cond-mat0105274/wpk.tex
1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2: %%% Kottos and Cohen (April 2001). 
3: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
4: 
5: % FIGURES:
6: % Fig.1: de_evol.eps
7: % Fig.2: wd_evol.eps
8: % Fig.3: pt_evol.eps
9: % Fig.4: str_profs.eps
10: 
11: \documentstyle[aps,prl,twocolumn,floats,epsfig]{revtex}
12: \textheight9.6in
13: \begin{document}
14: 
15: \newcommand{\tbox}[1]{\mbox{\tiny #1}}
16: \newcommand{\half}{\mbox{\small $\frac{1}{2}$}}
17: \newcommand{\sfrac}[1]{\mbox{\small $\frac{1}{#1}$}}
18: \newcommand{\mbf}[1]{{\mathbf #1}}
19: 
20: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
21: 
22: \title{Failure of random matrix theory 
23: to correctly describe quantum dynamics} 
24: 
25: \author
26: {Tsampikos Kottos$^1$ and Doron Cohen$^2$ \\
27: %
28: \footnotesize
29: %
30: $^1$
31: Max-Planck-Institut f\"ur Str\"omungsforschung,
32: 37073 G\"ottingen, Germany \\
33: %
34: $^2$
35: Department of Physics, Harvard University, Cambridge, MA 02138 \\
36: }
37: 
38: \date{April 2001}
39: 
40: \maketitle
41: 
42: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
43: 
44: \begin{abstract}
45: Consider a classically chaotic system which is described 
46: by a Hamiltonian ${\cal H}_0$. At $t=0$ the Hamiltonian 
47: undergoes a sudden-change ${\cal H}_0 \mapsto {\cal H}$. 
48: We consider the quantum-mechanical spreading of the evolving
49: energy distribution, and argue that it cannot be analyzed 
50: using a random-matrix theory (RMT) approach. RMT can be 
51: trusted only to the extend that it gives trivial results 
52: that are implied by first-order perturbation theory.  
53: Non-perturbative effects are sensitive to the underlying 
54: classical dynamics, and therefore the $\hbar\rightarrow 0$ 
55: behavior for effective RMT models is strikingly different 
56: from the correct semiclassical limit.    
57: \end{abstract}    
58:  
59: \ \\
60:   
61: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
62: 
63: Consider a system whose total Hamiltonian is ${\cal H}={\cal H}(Q,P;x)$, 
64: where $(Q,P)$ is a set of canonical coordinates, and $x$ is a 
65: constant parameter. We assume that the preparation and the representation 
66: of the system are determined by the Hamiltonian 
67: ${\cal H}_0={\cal H}(Q,P;x_0)$, and that both ${\cal H}_0$ and ${\cal H}$ 
68: generate classically chaotic dynamics of similar nature. 
69: Moreover, we assume that $\delta x\equiv (x{-}x_0)$ 
70: is {\em classically small}, meaning that it is possible to apply 
71: linear analysis in order to describe how the energy 
72: surfaces ${\cal H}(Q,P;x)=E$ are deformed as a result of changing 
73: the value of $x$. Physically, going from ${\cal H}_0$ to 
74: ${\cal H}$ may signify a change of an external field, or switching 
75: on a perturbation, or sudden-change of effective-interaction 
76: (as in molecular dynamics). Quantum mechanically, we can use a basis 
77: where ${\cal H}_0 = \mbf{E}_0$ has a diagonal representation, while 
78: %
79: \begin{eqnarray} \label{e1} 
80: {\cal H} \ \  = \ \ \mbf{E}_0 \ + \ \delta x \ \mbf{B}
81: \end{eqnarray}
82: %
83: For reasonably small $\hbar$, it follows from general semiclassical 
84: considerations \cite{mario}, that $\mbf{B}$ is a {\em banded matrix}.  
85: Generically, this matrix {\em looks random}, as if its off-diagonal 
86: elements were {\em independent} random numbers. 
87: 
88: It was the idea of Wigner \cite{wigner} forty years ago, 
89: to study a simplified model, where the Hamiltonian 
90: is given by Eq.~(\ref{e1}), and where $\mbf{B}$ 
91: is a Banded Random Matrix (BRM) \cite{felix}. 
92: This approach is attractive both analytically and numerically. 
93: Analytical calculations are greatly simplified by the 
94: assumption that the off-diagonal terms can be treated as 
95: independent random numbers. Also from numerical point 
96: of view it is quite a tough task to calculate the true 
97: matrix elements of the $\mbf{B}$ matrix. It requires 
98: a preliminary step where the chaotic ${\cal H}_0$ is 
99: diagonalized. Due to memory limitations one ends up with 
100: quite small matrices. We can think of Eq.~(\ref{e1}) as 
101: describing fictitious motion on a lattice. For the model 
102: below (Eq.~\ref{e2}) we were able to handle $N=5000$ sites 
103: maximum. This should be contrasted with BRM simulations, 
104: where using self-expanding algorithm \cite{slfex} we were able to 
105: handle $N=100000$ sites along with significantly reduced CPU time.   
106: 
107: 
108: However, the applicability of the RMT approach is a matter 
109: of conjecture. Obviously this conjecture should be tested. 
110: To be more specific, one should be aware that there is a hierarchy 
111: of challenges where the applicability of the RMT conjecture should 
112: be tested. Namely: The study of spectral statistics; 
113: The study of eigenstates; The study of quantum dynamics. 
114: While the issue of spectral statistics has become a major 
115: subject in "quantum chaos" studies \cite{smpl}, the two other 
116: issues are barely treated. In a previous study \cite{lds} we have 
117: demonstrated that the RMT approach is capable of giving the 
118: right qualitative picture of the parametric evolution of the 
119: eigenstates. As $\delta x$ is increased the eigenstates 
120: of Eq.~(\ref{e1}) change in a qualitative agreement with Wigner's theory. 
121: Still, RMT fails to capture non-universal system-specific features. 
122: 
123: 
124: In this Letter we turn to the study of quantum {\em dynamics}. 
125: Here we are going to end up with a much more alarming claim. 
126: Namely, the RMT approach fails to give the correct dynamical picture. 
127: RMT can be trusted only to the extend that it gives trivial results 
128: that are implied by first-order perturbation theory.  
129: Non-perturbative effects are sensitive to the underlying 
130: classical dynamics, and therefore the $\hbar\rightarrow 0$ 
131: behavior for effective RMT models is strikingly different 
132: from the correct semiclassical limit. 
133: In this Letter we are going to establish the failure of RMT for 
134: the case of dynamics which (for $t>0$) is generated by a time 
135: independent Hamiltonian. This we hope paves the way towards making 
136: an analogous statement regarding the response of driven systems \cite{rsp}.  
137: 
138: 
139: 
140: 
141: %% Fig1
142: %%%%%%%%%%%%%%%%%%%%%%%%%%%
143: \begin{figure}[t] 
144: \centerline{\epsfig{figure=de_evol.eps,width=\hsize}}
145: \vspace{.1in}
146: \caption{
147: (a) The classical energy spreading as a function of time. 
148: (b) The QM spreading for the 2DW Hamiltonian; 
149: (c) The QM spreading for the EBRM Hamiltonian. 
150: The energy in these simulations is $E\sim 3$, 
151: and $\delta x = 0.2123$.  
152: In (a) we see a crossover from ballistic spreading 
153: ($\delta E  \propto t$) to saturation ($\delta E \sim \mbox{const}$).
154: Only one time scale ($\tau_{\tbox{cl}} \sim 1$) is involved. 
155: The light dashed line has slope $1$ and is drawn to guide the eye. 
156: In (b) we see that the classical behavior is approached 
157: as $\hbar\rightarrow 0$. In (c) we see the opposite 
158: trend: as $\hbar\rightarrow 0$ an intermediate stage 
159: of diffusion ($\delta E  \propto \sqrt{t}$) develops.  
160: (Here the light dashed line has slope $1/2$).
161: Different lines correspond to different values 
162: of $\hbar$ as in (b), and additional curves 
163: ($\hbar=0.009,0.005$) have been added.
164: }
165: \end{figure}
166: %%%%%%%%%%%%%%%%%%%%%%%%%%%%
167: 
168: 
169: 
170: 
171: %% Fig2
172: %%%%%%%%%%%%%%%%%%%%%%%%%%%
173: \begin{figure}[t] 
174: \centerline{\epsfig{figure=wd_evol.eps,width=\hsize}}
175: \vspace{.1in}
176: \caption{
177: The width $\Delta \times N(t)$ as a function of time.  
178: Different lines correspond to different $\hbar$-values 
179: as in Fig.~1b. The heavy dashed line is the 
180: classical $\delta E(t)$. Having separation of scales 
181: ($\Delta {\times} N(t) \ll \delta E(t)$) is an 
182: indication for having a {\em perturbative} spreading profile.  
183: The upper panel is for the 2DW Hamiltonian 
184: while the lower panel is for the EBRM model. 
185: }
186: \end{figure}
187: %%%%%%%%%%%%%%%%%%%%%%%%%%%%
188: 
189: 
190: 
191: In order to test the RMT conjecture, 
192: we are going to use the same `direct' approach and the same model  
193: as in Ref.\cite{lds}. For the first time we are going to compare 
194: the {\em dynamics} which is generated by a `physical' Hamiltonian, 
195: with the corresponding dynamics that is obtained from an 
196: effective BRM model (EBRM). 
197: The latter is constructed by taking the matrix $\mbf{B}$ of 
198: the `physical' Hamiltonian, and then randomizing the signs of its 
199: off-diagonal elements. Such operation destroys any 
200: correlations between the matrix elements of $\mbf{B}$,  
201: while keeping the band profile unaffected \cite{ewbrm}.
202: %
203: %
204: %
205: %
206: We study the Hamiltonian \cite{lds}
207: %
208: \begin{eqnarray} \label{e2} 
209: {\cal H}(Q,P;x) = \half(P_1^2{+}P_2^2 + Q_1^2{+}Q_2^2)  
210: + x \cdot Q_1^2 Q_2^2
211: \end{eqnarray}
212: %
213: with $x=x_0+\delta x$ and $x_0=1$. This Hamiltonian describes the 
214: motion of a particle in a 2D well (2DW). The units are chosen 
215: such that the mass is equal to one, the frequency for small 
216: oscillations is one, and for $\delta x=0$ the coefficient of the 
217: anharmonic term is also one. The energy $E$ is the only 
218: dimensionless parameter of the classical motion. Our numerical 
219: study is focused on an energy window around $E \sim 3$ where 
220: the motion is mainly chaotic. Upon quantization we have 
221: a second dimensionless parameter, which is the scaled $\hbar$.
222: Associated with $\hbar$ are two energy scales. 
223: One is the mean level spacing $\Delta \propto \hbar^d$ 
224: with $d=2$, and the other is the bandwidth $\Delta_b \propto \hbar$. 
225: The second scale is further discussed below.   
226: 
227: 
228: It is useful to define a fluctuating quantity ${\cal F}(t) \equiv 
229: -(\partial {\cal H}/\partial x)$ which for the 2DW model equals 
230: ${\cal F}(t) = -Q_1^2(t) Q_2^2(t)$. The auto-correlation function 
231: of ${\cal F}(t)$ is denoted by $C(\tau)$. The associated correlation 
232: time is denoted by $\tau_{\tbox{cl}}$. The power spectrum of the 
233: fluctuations $\tilde{C}(\omega)$ is the Fourier transform of 
234: $C(\tau)$. The band profile of the matrix $\mbf{B}$ satisfies the 
235: semiclassical relation $|\mbf{B}_{nm}|^2 \approx (\Delta /(2\pi\hbar)) 
236: \tilde{C}((E_n{-}E_m)/\hbar)$.
237: See Fig.2 of \cite{lds} for numerical demonstration. 
238: It is implied by this relation that the bandwidth  
239: is $\Delta_b = 2\pi\hbar/\tau_{\tbox{cl}}$. 
240: For the quantum mechanical simulation the exact matrix 
241: $\mbf{B}$ has been calculated numerically. Memory constraints 
242: limit the maximum size ($N$) of the matrix that we can get. 
243: The EBRM Hamiltonian is obtained by randomizing the 
244: signs of the off-diagonal elements. A second, more `loose'  
245: strategy, is to generate the EBRM $\mbf{B}$ from scratch 
246: using the semiclassical band-profile as an input. 
247: The advantage of the latter strategy is that it opens 
248: the way to EBRM-model simulations with smaller $\hbar$, 
249: where the required $N$ is much larger.  We have verified 
250: that the latter strategy gives numerical results that 
251: agree with the sign-randomization approach. 
252: 
253: 
254: 
255:  
256: The initial preparation is assumed to be microcanonical. 
257: This means, in the classical case, an ergodic distribution 
258: of initial `points' on the energy surface ${\cal H}_0(Q,P) = E$ 
259: with $E \sim 3$.
260: In the quantum mechanical case we start each simulation 
261: with an initial eigenstate $m$ that has an energy $E_m \sim E$ 
262: where $2.75 < E < 3.2$. The time dependent evolution is 
263: determined by Schrodinger Equation. The probability distribution 
264: after time $t$ is $P_t(n|m)$. An average over 
265: initial state ($m$) is taken in order to get the average 
266: profile $P_t(n-m)$.  
267: We characterize the evolving distribution using three
268: different measures. The variance is $M(t)=\sum_r r^2 P_t(r)$, 
269: or in energy units it is $\delta E(t) = \Delta \times \sqrt{M(t)}$.   
270: The width $N(t)$ is defined as the $r$~region that 
271: contains $50\%$ of the probability. In case that we have 
272: a spreading profile that is characterized by a single 
273: energy scale, it is implied that $N(t)$ and $\sqrt{M(t)}$ 
274: would be the same (up to a numerical factor). 
275: The survival probability is $P(t)=P_t(r{=}0)$. 
276: The results of the simulations are presented in the Figs~{1-4}. 
277: The analysis of these results is discussed below. 
278: 
279: 
280:  
281: 
282: %% Fig3
283: %%%%%%%%%%%%%%%%%%%%%%%%%%%
284: \begin{figure}
285: \centerline{\epsfig{figure=pt_evol.eps,width=\hsize}}
286: \vspace{.1in}
287: \caption{
288: The width $N(t)$ and the survival probability $P(t)$ 
289: for the 2DW model.  Different lines correspond to 
290: different $\hbar$-values as in Fig.~1b. 
291: Having $N(t)=1$ or $P(t)\sim 1$ is the indication 
292: for having a {\em standard} perturbative spreading profile. 
293: }
294: \end{figure}
295: %%%%%%%%%%%%%%%%%%%%%%%%%%%%
296: 
297: 
298: 
299: Taking ${\cal H}$ to be a generator for the classical 
300: dynamics, the energy $E(t) = {\cal H}_0(Q(t),P(t))$ fluctuates. 
301: The fluctuations are characterized by the correlation time
302: $\tau_{\tbox{cl}}$, and by an amplitude $\delta E_{\tbox{cl}}$.
303: The initial preparation is assumed to be a microcanonical distribution 
304: that is supported by the energy-surface ${\cal H}_0(Q,P)=E(0)$. 
305: For $t>0$, the phase-space distribution 
306: spreads away from the initial surface. `Points' of the evolving
307: distribution move upon the energy-surfaces of ${\cal H}(Q,P)$.
308: We are interested in the distribution of the energy $E(t)$
309: of the evolving `points'. 
310: It is easily argued that for short times 
311: this distribution evolves in a ballistic fashion. 
312: Then, for $t \gg \tau_{\tbox{cl}}$, 
313: due to ergodicity, a `steady-state distribution' appears, 
314: where the evolving `points' occupies an `energy shell' in phase-space. 
315: The thickness of this energy shell equals $\delta E_{\tbox{cl}}$.
316: Thus we have a crossover from ballistic energy spreading 
317: to saturation. The dynamics in the classical limit 
318: is fully characterized by the two classical 
319: parameters $\tau_{\tbox{cl}}$ and $\delta E_{\tbox{cl}}$.
320: 
321: 
322: A quantitative description of the classical spreading 
323: is easily obtained. A straightforward derivation leads to 
324: the following result for the spreading 
325: %
326: \begin{eqnarray} \label{e3} 
327: \delta E_{\tbox{cl}}(t) = \delta x \times \sqrt{2(C(0)-C(t))}
328: \end{eqnarray} 
329: % 
330: As a particular result we get $\delta E_{\tbox{cl}} \equiv \delta E_{\tbox{cl}}(\infty)= 
331: \delta x \sqrt{2C(0)}$. The calculation of $\delta E_{\tbox{cl}}(t)$ for the model 
332: Hamiltonian is presented in Fig.~1a. 
333: %
334: It is implied by Eq.(\ref{e3}) that the spreading $\delta E(t)$, 
335: from semiclassical point of view, 
336: is just a property of the band-profile. Thus, one may get  
337: to the wrong conclusion that models with the same 
338: band-profile should lead to the same $\delta E(t)$, 
339: provided the off-diagonal elements look random.  
340: If this were the case, it would be implied that 
341: the EBRM model would be equivalent to the 2DW model as far as 
342: the spreading $\delta E(t)$ is concerned. Looking on Fig.1 
343: we see that this is not the case. As $\hbar\rightarrow 0$ 
344: the EBRM model further and further deviates from the 
345: (correct) semiclassical expectation. 
346: 
347: 
348: In order to understand the observed results we would 
349: like to recall some of the theory of \cite{crs,lds}. 
350: We already said that upon quantization we have 
351: the two energy scales $\Delta \propto \hbar^d$ and 
352: and $\Delta_b \propto  \hbar$. Actually there is also 
353: a semiclassical energy scale $\Delta_{\tbox{SC}} \propto \hbar^{2/3}$. 
354: Associated with these energy scales are three parametric 
355: scales $\delta x_c^{\tbox{qm}} \ll \delta x_{\tbox{prt}} \ll \delta x_{\tbox{SC}}$, 
356: where the strong inequalities hold in the $\hbar\rightarrow 0$ limit. 
357: For the 2DW model, assuming $E\sim 3$ we have \cite{lds}
358: the estimates $\delta x_c^{\tbox{qm}} \approx 3.8*\hbar^{3/2}$ 
359: and $\delta x_{\tbox{prt}} \approx 5.3*\hbar$ 
360: and $\delta x_{\tbox{SC}} \approx 4*\hbar^{2/3}$. 
361: In the standard perturbative regime ($\delta x < \delta x_c^{\tbox{qm}}$)  
362: the eigenstates of Eq.(\ref{e1}) have a simple perturbative 
363: structure. In the extended perturbative regime 
364: ($\delta x_c^{\tbox{qm}}  < \delta x < \delta x_{\tbox{prt}}$) 
365: the eigenstates of Eq.(\ref{e1}) have a core-tail 
366: structure that can be regarded as a generalization of 
367: Wigner's Lorentzian. In the non-perturbative regime 
368: ($\delta x > \delta x_{\tbox{prt}}$) the eigenstates have a 
369: purely non-perturbative structure. Depending on whether 
370: it is the EBRM Hamiltonian or the 2DW Hamiltonian, this 
371: `ergodic' non-perturbative structure is either semicircle-like 
372: or semiclassical-like respectively. The semiclassical 
373: regime ($\delta x > \delta x_{\tbox{SC}}$) is contained in the 
374: non-perturbative regime. It is only there that we 
375: can trust detailed quantal-classical correspondence (QCC).   
376: 
377: 
378: 
379: For the purpose of the present analysis it is convenient to specify 
380: the different regimes by regarding $\hbar$ as a free parameter. 
381: Thus, the standard perturbative regime is \mbox{$\hbar > C_{\tbox{cqm}}$}, 
382: the extended perturbative regime is \mbox{$C_{\tbox{prt}} < \hbar < C_{\tbox{cqm}}$}, 
383: and the non-perturbative regime is \mbox{$\hbar < C_{\tbox{prt}}$}. The latter 
384: contains the semiclassical regime \mbox{$\hbar < C_{\tbox{SC}}$}. Thus the 
385: semiclassical limit $\hbar\rightarrow 0$ is a non-perturbative limit.
386: In case of the 2DW model, the classical quantities 
387: are \mbox{$C_{\tbox{cqm}}=0.41*dx^{2/3}$} 
388: and \mbox{$C_{\tbox{prt}}=0.19*dx$} 
389: and \mbox{$C_{\tbox{SC}}=0.12*dx^{3/2}$}.
390: %
391: We have used in most of our numerical simulations 
392: $\delta x \sim 0.2$. Larger $\delta x$ may take us 
393: out of the classical linear regime. For this 
394: value of $\delta x$ we get $C_{\tbox{cqm}}=0.14$
395: and $C_{\tbox{prt}}=0.04$ and $C_{\tbox{SC}}=0.01$. 
396: The smallest $\hbar$ value that we could allow 
397: without having memory-overflow was $\hbar=0.015$.  
398: This means that we were able to access the 
399: non-perturbative regime, though the semiclassical 
400: regime was out of reach.  
401: 
402: 
403: As explained in \cite{wbr} the essential features 
404: of the spreading behavior in the perturbative regimes 
405: can be analyzed using first order perturbation theory (FOPT). 
406: Since correlations between off-diagonal elements are 
407: not important for FOPT, it follows that the EBRM-model and 
408: the 2DW model should be trivially equivalent in such case. 
409: It is only in the non-perturbative regime where 
410: the question of their equivalence becomes non-trivial. 
411: In case of the standard BRM we have witnessed \cite{wbr}
412: in the non-perturbative regime a premature departure 
413: from ballistic behavior, and appearance of an intermediate 
414: diffusive stage. We observe essentially the same 
415: behavior in case of the EBRM Hamiltonian (Fig.~1c).   
416: But with the 2DW Hamiltonian (Fig.~1b) we do not have 
417: such an effect:  
418: As $\hbar \rightarrow 0$ the correspondence 
419: with the classical behavior becomes better and better. 
420: Thus our simulations demonstrate that having 
421: diffusion in the non-perturbative regime is an 
422: artifact of the RMT approach. 
423: %
424: In our previous work \cite{wbr} we did not have 
425: a numerical proof to support such a strong statement. 
426: There, all we were able to do, was to argue that RMT 
427: should fail in the deep semiclassical regime
428: ($\hbar \ll C_{\tbox{SC}}$), 
429: thus leaving open the possibility for having an intermediate 
430: regime ($C_{\tbox{SC}} < \hbar < C_{\tbox{prt}}$) where RMT might 
431: be valid. As we see, our numerical results do not give 
432: any indication for the existence of such intermediate regime.
433: The failure of RMT happens {\em as soon as} we enter 
434: the non-perturbative regime ($\hbar < C_{\tbox{prt}}$). 
435: 
436: 
437: 
438: %% Fig4
439: %%%%%%%%%%%%%%%%%%%%%%%%%%%
440: \begin{figure}
441: \centerline{\epsfig{figure=str_profs.eps,width=\hsize}}
442: \vspace{.1in}
443: \caption{
444: The spreading profile $P_{\infty}(r)$   
445: in representative cases. 
446: The upper panel is an example for 
447: a standard perturbative profile 
448: ($P(t)\sim 1$). The middle panel 
449: is an example for a perturbative 
450: core-tail structure 
451: ($\Delta {\times} N  \ll \delta E$). 
452: The lower panel is an example for 
453: an ergodic-like non-perturbative structure  
454: ($\Delta {\times} N  \sim \delta E$).
455: }
456: \end{figure}
457: %%%%%%%%%%%%%%%%%%%%%%%%%%%%
458: 
459: 
460: As we make $\hbar$ smaller there is no indication in 
461: Fig.~1 for entering a non-perturbative regime. 
462: In \cite{crs,lds,wbr} we have made the important 
463: distinction between {\em detailed} QCC and {\em restricted} QCC. 
464: The former pertains to the whole spreading profile, 
465: while the latter pertains only to the variance. 
466: Restricted QCC is a robust type of correspondence that 
467: does not require a `very small $\hbar$'. If we want 
468: to have an indication for the crossover to a non-perturbative 
469: behavior we should look on other measures, such as 
470: $N(t)$ of Fig.~2.
471: If the spreading were classical-like, it would imply 
472: that the spreading profile is characterized by a single 
473: energy scale. In such case we would expect 
474: that $N(t)$ and $\sqrt{M(t)}$ would be the same 
475: (up to a numerical factor). 
476: Indeed this is the case for the $\hbar<0.04$ runs of Fig.2. 
477: However, this is definitely not the case in the 
478: perturbative regimes, where we have a separation of energy 
479: scales $N(t) \ll \sqrt{M(t)}$. 
480: In the perturbative regimes $M(t)$ is determined 
481: by the tails, and it is insensitive to the 
482: size of the `core' region.  The width $N(t)$ 
483: constitutes a practical estimate for the latter (see Fig.3a).   
484: It is $N(t)=1$ for a {\em standard} perturbative profile, 
485: and  $1 \ll N(t) \ll \sqrt{M(t)}$  for a fully developed 
486: core-tail structure. An alternate way to identify 
487: a standard perturbative profile is via the survival 
488: probability $P(t)$. Indeed for $\hbar<0.14$ we 
489: see in Fig.3b that we have $P(t)\sim 1$.    
490: 
491: 
492: 
493: The difference between the perturbative and the 
494: non-perturbative spreading profiles is further 
495: illustrated in Fig.4. Here we have plotted representative 
496: average saturation profiles, along with a comparison with 
497: the perturbative core-tail calculation (PRT for brevity). 
498: The saturation profile is given by the expression
499: % 
500: $P_{\tbox{$\infty$}}(n|m)=
501: \sum_{n'} |\langle n(x_0)|n'(x)\rangle|^2 |\langle n'(x)|m(x_0)\rangle|^2$.  
502: %
503: It can be regarded as the auto-convolution of 
504: $P_{\tbox{E}}(n|m)= |\langle n(x)|m(x_0)\rangle|^2$. 
505: Thus the {\em average} saturation profile $P_{\tbox{$\infty$}}(r)$ 
506: is approximately related to the {\em average}  
507: local density of states $P_{\tbox{E}}(r)$.   
508: The latter has been analyzed in \cite{lds}.  
509: In Fig.3 we have calculated the PRT of $P_{\tbox{$\infty$}}(r)$ 
510: via an auto-convolution of the PRT of $P_{\tbox{E}}(r)$. 
511: In the extended perturbative regime the major 
512: features of the saturation profile are captures 
513: by the PRT. The differences are mainly in the far tails 
514: where higher order perturbation theory is essential. 
515: There are also differences in the small scale details,  
516: where the non-perturbative mixing is important. 
517: In the non-perturbative regime the saturation 
518: profile becomes purely non-perturbative, and 
519: the PRT becomes useless. This is because there 
520: is no longer separation of energy scales, which is 
521: the working assumption of the core-tail theory.  
522:  
523: 
524: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
525: 
526: We thank Felix Izrailev for suggesting to study the 2DW model. 
527: 
528: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
529: 
530: \begin{thebibliography}{99}
531: 
532: \vspace*{-1cm}
533: 
534: \bibitem{mario} 
535: M. Feingold and A. Peres, Phys. Rev. A {\bf 34} 591, (1986).
536: M. Feingold, D. Leitner, M. Wilkinson, Phys. Rev. Lett. {\bf 66}, 986 (1991). 
537: 
538: 
539: \bibitem{wigner} 
540: E. Wigner, Ann. Math {\bf 62} 548 (1955); {\bf 65} 203 (1957);
541: Y.V. Fyodorov, O.A. Chubykalo, F.M. Izrailev and G. Casati, 
542: Phys. Rev. Lett. {\bf 76}, 1603 (1996).
543: 
544: 
545: \bibitem{felix} 
546: G. Casati, B.V. Chirikov, I. Guarneri, F.M. Izrailev, 
547: Phys. Rev. E {\bf 48}, R1613 (1993); \ Phys. Lett. A {\bf 223}, 430 (1996).  
548: 
549: 
550: \bibitem{slfex} 
551: F. M. Izrailev, T. Kottos, A. Politi and G. P. Tsironis, Phys.~Rev.~E., 
552: {\bf 55}, 4951 (1997).
553: 
554: 
555: \bibitem{smpl} In most of the RMT literature (including the later 
556: works by Wigner himself), it is assumed that for the purpose 
557: of `quantum chaos' studies one can consider full (rather than banded) 
558: matrices, and the first term ${\bf E}_0$ is generally neglected. 
559: In spite of these enormous simplifications, it turns out that the 
560: so-called Gaussian invariant ensembles (GOE,GUE) provide a valid 
561: description of some major spectral properties. 
562: 
563: 
564: \bibitem{rsp} 
565: D. Cohen and T. Kottos, 
566: Phys. Rev. Lett. 85, 4839 (2000). 
567: 
568: 
569: \bibitem{lds} 
570: D. Cohen and T. Kottos, 
571: Phys. Rev. E, {\bf 63} 36203, (2001).
572: 
573: 
574: \bibitem{ewbrm} Note that the standard BRM model, 
575: unlike our EBRM model, involves an additional simplification. 
576: Namely, in case of the former model one further assumes 
577: that $\mbf{B}$ has a rectangular band profile. 
578: 
579: 
580: \bibitem{crs} 
581: D. Cohen, Phys. Rev. Lett. {\bf 82}, 4951 (1999); 
582: Ann. Phys. {\bf 283}, 175-231 (2000).
583: D. Cohen and E.J. Heller, Phys. Rev. Lett. {\bf 84}, 2841 (2000).
584: D. Cohen, A. Barnett and E.J. Heller, Phys. Rev. E {\bf 63}, 46207 (2001).
585: D. Cohen, {\em in} Proceedings of the International School
586: of Physics `Enrico Fermi' Course CXLIII 
587: ``New Directions in Quantum Chaos'', 
588: Edited by G. Casati, I. Guarneri and U. Smilansky, 
589: IOS Press, Amsterdam, 2000.  
590:  
591:  
592: \bibitem{wbr} 
593: D. Cohen, F.M. Izrailev and T. Kottos,  
594: Phys. Rev. Lett. {\bf 84} 2052 (2000). 
595: 
596: 
597: \end{thebibliography}
598: 
599: 
600: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
601: \end{document}
602: 
603: