and A. C. Gossard, Phys. Rev. Lett. {\bf 69}, 506 (1992).

1737: 

1738: \end{thebibliography}

1739: 

1740: 

1741: \end{document}

1742: