1: \documentclass[aps,showpacs,psfig,floatfix]{revtex4}
2: \usepackage{graphicx}
3: \usepackage{epsfig}
4: \tolerance 10000
5:
6: \begin{document}
7:
8: \title{Brownian Motion Model of Quantization Ambiguity and Universality in
9: Chaotic Systems}
10: \author{L. Kaplan}
11: \affiliation{Department of Physics, Tulane University, New Orleans, LA 70118
12: USA}
13:
14: \begin{abstract}
15: We examine spectral equilibration of quantum chaotic spectra to universal
16: statistics, in the context of the Brownian motion model. Two competing time
17: scales, proportional and inversely proportional to the classical relaxation
18: time, jointly govern the equilibration process. Multiplicity of quantum
19: systems having the same semiclassical limit is not sufficient to obtain
20: equilibration of any spectral modes in two-dimensional systems, while in
21: three-dimensional systems equilibration for some spectral modes is possible if
22: the classical relaxation rate is slow. Connections are made with upper bounds
23: on semiclassical accuracy and with fidelity decay in the presence of a weak
24: perturbation.
25: \end{abstract}
26:
27: \pacs{05.45.Mt, 03.65.Sq}
28:
29: \maketitle
30:
31: \section{Introduction}
32: \label{secintro}
33:
34: Random Matrix Theory (RMT) was introduced by Wigner in the 1950s as a model
35: for describing the universal spectral behavior of complex many-body systems
36: such as compound nuclei. It was conjectured that a typical spectrum of
37: a complex system displays statistical properties similar to those of
38: a Hamiltonian chosen at random from a basis-independent ensemble of random
39: matrices in the same symmetry class. Subsequently, Dyson showed that
40: the statistical predictions of RMT could be reproduced within a Brownian
41: motion model, where Brownian motion of the energy levels results from a
42: stochastic perturbation of some initial, possibly nonrandom,
43: Hamiltonian~\cite{dyson}. Bohigas, Giannoni, and Schmit argued that the same
44: universal behavior of the spectrum occurs generically even in single-particle
45: systems, as long as the underlying classical dynamics is chaotic~\cite{bgs}.
46:
47: Wilkinson used the Brownian motion model to explain the connection between
48: quantum chaotic systems and RMT spectra, using the ``quantization ambiguity"
49: (i.e, multiplicity of quantum systems having the same semiclassical limit) as
50: the source of perturbation causing Brownian motion of the energy levels and
51: eventual equilibration to RMT spectral statistics~\cite{wilk}. More recently,
52: the Brownian motion approach has been extended to study the evolution of
53: eigenstates~\cite{wilkwalk} and to relate the smooth and oscillatory parts of
54: the spectral correlation function in diffusive and ballistic chaotic
55: systems~\cite{mehlwilk}.
56:
57: It well known that not all spectral modes of a typical quantum chaotic
58: spectrum obey universal statistics. It is therefore of interest in the context
59: of the Brownian motion model to investigate which modes are equilibrated in a
60: given dynamical system, for a given perturbation strength (such as that arising
61: from the quantization ambiguity), as a function of degree of chaos in the
62: system, and as a function of effective $\hbar$. Alternatively, one may ask
63: about the size of perturbation necessary to achieve equilibration to universal
64: statistics for a given spectral mode; this question is related to work by
65: Zirnbauer on energy level correlations in a quantum ensemble of classically
66: identical maps~\cite{zirnbauer}, as well as to the method developed by Gornyi
67: and Mirlin for studying wave function correlations in ballistic systems by
68: adding weak smooth disorder~\cite{gornyi}.
69:
70: In the present work, we show that the Brownian motion model can be used
71: successfully to predict the number of equilibrated spectral modes as a function
72: of perturbation strength, chaoticity of the classical system dynamics, system
73: dimension, and effective $\hbar$. The equilibration process is governed by two
74: competing time scales, allowing the number of modes equilibrated to RMT either
75: to increase or to decrease as the original system becomes more chaotic;
76: equilibration is reduced for systems with very short or very long classical
77: relaxation times and is maximized for systems with intermediate Lyapunov
78: exponent. We show that in two-dimensional quantum systems no equilibration to
79: universal behavior for any spectral modes can arise from the quantization
80: ambiguity in the $\hbar \to 0$ limit; in three dimensions some modes may be
81: equilibrated depending on the classical relaxation time of the underlying
82: classical dynamics; in four dimensions and higher, equilibration to
83: universality will occur for at least some modes, but maximal equilibration
84: requires the classical relaxation time to be much longer than the typical
85: ballistic time scale of the dynamics.
86:
87: The paper is organized as follows. In Sec.~\ref{secvar}, we review
88: semiclassical estimates for matrix element variance, which are required as
89: input for the Brownian motion model. In Sec.~\ref{secmaps}, we discuss several
90: example systems used for comparison between theory and numerics, and in
91: Sec.~\ref{secbrownian} we review the Brownian motion model itself. Two
92: competing conditions for equilibration of spectral modes are obtained in
93: Sec.~\ref{secequil} and applied to equilibration of the nearest level spacing
94: distribution in Sec.~\ref{secspac}. The implications for the effect of
95: quantization ambiguity on spectral universality are developed in
96: Sec.~\ref{secambig}. Finally, in Sec.~\ref{secsemi} we demonstrate the
97: relationship between results obtained in the Brownian motion model and recent
98: findings on semiclassical accuracy~\cite{sckaplan} and decay of quantum
99: fidelity~\cite{prosen}.
100:
101: \section{Semiclassical Calculation of Matrix Element Variance}
102: \label{secvar}
103:
104: This discussion follows~\cite{wilk,wilkwalk,eckh}. Consider the matrix
105: elements $B_{nn}=\langle \Psi_n |\hat B|\Psi_n \rangle$ of an operator $\hat
106: B$, given the eigenstates $|\Psi_n \rangle$ of Hamiltonian $\hat H$. It is
107: assumed that the operator $\hat B$ is ``independent" of $\hat H$ and has a
108: well-defined classical limit $B(q,p)$. We also assume that the phase-space
109: average of the corresponding classical observable vanishes, so that the matrix
110: elements fluctuate around zero, $\overline{B_{nn}} =0$. The variance of the
111: matrix elements at energy $E$ is
112: \begin{equation}
113: \sigma^2_B(E) = \overline{B_{nn}^2} = \nu(E)^{-1} \overline{\sum_n B_{nn}^2
114: \delta_\eta (E-E_n)} \,,
115: \end{equation}
116: where $\delta_\eta(E)=(2 \pi \eta^2)^{-1/2} e^{-E^2/2\eta^2}$,
117: $\nu(E)=\overline{\sum_n \delta_\eta(E-E_n)}$ is the density of states, and
118: $\overline{\cdots}$ denotes averaging over a ``suitable" ensemble. One may
119: approximate
120: \begin{equation}
121: \sigma^2_B(E) \approx {2 \sqrt{\pi} \,\epsilon \over \nu(E)}
122: \overline{\left [ \sum_n B_{nn} \delta_\epsilon(E-E_n) \right ]^2} \,,
123: \end{equation}
124: where $\epsilon = \sqrt{2}\, \eta$ is of the order of but smaller than the mean
125: level spacing $\Delta=1/\nu(E)$. In the semiclassical limit, the expression in
126: square brackets is approximated by a trace formula
127: \begin{equation}
128: \left [ \cdots \right ] \approx {1 \over 2 \pi\hbar}
129: {\rm Im} \sum_{p,r} T_{p} \,
130: B_p \, D_{p,r} \, e^{ i S_{p,r}/\hbar - i \mu_{p,r} \pi/2} \,
131: e^{-\epsilon^2 r^2 T_p^2 / 2 \hbar^2} \,.
132: \end{equation}
133: Here the sum is over all primitive periodic orbits $p$ at energy $E$ and their
134: repetitions $r$, $T_p$ is the primitive orbit period, $B_p$ is the classical
135: average of the observable $B$ over the orbit, $D_{p,r}= |{\rm
136: det}(M_{p,r}-1)|^{-1/2}$ is the square root of a classical focusing factor,
137: $M_{p,r}$ is the monodromy matrix of the orbit, $S_{p,r}$ is the action, and
138: $\mu_{p,r}$ is the Maslov index. Using $\overline{({\rm Im} f)^2} = {1 \over
139: 2} {\rm Re} \overline{f^\ast f}$ and neglecting repetitions, one obtains the
140: following estimate for the variance,
141: \begin{equation}
142: \sigma^2_B(E) \approx {2 \over \beta} {2 \sqrt{\pi} \, \epsilon \over \nu(E)}
143: (2 \pi \hbar)^{-2} \sum_p T_p^2 |B_p|^2 D_p^2 \, e^{-\epsilon^2 T_p^2 /
144: \hbar^2} \,,
145: \end{equation}
146: where as usual $\beta=1$, $2$ characterizes the presence or absence of time
147: reversal symmetry, respectively. Now, assuming $B_p$ to be real, and averaging
148: over many periodic orbits of period close to $T_p$,
149: \begin{equation}
150: B_p^2 = T_p^{-2} \int_0^{T_p} dt_1 \int_0^{T_p} dt_2 \,
151: B(t_1) B(t_2) \approx {1 \over T_p} \int_{-T_p/2}^{T_p/2} dt \,
152: \langle B(0) B(t) \rangle\,,
153: \end{equation}
154: where $\langle B(0) B(t) \rangle$ is the classical average of
155: $B(q,p)B(q(t),p(t))$ over the energy hypersurface at energy $E$. Making use of
156: the classical sum rule
157: \begin{equation}
158: \sum_p T_p \, f(T_p) \, D_p^2 \, e^{-\epsilon^2 T_p^2 /\hbar^2}
159: \approx 2\int_0^\infty dT \, f(T) \, e^{-\epsilon^2 T^2/\hbar^2} \,,
160: \end{equation}
161: one finally obtains
162: \begin{eqnarray}
163: \sigma^2_B(E) &\approx & {2 \over \beta} {4 \sqrt{\pi} \, \epsilon \over \nu(E)}
164: (2 \pi \hbar)^{-2} \, \int_0^\infty dT \int_{-T/2}^{T/2} dt \,
165: \langle B(t)B(0) \rangle e^{-\epsilon^2T^2/\hbar^2} \nonumber \\
166: &=& {2 \over \beta} { 1 \over \pi \nu(E) \hbar} \int_0^\infty dt \,
167: \langle B(t)B(0) \rangle f_\epsilon(t) \,,
168: \end{eqnarray}
169: with $f_\epsilon(t)=1-{\rm erf}(2\epsilon t/\hbar)$. This may be
170: approximated~\cite{artuso} as
171: \begin{equation}
172: \label{sigmaint}
173: \sigma^2_B(E) \approx {2 \over \beta} { \langle B^2 \rangle \over \pi \nu(E)
174: \hbar} \int_0^\infty dt \, P(t) f_\epsilon(t) \,,
175: \end{equation}
176: where
177: \begin{equation}
178: \label{pt}
179: P(t)={\langle \, B(t) \, B (0)\,\rangle \over \langle \, B^2 \, \rangle}
180: \end{equation}
181: is a classical correlation function. For a chaotic system, $P(t)$ typically
182: decays as a sum of exponentials~\cite{artuso}. The long-time behavior is then
183: governed by the smallest exponent $\lambda$, $P(t) \approx b \, e^{-\lambda
184: t}$, where $b$ is a classical constant of order unity. If the long-time
185: behavior dominates the integral, one obtains in the limit $\epsilon \ll \lambda
186: \hbar$
187: \begin{equation}
188: \label{longdom}
189: \sigma^2_B(E) \approx {2 \over \beta} { b\, \langle B^2 \rangle \over \pi
190: \nu(E) \hbar \lambda }
191: = {4 b\,\langle B^2 \rangle \over \beta \lambda T_H}
192: = {4 b\,\langle B^2 \rangle\, T_{\rm decay} \over \beta T_H}
193: \,,
194: \end{equation}
195: where $T_H=2\pi\hbar/\Delta=2\pi\hbar \nu(E)$ is the Heisenberg time, at which
196: individual eigenstates are resolved, and $T_{\rm decay}=\lambda^{-1}$ is a
197: classical correlation decay time. Since $\nu(E) \propto \hbar^{-d}$, the
198: variance decays as
199: \begin{equation}
200: \label{varhlam}
201: \sigma^2_B(E) \propto \hbar^{d-1}
202: \lambda^{-1}
203: \end{equation}
204: for small $\hbar$. This is consistent with Shnirelman's conjecture concerning
205: the ergodicity of chaotic eigenstates. The result is independent of the
206: smoothing function $\delta_\epsilon(E)$. In the limit of small $\lambda$, the
207: decay time $T_{\rm decay}=\lambda^{-1}$ becomes longer than the Heisenberg time
208: $T_H = 2 \pi \hbar \nu(E)$, and $\lambda^{-1}$ in Eq.~(\ref{varhlam}) should
209: be replaced with $2 \pi \hbar \nu(E)$. Then $\sigma^2_B(E) \propto \hbar^0$,
210: as expected for a quantum regular system. We can interpolate between the
211: extreme chaotic and effectively regular regimes by taking $T_{\rm decay}$
212: intermediate the period of the shortest orbit and the Heisenberg time.
213:
214: \section{Classically Chaotic Maps}
215: \label{secmaps}
216:
217: As an example, consider the class of maps
218: \begin{eqnarray}
219: p_{k+1} &=& p_k-V'(q_{k}) \nonumber
220: \\
221: q_{k+1}&=&q_k+T'(p_{k+1})
222: \,,
223: \end{eqnarray}
224: on the torus $[-\pi,\pi) \times [-\pi,\pi)$. Any such map may be thought of as
225: arising from the periodically kicked Hamiltonian~\cite{kick,kick2}
226: \begin{equation}
227: H(p,q,t) = { 1 \over T_{\rm kick}}
228: T(p) + V(q) \sum_{j=-\infty}^\infty \delta(t-j \, T_{\rm kick}) \,.
229: \end{equation}
230: The choice $T(p)=v \, p^2/2+K_2 \, v \sin p$ and $V(q)=-v \, q^2 /2 - K_1 \, v
231: \, \sin q$ corresponds to a perturbed sawtooth map. Here
232: \begin{equation}
233: \label{sawtooth}
234: \begin{array}{lll}
235: p_{k+1} &=& p_k+v\, q_k+ K_1 \, v \, \cos q_k \nonumber \\[5pt]
236: q_{k+1} &=& (1 + v^2) q_k + v \, p_k+K_1 \, v^2 \, \cos q_k
237: + K_2 \, v \, \cos p_{k+1}
238: \end{array}
239: \end{equation}
240: and linearized motion is governed by the monodromy matrix
241: \begin{equation}
242: M = \left [ \begin{array}{cc}
243: 1 & v\, (1- K_1 \sin q_k) \\ v\, (1-K_2 \sin p_{k+1}) &
244: 1+v^2\, (1-K_1 \sin q_k)(1-K_2 \sin p_{k+1})
245: \end{array} \right ] \,.
246: \end{equation}
247: In this case, strict hyperbolicity is guaranteed for $v \ne 0$, as long as
248: $|K_1|<1$ and $|K_2|<1$. The parameters $K_1$ and $K_2$ introduce nonlinearity
249: and symmetry breaking into the dynamics. The classical Lyapunov exponent and
250: correlation decay time $T_{\rm decay}$ may easily be controlled by adjusting
251: the parameter $v$.
252:
253: \begin{figure}[ht]
254: \centerline{
255: \psfig{file=fig1.ps,angle=270,width=4.5in}}
256: \vskip 0.2in
257: \caption{
258: The classical autocorrelation function $P(t)$ of Eq.~(\ref{pt}) is plotted for
259: the perturbed sawtooth map of Eq.~(\ref{sawtooth}), with $K_1=0.3$, $K_2=-0.2$,
260: and several values of $v$. $P(t)$ is computed for each of the six observables
261: $B(q,p)=\cos (2 m \pi q), \, \sin(2 m \pi q)$, where $m = 1$, $2$, $3$, and
262: then averaged to obtain the behavior for a typical observable. The thin
263: solid lines represent best fits to the long-time exponential behavior $P(t)=b
264: e^{-\lambda_v t}$, with $\lambda_{0.2}=0.0083$, $\lambda_{0.3}=0.0275$,
265: $\lambda_{0.4}=0.067$, and $\lambda_{0.5}=0.14$. } \label{fig_clas}
266: \end{figure}
267:
268: In Fig.~\ref{fig_clas}, the classical autocorrelation function $P(t)$ for the
269: perturbed sawtooth map is shown for several values of the $v$ parameter.
270: Clearly, the long-time decay exponent $\lambda$ decreases with decreasing $v$,
271: and empirically we find $\lambda_v \sim v^{3.1}$ for $v \ll 1$. Note that the
272: correlation decay rate $\lambda$ behaves very differently from the Lyapunov
273: exponent, which scales linearly with $v$.
274:
275: When a discrete-time map is quantized on an $N-$dimensional Hilbert space, we
276: have $\hbar=2 \pi/N$, $\nu(E)=(N/2 \pi)^2 \, T_{\rm kick}$, and $T_H=N \,
277: T_{\rm kick}$. Then Eq.~(\ref{sigmaint}) becomes
278: \begin{equation}
279: \label{sigmaintmap}
280: \sigma_B^2(E) \approx {2 \over \beta} {2 \, \langle B^2 \rangle \over N}
281: \left [ {1 \over 2} + \sum_{k=1}^{\infty} P(k) f_\epsilon(k)\right ] \,,
282: \end{equation}
283: and Eq.~(\ref{longdom}) takes the form
284: \begin{equation}
285: \sigma_B^2(E) \approx {2 \over \beta} {2 \, b\langle \, B^2 \rangle \over
286: \lambda \, N \, T_{\rm kick}} \,.
287: \end{equation}
288:
289: We may also consider a system with power-law decay of classical correlations,
290: e.g.,
291: \begin{eqnarray}
292: \label{clpower}
293: T(p) &=& {1 \over 2}p^2+\cos (2 \pi p) \nonumber \\
294: V(q) &=& \left \{ \begin{array}{ll} -{1 \over 2} \left [ q^2 - \pi^2/4 \right ]
295: \;\;\; & {\rm for} \; |q| <\pi/2 \\[5pt] 0 \;\;\; & {\rm otherwise} \end{array} \right .
296: \,.
297: \end{eqnarray}
298: Here, bouncing-ball modes for $|q|< \pi/2$ in the vicinity of $p=0$ dominate
299: the long-time classical correlation decay, resulting in $P(k) \sim b \,
300: k^{-\gamma}$, with $\gamma=1/3$. The sum over time steps $k$ in
301: Eq.~(\ref{sigmaintmap}) is effectively truncated on the scale of the Heisenberg
302: time $N$ by the smoothing function $f_\epsilon(k)$. Then $\sum_{k=1}^{N/2}
303: P(k)f_\epsilon(k) \sim N^{2/3}$ and thus
304: \begin{equation}
305: \sigma_B^2(E) \sim {\langle B^2 \rangle \over N^{1/3}} \,.
306: \end{equation}
307:
308: \section{Brownian Motion Model for Chaotic Energy Levels}
309: \label{secbrownian}
310:
311: The statistical properties of the energy levels $E_n$ can be obtained using a
312: Brownian motion model~\cite{dyson}. The discussion in this section follows
313: \cite{wilk,wilkwalk,mehlwilk}. In this model, the matrix elements of the
314: Hamiltonian undergo a diffusive evolution as a function of a fictitious time
315: variable $\tau$. We denote the infinitesimal change of $\hat H$ by $\delta
316: \hat H$. The off-diagonal elements of $\delta \hat H$ are assumed to satisfy
317: \begin{equation}
318: \overline{\delta H_{mn}}=0 \,, \;\;\;\; \overline{\delta H_{mn}
319: \delta H_{m'n'}^\ast} = \delta \tau \, C_{mn}^{\rm off} \delta_{nn'}
320: \delta_{mm'}\,,
321: \end{equation}
322: while the diagonal elements are taken to obey
323: \begin{equation}
324: \overline{\delta H_{nn}}=0 \,, \;\;\;\; \overline{\delta H_{mm}
325: \delta H_{nn}} = { 2 \over \beta} \, \delta \tau \, C_{mn}^{\rm diag} \,.
326: \end{equation}
327: Within RMT, $C_{mn}^{\rm diag} = \delta_{mn}$ and $C_{mn}^{\rm off} =1$.
328: Non-universal deviations from RMT are encoded in $C_{mn}^{\rm
329: diag}=C_{m-n}^{\rm diag}$ and $C_{mn}^{\rm off}=C_{m-n}^{\rm
330: off}$~\cite{mehlwilk}. The corresponding motion of the energy levels $E_0$ ,
331: ... , $E_{N-1}$ may be analyzed as follows. Using second order perturbation
332: theory, one obtains
333: \begin{equation}
334: \delta E_n = \sum_{m \ne n} {|\delta H_{mn}|^2 \over E_n -E_m}
335: +\delta H_{nn}
336: \end{equation}
337: for the energy level shifts $\delta E_n$. Thus
338: \begin{equation}
339: \label{deltaen}
340: \overline{\delta E_n} = \delta \tau \sum_{m \ne n}
341: {C_{m-n}^{\rm off} \over E_n -E_m}
342: \end{equation}
343: and
344: \begin{equation}
345: \label{deltaemn}
346: \overline{\delta E_m \delta E_n} = {2 \over \beta} \, \delta \tau \,
347: C^{\rm diag}_{m-n} \,.
348: \end{equation}
349:
350: It is convenient to express Eqs.~(\ref{deltaen}) and (\ref{deltaemn}) in terms
351: of the Fourier modes of $\Delta E_n \equiv E_n - n \Delta$. We enforce
352: periodic boundary conditions $\Delta E_N = \Delta E_0$ and define
353: \begin{equation}
354: a_k = {1 \over N} \sum_{n=0}^{N-1} \Delta E_n e^{-2 \pi i kn/N}
355: \end{equation}
356: so that
357: \begin{equation}
358: \Delta E_n = \sum_{k=-N/2}^{N/2-1} a_k \, e^{2 \pi i k n /N} =
359: \sum_{k=0}^{N/2-1} a_k \, e^{2 \pi i k n /N} + {\rm c.c.}
360: \end{equation}
361: Since the $\Delta E_n$ are real, $a_k=a^\ast_{-k}$. From Eq.~(\ref{deltaemn})
362: we have, using $\delta \Delta E_n = \delta E_n$,
363: \begin{equation}
364: \label{dakdap}
365: \overline{\delta a_k \delta a_p^\ast} =
366: \overline{\delta a_k^\ast \delta a_p}={2 \over \beta} \,
367: \delta \tau \, I_k \, \delta_{kp} \,,
368: \end{equation}
369: where $I_k=N^{-1} \sum_n C_n^{\rm diag} e^{2 \pi i kn/N}$, and
370: \begin{equation}
371: \overline{\delta a_k \delta a_p} = \overline{\delta a_k^\ast \delta a_p^\ast}
372: =0
373: \end{equation}
374: for $k,p \in (0,N/2)$. In general, the expectation value of $\delta a_k$ is a
375: complicated function of all $a_p$, which may be expanded as
376: \begin{equation}
377: \label{daexpand}
378: \overline{\delta a_k} = \delta \tau
379: \left [
380: A_1(k) a_k + \sum_q A_2(k,q) a_{k-q}a_q +\sum_{q,p} A_3(k,q,p) a_q a_p
381: a_{k-q-p} + \cdots \right ] \,.
382: \end{equation}
383: The coefficients $A_j$ are obtained as follows (assuming that $C_{m-n}^{\rm
384: off}=1$, see however Ref.~\cite{chalker}). Expanding the denominator $E_n-E_m$
385: in Eq.~(\ref{deltaen}) around the mean $(n-m)\Delta$,
386: \begin{equation}
387: \overline{\delta E_n} \approx -\delta \tau
388: \sum_{\ell \ne 0} { 1\over \ell \Delta}
389: \left [
390: 1- {1 \over \ell \Delta}(\Delta E_{n+\ell}-\Delta E_n)
391: + {1 \over 2 \ell^2 \Delta^2}(\Delta E_{n+\ell}-\Delta E_n)^2 +\cdots
392: \right ] \,.
393: \label{deltaenexp}
394: \end{equation}
395: Again using $\delta E_n = \delta \Delta E_n$ and the identity $ {1 \over N}
396: \sum_n e^{-2 \pi i kn/N} =\sum_j\delta_{k,j N} $, the first term in
397: Eq.~(\ref{deltaenexp}) implies
398: \begin{equation}
399: \overline{\delta a_k} = \delta \tau a_k \sum_{\ell \ne 0}
400: \left [
401: {1 \over \ell^2\Delta^2} \left ( e^{2\pi ik \ell /N} -1 \right ) + \cdots
402: \right ] \,.
403: \end{equation}
404:
405: In the limit of large $N$ one has to first order in $|k|/N$ (extending the
406: range of summation over $\ell$ from $-\infty$ to $\infty$)
407: \begin{equation}
408: \sum_{\ell \ne 0}
409: {1 \over \ell^2\Delta^2} \left ( e^{2\pi ik \ell /N} -1 \right )
410: \approx - {2 \pi^2 |k| \over N \Delta^2} \,,
411: \end{equation}
412: and thus
413: \begin{equation}
414: A_1(k)= - {2 \pi^2|k| \over N \Delta^2} \,.
415: \end{equation}
416: To lowest order in Eq.~(\ref{daexpand}), the equilibrium distribution of the
417: $a_k$ (corresponding to large fictitious time $\tau$) factorizes into a product
418: of Gaussians, where the variance is given by
419: \begin{equation}
420: \label{ak2}
421: \overline{|a_k|^2} = {2 \over \beta} {N \Delta^2 I_k \over
422: 4 \pi^2 k} \,,
423: \end{equation}
424: while higher-order cumulants are zero. The approximation of Eq.~(\ref{ak2}) is
425: appropriate for $k \ll N$. In this case, the $a_k$ for different $k$ are
426: uncorrelated. In RMT, $I_k=N^{-1}$, and
427: \begin{equation}
428: \label{ak2rmt}
429: \overline{|a_k|^2} = {2 \over \beta} {\Delta^2 \over 4 \pi^2 k}
430: \end{equation}
431: for $k \ll N$.
432:
433: In order to determine the fluctuations of $a_k$ for $k \sim N$, higher-order
434: terms in Eq.~(\ref{daexpand}) must be taken into account. This may be done
435: perturbatively, resulting in corrections to the variance and possibly non-zero
436: higher-order cumulants. Moreover, $a_k$ for larger values of $k$ may be
437: correlated.
438:
439: In the above derivation, we have started with initial conditions $a_k=0$ at
440: fictitious time $\tau=0$, i.e. we have assumed the initial unperturbed
441: Hamiltonian has a ``picket fence" spectrum, $E_n = n \Delta$. A typical
442: chaotic Hamiltonian, however, will not correspond naturally to a small
443: perturbation of such a ``picket fence" Hamiltonian. Therefore, in practice it
444: is more useful to observe spectral equilibration by comparing coefficients
445: $a_k$ of a given Hamiltonian $\hat H$ with the coefficients $a_k'$ of a
446: perturbed Hamiltonian $\hat H+\hat B$. Full equilibration implies that the
447: spectra of $\hat H$ and $\hat H +\hat B$ become independent of one another,
448: though drawn from the same random matrix ensemble. Then
449: \begin{equation}
450: \label{akdiffrmt}
451: \overline{|a_k-a_k'|^2} = {2 \over \beta} {\Delta^2 \over 2 \pi^2 k}
452: = {C_\beta \over k}
453: \end{equation}
454: for $k \ll N$.
455:
456: \section{Conditions for Energy Level Equilibration}
457: \label{secequil}
458:
459: We now address more carefully the question originally raised in the seminal
460: work by Wilkinson~\cite{wilk}. Specifically, we wish to understand under what
461: circumstances a class of classically small perturbations $\hat B$ of an initial
462: Hamiltonian $\hat H$ is sufficient to generate random matrix statistics in the
463: spectrum, at various energy scales. A related question is the size in
464: parameter space of a random chaotic ensemble necessary to average away
465: system-specific spectral properties and generate universal
466: statistics~\cite{zirnbauer,gornyi}.
467:
468: Two conditions are necessary for equilibration to universal statistics to
469: occur. First, if the perturbation is classically small, then equilibration to
470: RMT may only occur on time scales longer than the decay time of classical
471: correlations (otherwise, $I_k \ne N^{-1}$ for the corresponding modes $a_k$).
472: Thus, the modes $a_k$ may be equilibrated to RMT only for $k \gg T_{\rm
473: decay}/T_{\rm kick}$, independent of the perturbation, and correspondingly the
474: spectrum may only display universal statistics on energy scales ${\cal E} \ll
475: \hbar/T_{\rm decay}$. Secondly, the perturbation must be sufficiently strong
476: to equilibrate a given mode $a_k$ after the fictitious time $\tau$ during which
477: the Hamiltonian matrix elements undergo Brownian motion. In our units, this
478: fictitious time $\tau$ is simply equal to the variance~\cite{wilk}:
479: \begin{equation}
480: \tau = \sigma^2_B \,,
481: \end{equation}
482: and is given by Eq.~(\ref{longdom}). On the other hand, the characteristic
483: response time of the $k$-th mode is
484: \begin{equation}
485: \tau_k = {\hbar \over \pi \nu(E) (k T_{\rm kick})} = {2 \hbar^2 \over
486: (k T_{\rm kick}) T_H} \,,
487: \end{equation}
488: as may easily be seen by comparing Eqs.~(\ref{dakdap}) and (\ref{ak2}) and
489: then noting that the Heisenberg time $T_H$ is given by
490: $2\pi\hbar/\Delta=2\pi\hbar \nu(E) =N T_{\rm kick}$.
491: Equilibration will therefore happen for a given mode $a_k$ when $\tau \gg
492: \tau_k$, i.e.
493: \begin{equation}
494: \label{kcond}
495: k \gg {\hbar^2 \over \langle B^2 \rangle T_{\rm decay} T_{\rm kick} } \,,
496: \end{equation}
497: and transforming to the energy domain we find equilibration on scales
498: \begin{equation}
499: \label{econd}
500: {\cal E} \ll {\langle B^2 \rangle T_{\rm decay} \over \hbar} \,.
501: \end{equation}
502: The condition of Eq.~(\ref{econd}) must be satisfied simultaneously with the
503: first condition ${\cal E} \ll \hbar /T_{\rm decay}$, i.e. equilibration to
504: universal statistics occurs on all energy scales $\cal E$ satisfying
505: \begin{equation}
506: \label{twoconde}
507: {\cal E } \ll {\cal E}_{\rm equil} \sim {\rm min} \left (
508: {\hbar \over T_{\rm decay}}, {\langle B^2 \rangle T_{\rm decay} \over \hbar}
509: \right ) \,.
510: \end{equation}
511: Note that the two upper limits on the equilibration energy scale have opposite
512: dependence on the classical correlation scale $T_{\rm decay}$.
513:
514: Let us consider the behavior of the system as the size of the perturbation
515: $\hat B$ is varied, for a given initial Hamiltonian $\hat H$, at a given
516: classical energy. We assume $T_{\rm decay} < T_H$, so that the system is in
517: the quantum chaotic regime. (i) For very small perturbations, $\langle B^2
518: \rangle \ll \hbar^2 / (T_{\rm decay} T_H)$, there is no equilibration on any
519: time scale before the Heisenberg time $T_H$ and consequently no equilibration
520: on any energy scale larger than a mean level spacing $\Delta$. (ii) For larger
521: perturbations, i.e. $\hbar^2 / (T_{\rm decay} T_H) \ll \langle B^2 \rangle \ll
522: \hbar^2/T_{\rm decay}^2$, all modes $k \gg k_{\rm min} \sim \hbar^2 / (T_{\rm
523: kick} T_{\rm decay} \langle B^2 \rangle)$ equilibrate to their RMT values. In
524: the energy domain, the corresponding condition is ${\cal E} \ll {\cal E}_{\rm
525: equil} \sim \langle B^2 \rangle T_{\rm decay} / \hbar$. (iii) Finally, for the
526: largest (still classically small) perturbations $\hbar^2 / T_{\rm decay}^2 \ll
527: \langle B^2 \rangle \ll E^2$, all time scales beyond $T_{\rm decay}$ and all
528: energy scales below ${\cal E}_{\rm equil} \sim \hbar/T_{\rm decay}$ equilibrate
529: to universal statistics.
530:
531: What happens if we instead fix the size of perturbation $\hat B$, and consider
532: a variety of classical systems, with different classical time scales $T_{\rm
533: decay}$? (i) For the most strongly chaotic systems, $T_{\rm decay} \ll \hbar
534: {\langle B^2 \rangle }^{-1/2}$, and equilibration is limited by the condition
535: of Eq.~(\ref{econd}). (ii) As the classical system becomes less chaotic,
536: $k_{\rm min}$ falls, more and more modes $a_k$ come into equilibrium, and the
537: energy scale ${\cal E}_{\rm equil}$ increases. (iii) Maximum equilibration to
538: RMT is attained when $T_{\rm decay} \sim \hbar {\langle B^2 \rangle}^{-1/2}$,
539: where all modes $a_k$ for $k > k_{\rm min} \sim \hbar T_{\rm kick}^{-1}
540: {\langle B^2 \rangle}^{-1/2}$ equilibrate, and ${\cal E}_{\rm equil} \sim
541: {\langle B^2 \rangle}^{1/2}$. (iv) Then, as the degree of chaoticity of the
542: system continues to decrease, $k_{\rm min}$ begins to increase, modes again
543: move away from RMT equilibrium and the energy ${\cal E}_{\rm equil}$ begins to
544: drop. (v) Eventually, we reach the border $T_{\rm decay} \sim T_H$ between
545: quantum chaos and quantum regularity, where all equilibration to universal
546: statistics is again absent. Thus, absence of equilibration to RMT at a given
547: time or energy scale may be consistent with either very strongly chaotic or
548: very weakly chaotic (or regular) systems, while maximum possible equilibration
549: is attained for moderately chaotic systems between these two extremes.
550:
551: \begin{figure}[ht]
552: \centerline{
553: \psfig{file=fig2.ps,angle=270,width=4.5in}}
554: \vskip 0.2in
555: \caption{
556: The effect of a perturbation on the quantum spectrum as a function of Fourier
557: mode $k$ is plotted for four classical systems, distinguished by the parameter
558: $v$ of Eq.~(\ref{sawtooth}). The values of $K_1$ and $K_2$ are as in
559: Fig.~\ref{fig_clas}, and the Hilbert space dimension is $N=512$. As in the
560: previous figure, the data are averaged over an ensemble of perturbations
561: $B(q,p)=\sum_{m=1}^3 \left [ x_m \sin(2\pi m q)+ y_m \cos(2 \pi m q) \right ]$
562: with $\langle B^2 \rangle={1 \over 2} \sum_{m=1}^3 (x_m^2+y_m^2)=0.6
563: N^{-3}=4.51\cdot 10^{-9} $. The constant $C_\beta$, is given by
564: Eq.~(\ref{akdiffrmt}), where $\beta=2$ in the present case. The thin dotted
565: curve indicates the limit of full equilibration on all energy scales, as
566: obtained by choosing random and uncorrelated Hamiltonians $\hat H$ and $\hat
567: H'=\hat H+\hat B$.
568: }
569: \label{fig_a}
570: \end{figure}
571:
572: The discussion of the preceding paragraph is illustrated for the perturbed
573: sawtooth map in Fig.~\ref{fig_a}, where we show the effect on the spectrum at
574: various energy scales of a fixed-size perturbation, for four classical systems
575: characterized by different decay times $T_{\rm decay}$. The classical systems
576: treated here are the same as in Fig.~\ref{fig_clas}, while the family of
577: perturbations used corresponds precisely to the family of classical functions
578: used in that earlier figure. The squared change in each Fourier spectral
579: coefficient $a_k$ has been normalized by the constant $C_\beta$ of
580: Eq.~(\ref{akdiffrmt}), so that full equilibration implies $k |a_k - a_k'|^2
581: /C_\beta =1$ for $k \ll N$. For $k \sim N$, this result is modified due to
582: higher-order terms in Eq.~(\ref{daexpand}), as discussed in
583: Section~\ref{secbrownian} and illustrated by the thin dotted curve in
584: Fig.~\ref{fig_a}. From $v=0.70$ to $v=0.20$, the four systems clearly
585: demonstrate the effect of gradually reducing the Lyapunov exponent and
586: increasing the classical parameter $T_{\rm decay}$.
587:
588: Looking first at the $v=0.70$ curve in Fig.~\ref{fig_a}, we find equilibration
589: of the spectrum only for $k \ge 100$, corresponding approximately to energy
590: scales of three levels spacings or fewer. When the degree of chaos is reduced
591: ($v=0.40$) and $T_{\rm decay}$ correspondingly increases, the same perturbation
592: strength is sufficient to equilibrate the spectrum up through much larger
593: energy scales, namely $k \ge k_{min} \approx 50$. However, further reducing
594: the degree of chaos by tuning $v$ down below $0.4$ and thereby increasing
595: $T_{\rm decay}$ serves to {\it reduce} the range of equilibrated energies for
596: the same perturbation strength, as equilibration is now governed by the
597: condition $k \gg T_{\rm decay}/T_{\rm kick}$. Finally, for $v=0.20$, $T_{\rm
598: decay}/T_{\rm kick} =120$, and once again equilibration to universal behavior
599: is not attained on any energy scale for our system size.
600:
601: \section{Equilibration of Level Spacings} \label{secspac}
602:
603: In the previous section we considered spectral equilibration at various time
604: scales $k T_{\rm kick}$ and energy scales $E$. We now focus specifically on
605: the conditions for equilibration at the energy scale $\Delta$, necessary for
606: example to reproduce the Wigner-Dyson distribution of nearest neighbor level
607: spacings. Assuming $T_{\rm decay} < T_H$ as before, we need to satisfy
608: Eq.~(\ref{econd}) for ${\cal E} \sim \Delta$:
609: \begin{equation}
610: \langle B^2 \rangle \gg {\hbar \Delta \over T_{\rm decay}} \propto
611: {\hbar^{d+1} \over T_{\rm decay}} \,.
612: \end{equation}
613: The size of the perturbation $\hat B$ must generically be at least of order
614: $\hbar^{(d+1)/2}$ to produce equilibration of the level spacings. Less chaotic
615: systems, however, as measured by a longer decay time, equilibrate more
616: efficiently. For example, in a two-dimensional Hamiltonian system, $B \propto
617: \hbar^{3/2}$ is needed for a fixed $T_{\rm decay}$, but $B \propto \hbar^{7/4}$
618: would be sufficient if $T_{\rm decay} \propto T_H^{1/2}$, and $B \propto
619: \hbar^{2 - \epsilon /2}$ suffices if $T_{\rm decay} \propto T_H^{1-\epsilon}$.
620:
621: Equilibration at the scale of the mean level spacing may be measured by
622: focusing on the Fourier coefficient $a_{N/2}$ (notice that $a_N=a_0$).
623: Equilibration of this coefficient is studied in Fig.~\ref{fig_str} as a
624: function of perturbation strength, for several classical Hamiltonians.
625: Specifically, the mean squared fluctuation in the $a_{N/2}$ coefficient is
626: plotted as a function of perturbation strength $\langle B^2 \rangle$. For each
627: classical system, we clearly observe the proportionality between
628: $|a_{N/2}-a'_{N/2}|^2$ and $\langle B^2 \rangle$ when $|a_{N/2}-a'_{N/2}|^2 \ll
629: N^{-1}$, and the eventual saturation at the system-independent equilibrium
630: value at large perturbation size, just as predicted by the Brownian motion
631: model. Furthermore, at a given perturbation strength, we see faster
632: equilibration for systems with slower classical correlation falloff, as
633: indicated by smaller classical parameter $v$. Semiclassically, we expect this
634: increase in equilibration rate to be controlled by the integral $\int_0^\infty
635: dt \, P(t) \approx b T_{\rm decay}$ as in Eqs.~(\ref{sigmaint}) and
636: (\ref{longdom}); the classical predictions are indicated by dotted lines in
637: Fig.~\ref{fig_str}. The slight discrepancy between the quantum data and the
638: semiclassical prefactors may be explained by the finite system size, $N=256$,
639: since the classical expressions assume $N \gg T_{\rm decay}/T_{\rm kick}$,
640: while in our case $N=256$ and $T_{\rm decay}/T_{\rm kick}$ reaches $120$ when
641: $v=0.2$.
642:
643: \begin{figure}[ht]
644: \centerline{
645: \psfig{file=fig3.ps,angle=270,width=4.5in}}
646: \vskip 0.2in
647: \caption{
648: Equilibration of the highest-frequency mode in the spectrum is studied for
649: $N=256$ as a function of perturbation strength for several initial
650: Hamiltonians. The classical systems as well as the ensemble of perturbations
651: are the same as in Fig.~\ref{fig_a}. The thin dotted lines correspond to the
652: classical prediction $|a_{N/2}-a'_{N/2}| \propto \langle B^2 \rangle
653: \int_0^\infty dt \,P(t)$.
654: }
655: \label{fig_str}
656: \end{figure}
657:
658: Fig.~\ref{fig_c} shows explicitly the dependence of equilibration rate on the
659: classical system dynamics, by varying the classical parameter $v$ while the
660: perturbation strength is held fixed. Again, we see an order of magnitude
661: change in the equilibration rate as $v$ is varied, in agreement with the
662: semiclassical prediction indicated by the dashed curve.
663:
664: \begin{figure}[ht]
665: \centerline{
666: \psfig{file=fig4.ps,angle=270,width=4.5in}}
667: \vskip 0.2in
668: \caption{
669: Equilibration of the highest-frequency mode in the spectrum is studied, for
670: $N=1024$ and perturbation strength $\langle B^2 \rangle =6.4 \cdot 10^{-7}$, as
671: a function of the classical parameter $v$ (solid curve). The classical systems
672: and the ensemble of perturbations are the same as in Figs.~\ref{fig_a} and
673: \ref{fig_str}. The classical prediction based on Eq.~(\ref{sigmaint}) is
674: indicated by the dashed curve.
675: }
676: \label{fig_c}
677: \end{figure}
678:
679: \section{Quantization Ambiguity and Spectral Equilibration}
680: \label{secambig}
681:
682: The original motivation behind Wilkinson's work~\cite{wilk} was to understand
683: the relationship between spectral equilibration and the quantization ambiguity,
684: thus using the latter to explain the approach of generic chaotic systems to RMT
685: behavior on short energy scales. Generally, quantization ambiguities come in
686: two types. The first are gauge or boundary condition ambiguities that allow
687: many different semiclassical theories to be associated with the same classical
688: dynamics, as in the Aharonov-Bohm effect or in a change from Dirichlet to
689: Neumann boundary conditions for a billiard system. These ambiguities
690: correspond to $O(\hbar_{\rm eff})$ terms in the quantum Hamiltonian, where
691: $\hbar_{\rm eff}=\hbar/S_{\rm typ} \sim \hbar/ET_{\rm kick} \sim \hbar/pL$,
692: $S_{\rm typ}$ is the typical action of a short classical orbit, $p$ is the
693: typical momentum, and $L$ is the system size. Secondly, there are operator
694: ordering ambiguities that allow multiple quantum Hamiltonians to have the same
695: semiclassical limit, including identical action phases at leading order.
696: Ambiguities of this second class are $O(\hbar_{\rm eff}^2)$ and result, for
697: example, from canonically quantizing the same classical Hamiltonian in two
698: coordinate systems related by a nonlinear canonical transformation~\cite{wilk};
699: they also appear naturally when different limiting procedures are used to
700: define a quantum dynamics on a constrained surface~\cite{ambig}.
701:
702: We first examine spectral equilibration due to the $O(\hbar_{\rm eff}^2)$
703: quantization ambiguity, considered by Wilkinson in Ref.~\cite{wilk}.
704: Substituting $B \sim \hbar_{\rm eff}^2 E$ into Eq.~(\ref{twoconde}), we find
705: equilibration on energy scales
706: \begin{equation}
707: \label{h2e}
708: {\cal E} \ll {\cal E}_{\rm equil} = E \,{\rm min} \left ( \hbar_{\rm eff}
709: {T_{\rm kick} \over T_{\rm decay}}, \hbar_{\rm eff}^3 {T_{\rm decay} \over
710: T_{\rm kick}} \right )\,,
711: \end{equation}
712: or equivalently for Fourier modes
713: \begin{equation}
714: \label{h2k}
715: k \gg k_{\rm min} = {\rm max} \left ( {T_{\rm decay} \over T_{\rm kick}},
716: \hbar_{\rm eff}^{-2} {T_{\rm kick} \over T_{\rm decay}} \right ) \,.
717: \end{equation}
718: For a fixed classical dynamics, the second expression in the parentheses in
719: Eq.~(\ref{h2e}) or Eq.~(\ref{h2k}) always dominates in the semiclassical limit
720: $\hbar_{\rm eff} \to 0$. For a $d$-dimensional Hamiltonian system, the total
721: number of available fluctuating modes in the spectrum is $O(\hbar_{\rm
722: eff}^{1-d})$, and all but the first $O(\hbar_{\rm eff}^{-2} T_{\rm kick}/T_{\rm
723: decay})$ of these are equilibrated. Thus, full equilibration is not possible
724: for any spectral mode in the case $d=2$, while for $d=3$ full equilibration is
725: possible for the highest-$k$ modes, assuming slow decay of classical
726: correlations, $T_{\rm decay}/T_{\rm kick} \gg 1$. In the energy domain, this
727: implies equilibration only on scales up to $O(T_{\rm decay}/T_{\rm kick})$ mean
728: level spacings $\Delta$ in the $3$-dimensional case. The case $d=4$ (e.g., two
729: interacting particles in two dimensions) is the first for which equilibration
730: generically extends to scales much larger than a mean level spacing, ${\cal
731: E}_{\rm equil} \gg \Delta$. Finally, in the many-body limit $d \to \infty$, an
732: ever-increasing number of modes are equilibrated by the $O(\hbar_{\rm eff}^2)$
733: ambiguity, however, the first $O(\hbar_{\rm eff}^{-2} T_{\rm kick}/ T_{\rm
734: decay})$ modes are never equilibrated. Correspondingly the range of energies
735: over which equilibration may occur always remains a factor of $O(\hbar_{\rm
736: eff}^{2} T_{\rm decay}/ T_{\rm kick})$ smaller than the ballistic Thouless
737: energy $E_{\rm Thouless} \sim \hbar_{\rm eff}E$, independent of dimension.
738:
739: We now turn to the $O(\hbar_{\rm eff}^{-1})$ quantization ambiguity,
740: associated with external gauge fields or boundary conditions. A similar
741: analysis shows that equilibration now occurs on energy scales ${\cal E} \ll
742: {\cal E}_{\rm equil} \sim \hbar_{\rm eff}E T_{\rm kick}/T_{\rm decay}$ and for
743: Fourier modes $k \gg k_{\rm min} \sim T_{\rm decay}/T_{\rm kick}$. Thus, all
744: modes equilibrate in any dimension for a chaotic system, except for the first
745: few that encode non-universal short-time dynamics for a classical system with
746: slow correlation decay.
747:
748: \begin{figure}[ht]
749: \centerline{
750: \psfig{file=fig5.ps,angle=270,width=4.5in}}
751: \vskip 0.2in
752: \caption{
753: Equilibration of the highest-frequency mode in the spectrum of a perturbed
754: sawtooth map is shown for $O(\hbar_{\rm eff}^2)$ perturbation strength $\langle
755: B^2 \rangle =(0.66/N^2)^2$ (thick curves) and $O(\hbar_{\rm eff})$ perturbation
756: strength $\langle B^2 \rangle =(0.16/N)^2$ (thin curves) as a function of $N$.
757: The thin dotted lines indicate the predicted $N^{-1}$ behavior for the
758: $O(\hbar_{\rm eff}^2)$ case in the semiclassical limit $N \to \infty$.
759: }
760: \label{fig_b}
761: \end{figure}
762:
763: In Fig.~\ref{fig_b}, we again focus on equilibration of the highest-frequency
764: Fourier mode $a_{N/2}$, corresponding to energy scales of order $\Delta$, for
765: our quantum map numerical model (which exhibits the scaling of a $d=2$
766: autonomous Hamiltonian system). Consistently with the above discussion, an
767: $O(\hbar_{\rm eff}^2)$ ambiguity is not sufficient to equilibrate even this
768: highest-frequency Fourier mode in the semiclassical limit $N \to \infty$
769: ($\hbar_{\rm eff} \to 0$); lower-frequency modes will be even further away from
770: equilibration. On the other hand, an $O(\hbar_{\rm eff})$ (gauge-size)
771: ambiguity provides full equilibration independent of $\hbar_{\rm eff}$.
772:
773: The above analysis and numerical simulation assume a constant $T_{\rm decay}$
774: in the $\hbar_{\rm eff} \to 0$ limit. However, a system remains in the quantum
775: chaotic regime as long as $T_{\rm decay}$ is shorter than the Heisenberg time,
776: i.e. when $T_{\rm decay}/T_{\rm kick} < \hbar_{\rm eff}^{1-d}$. In general,
777: the rate of equilibration to RMT behavior is increased by choosing a system
778: with very slow classical relaxation, $T_{\rm decay} \gg T_{\rm kick}$. For the
779: $d=2$ case, it is impossible to attain complete equilibration to RMT for any
780: spectral mode with an $O(\hbar_{\rm eff}^2)$ perturbation, regardless of the
781: choice of classical system. For $d\ge 3$, on the other hand, Eqs.~(\ref{h2e})
782: and (\ref{h2k}) clearly indicate that full equilibration for some modes can be
783: achieved with an $O(\hbar_{\rm eff}^2)$ perturbation for systems with $T_{\rm
784: decay}/T_{\rm kick} \gg 1$ (where a generic chaotic system with $T_{\rm
785: decay}/T_{\rm kick} \sim 1$ would not display equilibrated behavior). Optimal
786: equilibration is attained for $T_{\rm decay}/ T_{\rm kick} \sim \hbar_{\rm
787: eff}^{-1}$, where all but the lowest $O(\hbar_{\rm eff}^{-1})$ modes are
788: equilibrated in any dimension $d \ge 3$.
789:
790: \begin{figure}[ht]
791: \centerline{
792: \psfig{file=fig6.ps,angle=270,width=4.5in}}
793: \vskip 0.2in
794: \caption{
795: Equilibration of the highest-frequency mode in the spectrum is shown for
796: $O(\hbar^2)$ perturbation strength $\langle B^2 \rangle =(0.66/N^2)^2$, for a
797: system with power-law classical correlation decay, defined by
798: Eq.~(\ref{clpower}), solid curve, and for the perturbed sawtooth map with
799: $v=0.7$, dashed curve. The thin dotted lines indicate the predicted
800: $N^{-\gamma}$ behavior, for $\gamma=1/3$ and $\gamma=1$.
801: }
802: \label{fig_bb}
803: \end{figure}
804:
805: Finally, in Fig.~\ref{fig_bb} we compare equilibration in the hard chaotic
806: perturbed sawtooth map with equilibration in a system governed by a slow
807: power-law classical decay, defined by Eq.~(\ref{clpower}). In both cases a
808: perturbation of order $\hbar_{\rm eff}^2 \sim N^{-2}$ is used. Although no
809: equilibration is possible for any system in $d=2$, we see clearly that the
810: power-law classical system comes closer to equilibration than the hard chaotic
811: system for any large $N$, and the two systems are governed by different
812: exponents in the $N \to \infty$ limit.
813:
814: \section{Fidelity and Semiclassical Accuracy}
815: \label{secsemi}
816:
817: The fidelity of quantum evolution in the presence of a small perturbation,
818: also known as the Loschmidt echo, has received much attention recently,
819: particularly in the context of classical--quantum correspondence and the very
820: different behaviors that the quantum fidelity can exhibit for classically
821: regular and chaotic systems~\cite{emerson}. For sufficiently large
822: perturbations, the decay of quantum fidelity with time can be governed by the
823: classical Lyapunov exponent~\cite{jalabert,jacquod}. Here we wish to mention
824: an interesting connection between spectral equilibration and fidelity decay in
825: quantum chaotic systems due to a {\it small} perturbation, as analyzed recently
826: by Prosen and collaborators~\cite{prosen}. It was shown that for a static
827: perturbation $\hat B$ with $\langle B^2 \rangle < \hbar^2/T_{\rm decay}^2$, the
828: fidelity decays exponentially on a time scale varying inversely with $T_{\rm
829: decay}$, namely $T_{\rm fidelity} \sim \hbar^2/\langle B^2 \rangle T_{\rm
830: decay}$. Dividing through by the one-step time scale $T_{\rm kick}$, we find
831: that $T_{\rm fidelity}$ corresponds to a dimensionless mode number
832: \begin{equation}
833: k_{\rm fidelity} \sim {\hbar^2 \over \langle B^2 \rangle T_{\rm decay} T_{\rm
834: kick}} \,, \end{equation}
835: which precisely agrees with the boundary between equilibrated and
836: non-equilibrated modes given by Eq.~(\ref{kcond}), valid as long as $k_{\rm
837: fidelity} > T_{\rm decay}/T_{\rm kick}$. In other words, a mode $k$ is
838: equilibrated to universal behavior by a given class of perturbations if and
839: only if two conditions hold simultaneously: the quantum fidelity has decayed to
840: a value much less than unity by the corresponding time $k T_{\rm kick}$ {\it
841: and} this time scale $k T_{\rm kick}$ is larger than the classical relaxation
842: time $T_{\rm decay}$.
843:
844: Another important connection is between spectral equilibration behavior
845: discussed in the present work and the decay of semiclassical accuracy. Since
846: different quantizations of the same semiclassical dynamics differ by
847: $O(\hbar_{\rm eff}^2)$ in the Hamiltonian, the difference between a typical
848: quantization and the semiclassical approximation must be at least of this order
849: in the $\hbar_{\rm eff} \to 0$ limit. Thus the error in the semiclassical
850: approximation must be at least of the same size as the error caused by an
851: $O(\hbar_{\rm eff}^2)$ perturbation in quantum mechanics, assuming of course
852: that the physically correct quantization is ``typical". Clearly, hard quantum
853: effects such as diffraction can increase the error in the semiclassical
854: approximation, so the $O(\hbar_{\rm eff}^2)$ quantization ambiguity only
855: provides a lower bound on the size of the semiclassical error, or equivalently
856: an upper bound on the breakdown time of semiclassical validity. Previous work
857: has shown that this bound on the semiclassical error is saturated for the case
858: of a smooth Hamiltonian in the absence of caustics~\cite{sckaplan}, and
859: furthermore it was demonstrated that the semiclassical error obeys different
860: scaling laws with time and $\hbar$ for regular as opposed to chaotic classical
861: systems, just as one would predict using the quantization ambiguity approach.
862: The semiclassical accuracy problem may also be considered as an example of a
863: generalized quantum fidelity problem, if non-Hermitian perturbations are
864: considered (since semiclassical evolution is in general non-unitary).
865:
866: The results of the present work, particularly those of Sec.~\ref{secambig},
867: imply an upper bound on the breakdown time scale of semiclassical accuracy:
868: $T_{\rm semiclassical} \sim \hbar_{\rm eff}^{-2} T_{\rm kick}^2/T_{\rm decay}$
869: and a lower bound on the breakdown energy scale of semiclassical accuracy:
870: \begin{equation}
871: {\cal E}_{\rm semiclassical} \sim E \hbar_{\rm eff}^3 {T_{\rm decay} \over
872: T_{\rm kick}} \sim E_{\rm Thouless} \hbar_{\rm eff}^2 {T_{\rm decay} \over
873: T_{\rm kick}} \,,
874: \end{equation}
875: where $E_{\rm Thouless}$ is a ballistic Thouless energy. In particular,
876: semiclassical accuracy will persist beyond the Heisenberg time $T_H \sim
877: \hbar_{\rm eff}^{-1} T_{\rm kick}$ in any two-dimensional chaotic system
878: (assuming diffraction and caustics are properly accounted for), allowing
879: individual energy levels and wave functions to be semiclassically resolved. In
880: three-dimensional systems, $T_H \sim \hbar_{\rm eff}^{-2} T_{\rm kick}$, and
881: $T_{\rm semiclassical} \sim T_H T_{\rm kick}/T_{\rm decay}$. Here, the
882: semiclassical breakdown time may be comparable to the Heisenberg time for the
883: most chaotic systems ($T_{\rm decay}/T_{\rm kick} \sim 1$), but any slowdown in
884: the classical relaxation rate will cause a faster breakdown in the
885: semiclassical approximation. Finally, given four or more classical degrees of
886: freedom, semiclassical accuracy inevitably breaks down well before the
887: Heisenberg time for any generic dynamics, even the most chaotic, and
888: semiclassical reproduction of individual spectral levels is never possible. It
889: may be of interest to investigate the manner in which improved semiclassical
890: approximations beyond leading order in $\hbar$~\cite{weibert} may produce
891: different scaling of the semiclassical error and possibly permit quantum
892: spectra to be semiclassically resolved in four dimensions and higher.
893:
894: \section{Summary}
895: \label{secsummary}
896:
897: A careful examination of the Brownian motion model for spectral equilibration
898: to universal statistics shows that equilibration is strongly dependent on the
899: classical relaxation rate as well as on the dimensionality of the system. Two
900: competing time scales, proportional and inversely proportional to the classical
901: relaxation time $T_{\rm decay}$, jointly govern the equilibration process.
902: Balancing of these two time scales implies that for a given perturbation $B$,
903: equilibration of the maximum number of spectral modes is achieved when $T_{\rm
904: decay} \sim \hbar \langle B^2 \rangle^{-1/2}$. For small perturbations, $B <
905: \hbar/T_{\rm decay}$, a relation exists between spectral equilibration as a
906: function of mode number $k$ and the decay of the quantum fidelity as a function
907: of time.
908:
909: Focusing on the effect of $O(\hbar_{\rm eff}^2)$ perturbations, associated
910: with the ambiguity of quantization, we find that no equilibration to universal
911: statistics is ever possible for dynamics in two dimensions. In three
912: dimensions, equilibration to universal statistics occurs for some modes, but
913: only if the classical relaxation time $T_{\rm decay}$ is sufficiently long,
914: while in four dimensions and higher, equilibration of at least some modes
915: occurs for any chaotic system. However, optimal equilibration only occurs for
916: very long classical relaxation times, $T_{\rm decay} \propto \hbar_{\rm
917: eff}^{-1}$. These predictions of the Brownian motion model are also entirely
918: consistent with results for semiclassical accuracy in smooth chaotic systems.
919:
920: \section*{Acknowledgments} The author is grateful to B.~Mehlig for invaluable
921: contributions in the early stages of this work, and to M.~Wilkinson and
922: U.~Smilansky for very useful discussions. This work was supported in part by
923: the U.S. Department of Energy Grant No.\ DE-FG03-00ER41132 and the Louisiana
924: Board of Regents Support Fund Contract LEQSF(2004-07)-RD-A-29.
925:
926: \begin{thebibliography}{99}
927:
928: \bibitem{dyson} F.~J.~Dyson, {\it J. Math. Phys.} {\bf 3}, 1191 (1962);
929: F.~J.~Dyson, {\it J. Math. Phys.} {\bf 13}, 90 (1972).
930:
931: \bibitem{bgs} O.~Bohigas, M.-J.~Giannoni, and C.~Schmit, {\it J. Physique
932: Lett.} {\bf 45}, L-1015 (1984).
933:
934: \bibitem{wilk} M.~Wilkinson, {\it J. Phys. A} {\bf 21}, 1173 (1988).
935:
936: \bibitem{wilkwalk} M.~Wilkinson and P.~N.~Walker, {\it J. Phys. A}
937: {\bf 28}, 6143 (1996).
938:
939: \bibitem{mehlwilk} B.~Mehlig and M.~Wilkinson, {\it Phys. Rev. E} {\bf 63},
940: 045203(R) (2001).
941:
942: \bibitem{zirnbauer} M.~R.~Zirnbauer, in {\it Supersymmetry and Trace Formulae:
943: Chaos and Disorder}, ed. by I.~V.~Lerner, J.~P.~Keating, and
944: D.~E.~Khmelnitskii (Plenum, New York, 1999).
945:
946: \bibitem{gornyi} I.~V.~Gornyi and A.~D.~Mirlin, {\it Phys. Rev. E} {\bf 65},
947: 025202(R) (2002); {\it J. Low Temp. Phys.} {\bf 126}, 1339 (2002).
948:
949: \bibitem{sckaplan} L.~Kaplan, {\it Phys. Rev. E} {\bf 70} 026223 (2004); {\bf
950: 58}, 2983 (1998).
951:
952: \bibitem{prosen} T.~Prosen and M.~Znidaric, {\it J. Phys. A} {\bf 35}, 1455
953: (2002); T.~Prosen and T.~H.~Seligman, {\it ibid.} {\bf 35}, 4707 (2002).
954:
955: \bibitem{eckh} B.~Eckhardt, S.~Fishman, J.~Keating, O.~Agam, J.~Main, and
956: K.~M\"uller, {\it Phys. Rev. E} {\bf 52}, 5893 (1995).
957:
958: \bibitem{artuso} R.~Artuso and A.~Prampolini, {\it Phys. Lett. A} {\bf 246},
959: 407 (1998); R.~Artuso, {\it Physica D} {\bf 131}, 68 (1999).
960:
961: \bibitem{kick} S.~Fishman, D.~R.~Grempel, and R.~E.~Prange, {\it Phys. Rev.
962: Lett.} {\bf 49}, 509 (1984).
963:
964: \bibitem{kick2} G.~Casati, B.~V.~Chirikov, F.~M.~Izraelev, and J.~Ford, in {\it
965: Stochastic Behavior in Classical and Quantum Hamiltonian Systems,} ed. by
966: G.~Casati and J.~Ford, Lecture Notes in Physics Vol. 93 (Springer, Berlin,
967: 1979).
968:
969: \bibitem{chalker} J.~T.~Chalker, I.~V.~Lerner, and R.~A.~Smith,
970: {\it Phys. Rev. Lett.} {\bf 77}, 554 (1996).
971:
972: \bibitem{ambig} K.~A.~Mitchell, {\it Phys. Rev. A} {\bf 63}, 042112 (2001);
973: P.~Maraner, {\it J. Phys. A} {\bf 28}, 2939 (1995).
974:
975: \bibitem{emerson} J.~Emerson, Y.~S.~Weinstein, S.~Lloyd, and D.~G.~Cory,
976: {\it Phys. Rev. Lett.} {\bf 89}, 284102 (2002).
977:
978: \bibitem{jalabert} R.~A.~Jalabert and H.~M.~Pastawski, {\it Phys. Rev. Lett.}
979: {\bf 86}, 2490 (2001)
980:
981: \bibitem{jacquod} P.~Jacquod, P.~G.~Silvestrov, and C.~W.~J.~Beenakker, {\it
982: Phys. Rev. E} {\bf 64}, 055203(R) (2001).
983:
984: \bibitem{weibert} K.~Weibert, J.~Main, and G.~Wunner, {\it Eur. Phys. J. D}
985: {\bf 19}, 379 (2002).
986:
987: \end{thebibliography}
988:
989: \end{document}
990: