cond-mat0408240/rng.tex
1: \documentclass[aps,prl,twocolumn,floats]{revtex4}
2: 
3: % special 
4: \usepackage{ifthen}
5: \usepackage{ifpdf}
6: 
7: \ifpdf
8: \usepackage{graphicx}
9: \usepackage{epstopdf}
10: \else
11: \usepackage{graphicx}
12: \usepackage{epsfig}
13: \fi
14: 
15: 
16: \usepackage{times}
17: \usepackage{float}
18: \usepackage{amsmath}
19: \usepackage{amssymb}
20: \usepackage{bm}
21: \usepackage{latexsym}
22: 
23: 
24: 
25: \begin{document}
26: \newcommand{\hide}[1]{}
27: \newcommand{\tbox}[1]{\mbox{\tiny #1}}
28: \newcommand{\half}{\mbox{\small $\frac{1}{2}$}}
29: \newcommand{\sinc}{\mbox{sinc}}
30: \newcommand{\const}{\mbox{const}}
31: \newcommand{\trc}{\mbox{trace}}
32: \newcommand{\intt}{\int\!\!\!\!\int }
33: \newcommand{\ointt}{\int\!\!\!\!\int\!\!\!\!\!\circ\ }
34: \newcommand{\eexp}{\mbox{e}^}
35: \newcommand{\bra}{\left\langle}
36: \newcommand{\ket}{\right\rangle}
37: \newcommand{\EPS} {\mbox{\LARGE $\epsilon$}}
38: \newcommand{\ar}{\mathsf r}
39: \newcommand{\im}{\mbox{Im}}
40: \newcommand{\re}{\mbox{Re}}
41: \newcommand{\bmsf}[1]{\bm{\mathsf{#1}}}
42: \newcommand{\mpg}[3][b]{\begin{minipage}[#1]{#2}{#3}\end{minipage}}
43: 
44: 
45: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
46: \title{ The twilight zone in the parametric evolution 
47: of eigenstates: beyond perturbation theory and semiclassics}
48: 
49: 
50: \author{\small
51: J. A. M\'endez-Berm\'udez$^{1}$,
52: Tsampikos Kottos$^{1}$,
53: Doron Cohen$^{2}$
54: }
55: 
56: 
57: \affiliation{
58: $^{1}$ Max-Planck-Institut f\"ur Dynamik und Selbstorganisation, \\
59: Bunsenstra\ss e 10, D-37073 G\"ottingen, Germany
60: $^{2}$ Department of Physics, Ben-Gurion University, Beer-Sheva 84105, Israel 
61: }
62: 
63: 
64: %\pacs{03.65.-w}   {Quantum mechanics}
65: %\pacs{05.45.Mt}   {Quantum chaos}
66: %\pacs{73.23.-b}   {Mesoscopic systems}
67: 
68: 
69: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
70: 
71: \begin{abstract}
72: 
73: Considering a quantized chaotic system, 
74: we analyze the evolution of its eigenstates 
75: as a result of varying a control parameter. 
76: As the induced perturbation becomes larger,  
77: there is a crossover from a perturbative 
78: to a non-perturbative regime, which is reflected 
79: in the structural changes 
80: of the local density of states.
81: For the first time the {\em full} scenario 
82: is explored for a physical system: 
83: an Aharonov-Bohm cylindrical billiard. 
84: As we vary the magnetic flux,  
85: we discover an intermediate twilight regime 
86: where perturbative and semiclassical features co-exist.
87: This is in contrast with the {\em simple} crossover 
88: from a Lorentzian to a semicircle line-shape 
89: which is found in random-matrix models. 
90: 
91: \end{abstract}
92: 
93: \maketitle
94: 
95: %%%%%%%%%%%%%%%%%%%%%%%%%%%%     Introduction     %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
96: 
97: The analysis of the evolution of eigenvalues and of the structural changes that the 
98: corresponding eigenstates of a chaotic system exhibit as one varies a parameter $\phi$ 
99: of the Hamiltonian ${\cal H}(\phi)$ has sparked a great deal of research activity for 
100: many years. Physically the change of $\phi$ may represent the effect of some externally 
101: controlled field (like electric field, magnetic flux, gate voltage) or a change of an 
102: effective-interaction (as in molecular dynamics). Thus, these studies are relevant for 
103: diverse areas of physics ranging from nuclear \cite{HZB95,W55} and atomic physics 
104: \cite{TAA95,FGGP99} to quantum chaos \cite{W88,CK01,MLI04,B03} and mesoscopics 
105: \cite{VLG02,TA93}. 
106: 
107: 
108: Up to now the majority of this research activity was focused 
109: on the study of eigenvalues, where a good understanding has been 
110: achieved, while much less is known about eigenstates.
111: The pioneering work in this field has been done by Wigner \cite{W55}, 
112: who studied the parametric evolution of eigenstates 
113: of a simplified Random Matrix Theory (RMT) 
114: model of the type ${\cal H}=\bm{E}+\phi \bm{B}$. 
115: The elements of the diagonal matrix $\bm{E}$ are the 
116: ordered energies $\{E_n\}$, with mean level spacing $\Delta$, 
117: while $\bm{B}$ is a banded {\it random} matrix. 
118: Wigner found that as the parameter $\phi$ increases the 
119: eigenstates undergoes a transition from 
120: a perturbative {\em Lorentzian-type line shape} 
121: to a non-perturbative {\em semicircle line-shape}. 
122: 
123: 
124: For many years the study of parametric evolution for {\it canonically quantized systems} 
125: was restricted to the exploration of the crossover from integrability to chaos \cite{MLI04,B03}. 
126: Only later \cite{CK01} it has been realized that a theory is lacking for systems 
127: that are chaotic to begin with. 
128: Inspired by Wigner theory, the natural prediction was that the local density of 
129: states (LDOS) should exhibit a crossover from a regime where a perturbative 
130: treatment is applicable, to a regime where semiclassical approximation is valid. However, 
131: despite a considerable amount of numerical efforts \cite{CK01}, there was no clear-cut 
132: demonstration of this crossover. Neither a theory has been developed describing how the 
133: transition from the perturbative to the non-perturbative regime takes place.
134: 
135: 
136: 
137: %%%%%%%%%%%%%%%%%%%%%%%%%%    Main Results    %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
138: 
139: It is the purpose of this Letter to present, for the first time, 
140: a complete scenario of parametric evolution, 
141: in case of a physical system that exhibits {\it hard} chaos. 
142: We explore the validity of perturbation theory and semiclassics,  
143: and we discover the appearance of an intermediate regime 
144: (``twilight zone'') where {\it both perturbative and semiclassical 
145: features co-exist}.  Without loss of generality we consider 
146: as an example a billiard system whose classical dynamics 
147: is characterized by a correlation time $\tau_{\tbox{cl}}$,   
148: which is simply the ballistic time. 
149: Associated with $\tau_{\tbox{cl}}$ is the energy scale 
150: $\hbar/\tau_{\tbox{cl}}$. Next we look on a {\em similar} billiard, 
151: but with a rough boundary. This roughness is characterized by 
152: a length scale which is $\ell$ times smaller, 
153: hence we can associate with it an energy scale  
154: $\delta E_{\tbox{NU}} = (\hbar/\tau_{\tbox{cl}})\times \ell$. 
155: The roughness does not affect the chaoticity:  
156: the correlation time $\tau_{\tbox{cl}}$ 
157: as well as the whole power spectrum are barely affected. 
158: Consequently we explain that $\delta E_{\tbox{NU}}$ 
159: is not reflected in the RMT modeling of the Hamiltonian.
160: Still in the LDOS analysis we find that 
161: non-universal (system specific) features appear.  
162: The appearance of such features is a {\em generic} phenomenon 
163: in quantum chaos studies. It introduces a new ingredient into 
164: the theory of parametric evolution {\em which goes beyond RMT}.         
165: 
166: 
167: \hide{
168: The resulting structures can be understood using 
169: a phase-space picture, and are related to the non-semiclassical 
170: features of the wavefunctions. In other words, on the one hand 
171: we demonstrate the limitations of random matrix modeling, 
172: while on the other hand we show the consequences of having failure 
173: of the so-called ``Berry conjecture" \cite{berry}.  
174: }
175: 
176: 
177: 
178: %%%%%%%
179: \begin{figure}[b]
180: \includegraphics[clip,width=0.8\hsize]{rng_fig1}
181: \caption{
182: Left: 2D billiard with $\ell=1$. Right: Corresponding Aharonov-Bohm cylinder.} 
183: \end{figure}
184: %%%%%%%
185: 
186: 
187: 
188: %%%%%%%%%%%%%%%%%%%%%%%%%%%   model Hamiltonian  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
189: 
190: 
191: The model that we will use in our analysis 
192: is a particle confined to an Aharonov-Bohm (AB)
193: cylindrical billiard (see Fig.1) where one can control 
194: the magnetic flux $\Phi$.
195: The cylindrical billiard is constructed by wrapping 
196: a 2D billiard with hard wall boundaries.
197: The lower boundary at $y=0$ is flat, while the 
198: upper boundary $y=L_y+W\xi(x)$ is deformed. 
199: The deformation is described by $\xi(x)=\sum_{n=1}^{\ell} A_n \cos(nx)$
200: where $A_n$ are random numbers in the range [-1,1].
201: The illustration in Fig.1 assumes a smooth boundary ($\ell=1$).
202: The Hamiltonian of a particle in the cylindrical AB billiard is
203: %
204: \begin{equation}
205: {\cal H}(\phi) = 
206: \frac{1}{2\mathsf{m}} 
207: \left[ \left( p_x - \frac{e}{L_x}\Phi \right)^2 + p_y^2 \right]
208: \label{Hcl}
209: \end{equation}
210: %
211: supplemented by $L_x$ periodic boundary conditions
212: in the horizontal direction, and hard wall boundary
213: conditions along the lower and upper boundaries.
214: $p_x$ and $p_y$ are the momenta. Later we shall use
215: the notation $\phi=e\Phi/\hbar$.
216: We consider the chaotic ${\cal H}(\phi=0)$ 
217: as the unperturbed Hamiltonian.
218: 
219: 
220: 
221: 
222: %%%%%%%%%%%%%%%%%%%%%%%%%%  transformed Hamiltonian   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
223: 
224: 
225: 
226: After conformal transformation \cite{MLI04} the billiard
227: is mapped into a rectangular, with a mass tensor which is space
228: dependent. Then it is possible to compute the matrix representation
229: of the Hamiltonian in the plane wave basis $|\nu\mu\rangle$
230: of the rectangular. The result is:
231: %
232: \begin{eqnarray} 
233: && \hspace*{-0.5cm}
234: {\cal H}_{\nu\mu,\nu'\mu'}(\phi) =
235: \frac{\hbar^2}{2\pi \mathsf{m}}
236: \left\{
237: \pi \left( \nu-\frac{\phi}{2\pi} \right) ^2 \delta_{\nu,\nu'}
238: \delta_{\mu,\mu'}
239: \right.
240: \nonumber \\  
241: && +
242: \left[
243: \frac{\mu^2}{8\alpha^2}J^{(0,2)}_{\nu'\nu} +
244: \epsilon^2 J^{(2,2)}_{\nu'\nu}
245: (\frac{1}{8}{+}\frac{\pi^2 \mu^2 }{6}) \right] \delta_{\mu,\mu'}
246: + (-1)^{\mu{+}\mu'}\mu \mu'
247: \nonumber \\
248: && \times
249: \left.
250: \left[
251: \epsilon^2\frac{2(\mu^2+\mu'^2)}
252: {(\mu^2-\mu'^2)^2}J^{(2,2)}_{\nu'\nu}
253: -i\epsilon
254: \frac{\nu+\nu'-\phi/\pi} {\mu^2-\mu'^2} J^{(1,1)}_{\nu'\nu}
255: \right]
256: \right\}
257: \label{hmn}
258: \end{eqnarray}
259: %
260: where
261: %
262: \begin{eqnarray} \nonumber
263: J^{(l,k)}_{\nu'\nu} =
264: \int^{L_x}_0 dx \, \eexp{i(\nu'-\nu)2\pi x/L_x}
265: \left(\frac{d\xi}{dx}\right)^l
266: \frac{1}{(1+\epsilon\xi(x))^k}
267: \end{eqnarray}
268: %
269: The classical dimensionless parameters
270: of the model are the aspect ratio $\alpha=L_y/L_x$,
271: the tilt relative amplitude $\epsilon=W/L_y$,
272: and the roughness parameter $\ell$.
273: Upon quantization we have $\hbar$ that together with $\mathsf{m}$ and $E$
274: determines the De-Broglie wavelength of the particle,
275: and hence leads to an additional dimensionless parameter
276: $n_E=[L_xL_y/(2\pi\hbar^2)]\mathsf{m}E$.
277: For 2D billiards the mean level spacing $\Delta$ is constant,
278: and hence $n_E=E/\Delta \propto 1/\hbar^2 $
279: can be interpreted as either the scaled energy
280: or as the level index. Optionally we define
281: a semiclassical parameter $\hbar_{\tbox{scaled}}=1/\sqrt{n_E}$.  
282:  
283: 
284: 
285: %%%%%%%%%%%%%%%%%%%%%%%%%%  numerics %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
286: 
287: In the numerical study we have taken $\epsilon{=}0.06$ 
288: and $\alpha{=}1$, for which the classical dynamics  
289: is completely chaotic (for any~$\phi$). 
290: We consider either $\ell{=}1$ for smooth boundary,  
291: or $\ell{=}100$ for rough boundary.  
292: The eigenstates $|n(\phi)\rangle$ of the Hamiltonian ${\cal H}(\phi)$ 
293: were found numerically for various values of 
294: the flux (\mbox{$0.0006 < \phi < 60$}). 
295: We were interested in the states within 
296: an energy window $\delta E {\approx} 45$ that contains $\delta n_E {\sim} 200$ levels 
297: around the energy $E {\approx} 400 $. Note that the size 
298: of the energy window is classically small (\mbox{$\delta E \ll E$}),  
299: but quantum mechanically large (\mbox{$\delta E \gg \Delta$}).
300: 
301: 
302: 
303: 
304: 
305: %%%%%%%%%%%%%%%%%%%%%%%%%%  LDOS definition  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
306: 
307: The object of our interest are the overlaps 
308: of the eigenstates $|n(\phi)\rangle$ 
309: with a given eigenstate $|m(0)\rangle$ 
310: of the unperturbed Hamiltonian:
311: %
312: \begin{eqnarray} \label{e3}
313: \hspace*{-0.5cm}
314: P(n|m) = |\langle n(\phi)|m(0)\rangle|^2 
315: = \int \frac{dx dy dp_x dp_y}{(2\pi\hbar)^2} \rho^{(n)}\rho^{(m)} 
316: \end{eqnarray}
317: %
318: The overlaps $P(n|m)$ can be regarded as a distribution 
319: with respect to $n$. Up to some trivial scaling it
320: is essentially the local density of states (LDOS). 
321: The associated dispersion is defined as  
322: $\delta E = [\sum P(n|m) (E_n-E_m)^2]^{1/2}$
323: %
324: In practice we plot $P(n|m)$ as a function 
325: of $r=n-m$ or as a function of $(E_n{-}E_m)$, 
326: and average over the reference state $m$. 
327: %
328: The second equality in (\ref{e3}) is useful 
329: for the semiclassical analysis. It involves 
330: the Wigner functions $\rho^{(n)}(x,y,p_x,p_y)$ 
331: which are associated with 
332: the eigenstates $|n(\phi)\rangle$. 
333: The semiclassical approximation is based 
334: on the microcanonical approximation 
335: $\rho^{(n)} \propto \delta(E_n{-}{\cal H}(x,y,p_x,p_y))$. 
336: With this approximation the integral 
337: can be calculated analytically leading to  
338: % 
339: \begin{eqnarray}
340: P_{\tbox{cl}}(n|m) = 
341: \frac{\Delta}
342: {\pi\sqrt{2(\delta E_{\tbox{cl}})^2 - 
343: \left[ (E_n{-}E_m) - \delta E_{\tbox{cl}}^2/(2E_m) \right]^2}}
344: \label{Pcl}
345: \end{eqnarray}
346: %
347: where $\delta E_{\tbox{cl}} = (\hbar v_E / L_x)\phi$ 
348: with $v_E = (2E/\mathsf{m})^{1/2}$. 
349: It is implicit in (\ref{Pcl}) 
350: that $P_{\tbox{cl}}(n|m){=}0$ outside of the allowed range,  
351: which is where the expression under the square~root
352: is negative: For large $|E_n{-}E_m|$ there 
353: is no intersection of the corresponding energy 
354: surfaces, and hence no classical overlap. 
355: 
356: 
357: 
358: 
359: %%%%%%%%%%%%%%%%%%%%%%%%%%  QCC  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
360: 
361: A few words are in order regarding quantum 
362: to classical correspondence (QCC).  
363: Whenever $P(n|m)\approx P_{\tbox{cl}}(n|m)$  
364: we call it ``detailed QCC", 
365: while $\delta E \approx \delta E_{\tbox{cl}}$
366: is referred to as ``restricted QCC" \cite{CK01}. 
367: It is remarkable that (the robust) 
368: restricted QCC holds even if 
369: (the fragile) detailed QCC fails completely. 
370: We have verified \cite{MCK04}
371: that also in the present system  $\delta E$ 
372: is numerically indistinguishable from $\delta E_{\tbox{cl}}$.   
373: 
374: 
375: 
376: %%%%%%%%%%%%%%%%%%%%%%%%%%  perturbation theory %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
377:  
378: A fixed assumption of this work is that $\phi$ 
379: is classically small. But quantum mechanically it 
380: can be either `small' or `large'.  
381: Quantum mechanically {\em small} $\phi$ means  
382: that perturbation theory do provide 
383: a valid approximation for $P(n|m)$. 
384: What is the {\em border} between the perturbative regime 
385: and the non-perturbative regime, we discuss later. 
386: First we would like to show that the prediction which 
387: is based on perturbation theory, 
388: to be denoted as $P_{\tbox{prt}}(n|m)$, 
389: is very different from the semiclassical approximation. 
390: 
391: 
392: In order to write the expression for $P_{\tbox{prt}}(n|m)$
393: we have first to clarify how to apply perturbation 
394: theory in the context of the present model. 
395: To this end, we write the perturbed Hamiltonian ${\cal H}(\phi)$ 
396: in the basis of ${\cal H}(\phi=0)$. 
397: Since we assume that the perturbation 
398: is classically small, it follows that we can linearize 
399: the Hamiltonian with respect to $\phi$. 
400: Consequently the perturbed Hamiltonian is written 
401: as ${\cal H} = \bm{E} + \phi \bm{B}$, 
402: where $\bm{E}=\mbox{diag}\{E_n\}$ is a diagonal matrix, 
403: while $\bm{B} = \{ -(\hbar/e){\cal I}_{nm} \}$.
404: %
405: The current operator is conventionally defined as
406: %
407: \begin{eqnarray} \nonumber
408: {\cal I}\equiv -{\partial {\cal H}/\partial \Phi} = (e/(\mathsf{m}L_x))p_x
409: \end{eqnarray}
410: %
411: Its matrix elements can be found using a semiclassical recipe \cite{FP86},
412: namely $|{\cal I}_{nm}|^2 \approx (\Delta /(2 \pi\hbar)) \tilde{C}((E_n{-}E_m)/\hbar)$,
413: where $\tilde{C}(\omega)$ is the Fourier transform
414: of the current-current correlation function $C(\tau)$.
415: Conventional condensed matter calculations are done
416: for {\em disordered} rings where one assumes
417: $C(\tau)$ to be exponential, with time constant $\tau_{\tbox{cl}}$
418: which is essentially the ballistic time.
419: Hence $\tilde{C}(\omega) \propto 1/(\omega^2+(1/\tau_{\tbox{cl}})^2)$
420: is a Lorentzian. This Lorentzian approximation
421: works well also for the chaotic ring that we consider.
422: In fact we can do better by exploiting a relation
423: between ${\cal I}(t)$ and the force ${\cal F}(t)=-{\dot p_x}$,
424: leading to $\tilde{C}(\omega) = (e/(\mathsf{m}L_x))^2 
425: {\tilde C}_{\cal F}(\omega)/\omega^2$.
426: The force ${\cal F}(t)$ is a train of spikes corresponding
427: to collisions with the boundaries. Assuming that the collisions 
428: are uncorrelated on short times we have
429: ${\tilde C}_{ \cal F} (\omega) \approx (8/3\pi)\mathsf{m}^2v_E^3/L_y$,
430: for $\omega \gg (1/\tau_{\tbox{cl}})$.  
431: This is known as the ``white noise" approximation \cite{BC}.
432: %
433: We have checked the validity of this approximation
434: in the present context by a direct numerical evaluation
435: of ${\tilde C}(\omega)$, and also verified the validity of 
436: the above recipe by direct evaluation of the matrix elements of 
437: $\bm{B}$ via Eq.(\ref{hmn}), see Fig.~2(a). 
438: %
439: The classical ${\tilde C}(\omega)$ was numerically evaluated 
440: by Fourier analysis of the fluctuating current $\mathcal{I}(t)$ 
441: for a very long ergodic trajectory that covers densely 
442: the whole energy surface ${\cal H}(0)=E$.  
443: 
444: 
445: 
446: %%%%%%%
447: \begin{figure}
448: \includegraphics[clip,width=1.0\hsize]{rng_fig2}
449: \caption{(a) The classical power spectrum $C(\omega)$ plotted (in grey) 
450: together with the quantum mechanical 
451: band-profile $(2\pi e^2/\Delta \hbar)|{\bf B}_{nm}|^2$ 
452: for $\ell=1$, and $\hbar_{\rm scaled}\approx 0.018$. 
453: (b) The LDOS kernel $P(n|m)$ in the perturbative regime 
454: for a billiard with $\ell=100$ and perturbation $\phi=2.7$. 
455: (c) Same as (b) but zoomed normal scale. 
456: The width of the non-perturbative component is $\Gamma/\Delta=36$. 
457: Note that in this regime the variance $\Delta E/\Delta \approx 58$ 
458: is still dominated by the (perturbative) tails. For comparison we display the 
459: calculated $P_{\tbox{prt}}$, $P_{\tbox{cl}}$, and $P_{\tbox{RMT}}$.}
460: \end{figure}
461: %%%%%%%
462: 
463: 
464: 
465: Perturbation theory to infinite order
466: with the Hamiltonian
467: ${\cal H} = \bm{E} + \phi \bm{B}$
468: leads to a Lorentzian-type approximation
469: for the LDOS \cite{W55} 
470: (see also Section 18 of \cite{CK01}c). 
471: It is an approximation because
472: all the higher orders are treated within
473: a Markovian-like approach (by iterating
474: the first order result) and convergence
475: of the expansion is pre-assumed, leading to
476: %
477: %
478: $P_{\tbox{prt}}(n|m) =
479: \phi^2 |\bm{B}_{nm}|^2 / [\Gamma^2 + (E_n{-}E_m)^2]$.
480: %
481: %
482: In practice the parameter $\Gamma(\phi)$ can be determined
483: (for a given $\phi$) by imposing the requirement of
484: having $P_{\tbox{prt}}(r)$ normalized to unity.
485: %
486: %
487: Substituting the expression for the matrix elements we get
488: %
489: \begin{eqnarray}
490: \label{pert}
491: P_{\tbox{prt}}(n|m) =
492: \frac{8\hbar^2(\hbar v_E)^3/(3\pi \mathsf{m}L_y^2 L_x^3)}
493: {(E_n{-}E_m)^2+(\hbar/\tau_{\tbox{cl}})^2}
494: \frac{\phi^2}{(E_n{-}E_m)^2+\Gamma^2}
495: \end{eqnarray}
496: %
497: %
498: %
499: %
500: %
501: %
502: By comparing the exact $P(r)$ to the approximation Eq.(\ref{pert})
503: we can determine the regime $\phi < \phi_{\tbox{prt}}$ 
504: for which the approximation $P(r) \approx P_{\tbox{prt}}(r)$ makes sense.
505: The practical procedure to determine $\phi_{\tbox{prt}}$
506: is to plot $\delta E_{\tbox{prt}}$ and to see where 
507: it departs from $\delta E_{\tbox{cl}}$. 
508: The latter is a linear function of $\phi$ 
509: while the former becomes sublinear for large enough $\phi$, 
510: (and even would exhibit saturation if we had a finite bandwidth). 
511: In case of Eq.(\ref{pert}) this reasoning leads 
512: to a crossover when ${\delta E_{\tbox{cl}}(\phi) \sim \hbar/\tau_{\tbox{cl}}}$. 
513: Hence we get that the border of the perturbative regime 
514: (see footnote \footnote{Optionally $\phi_{\tbox{prt}}$
515: is determined by ${ \Gamma(\phi) \sim \hbar/\tau_{\tbox{cl}} }$. 
516: It should be distinguished from the border of the first order 
517: perturbative regime which is determined by ${\Gamma(\phi) \sim \Delta}$, 
518: leading to $\phi_{\tbox{FOPT}} \sim \phi_{\tbox{prt}}/\sqrt{b}$ 
519: where ${b=(\hbar/\tau_{\tbox{cl}})/\Delta \gg 1}$. 
520: In other words $\phi_{\tbox{FOPT}}$ is the 
521: perturbation which is needed to mix neighboring levels.} )
522: is ${\phi_{\tbox{prt}} = L_x / (v_E \tau_{\tbox{cl}}) \sim 1}$. 
523: 
524: 
525: 
526: %%%%%%%
527: \begin{figure}
528: \includegraphics[clip,width=1.0\hsize]{rng_fig3}
529: \caption{
530: (a) The LDOS kernel $P(n|m)$ for $\phi=31.4$, where $\ell=100$.  
531: (b) The same parameters but $\ell=1$. In panel (a) 
532: we observe co-existence of perturbative and SC structures
533: while in panel (b) we witness detailed QCC.}
534: \end{figure}
535: %%%%%%%
536: 
537: 
538: 
539: 
540: %%%%%%%%%%%%%%%%%%%%%%%%%%  The observation %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
541: 
542: 
543: What happens to $P(r)$ in practice?  If we take the Wigner RMT model 
544: as an inspiration, we expect to have at $\phi\sim\phi_{\tbox{prt}}$ 
545: a simple crossover from a $P_{\tbox{prt}}$ line-shape to a $P_{\tbox{cl}}$ line-shape. 
546: The latter is regarded as the semiclassical analogue 
547: of the (artificial) semicircle line shape. 
548: Indeed for the smooth billiard ($\ell=1$) we have verified that 
549: this naive expectation is realized \cite{MCK04}.
550: %
551: But for the rough billiard ($\ell=100$) we witness a more 
552: complicated scenario. In Fig.~2(b,c) we show the LDOS for $\phi<\phi_{\tbox{prt}}$, 
553: where it (still) agrees quite well with $P_{\tbox{prt}}$. 
554: In Fig.~3 we show the LDOS for $\phi>\phi_{\tbox{prt}}$, 
555: where we would naively expect agreement with $P_{\tbox{cl}}$. 
556: Rather we witness a three peak structure, where the $r\sim0$ peak 
557: is of perturbative nature, while the other are the fingerprint of semiclassics.
558: For sake of comparison we show the corresponding results 
559: for a smooth billiard ($\ell=1$) and otherwise the same parameters.
560: There we have detailed QCC as is naively expected.
561: The co-existence of perturbative and semiclassical features  
562: persists within an intermediate regime of $\phi$ values, 
563: to which we refer as the ``twilight zone". 
564: 
565: 
566: 
567: 
568: %%%%%%%%%%%%%%%%%%%%%%%%%%  RMT %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
569: 
570: 
571: Before we adopt a phase space picture in order to explain 
572: the above observations, we would like to verify that indeed random matrix 
573: modeling does not lead to a similar effect: 
574: After all the standard Wigner model, 
575: that gives rise to a simple crossover from a Lorentzian 
576: to a semicircle line shape, assumes a simple banded matrix,  
577: which is {\em not} the case in our model.  As argued above 
578: the matrix elements of $\bm{B}$ decay as $1/|n-m|^2$ from 
579: the diagonal. This implies that $P_{\tbox{prt}}(r)$ is in fact not 
580: a Lorentzian, and also may imply that the crossover 
581: to the non-perturbative regime is more complicated. 
582: %
583: In order to resolve this subtlety we have taken a randomized 
584: version of the Hamiltonian  ${\cal H} = \bm{E} + \phi \bm{B}$.  
585: Namely, we have randomized the signs of the off-diagonal 
586: elements of the $\bm{B}$ matrix. Thus we get an RMT model 
587: with the same band profile as in the physical model.
588: This means that $P_{\tbox{prt}}$ is the same for both models 
589: (the physical and the randomized), but still they can 
590: differ in the non-perturbative regime. Indeed, 
591: looking at the LDOS of the randomized model we observe 
592: that the semiclassical features are absent:  
593: $P_{\tbox{RMT}}(r)$ unlike $P(r)$ exhibits a simple crossover 
594: from perturbative to non-perturbative lineshape.  
595: 
596: 
597: 
598: 
599: %%%%%%%%%%%%%%%%%%%%%%%%%%  phase space picture %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
600: 
601: %%%%%%%
602: \begin{figure}
603: \includegraphics[clip,width=0.9\hsize]{rng_fig4}
604: \caption{
605: The probability distribution $|\langle \nu,\mu |n \rangle|^2$ for 
606: (a) The $n=2423$ eigenstate of the smooth ($\ell=1$) billiard; 
607: (b) The $n=1000$ eigenstate of the rough ($\ell=100$) billiard. 
608: Note that this is essentially the $(p_x,p_y)$ momentum distribution. 
609: The state in panel (a), unlike the state in panel (b), 
610: is a typical semiclassical state. Namely it is well concentrated 
611: on the energy shell.}
612: \end{figure}
613: %%%%%%%
614: 
615: 
616: In what follows we would like to argue that the structure of $P(r)$, 
617: both perturbative and non-perturbative components, 
618: can be explained using a {\em phase space picture}. 
619: [For phrasing purpose we find the ``Wigner function language'' 
620: most convenient, still the reader should notice that 
621: we do not need or use this representation in practice]. 
622: We recall the $P(n|m)$ is determined by the overlap of two Wigner functions.
623: In the present context the Wigner functions $\rho^{(n)}$ are supported 
624: by shifted circles $(p_x-(\phi/(2\pi))^2+p_y^2 = 2\mathsf{m}E_n$. 
625: We are looking for their overlap with a reference Wigner function which 
626: is supported by the circle $p_x^2+p_y^2 = 2\mathsf{m}E_m$. 
627: The question is whether the overlaps of the Wigner functions 
628: $\rho^{(n)}$ and $\rho^{(m)}$ can be approximated by a 
629: classical calculation, and under what circumstances we 
630: need perturbation theory. 
631: 
632: 
633: Generically the Wigner function has a transverse Airy-type 
634: structure. If the ``thickness" of the Wigner function is 
635: much smaller compared with the separation $|E_n{-}E_m|$ of the energy surfaces 
636: then we can trust the semiclassical approximation. 
637: This will always be the case if $\hbar$ is small enough, 
638: or equivalently if we can make $\phi$ large enough. In such case 
639: the dominant contribution comes from the intersection of 
640: the energy surfaces, which is the phase space analogue of stationary 
641: phase approximation. 
642: %
643: %
644: %
645: The other extreme is the case where the ``thickness" of Wigner function 
646: is larger compared with the separation of the energy surfaces 
647: (namely $\delta E_{\tbox{cl}}(\phi) < \hbar/\tau_{\tbox{cl}}$). 
648: Then the contribution to the overlap 
649: comes ``collectively" from all the regions of 
650: the Wigner (quasi) distribution, not just from the intersections. 
651: In such case we expect perturbation theory to work. 
652: 
653: 
654: The above reasoning assumes that the wavefunction is 
655: concentrated in an ergodic-like fashion in the vicinity of 
656: the energy surface. This is known as ``Berry conjecture" \cite {berry}. 
657: In case of billiards it implies that the wavefunction 
658: looks like a random superposition 
659: of plane waves with \mbox{$|p|=(2\mathsf{m}E)^{1/2}$}. 
660: We find (see Fig.~4) that this does not hold in case of a rough billiard 
661: (unless $\hbar$ were extremely small, 
662: so as to make the De-Broglie wavelength very short). 
663: Namely, in the case of a rough billiard 
664: there are eigenstates that have a lot of weight 
665: in the region \mbox{$|p|<(2\mathsf{m}E)^{1/2}$}.   
666: Consequently there are both semiclassical and non-semiclassical overlaps. 
667: Specifically, if we have non-semiclassical wavefunctions, 
668: and $|E_n-E_m|\sim 0$, then the {\em collective} contribution 
669: dominates, which give rise to the perturbative-like peak in the LDOS.
670: 
671: 
672: Our findings apply to systems, such as the rough billiard,  
673: where there is an additional (large) {\em non-universal} energy 
674: scale $\delta E_{\tbox{NU}}$. This is defined as an energy scale 
675: which is {\em not related} to the bandprofile, 
676: and hence does not emerge in the RMT modeling. 
677: Hence in general there is a distinct twilight regime  
678: \mbox{$\hbar/\tau_{\tbox{cl}} < \delta E_{\tbox{cl}}(\phi) < \delta E_{\tbox{NU}}$}, 
679: which is neither ``perturbative" nor ``semiclassical". 
680: [In our numerics $\ell{=}100$ is so large that $\delta E_{\tbox{NU}}{\sim} E$.]
681: 
682: 
683: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
684: %\clearpage
685: 
686: {\bf Summary:} We have analyzed the parametric evolution of the 
687: eigenstates of an Aharonov-Bohm cylindrical billiard, 
688: as the flux is changed. For the first time the full 
689: crossover from the perturbative to the non-perturbative 
690: regime is demonstrated. Random matrix theory suggests 
691: a {\em simple} crossover. Instead, we discover an intermediate 
692: twilight regime where perturbative and semiclassical features 
693: co-exist. This can be understood by adopting a phase space picture,  
694: and taking into account the inapplicability of the Berry conjecture 
695: regarding the semiclassical structure of the wavefunctions.  
696: 
697: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
698:  
699: This research was supported by a grant from the GIF, 
700: the German-Israeli Foundation for Scientific Research and Development, 
701: and by the Israel Science Foundation (grant No.11/02).
702: 
703: 
704: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
705: 
706: \begin{thebibliography}{99}
707: 
708: \bibitem{HZB95} V. K. B. Kota, Phys. Rep. {\bf 347}, 223 (2001); V. Zelevinsky et al.,
709: Phys. Rep. {\bf 276}, 85 (1996). 
710: 
711: \bibitem{W55} E. Wigner, Ann. Math. {\bf 62}, 548 (1955); {\bf 65}, 203 (1957).
712: 
713: \bibitem{TAA95} N. Taniguchi, A. V. Andreev and B. L. Altshuler, Europhys. Lett. {\bf 29},
714: 515 (1995).
715: 
716: \bibitem{FGGP99} 
717: L. Kaplan and T. Papenbrock, Phys. Rev. Lett. {\bf 84}, 4553 (2000);
718: V. V. Flambaum, A. A. Gribakina, G. F. Gribakin, M. G. Kozlov, Phys. Rev. A {\bf 50},
719: 267 (1994).
720: 
721: \bibitem{W88} M. Wilkinson, J. Phys. A {\bf 21}, 4021 (1988); O. Agam, A. V. Andreev
722: and B. L. Altshuler, Phys. Rev. Lett. {\bf 75}, 4389 (1995).
723: 
724: \bibitem{CK01} 
725: D. Cohen and T. Kottos, Phys. Rev. E {\bf 63}, 036203 (2001); 
726: % D. Cohen, A. Barnett, and E. J. Heller, ibid, {\bf 63}, 046207 (2001); 
727: D. Cohen and E. J. Heller, Phys. Rev. Lett. {\bf 84}, 2841 (2000); 
728: D. Cohen, Ann. Phys. {\bf 283}, 175-231 (2000).
729: 
730: \bibitem{MLI04} 
731: J. A. M\'endez-Berm\'udez, G. A. Luna-Acosta, and F. M. Izrailev,
732: Physica E {\bf 22}, 881 (2004); Phys. Rev. E {\bf 68}, 066201 (2003). 
733: 
734: \bibitem{B03}L. Benet et al., J. Phys. A: Math. Gen. {\bf 36}, 1289 (2003); L. Benet et al.
735: Phys. Lett. A {\bf 277}, 87 (2000);  F. Borgonovi, I. Guarneri, F. M. Izrailev, Phys. Rev. E {\bf 57},
736: 5291 (1998). 
737: 
738: \bibitem{VLG02} R. O. Vallejos, C. H. Lewenkopf, and Y. Gefen, Phys. Rev. B
739: {\bf 65}, 085309 (2002); G. Murthy, {\it et al.}, ibid, {\bf 69}, 075321 (2004);
740: L G. G. V. Dias da Silva, {\it et al.}, ibid, {\bf 69}, 075311 (2004).
741: 
742: \bibitem{TA93} N. Taniguchi and B. L. Altshuler, Phys. Rev. Lett. {\bf 71},
743: 4031 (1993); B. L. Altshuler and B. Simons, Phys. Rev. B {\bf 48}, 5422 (1993).
744: 
745: \bibitem{FP86} M. Feingold and A. Peres, Phys. Rev. A {\bf 34} 591, (1986);
746: M. Feingold, D. Leitner, M. Wilkinson, Phys. Rev. Lett. {\bf 66}, 986 (1991).
747: 
748: \bibitem{BC} 
749: A. Barnett, D. Cohen and E.J. Heller, 
750: Phys. Rev. Lett. {\bf 85}, 1412 (2000); 
751: J. Phys. A {\bf 34}, 413 (2001).
752: 
753: \bibitem{MCK04} J. A. M\'endez-Berm\'udez, D. Cohen and T. Kottos, 
754: in preparation (2005).
755: 
756: \bibitem{berry} M. V. Berry, J. Phys. A {\bf 10}, 2081 (1977). 
757: 
758: \end{thebibliography}
759: 
760: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
761: 
762: 
763: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
764: \end{document}
765: