cond-mat0104450/qff.tex
1: 
2: \documentstyle[prl,aps,multicol]{revtex}
3: %\documentstyle[12pt]{article}
4: \input{epsf}
5: \begin{document}
6: \draft
7: \title{Quantum Fractal Fluctuations}  
8: \author{Giuliano Benenti$^{(a,b)}$, Giulio Casati$^{(a,b,c)}$, 
9: Italo Guarneri$^{(a,b,d)}$, and Marcello Terraneo$^{(a,b)}$}  
10: \address{$^{(a)}$International Center for the Study of Dynamical 
11: Systems,} 
12: \address{
13: Universit\`a degli Studi dell'Insubria, Via Valleggio 11,  
14: 22100 Como, Italy} 
15: \address{$^{(b)}$Istituto Nazionale per la Fisica della Materia, 
16: Unit\`a di Milano, Via Celoria 16, 20133 Milano, Italy}   
17: \address{$^{(c)}$Istituto Nazionale di Fisica Nucleare, 
18: Sezione di Milano, Via Celoria 16, 20133 Milano, Italy}   
19: \address{$^{(d)}$Istituto Nazionale di Fisica Nucleare, 
20: Sezione di Pavia, Via Bassi 6, 27100 Pavia, Italy}   
21: \date{\today}
22: \maketitle
23: 
24: \begin{abstract}
25: We numerically analyse quantum survival probability fluctuations 
26: in an open, classically chaotic system. In a quasi-classical regime, and in
27: the presence of classical mixed phase space, such fluctuations are 
28: believed to exhibit a fractal pattern, on the grounds of semiclassical 
29: arguments.
30: In contrast, we work in a classical regime of complete chaoticity, 
31: and in a deep quantum regime of strong localization. We provide 
32: evidence that fluctuations are still fractal,  due to 
33: the slow, purely quantum algebraic  decay in time produced by 
34: dynamical localization.
35:  Such findings considerably enlarge  the scope of the existing theory.
36: 
37: \end{abstract}
38: \pacs{PACS numbers: 05.45.Df, 05.40.-a, 05.45.Mt}  
39: %05.45.Df - Fractal 
40: %05.40.-a - Fluctuation phenomena, random processes, noise, and 
41: %Brownian motion 
42: %05.45.Mt - Semiclassical chaos ("quantum chaos") 
43: 
44: \begin{multicols}{2}
45: \narrowtext
46: 
47: The study of mesoscopic conductance fluctuations has gained important 
48: insights from the analysis of the underlying classical dynamics
49: \cite{jal}.
50: In particular, it has been found that some statistical properties 
51: of quantum fluctuations are semiclassically 
52: determined by the classical law which rules the decay in time of the 
53: survival probability inside the open 
54: chaotic system\cite{smila}. Building on this general relation, it was eventually 
55: predicted that, if the classical decay law is algebraic, due to 
56: the presence of residual stable islands in the classical phase space, 
57: then the quantum fluctuation pattern should display a fractal structure 
58: with a fractal dimension directly related to the classical decay exponent
59: \cite{ketz}. 
60: Such fractal structure of quantum 
61: fluctuations has been interpreted as a quantal manifestation of
62: classical fractality  associated  with hierarchical 
63: structures of surviving stable islands.
64: This theoretical prediction has received experimental support\cite{expe}, 
65: and has 
66: been numerically confirmed on a dissipative model system, 
67: where coupling to continuum is simulated by complete absorption of that part of the
68: wave packets which propagates outside the interaction region\cite{casa,spain}.
69: 
70: In such a theoretical frame, quasi-classicality of the quantum dynamics 
71: appears essential in order that slow classical decay be quantally 
72: reproduced over sufficiently long time scales. In contrast, in a deep quantum 
73: regime, localization effects would become dominant, rapidly effacing any 
74: memory of the underlying classical dynamics.
75: 
76: In this Letter we provide evidence that  fractality of quantum fluctuations 
77: nevertheless survives even in such strongly  non-classical 
78: situations. In the model systems presently to be described, the classical 
79: phase space has no 
80: significant stable islands, the classical motion is diffusive, and  
81:  the classical decay is exponential; therefore, in the quantum weakly 
82: localized regime, one observes non-fractal, Lorentz-correlated 
83: conductance fluctuations\cite{borgo}. In contrast, we work in a 
84: strongly 
85: localized regime, and observe fractal fluctuations, apparently unrelated to 
86: any classical phase-space structure. Decisive in this respect is the fact 
87: that the {\it quantum} decay, albeit highly non-classical, is algebraic 
88:  due to purely quantum localization effects, as 
89: recently predicted\cite{dima,hatom}. As a matter of fact, we find that 
90:  the fractal dimension of quantum fluctuations is connected to the algebraic
91: decay exponent in exactly the same way predicted by the semiclassical theory
92: \cite{ketz,casa}.
93:  It therefore appears that the quantum 
94: fluctuation pattern bears the same relation to the {\it quantum} 
95: probability decay, which was originally predicted by semiclassical arguments,
96: even though such decay is not by any means quasi-classical.
97: 
98: We consider the kicked rotator model \cite{fishman} with absorbing boundary 
99: conditions. Classical dynamics is described by the  standard map:
100: \begin{equation}
101: \label{clrot}
102: \bar{I}  =  I + k \sin \theta, \quad 
103: \bar{\theta}  =  \theta + T \bar{I}, 
104: \end{equation} 
105: with absorption for $\bar{I} < 0 $ and for $\bar{I} > N$.
106: Classical dynamics only depends on the scaling parameter $K=kT$.
107: The corresponding quantum map is obtained by the substitution $ I \to
108: \hbar \hat{n} = -i \hbar \partial / \partial \theta$. 
109:  We denote $\psi_t$ the state vector immediately before 
110: the $t-$th kick, $T$ the kicking period, and we set $\hbar=1$.
111: Throughout this Letter, we will treat the kick counter
112: $t$ as a discrete time variable in units of $T$. 
113: The discrete-time evolution from 
114: time $t$ to time $t+1$ is described in the classical case by 
115: (\ref{clrot}) and in the quantum case by :
116: \begin{equation} 
117: \label{rot}
118: {\psi_{t+1}}=\hat{P}\tilde\psi_{t+1}=\hat{P}e^{-iT(\hat{n}+\phi)^2/2} 
119: e^{-ik\cos\hat{\theta}}\psi_t.  
120: \end{equation} 
121: The state   
122:    $\tilde\psi_{t+1}$ is obtained from $\psi_{t}$ by means of the unitary 
123: one-period propagator of the kicked rotor. The operator  
124:  $\hat{P}$ then projects  $\tilde\psi_{t+1}$  over states in a fixed 
125: interval of $n$ values ($0 \le n \le N$), so it describes complete
126: deletion (absorption) of the  
127: part of the wave packet which propagates outside the given interval. 
128:  The parameter $\phi$ can be interpreted as an  
129: Aharonov-Bohm flux
130: through the ring parametrized by the coordinate $\theta$.
131: 
132: We consider the case $K=kT=7$,  $k=5$, with 
133: absorption for $n< 0$ and $n>N=250$. With such parameter values, 
134: the classical phase space has no  
135: significant island of stability \cite{note}. We consider a statistical 
136: ensemble of orbits starting at $t=0$ with $I=0$ and randomly 
137: distributed phases $\theta$. The survival probability 
138: $P(t)$ is the fraction of the ensemble which hasn't been absorbed 
139: at time $t$; equivalently, $P(t)$ is the integrated distribution of 
140: exit times of orbits in the ensemble.
141: 
142: \begin{figure}
143: \centerline{\epsfxsize=7.2cm\epsffile{fig1.eps}}
144: \caption{Classical (dashed line) and quantum (thick solid line) survival 
145: probability for the kicked rotator model with $K=7$, $k=5$, $\phi=0$. 
146: The evolution starts at $n_0=0$ and the probability is absorbed for 
147: $n< 0$ and $n>250$. The thin solid line has slope one.  
148: Here and in the following figures the logarithms are decimal. 
149: } 
150: \label{fig1} 
151: \end{figure}
152:  
153: \vskip -0.5cm 
154: \begin{figure}
155: \centerline{\epsfxsize=7.5cm\epsffile{fig2.eps}}
156: \caption{Survival probability vs flux $\phi$ at fixed time 
157: $t=10^3$ (above) and $t=10^4$ (below), with parameter values as 
158: in Fig.\ref{fig1}. 
159: } 
160: \label{fig2} 
161: \end{figure}
162: 
163: The classical decay of $P(t)$  is well described 
164: by a diffusion process, practically insensitive to the chosen 
165: value of $\phi$. As 
166: shown in Fig.\ref{fig1}, after a diffusion time 
167: $t_D\approx N^2/D\propto N^2/k^2$ ($D$ diffusion rate), 
168: the survival probability exponentially decays  with time:
169: $P(t)\propto\exp(-\gamma t)$, with 
170: $\gamma\approx 1/t_D\approx 2.2\times 10^{-3}$. In contrast, quantum dynamics 
171: is strongly localized, as the localization length $\ell\approx k^2$  
172: is much less than the ``sample size'' $N$. Such strong localization 
173: causes a slow decay of the quantum survival probability $P_{\phi}(t)$ 
174:   The latter is  
175: defined as the total probability on states $0<n<250$ at time $t$, 
176: given that at $t=0$ the state was $n_0=0$. After the ``Heisenberg time'' 
177:  $t_H
178: $, it 
179: decays proportional to $1/t$ up to a time $t_{max}\sim\exp(2N/{\ell})$ 
180: \cite{dima}, as shown in 
181: Fig.\ref{fig1} (in this case, $t_{max}\sim 5\times 10^8$). 
182: 
183: Though the exponent $1$ of algebraic quantum decay is independent 
184: of the value of $\phi$, the quantum survival probability 
185: $P_{\phi}(t)$ at a fixed 
186: time $t$ is sensitively dependent on $\phi$. The fluctuation pattern 
187: exhibited by 
188: the graph of 
189: $P_{\phi}(t)$ versus $\phi$ at fixed $t$ (two examples are given 
190: in Fig.\ref{fig2})  
191: is precisely the object of our analysis, at values of $t$ up to 
192: $10^4$, and for $10^4$ values of $\phi$ in the interval 
193: $[0,0.1]$. 
194: First we have computed autocorrelation functions:
195: \begin{equation} 
196: \label{corr0}
197: C(\delta\phi)=<P_\phi(t)P_{\phi+\delta\phi}(t)>_{\phi}.  
198: \end{equation}  
199: Such correlations play a key role in the existing semiclassical theory of
200: fractal fluctuations, which  consists of three steps:
201: 
202: \noindent
203: (i)  correlation functions are related 
204: via Fourier transform 
205: to the classical decay of the canonical variable
206: conjugate to $\phi$\cite{smila,casa}, notably, to the 
207: function
208: $P_{cl}(\Theta)$ yielding the probability distribution of the angle $\Theta$ accumulated 
209: by the rotor at the time of exit from the finite sample (``accumulated'' 
210: means multiples of $2\pi$ included). In particular, if 
211:  $P_{cl}(\Theta)$ decays as $\Theta^{-\alpha}$ at large 
212: $\Theta$, then 
213: \begin{equation} 
214: \label{corr}
215: C(\delta\phi)\sim C(0)-\hbox{const}|\delta\phi|^\alpha 
216: \end{equation} 
217: at small $\delta\phi$.
218: 
219: \noindent
220: (ii) It is physically intuitive, and 
221: numerically confirmed \cite{casa}, that the decay exponent $\alpha$  
222: coincides with the decay exponent of the classical survival probability. 
223: 
224: \noindent
225: (iii) a signal-theoretic argument \cite{mandelbrot} 
226: allows to conclude from (i) and (ii)
227: that the 
228: fluctuation graph has a fractal dimension given by
229: \begin{equation} 
230: \label{frac}
231: D=2-\frac{\alpha}{2} \quad (\hbox{for }\alpha\leq 2),   
232: \end{equation} 
233: where $\alpha$ is the time decay exponent of the classical survival 
234: probability.
235: Step (iii) actually invokes certain statistical properties of 
236: fluctuations, which were assumed to be ensured by the chaotic nature 
237: of classical dynamics\cite{ketz1}.
238: 
239: In our case, semiclassical arguments are invalid,  
240: because of strong quantum localization. In addition, there is no 
241: classical decay exponent $\alpha$, because classical 
242: decay is exponential. So the above theory doesn't apply; still,   
243: numerical results to be described below demonstrate that (i), (ii)
244: and (iii) are 
245: still legitimate in our 
246: highly non semiclassical case, too, provided one replaces the classical 
247: distribution functions $P(t)$,$P_{cl}(\Theta)$ by the corresponding 
248: quantum distributions $P_{\phi}(t)$,$P_q(\Theta)$, and 
249: $\alpha$ by the exponent of the {\it quantum} algebraic decay. 
250: In order to compute $P_q(\Theta)$ we perform a lifting of our model
251: (\ref{rot}), that is, we translate the 
252:  model (\ref{rot}) of a kicked particle on a circle into the model of a
253: kicked particle on the line, described by the coordinate $\Theta$. To simulate
254: the corresponding dynamics we restrict $\Theta$ inside a large, yet finite 
255: box, with periodic boundary conditions.  
256: At every 
257: evolution step $t$ we compute the function $|\psi_t(\Theta)-
258: \tilde\psi_t(\Theta)|^2$, which yields the  probability loss occurring at 
259: time $t$ and position $\Theta$ due to absorption. 
260: Summing this quantity over $t$ gives the 
261: total probability lost at position $\Theta$, that is, the probability 
262: distribution of the  angle accumulated at the exit time. The corresponding 
263: integrated distribution is  shown 
264: in Fig.\ref{fig3} and displays a decay $\propto \Theta^{-1}$. 
265: The faster decay 
266: in the rightmost part of Fig.\ref{fig3} is due to the finite total time 
267: and to the finite $\Theta-$box used in the integration. Thus, at any given 
268: time $t$, the distribution of accumulated exit angles up to that time 
269: can be assumed to decay proportional to $\Theta^{-1}$ in a range 
270: $\Theta_{min}<\Theta<\Theta_{max,t}$, with $\Theta_{max,t}$ increasing with 
271: $t$ as long as $t<t_{max}$. 
272: One can estimate $\Theta_{min}$ as the angle accumulated until  
273: the time $t_H\approx\ell$. 
274:  This gives     
275: $\Theta_{min}\sim \int_0^{t_H T} \sqrt{(D/T) \tau} d \tau 
276: \sim (2/3)t_H^{2}T$. 
277: Assuming a maximal momentum 
278: $\sim\ell$ for the rotor started at $n_0=0$, 
279: a rough estimate for $\Theta_{max,t}$ is provided by $\ell tT$ 
280: ($\Theta_{min}\sim 2 \times 10^2$ and $\Theta_{max,t}\approx 
281: 2.1\times 10^4$ for the case of Fig.3).  
282: 
283: \begin{figure}
284: \centerline{\epsfxsize=7.2cm\epsffile{fig3.eps}}
285: \caption{Quantum integrated distribution of the accumulated angle $\Theta$ 
286: at the exit time, computed over a total time $t=600$, parameter 
287: values as in Fig.\ref{fig1}.  The straight line has slope one.  
288: } 
289: \label{fig3} 
290: \end{figure}
291: 
292: 
293: 
294: Numerically computed correlations (\ref{corr0}) shown in the inset of 
295: Fig.\ref{fig4}, 
296: at two different times, indicate 
297:  that relation (\ref{corr}) is still valid in our 
298: strongly localized case\cite{borgo1}. 
299: We have computed the fractal dimensions   
300: of the corresponding graphs of $P_{\phi}(t)$ versus $\phi$ shown 
301: in Fig.\ref{fig2} using the box-counting algorithm described in 
302: \cite{expe,casa}. The results are presented in  
303: Fig.\ref{fig4} and demonstrate that the graph 
304: has a fractal dimension 
305: $\approx 1.5$ over a significant range of $\delta\phi$ scales, 
306: as predicted by equation (\ref{frac}).
307: This range roughly lies in between scales 
308: $\delta\phi_{min,t}$,
309: $\delta\phi_{max}$. The latter scale is inverse to 
310: $\Theta_{min}$. The former scale can be estimated 
311: by  $\Theta_{max,t}^{-1}$ (hence it decreases as $t$ increases),
312: as long as it remains above a minimal $\delta\phi_0\sim 
313: 2\pi/(N^2T)$ scale. This scale is imposed 
314:  by the finiteness of the sample, and can be explained as follows. 
315: The evolution operator (\ref{rot}) has complex eigenvalues inside the unit 
316: circle. Varying $\phi$ causes these eigenvalues to move, and their maximal 
317: shift under a change $\delta\phi$ is estimated by $TN\delta\phi$. 
318:  Any time one moving eigenvalue  comes close to the unit circle, a local maximum  in the graph of $P_{\phi}(t)$ 
319: vs $\phi$ appears. The number of peaks produced by a single moving 
320: eigenvalue as $\phi$ changes from $0$ to $1$  is at most $\sim TN/(2\pi)$. 
321: As there are $N$ eigenvalues, the total number of peaks is $\sim 
322: TN^2/(2\pi)$, with an average $\phi-$ spacing $\sim \delta\phi_0=
323: (2\pi)/(N^2T)$. 
324: As the fractal structure of the graph is due to the superposition of many 
325: tiny peaks originated in this way, no fractality can  
326: be expected below the $\delta\phi_0$ scale. 
327: 
328: 
329: \begin{figure}
330: \centerline{\epsfxsize=7.5cm\epsffile{fig4.eps}}
331: \caption{Fractal analysis for the kicked rotator model, with 
332: parameter values as in Fig.\ref{fig1}. The graph of $P_{\phi}(t)$ 
333: vs $\phi$ was covered by boxes of side $\delta\phi$, and the largest 
334: excursion of $P_\phi(t)$ in each strip was recorded; summing over all 
335: strips and dividing the result by $\delta\phi$, we obtained $N(\delta\phi)$. 
336: Data are shown for $t=10^3$ (circles) 
337: and $t=10^4$ (diamonds). The straight lines correspond 
338: to the fractal dimension $D=1.5$ expected for $\alpha=1$.  
339: The inset shows the correlation function $C(\delta\phi)$ at the 
340: same times. The straight lines correspond to 
341: $1-C(\delta\phi)/C(0)\propto |\delta\phi|^\alpha$, with $\alpha=1$.  
342: } 
343: \label{fig4} 
344: \end{figure}
345: 
346: Further confirmation for the validity of the above illustrated scenario was 
347: obtained 
348: by analyzing the following variant of the basic model (\ref{rot}): 
349: \begin{equation}
350: \label{disc}  
351: \psi_{t+1}=\hat{P} e^{-iT(\hat{n}+\phi)^2/2} 
352: e^{-ik|\cos\hat{\theta}|}\psi_t.  
353: \end{equation} 
354: which differs from (\ref{rot}) because the kicking potential has now a
355: discontinuous derivative. Localization is in this case algebraic, 
356: as eigenfunctions decay like $1/n^2$. The argument used to predict 
357: the $t^{-1}$ decay of the survival probability in the case of 
358: exponential localization \cite{dima}
359: yields an
360: asymptotic decay $\propto t^{-3/4}$. The numerical analysis 
361: illustrated above for the case of exponential localization was replicated 
362: for this model, too, yielding the results shown  in Fig.\ref{fig5}.
363: Fractal analysis of the fluctuation graph yields good agreement 
364: over a broad range with the theoretical $D$  
365: value of $13/8$ predicted using equation (\ref{frac}) and 
366: the decay exponent $\alpha=3/4$.
367: 
368: \begin{figure}
369: \centerline{\epsfxsize=7.5cm\epsffile{fig5.eps}}
370: \caption{
371: Same as in Fig.\ref{fig1} (upper inset) and Fig.\ref{fig4} 
372: but for the discontinuous map (\ref{disc}). 
373: The straight lines correspond to a power law decay of the 
374: quantum survival probability with $\alpha=3/4$ (upper inset), 
375: a correlation function $C(\delta\phi)\propto |\delta\phi|^{\alpha}$ 
376: (lower inset) and a fractal dimension $D=2-\alpha/2=13/8$. 
377: The $x$-axis in the lower inset is the same as in the 
378: main figure. 
379: } 
380: \label{fig5} 
381: \end{figure}
382: 
383: In conclusion, we have provided numerical evidence for fractal fluctuations 
384: of the quantum survival probability in a classically chaotic system 
385: in the regime of strong localization, and in the absence of significant 
386: classical critical structures.  
387: Whereas the original theory \cite{ketz} predicted fractal {\it conductance} 
388: fluctuations, our present results are about fluctuations 
389: of a different quantity, namely  {\it survival probability}. 
390: As both quantities ultimately  reflect the fluctuations  of the 
391: scattering matrix, their fluctuations should be similar in nature. As a 
392: matter of fact, the very same semiclassical arguments used to predict  
393: conductance  fluctuations could be adapted to the survival 
394: probability, too \cite{casa}.  The main difference is that a scan of  $P_{\phi}(t)$ 
395: at finite however large $t$ cannot achieve the $\phi-$resolution 
396: exhibited by conductance at fixed (quasi-)energy. On account of 
397: the  effect of increasing $t$ reported earlier in this Letter, such  
398: coarsening is unlikely to  ``fractalize'' otherwise non-fractal 
399: fluctuation patterns. 
400: In any case,  
401: survival probability is a meaningful quantity in a broader class 
402: of problems than electronic transport in semiconductor structures. 
403:  The scenario we have analyzed in this Letter 
404: can also be exported to  other realistic problems, where 'conductance' 
405: is instead  a problematic concept. As an example we quote  
406: microwave ionization of Rydberg atoms \cite{hatom}. This may open a way for 
407: experimental observation of localization-induced fractal fluctuations in 
408: atomic physics. 
409: 
410: Our results show that quantum fluctuations may be fractal even in 
411: situations where the existing semiclassical theory does not apply.
412: This does not command a reinterpretation of existing experimental data, 
413: which were obtained in situations far from the strongly localized regime 
414: considered in this Letter. Nevertheless it signals that the 
415: current understanding of the deep quantum mechanisms 
416: responsible for fractal fluctuations is still far from complete.
417: 
418: 
419: 
420: 
421: 
422: Support from the Progetto Avanzato INFM ``Quantum transport 
423: and classical chaos'' is gratefully acknowledged. 
424: 
425: \begin{references}
426: \bibitem{jal} For a review see 
427: R.A. Jalabert, {\it The Semiclassical Tool in Mesoscopic 
428: Physics}, in {\it New Directions in Quantum Chaos},   
429: Proceedings of the International School of Physics 
430: ``Enrico Fermi'', Course CXLIII, edited by G.Casati, I.Guarneri, 
431: and U.Smilansky (IOS Press, Amsterdam, 2000).
432: \bibitem{smila} R.Bl\"umel and U.Smilansky, Phys. Rev. Lett. {\bf 60}, 
433: 477 (1988); Physica D {\bf 36}, 111 (1989). 
434: \bibitem{ketz} R. Ketzmerick, Phys. Rev. B {\bf 54}, 10841 (1996).
435: \bibitem{expe} H. Hegger, B. Huckestein, K. Hecker, M. Janssen, 
436: A. Freimuth, G. Reckziegel, and R. Tuzinski, Phys. Rev. Lett. {\bf 77}, 
437: 3885 (1996); A.S. Sachrajda, R. Ketzmerick, C. Gould, Y. Feng, P.J. Kelly, 
438: A. Delage, and Z. Wasilewski, Phys. Rev. Lett. {\bf 80}, 1948 (1998).
439: \bibitem{casa} G. Casati, I. Guarneri, and G. Maspero, Phys. Rev. Lett.
440: {\bf 84}, 63 (2000).
441: \bibitem{spain} Fractal conductance fluctuations were also numerically exposed 
442: in a two-dimensional tight-binding model of a quantum dot:
443: E. Louis and J.A. Verges, Phys. Rev. B {\bf 61}, 13014 (2000). 
444: The connection of that model to the semiclassical theory is unclear, though. 
445: \bibitem{borgo} F. Borgonovi, I. Guarneri, and L. Rebuzzini, Phys. Rev. Lett.
446: {\bf 72}, 1463 (1994).
447: \bibitem{dima} G. Casati, G. Maspero, and D.L. Shepelyansky, 
448: Phys. Rev. Lett. {\bf 82}, 524 (1999). 
449: \bibitem{hatom} G. Benenti, G. Casati, G. Maspero, and D.L. Shepelyansky, 
450: Phys. Rev. Lett. {\bf 84}, 4088 (2000).    
451: \bibitem{fishman} For a review see  
452: F.M. Izrailev, Phys. Rep. {\bf 196}, 299 (1991). 
453: \bibitem{note} Even though accelerator modes are present at $K=7$, 
454: they do not significantly affect the transport properties discussed in 
455: this Letter. 
456: \bibitem{mandelbrot} B.B. Mandelbrot, {\it The Fractal Geometry of 
457: Nature} (Freeman, San Francisco, 1982).  
458: \bibitem{ketz1} For further refinements of the theory, see 
459: L. Hufnagel, R. Ketzmerick, and M. Weiss, cond-mat/0009010. 
460: \bibitem{borgo1} A similar behaviour for energy-energy correlations  
461: of scattering matrices in a strongly localized regime was first 
462: observed by F. Borgonovi and I. Guarneri, 
463: Phys. Rev. E {\bf 48}, R2347 (1993). 
464: 
465: \end{references}
466: 
467: \end{multicols}
468: 
469: \end{document}
470: 
471: