1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2: %%% Parametric Evolution for a Deformed Cavity (July 2000)
3: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
4:
5: % FIGURES:
6: % FIG.1: potential.eps
7: % FIG.2: image_box.eps; profile_box.eps
8: % FIG.3: bandprof_rota.eps; bandprof_defo.eps
9: % FIG.4: image_rota.eps; image_defo.eps
10: % FIG.5: prof_rota_1.eps; prof_rota_2.eps; prof_rota_3.eps
11: % FIG.6: prof_defo_1.eps; prof_defo_2.eps; prof_defo_3.eps
12: % FIG.7: prt_rota.eps; prt_defo.eps; prt_defo_core.eps
13: % FIG.8: probs_rota.eps; probs_defo.eps; regions_defo.eps
14:
15: \documentstyle[aps,pre,twocolumn,floats,epsfig]{revtex}
16: \begin{document}
17:
18: \newcommand{\tbox}[1]{\mbox{\tiny #1}}
19: \newcommand{\half}{\mbox{\small $\frac{1}{2}$}}
20: \newcommand{\mbf}[1]{{\mathbf #1}}
21:
22: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
23:
24: \title{Parametric Evolution for a Deformed Cavity}
25:
26: \author{Doron Cohen, Alex Barnett and Eric J. Heller}
27:
28: \date{July 2000}
29:
30: \address{Department of Physics, Harvard University}
31:
32: \maketitle
33:
34: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
35:
36: \begin{abstract}
37: We consider a classically chaotic system that is described
38: by a Hamiltonian ${\cal H}(Q,P;x)$, where $(Q,P)$ describes
39: a particle moving inside a cavity, and $x$ controls
40: a deformation of the boundary.
41: The quantum-eigenstates of the system are $|n(x)\rangle$.
42: We describe how the parametric kernel
43: $P(n|m)= |\langle n(x)|m(x_0)\rangle|^2$,
44: also known as the local density of states,
45: evolves as a function of $\delta x=x{-}x_0$.
46: We illuminate the non-unitary nature of this parametric
47: evolution, the emergence of non-perturbative features,
48: the final non-universal saturation, and the limitations
49: of random-wave considerations.
50: The parametric evolution is demonstrated numerically
51: for two distinct representative deformation processes.
52: \end{abstract}
53:
54:
55: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
56: \section{Introduction}
57:
58: \subsection{The local density of states}
59:
60: Consider a system that is described by an Hamiltonian
61: ${\cal H}(Q,P;x)$ where $(Q,P)$ are canonical variables and
62: $x$ is a constant parameter. Our interest in this paper
63: is in the case where $(Q,P)$ describe the motion of
64: a particle inside a cavity, and $x$ controls the deformation
65: of the confining boundary. The 1D version of a cavity,
66: also known as `potential well', is illustrated in Fig.1.
67: However, we are mainly interested in the case of chaotic
68: cavities in $d>1$ dimensions. Cavities in $d=2$ dimensions,
69: also known as billiard systems, are prototype examples
70: in the studies of classical and quantum chaos, and we shall
71: use them for the purpose of numerical illustrations.
72:
73:
74: \begin{figure}[hb]
75: %%%%%%%%%%%%%%%%%%%%%%%%%%%
76: \centerline{\epsfig{figure=potential.eps,width=\hsize}}
77: \vspace{.1in}
78: \caption{
79: The shape of a cavity in $d$ dimensions
80: is defined by its $d-1$ boundary.
81: The confining potential is $V(Q)$.
82: The figure illustrates $V(Q)$ for 1D well.
83: It also can be regarded as a cross section of
84: $d>1$ cavity.
85: The kinetic energy of the particle is $E= \half mv^2$.
86: The walls of the cavity exert a field of force $f$
87: on the bouncing particle. The hard wall
88: limit corresponds to $f\rightarrow\infty$
89: and $V_0\rightarrow\infty$.
90: For theoretical considerations it is convenient
91: to assume that $f$ and $V_0$ are large but finite.
92: Mathematically it is also convenient to think
93: of the embedding space as having some huge but
94: finite volume (not illustrated).}
95: %%%%%%%%%%%%%%%%%%%%%%%%%%%
96: \end{figure}
97:
98:
99: The eigenstates of the quantized Hamiltonian are $|n(x)\rangle$
100: and the corresponding eigen-energies are $E_n(x)$.
101: The eigen-energies are assumed to be ordered, and the
102: mean level spacing will be denoted by $\Delta$.
103: We are interested in the parametric kernel
104: %
105: \begin{eqnarray} \label{e1}
106: P(n|m) \ = \ |\langle n(x)|m(x_0)\rangle|^2
107: \ = \ \mbox{trace}(\rho_n\rho_m)
108: \end{eqnarray}
109: %
110: In the equation above $\rho_m(Q,P)$ and $\rho_n(Q,P)$
111: are the Wigner functions that correspond to the
112: eigenstates $|m(x_0)\rangle$ and $|n(x)\rangle$
113: respectively. The trace stands for $dQdP/(2\pi\hbar)^d$
114: integration. The difference $x-x_0$ will be denoted by $\delta x$.
115: We assume a dense spectrum, so that our interest is in
116: `classically small' but `quantum mechanically large'
117: energy scales. It is important to realize
118: that the kernel $P(n|m)$ has a well defined classical limit.
119: The classical approximation (see remark \cite{rmrk1})
120: is obtained by using microcanonical distributions instead
121: of Wigner functions.
122:
123:
124:
125:
126: Fixing $n$, the vector $P(n|m)$ describes
127: the shape of the $n$-th eigenstate in the
128: ${\cal H}_0 = {\cal H}(Q,P;x_0)$ representation.
129: By averaging over several eigenstates one obtains the
130: average shape of the eigenstate (ASOE).
131: We can also identify $P(n|m)$ as the
132: local density of states (LDOS),
133: by regarding it as a function of $n$,
134: where $m$ is considered to be a fixed
135: reference state. In the latter case
136: an average over few $m$-states is assumed.
137: We shall denote the LDOS by $P(r)$ where
138: $r=(n-m)$. The ASOE is just $P(-r)$.
139: Note that the ASOE and the LDOS are
140: given by the same function. One would
141: have to be more careful with these
142: definitions if ${\cal H}_0$ were
143: integrable while ${\cal H}$ non-integrable.
144:
145:
146:
147: A few words are in order regarding the definition
148: of the LDOS, and its importance in physical applications.
149: The LDOS, also known as strength function \cite{casati,flamb,felix},
150: describes an energy distribution. Conventionally
151: it is defined as follows:
152: %
153: \begin{eqnarray} \label{e_2}
154: \mbox{LDOS}(E) \ \ &=& \ \
155: -\frac{1}{\pi} \ \langle m | \Im \mbf{G}(E) | m \rangle
156: \nonumber \\
157: \ \ &=& \ \ \sum_n P(n|m) \ \delta(E-E_n)
158: \end{eqnarray}
159: %
160: where $\mbf{G}(E)=1/(E-{\cal H}+i0)$ is the
161: retarded Green function. We are interested in
162: chaotic systems, so it should be clear that
163: our $P(r)$ is related by trivial change of variable
164: ($E \mapsto r$) to the above defined $\mbox{LDOS}(E)$.
165: Our $P(r)$ also incorporates an average over
166: the reference state. The LDOS is important in studies
167: of either chaotic or complex conservative quantum
168: systems that are encountered in nuclear physics
169: as well as in atomic and molecular physics. Related
170: applications may be found in mesoscopic physics.
171: Going from ${\cal H}_0$ to ${\cal H}$ may signify
172: a physical change of an external field,
173: or switching on of a perturbation,
174: or a sudden-change of an effective-interaction
175: (as in molecular dynamics \cite{heller}).
176: The so called `line shape' of the LDOS is important
177: for the understanding of the associated dynamics.
178: It is also important to realize that the LDOS
179: is the Fourier transform of the so-called
180: `survival probability amplitude' \cite{heller}
181: (see \cite{wls} for concise presentation of this point).
182:
183:
184: \subsection{Parametric Evolution}
185:
186: Textbook \cite{ct} formulations of
187: perturbation theory can be applied in order to find
188: the LDOS. Partial summations of diagrams to infinite order
189: can be used in order to get an improved
190: Lorentzian-type approximation. However most
191: textbooks do not illuminate the limitations
192: and the subtleties which are involved in
193: using the conventional perturbative schemes.
194: It is therefore interesting to take a
195: somewhat different approach to the
196: study of LDOS. The roots of this alternate approach
197: can be traced back to the work of Wigner \cite{wigner}
198: regarding a simple banded random matrix (BRM) model
199: ${\cal H}=\mbf{E}+\delta x\mbf{B}$.
200: Here $\mbf{E}$ is a diagonal matrix
201: whose elements are the ordered energies $\{E_n\}$,
202: and $\mbf{B}$ is a banded matrix. The study of this
203: model can be motivated by the realization that
204: in generic circumstances it is possible to write
205: ${\cal H}(Q,P;x) \approx {\cal H}_0(Q,P) + \delta x {\cal F}(Q,P)$.
206: Using a simple semiclassical argument \cite{mario}
207: it turns out that the matrix representation of any
208: generic ${\cal F}$, in the eigen-basis that is
209: determined by the chaotic Hamiltonian ${\cal H}_0$,
210: is a banded matrix.
211:
212:
213: The important ingredient (from our point of view)
214: in the original work by Wigner, is the emphasis
215: on the parametric-evolution (PE) of the LDOS.
216: The LDOS describes an energy distribution:
217: For $\delta x=0$ the kernel $P(r)$ is simply
218: a Kroneker delta function. As $\delta x$ becomes
219: larger, the width as well as the whole profile of
220: this distribution `evolves'.
221: Wigner has realized that for his WBR model there
222: are three parametric regimes. For very small
223: $\delta x$ we have the standard perturbative
224: structure where most of the probability is
225: concentrated in $r=0$. For larger $\delta x$
226: we have a Lorentzian line shape. But this
227: Lorentzian line shape does not persist if we
228: further increase $\delta x$. Instead we get
229: a semi-circle line shape. Many works about
230: the LDOS have followed \cite{casati,flamb,felix},
231: but the issue of PE has not been further
232: discussed there. The emphasis in those works
233: is mainly on the case where ${\cal H}_0$ is
234: an integrable or non-interacting system, while
235: ${\cal H}$ is possibly (but not necessarily)
236: chaotic due to some added perturbation term.
237:
238:
239: The line of study which is pursued in the present
240: work has been originated and motivated by studies
241: of quantum dissipation \cite{crs,vrn,frc}.
242: Understanding PE can be regarded as a preliminary
243: stage in the analysis of the energy spreading
244: process in driven mesoscopic systems. Note that
245: the LDOS gives the energy re-distribution due to
246: a `sudden' (very fast) change of the Hamiltonian.
247: Unlike the common approach for studies of LDOS,
248: we assume both ${\cal H}$ and ${\cal H}_0$ to be
249: chaotic. Both correspond to the same
250: parametrically dependent Hamiltonian
251: ${\cal H}(Q,P;x)$, and there is nothing
252: special in choosing a particular value $x=x_0$
253: as a starting point for the PE analysis.
254:
255:
256:
257: \subsection{Main results}
258:
259:
260: The theory of PE, as discussed
261: in general in \cite{crs,vrn,frc} and in
262: particular in \cite{wls,lds} takes us beyond
263: the random-matrix-theory considerations
264: of Wigner. There appear five (rather than two)
265: different parametric scales (see remark \cite{rmrk2}).
266: These are summarized by Table 1.
267:
268:
269: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
270: \begin{table}[h]
271: %
272: \begin{tabular}{lll}
273: %
274: $\delta x_c^{\tbox{cl}}$ & Is it possible to linearize ${\cal H}(x+\delta x)$? \\
275: $\delta x_c^{\tbox{qm}}$ & Is it possible to use standard perturbation theory?\\
276: $\delta x_{\tbox{prt}}$ & Do perturbative tail regions survive? \\
277: $\delta x_{\tbox{NU}}$ & Do non-universal core features show up? \\
278: $\delta x_{\tbox{SC}}$ & Is it possible to use semiclassical approximation? \\
279: %
280: \end{tabular}
281: %
282: \vspace{.1in}
283: %
284: \caption{
285: The parametric scales in the general theory of PE
286: are listed (left column) along with the questions
287: that motivate their introduction.
288: The distribution $P(r)$ may contain perturbative tail regions
289: (for $\delta x \ll \delta x_{\tbox{prt}}$),
290: and non-perturbative core regions
291: (for $\delta x > \delta x_c^{\tbox{qm}}$).
292: Non-universal (system specific) features may manifest themselves in
293: the core structure for $\delta x \gg \delta x_{\tbox{NU}}$.
294: In generic examples $\delta x \gg \delta x_{\tbox{SC}}$
295: allows classical approximation for $P(r)$.
296: We are going to explain that only two independent
297: parametric scales survive in the hard wall limit. }
298: %
299: \end{table}
300: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
301:
302:
303:
304: In the present paper we consider cavities with hard walls.
305: We are going to explain that because of the
306: `hard wall limit' there are only {\em two}
307: independent parametric scales:
308: One is $\delta x_c^{\tbox{qm}}$ and the others
309: (see remark \cite{rmrk3})
310: coincide with $\delta x_{\tbox{NU}}$.
311: Assuming that $\delta x_c^{\tbox{qm}}$ and
312: $\delta x_{\tbox{NU}}$ are well separated,
313: it follows that there are {\em three} distinct
314: parametric regimes in the PE of our system.
315: These are the standard perturbative regime
316: ($\delta x \ll \delta x_c^{\tbox{qm}}$),
317: the core-tail regime
318: ($\delta x_c^{\tbox{qm}} \ll \delta x \ll \delta x_{\tbox{NU}}$),
319: and the non-universal regime
320: ($\delta x \gg \delta x_{\tbox{NU}}$).
321:
322:
323: The exploration of the three parametric regimes
324: in the PE of a deformed cavity with hard walls is
325: the main issue of the present paper. To the best
326: of our knowledge such detailed exploration has
327: not been practical in the past. We owe our
328: ability to carry out this task to a new
329: powerful technique for finding clusters
330: of billiard eigenstates \cite{vergini,alexthesis}.
331: %
332: There are also some secondary
333: issues that we are going to address:
334:
335: {\bf (a)} In the strict limit of hard walls the PE becomes
336: non-unitarity. We shall use the 1D well example
337: in order to shed light on this confusing issue.
338: In particular we demonstrate that any truncation
339: of the PE equation leads to false unitarity due to
340: a finite-size edge effect.
341:
342: {\bf (b)} For special deformations, namely those that constitute
343: linear combination of translations rotations and dilations,
344: the parametric scales $\delta x_c^{\tbox{qm}}$ and
345: $\delta x_{\tbox{NU}}$ coincide. Consequently
346: there is no longer distinct core-tail regime, and
347: the PE features a quite sharp transition from the
348: standard perturbative regime to the non-universal regime.
349:
350: {\bf (c)} In the non-universal regime we demonstrate
351: that our numerical results are in accordance with
352: our theoretical expectation \cite{wls}.
353: Namely, the width of the LDOS profile is determined
354: by time-domain semiclassical considerations, rather
355: then by phase-space or random-wave considerations.
356:
357: {\bf (d)} The last section puts our specific study
358: in a larger context. We explain why Wigner's scenario
359: of PE is not followed once hard walls are considered.
360:
361:
362: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
363: \section{The cavity system}
364:
365:
366: We consider a particle moving inside $d$ dimensional cavity
367: whose volume is $\mathsf{V}$.
368: The kinetic energy of the particle is $E=\half mv^2$,
369: where $m$ is its mass, and $v$ is its velocity.
370: It is assumed that this motion is classically chaotic.
371: The ballistic mean free path is $\ell_{\tbox{bl}}$.
372: One can use the estimate
373: $\ell_{\tbox{bl}} \sim \mathsf{V}/\mathsf{A}$,
374: where $\mathsf{A}$ is the total area of the walls.
375: The associated time scale is
376: $\tau_{\tbox{bl}} = \ell_{\tbox{bl}}/v$.
377:
378:
379:
380: The penetration distance upon a collision is $\ell=E/f$,
381: where $f$ is the force that is exerted by the wall.
382: Upon quantization we have an additional length
383: scale, which is the De-Broglie wavelength
384: $\lambda_{\tbox{B}}=2\pi\hbar/(mv)$. We shall distinguish
385: between the {\em hard walls} case where we assume
386: $\ell < \lambda_{\tbox{B}} \ll \ell_{\tbox{bl}}$,
387: and {\em soft walls} for which $\lambda_{\tbox{B}} \ll \ell$.
388: Note that taking $\hbar\rightarrow 0$ implies soft walls.
389:
390:
391:
392: There is a class of special deformations that are
393: shape-preserving. These are generated by translations,
394: rotations and dilations of the cavity.
395: A general deformation need not preserve the
396: billiard shape nor its volume.
397: We can specify any deformation by a function $D(\mbf{s})$,
398: where~$\mbf{s}$ specifies the location of a wall element
399: on the boundary (surface) of the cavity,
400: and $D(\mbf{s})\delta x$ is the normal displacement of
401: this wall element.
402: In many practical cases it is possible to use the convention
403: $|D(\mbf{s})| \sim 1$. With this convention $\delta x$ has
404: units of length, and its value has the meaning of typical
405: wall displacement.
406:
407:
408: The eigen-energies of a particle inside the cavity
409: are in general $x$-dependent, and can be written
410: as $E_n = (\hbar k_n)^2/(2m)$. The mean level spacing is
411: %
412: \begin{eqnarray} \label{e2}
413: \Delta = \hbar v \times
414: \frac{2\pi}{\Omega_d}
415: \ \frac{1}{{\mathsf V}}
416: \lambda_{\tbox{B}}^{d{-}1}
417: \end{eqnarray}
418: %
419: where $\Omega_d=2\pi,4\pi,\cdots$ for $d=2,3,\cdots$.
420: In our numerical study we shall consider a quarter
421: stadium with curved edge of radius $1$ and straight
422: edge of length $1$. The `volume' of the quarter stadium
423: is ${\mathsf V}=1+\pi/4$. The 'area' of its boundary
424: ${\mathsf A}=4+\pi/2$ is just the perimeter.
425: We shall look on the parametric evolution of eigenstates around
426: $k\sim 400$ where the mean level spacing in $k$ units
427: is $\tilde{\Delta}=\Delta/(\hbar v)\approx 0.0088$.
428:
429:
430:
431:
432: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
433: \section{Parametric evolution}
434:
435:
436:
437: Consider the quantum-mechanical state $\psi=|m(x_0)\rangle$.
438: We can write $\psi = \sum_n a_n(x) |n(x)\rangle$.
439: The parametric kernel can be written as
440: $P(n|m)=|a_n(x)|^2$. It is a standard exercise
441: to obtain (from $d\psi/dx=0$ and differentiating by parts)
442: the following equation for the amplitudes:
443: %
444: \begin{eqnarray} \label{e3}
445: \frac{da_n}{dx} \ = \
446: -\frac{i}{\hbar}\sum_{m} \mbf{W}_{nm}(x) \ a_{m}
447: \end{eqnarray}
448: %
449: In order to get $P(n|m)$ one should solve this
450: equation with the initial conditions $a_n(x_0)=\delta_{nm}$.
451: The transitions between
452: levels are induced by the matrix elements
453: %
454: \begin{eqnarray} \label{e4}
455: {\mathbf W}_{nm} =
456: \frac{i\hbar}{E_n{-}E_m}
457: \left(\frac{\partial{\cal H}}{\partial x}\right)_{nm}
458: \end{eqnarray}
459: %
460: and we use the `gauge' convention
461: ${\mathbf W}_{nm}{=}0$ for $n{=}m$.
462: (Only one parameter is being changed and
463: therefore Berry's phase is not an issue).
464:
465:
466: Eq.(\ref{e3}) is a possible starting point for
467: constructing a perturbation theory for the PE of $P(r)$.
468: See Ref.\cite{frc} for more details.
469: As an input for this equation we need the
470: matrix elements of ${\partial{\cal H}}/{\partial x}$.
471: These can be calculated using a simple
472: boundary integral formula \cite{berry} whose
473: simplest derivation \cite{frc} is as follows:
474: %
475: The position of the particle in the vicinity
476: of a wall element is $Q=(z,\mbf{s})$,
477: where $\mbf{s}$ is a surface coordinate
478: and $z$ is a perpendicular `radial' coordinate.
479: We take $f=\infty$ so that
480: %
481: \begin{eqnarray} \label{e5}
482: \frac{\partial {\cal H}}{\partial x} \ = \
483: - D(\mbf{s}) \ V_0 \ \delta(z)
484: \end{eqnarray}
485: %
486: The logarithmic derivative of the wavefunction
487: on the boundary is $\varphi(\mbf{s})/\psi(\mbf{s})$ where
488: $\varphi(\mbf{s})=\mbf{n}{\cdot}\nabla\psi$, and
489: $\mbf{n}$ is a unit vector in the $z$ direction.
490: For $z>0$ the wavefunction $\psi(Q)$ is a decaying
491: exponential. If $V_0$ is large enough, then the
492: exponential decay is fast, and we can treat
493: the boundary as if it were locally flat.
494: It follows that the logarithmic derivative of the wavefunction
495: on the boundary should be equal to $-\sqrt{2mV_0}/\hbar$.
496: Consequently one obtains the following expression for the
497: matrix elements:
498: %
499: \begin{eqnarray} \label{e6}
500: \left(\frac{\partial {\cal H}}{\partial x}\right)_{nm}
501: \ = \
502: -\frac{\hbar^2}{2m}\oint \varphi_n(\mbf{s})\varphi_m(\mbf{s})
503: \ D(\mbf{s}) d\mbf{s}
504: \end{eqnarray}
505: %
506: In the one-dimensional case the boundary integral
507: is replace by the sum $\sum_s \varphi_n(s) \varphi_m(s) D(s) $
508: where $s=1,2$ are the two turning points of the
509: potential well.
510:
511:
512:
513: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
514: \section{Hard walls and non-Unitarity}
515:
516: For the purpose of the following argumentation
517: it is convenient to take $f=\infty$, but to
518: keep $V_0$ large but finite.
519: Mathematically it is also convenient to think
520: of the embedding space as having some huge but
521: finite volume.
522: We would like to illuminate a subtlety which is associated
523: with the hard wall limit $V_0\rightarrow\infty$.
524: For any finite $V_0$ the parametric kernel satisfies
525: %
526: \begin{eqnarray} \label{e7}
527: \sum_n P(n|m) = p_{\tbox{total}}
528: \end{eqnarray}
529: %
530: with $p_{\tbox{total}}=1$. This follows from the
531: fact that $|n(x)\rangle$ is a complete orthonormal
532: basis for any $x$. However, for hard walls ($V_0=\infty$)
533: this statement is not true. This implies that
534: for hard walls the PE in non-unitary.
535: We are going to explain this point below.
536:
537:
538: Let us denote the volume of the original cavity
539: by ${\mathsf V}_0$ and of the deformed cavity
540: by ${\mathsf V}$. The volume shared by the deformed
541: and the undeformed cavities will be denoted by
542: ${\mathsf V}_0\cap{\mathsf V}$ and we shall use
543: the notation $\eta=({\mathsf V}_0\cap{\mathsf V})/{\mathsf V}_0$.
544: For the purpose of the following argumentation
545: let us consider a reference state $m$ whose energy
546: $E_m$ is well below $V_0$. Let us also assume
547: that the wall displacement is large compared to De-Broglie
548: wavelength. Consequently the expression for
549: $P(n|m)$ has the following semiclassical structure:
550: %
551: \begin{eqnarray} \label{e_7}
552: P(n|m) \ \approx \ \eta \times f(E_n{-}E_m)
553: + (1{-}\eta)\times g(E_n{-}E_c)
554: \end{eqnarray}
555: %
556: where $E_c = (V_0+E_m)\sim V_0$. The above result
557: can be deduced by assuming that the wavefunctions
558: look ergodic in space, but still that they are characterized
559: by a well-defined {\em local} wavelength.
560: An equivalent derivation is obtained by using
561: the phase-space picture of \cite{wls}.
562: %
563: Both $f$ and $g$ in the above expression have unit
564: normalization, and therefore $p_{\tbox{total}}=1$
565: for any finite $V_0$. However, for hard walls
566: ($V_0=\infty$) we have $E_c=\infty$ and therefore
567: $p_{\tbox{total}}=\eta$. We may say the the operation
568: of taking the hard wall limit does not commute
569: with the summation in Eq.(\ref{e7}).
570: An analogous statement can be derived regarding
571: the summation $\sum_m P(n|m)$, with the respective
572: definition $\eta=({\mathsf V}_0\cap{\mathsf V})/{\mathsf V}$.
573:
574:
575:
576: The correctness of the above observation
577: becomes less trivial if we consider (\ref{e3})
578: with expressions (\ref{e4}) and (\ref{e6})
579: substituted for the matrix elements.
580: Looking on (\ref{e4}) with (\ref{e6}) it looks
581: as if the matrix ${\mathbf W}_{nm}$ is
582: hermitian, and therefore should generate
583: unitary PE. But this statement is
584: mathematically correct only for (any) finite
585: truncation $N$ of the PE equation.
586: For $N=\infty$ the matrix ${\mathbf W}_{nm}$
587: becomes non-hermitian. It turns out that
588: for any finite $N$, there is a pile-up
589: of probability in the edges of the spreading
590: profile, due to finite-size effect.
591: We shall demonstrate this effect in the next
592: section using a simple 1D example.
593: In other words, if we solve Eq.(\ref{e3})
594: for hard-walled cavity, we get as a result Eq.(\ref{e_7})
595: with $E_c=E_N$. For $N=\infty$ we get $E_c=\infty$
596: and therefore $p_{\tbox{total}}=\eta$ in accordance
597: with the conclusion of the previous paragraph.
598:
599:
600:
601: Thus if either $V_0<\infty$ or $N<\infty$ then we have
602: unitary PE. But for hard walls, meaning
603: $V_0=\infty$ with $N=\infty$, we have non-unitary PE.
604: The lost probability is associated with the second
605: term in Eq.(\ref{e_7}). This term is peaked around
606: a high energy $E_c$. For hard walls $E_c=\infty$ and
607: consequently some probability is lost.
608: The above picture is supported by the simple
609: pedagogical example of the next section.
610:
611:
612:
613: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
614: \section{Parametric evolution for a 1D box}
615:
616:
617: Consider a 1D box with hard walls,
618: where the free motion of the
619: particle is within $0<Q<{\mathsf a}$.
620: The eigenstates of the Hamiltonian are
621: %
622: \begin{eqnarray} \label{e8}
623: |n({\mathsf a})\rangle \ \ \longrightarrow \ \
624: (-1)^n \sqrt{\frac{2}{{\mathsf a}}} \ \sin(k_nQ)
625: \end{eqnarray}
626: %
627: where $k_n=n\times(\pi/{\mathsf a})$ is the wavenumber,
628: and $n=1,2,...$ is the level index.
629: The phase factor $(-1)^n$ has been introduced
630: for convenience. We consider now the
631: parametric evolution as a function of ${\mathsf a}$.
632: One easily obtains
633: %
634: \begin{eqnarray} \label{e9}
635: \langle n({\mathsf a}) | m({\mathsf a}_0) \rangle \ = \
636: (-1)^{n} \sqrt{\eta} \
637: \frac{\sin(\pi\eta n)}{\pi}
638: \frac{2m}{\eta^2 n^2 - m^2}
639: \end{eqnarray}
640: %
641: where $\eta\equiv{\mathsf a}_0/{\mathsf a}$ is assumed to be
642: smaller than $1$, corresponding to expansion
643: of the box. The probability kernel is
644: $P(n|m)=|\langle n | m \rangle|^2$.
645: One can verify that the parametric
646: evolution in the ${\mathsf a}_0\mapsto{\mathsf a}$ direction
647: is unitary, meaning that $\sum_n P(n|m) = 1$.
648: On the other hand in the ${\mathsf a}\mapsto{\mathsf a}_0$
649: direction the parametric evolution is
650: non-unitary, because $\sum_m P(n|m) = \eta$.
651: The profile of $P(n|m)$ for fixed $n$ is illustrated
652: by a dashed line in Fig.2.
653:
654:
655: \begin{figure}
656: %%%%%%%%%%%%%%%%%%%%%%%%%%%
657: \centerline{\epsfig{figure=image_box.eps,width=\hsize}}
658: \centerline{\epsfig{figure=profile_box.eps,width=\hsize}}
659: \vspace{.1in}
660: \caption{
661: {\em Upper subfigure}: An image of the kernel $P(n|m)$
662: for $13\%$ expansion of the box (ie ${\mathsf a}/{\mathsf a}_0=1.13$).
663: The kernel has been calculated numerically using
664: Eq.(\ref{e11}) with finite truncation $N=256$.
665: {\em Lower subfigure}: The profile of a representative
666: row of $P(n|m)$. The dashed line is the $N=\infty$
667: analytical result using Eq.(\ref{e9}).
668: }
669: %%%%%%%%%%%%%%%%%%%%%%%%%%%
670: \end{figure}
671:
672:
673:
674: We can restore unitarity by making $V_0$ large
675: but finite. In such case, a variation of the above
676: calculation leads to the following picture:
677: Consider the overlap of a reference level $n({\mathsf a})$
678: with the levels $m({\mathsf a}_0)$. As in the
679: case $V_0=\infty$ there is a probability $\eta$
680: which is located in the levels whose energies are
681: $E_m \approx E_n$. But now the ``lost'' probability
682: $(1-\eta)$ is located in the levels
683: whose energies are $E_m \approx E_n + V_0$.
684: Thus we have $\sum_m P(n|m) = 1$ rather than
685: $\sum_m P(n|m) = \eta$.
686:
687:
688: We consider again the case $V_0=\infty$.
689: The normal derivative on the
690: boundary is $\varphi_n({\mathsf a})=\sqrt{(2/{\mathsf a})}\ k_n$.
691: Hence we can easily get the following result
692: %
693: \begin{eqnarray} \label{e10}
694: \frac{1}{\hbar}\mbf{W}_{nm} \ = \
695: \frac{-i}{k_n^2-k_m^2}\varphi_n({\mathsf a})\varphi_m({\mathsf a}) \ = \
696: -i \frac{1}{{\mathsf a}}\frac{2nm}{n^2-m^2}
697: \end{eqnarray}
698: %
699: It is more convenient to use
700: $\alpha=\ln({\mathsf a})$ for parameterization.
701: Hence the equation that describes the
702: parametric evolution is
703: %
704: %
705: \begin{eqnarray} \label{e11}
706: \frac{da_n}{d\alpha} \ = \
707: -\sum_m \frac{2nm}{n^2-m^2} \ a_m
708: \end{eqnarray}
709: %
710: For any finite truncation $N<\infty$ this equation
711: manifestly generates unitary parametric evolution.
712: It is only for $N=\infty$ that it becomes
713: equivalent to the non-unitary evolution of
714: the 1D box. Again, one can wonder where the
715: `lost' probability is located if $N<\infty$.
716: The answer is illustrated in Fig.2. We see that
717: the `lost' probability piles up at the edge
718: of the (truncated) tail.
719:
720:
721:
722: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
723: \section{Matrix elements for chaotic cavity}
724:
725:
726:
727: It is possible to use semiclassical considerations \cite{mario}
728: in order to determine the band profile of the matrix Eq.(\ref{e6}).
729: The application to the cavity example has been introduced
730: in \cite{frc}, and numerically demonstrated in \cite{dil}.
731: The accuracy of this semiclassical estimate is remarkable.
732: Here we summarize the recipe. First one should generate
733: a very long (ergodic) trajectory,
734: and define for it the fluctuating quantity
735: %
736: \begin{eqnarray} \label{e12}
737: {\cal F}(t) = -\frac{\partial {\cal H}}{\partial x} =
738: \sum_{\tbox{col}} 2mv \ \cos(\theta_{\tbox{col}})
739: \ D_{\tbox{col}} \ \delta(t-t_{\tbox{col}})
740: \end{eqnarray}
741: %
742: where $t_{\tbox{col}}$ is the time of a collision,
743: $D_{\tbox{col}}$ stands for $D(\mbf{s})$
744: at the point of the collision, and $v\cos(\theta_{\tbox{col}})$
745: is the normal component of the particle's velocity.
746: Each delta spike (for soft walls it is actually a narrow
747: rectangular spike) corresponds to one collision.
748: Now one can calculate the correlation function
749: $C(\tau)$ of the fluctuating quantity ${\cal F}(t)$,
750: and its Fourier transform $\tilde{C}(\omega)$.
751: The semiclassical estimate for the band profile is
752: %
753: \begin{eqnarray} \label{e13}
754: \left\langle\left|\left(
755: \frac{\partial {\cal H}}{\partial x}\right)_{nm}
756: \right|^2\right\rangle
757: \ \ \approx \ \
758: \frac{\Delta}{2\pi\hbar} \
759: \tilde{C}\left(\frac{E_n{-}E_m}{\hbar}\right)
760: \end{eqnarray}
761:
762:
763:
764: Ref.\cite{wlf} contains a systematic study of the
765: function $\tilde{C}(\omega)$. For large $\omega$,
766: meaning $\omega \gg 1/\tau_{\tbox{bl}}$, one can use
767: %
768: \begin{eqnarray} \label{e14}
769: \tilde{C}(\omega) \ \approx \
770: 2 m^2 v^3 \langle|\cos\theta|^3\rangle
771: \frac{1}{{\mathsf V}} \oint [D(\mbf{s})]^2 d\mbf{s}
772: \end{eqnarray}
773: %
774: where the geometric factor is
775: $\langle|\cos(\theta)|^3\rangle = 1, 4/3\pi, \cdots$
776: for $d=1,2, \cdots$.
777: %
778: A lengthy calculation \cite{frc} reveals that
779: Eq.(\ref{e13}) with (\ref{e14}) substituted,
780: is an exact global result if we could assume
781: that the cavity eigenfunctions look like `random waves',
782: and that different wavefunctions are uncorrelated.
783: However, it turns out that to take this `random wave'
784: result as a global approximation is an over-simplification.
785: For $\omega \ll 1/\tau_{\tbox{bl}}$, using the semiclassical
786: recipe and assuming strongly chaotic cavity, one obtains
787: %
788: \begin{eqnarray} \label{e15}
789: \tilde{C}(\omega) \ \approx \
790: \tilde{C}(\infty) \times (\tau_{\tbox{bl}} \omega)^{\gamma}
791: \end{eqnarray}
792: %
793: with $\gamma=4$ for dilations and translations,
794: $\gamma=2$ for rotations, and $\gamma=0$ for
795: normal deformations. We use the term
796: 'special deformations' \cite{wlf} in order
797: to distinguish those deformations that has
798: the property $\tilde{C}(\omega) \rightarrow 0$
799: in the limit $\omega\rightarrow0$.
800: Any combination of dilations, translations and
801: rotations is a special deformation.
802: %
803: Around the bouncing frequency ($\omega \sim 1/\tau_{\tbox{bl}}$)
804: the function $\tilde{C}(\omega)$ typically displays
805: some non-universal (system and deformation specific) structure.
806: This is true for any typical deformation, but for some
807: deformations the non-universal features are more pronounced.
808: If the cavity has bouncing ball modes, we may get also a modified
809: (non-universal) behavior in the small frequency limit.
810: %
811: For the purpose of general discussion
812: it is convenient to assume that the interpolation
813: between (\ref{e15}) and (\ref{e14}) is
814: smooth, but in actual numerical calculation the
815: actual $\tilde{C}(\omega)$ is computed (see below).
816:
817:
818:
819:
820: \begin{figure}
821: %%%%%%%%%%%%%%%%%%%%%%%%%%%
822: \centerline{\epsfig{figure=bandprof_rota.eps,width=\hsize}}
823: \centerline{\epsfig{figure=bandprof_defo.eps,width=\hsize}}
824: \vspace{.1in}
825: \caption{
826: Band profiles for deformations of the quarter stadium
827: as defined by Eq.(\ref{e16}).
828: {\em Upper plot:} Rotation around the stadium center.
829: {\em Lower plot:} Generic (non-special) deformation
830: involving displacement of the curved edge.
831: It is important to notice that for the special
832: deformation we have $\tilde{C}(\omega)\rightarrow0$
833: in the limit $\omega\rightarrow0$. }
834: %%%%%%%%%%%%%%%%%%%%%%%%%%%
835: \end{figure}
836:
837:
838:
839: As a numerical example we have picked the
840: stadium billiard. We have found the eigenstates
841: of a de-symmetrized (quarter) stadium as described
842: in Ref.\cite{dil}. We have selected those
843: eigenstates whose eigen-energies $k_n$ are
844: in the vicinity of $k=400$.
845: %
846: Our two representative deformations are:
847: {\bf (a)} rotation around the stadium center;
848: and {\bf (b)} generic (non-special) deformation
849: involving the curved edge. In the latter
850: case the curved edge of the quarter stadium
851: ($0<s<\pi/2$) is pushed outwards with $D(s)=(\cos(s))^2$,
852: while for the straight edges $D(s)=0$.
853: (The corner $s=0$ is the $90^{\circ}$ intersection
854: of the curved edge with the long straight edge).
855: The respective band profiles are displayed in Fig.3.
856: The band profile has been defined as
857: %
858: \begin{eqnarray} \label{e16}
859: B(\kappa) \ = \ \frac{1}{4k^2} \
860: \left\langle \left|
861: \oint \varphi_n(\mbf{s})\varphi_m(\mbf{s})
862: \ D(\mbf{s}) d\mbf{s}
863: \right|^2\right\rangle
864: \end{eqnarray}
865: %
866: where $\kappa=(k_n-k_m)$ is the distance from
867: the diagonal. Note that $B(\kappa)$ is just
868: a scaled version of the semiclassical $\tilde{C}(\omega)$
869: as implied by Eq.(\ref{e13}) with (\ref{e6}).
870: The remarkable agreement of $B(\kappa)$ with the
871: semiclassical calculation has been demonstrated
872: in \cite{dil,wlf}.
873:
874:
875: It is important to realize that in the hard wall limit
876: (which is assumed here) the matrix
877: $({\partial {\cal H}}/{\partial x})_{nm}$
878: is not a banded matrix. It would become banded
879: if we were assuming {\em soft} walls.
880: For soft walls $\tilde{C}(\omega)$ becomes vanishingly small
881: for $\omega \gg v/\ell$. The bandwidth in energy units is
882: $\Delta_b=\hbar v/\ell$, and in dimensionless
883: units it is
884: %
885: \begin{eqnarray} \label{e17}
886: b \ \ = \ \ \frac{\Delta_b}{\Delta} \ \ = \ \
887: \frac{{\mathsf V}}{\ell \ \lambda_{\tbox{B}}^{d{-}1}}
888: \end{eqnarray}
889: %
890: Unless stated otherwise we have $b=\infty$.
891:
892:
893:
894: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
895: \section{Parametric evolution - numerical results}
896:
897:
898: The parametric evolution of $P(r)$ for rotation
899: and for generic deformation of the stadium is illustrated
900: by the images of Fig.4 and by the plots of Figs.5-6.
901: The calculation of each $P(r)$ profile is carried
902: out as follows: Given $\delta x$ we use the method
903: which is described in \cite{alexthesis} in order to calculate
904: the matrix $P(n|m)$. Then we plot the elements
905: of $P(n|m)$ versus $\kappa = (k_n(x)-k_m(0))$.
906: In order to obtain the average profile the plot
907: is smeared using standard procedure (see remark \cite{note}).
908: The transformation from $\kappa$ to $r=(n-m)$ is
909: done using the relation (see remark \cite{note}):
910: %
911: \begin{eqnarray} \label{e18}
912: \kappa \ \ = \ \ \tilde{\Delta} \cdot r \ - \
913: \frac{1}{d} k \times \frac{\delta {\mathsf V}}{{\mathsf V}}
914: \end{eqnarray}
915: %
916: Above $\tilde{\Delta}$ is the mean level spacing
917: of the $\{k_n\}$ spectrum, and $\delta {\mathsf V}$
918: is the volume change that is associated with
919: the deformation (it is approximately proportional
920: to $\delta x$). If the deformation is volume preserving
921: (as in the case of rotation) then the second term
922: equals zero. But for the generic deformation that
923: we have picked in our second numerical example,
924: the volume is not preserved, and the systematic
925: `downwards' shift of the levels should be taken
926: into account.
927:
928:
929:
930: \begin{figure}
931: %%%%%%%%%%%%%%%%%%%%%%%%%%%
932: %\epsfysize=2.7in
933: %\epsffile{image_avprof_bump.eps}
934: \centerline{\epsfig{figure=image_rota.eps,width=\hsize}}
935: \centerline{\epsfig{figure=image_defo.eps,width=\hsize}}
936: %
937: \vspace{.1in}
938: \caption{
939: Each column is an image of $P(r)$ versus $\kappa$
940: for a different value of $\delta x$. There are 41 columns.
941: The value $\delta x \sim 1$ corresponds roughly
942: to $\lambda_{\tbox{B}}$ displacement of the boundary.
943: The upper figure is for rotation, and the lower
944: is for generic deformation. Note that the $r=0$
945: component is excluded from the image. Instead
946: we have plotted over the image an $r=0$ line
947: wherever this component contains more than $50\%$
948: of the probability. In the lower figure this line
949: cannot be resolved from the developing core region.}
950: %%%%%%%%%%%%%%%%%%%%%%%%%%%
951: \end{figure}
952:
953:
954:
955: Looking first in the case of rotation,
956: we see clearly two parametric regimes:
957: The standard perturbative regime ($\delta x<0.2$),
958: and the non-universal regime ($\delta x>0.2$).
959: Let us clarify this observation.
960: %
961: We see that for $\delta x<0.2$ most of the
962: probability is well concentrated in $r=0$.
963: This implies that we can use standard perturbation
964: theory in order to estimate the $r \ne 0$ probabilities.
965: On the other hand for $\delta x>0.2$ the perturbative
966: nature of $P(r)$ is destroyed. Now $P(r)$ becomes smoother,
967: and eventually (for $\delta x>0.5$) there is a very good
968: fitting with Lorentzian (see lower plot in Fig.5).
969:
970:
971:
972: The qualitative explanation for the
973: Lorentzian profile is as follows:
974: For $\delta x>0.5$ the typical displacement
975: of the walls is of the order of $\lambda_{\tbox{B}}$.
976: Therefore the $|n(x)\rangle$ eigenstates
977: become uncorrelated with the $|m(0)\rangle$
978: eigenstates. Consequently $P(r)$ becomes
979: $\delta x$ independent. The Lorentzian profile
980: agrees with the assumption of uncorrelated
981: random waves as explained in Appendix A.
982:
983:
984: Let us look now in the case of generic deformation.
985: Here we see clearly three parametric regimes:
986: The standard perturbative regime ($\delta x<0.004$),
987: the core-tail regime ($0.004< \delta x < 0.2$),
988: and the non-universal regime ($\delta x>0.2$).
989: Let us clarify this observation.
990: %
991: As in the case of rotation there is a standard
992: perturbative regime ($\delta x<0.004$) where most
993: of the probability is well concentrated in $r=0$.
994: %
995: For larger deformation, namely for $\delta x>0.004$,
996: standard perturbation theory is no longer
997: applicable because the $n=m$ level is mixed
998: non perturbatively with other (neighboring) levels.
999: As a result $P(r)$ contains a non-perturbative
1000: `core' component. However, for $0.004< \delta x < 0.2$
1001: we definitely do not get a Lorentzian. Rather the
1002: tails of $P(r)$ keep growing in the same way
1003: as in the standard perturbative regime.
1004:
1005:
1006: In the case of the generic deformation,
1007: as in the case of rotation,
1008: we enter the non-universal regime,
1009: and eventually (for $\delta x>0.6$)
1010: we get a smooth Lorentzian-like distribution.
1011: However, the Lorentzian-like distribution
1012: is not identical with that of the rotation case.
1013: Also the similarity to proper Lorentzian is
1014: far from being satisfactory. (see lower plot in Fig.6).
1015: This means that the random-wave picture of Appendix A
1016: is an oversimplification.
1017:
1018:
1019: In the following sections we are going to
1020: summarize the theoretical considerations \cite{wls}
1021: that explain the observed parametric scenario.
1022: In particular we are going to illuminate the way
1023: in which non-perturbative features emerge;
1024: to clarify the crossover to the non-universal
1025: regime; and to explain the specific nature of
1026: the non-universal distribution.
1027:
1028:
1029:
1030: \begin{figure}[p]
1031: %%%%%%%%%%%%%%%%%%%%%%%%%%%
1032: %\epsfysize=2.7in
1033: %\epsffile{image_avprof_bump.eps}
1034: \centerline{\epsfig{figure=prof_rota_1.eps,width=\hsize}}
1035: \centerline{\epsfig{figure=prof_rota_2.eps,width=\hsize}}
1036: \centerline{\epsfig{figure=prof_rota_3.eps,width=\hsize}}
1037: %
1038: \vspace{.1in}
1039: \caption{
1040: Representative plots of $P(r)$ for the case of
1041: rotation. The upper subfigure is for
1042: $0.0010 \le \delta x \le 0.2512$. The $r=0$
1043: component is excluded. In the lower subfigure the
1044: fitted Lorentzian is indistinguishable from the
1045: actual profile.}
1046: %%%%%%%%%%%%%%%%%%%%%%%%%%%
1047: \end{figure}
1048:
1049:
1050:
1051: \begin{figure}[p]
1052: %%%%%%%%%%%%%%%%%%%%%%%%%%%
1053: %\epsfysize=2.7in
1054: %\epsffile{image_avprof_bump.eps}
1055: \centerline{\epsfig{figure=prof_defo_1.eps,width=\hsize}}
1056: \centerline{\epsfig{figure=prof_defo_2.eps,width=\hsize}}
1057: \centerline{\epsfig{figure=prof_defo_3.eps,width=\hsize}}
1058: %
1059: \vspace{.1in}
1060: \caption{
1061: Representative plots of $P(r)$ for the case of
1062: generic deformation. The upper subfigure is for
1063: $0.0010 \le \delta x \le 0.2512$. In the lower
1064: subfigure a fitted Lorentzian is overlayed for
1065: the purpose of comparison.}
1066: %%%%%%%%%%%%%%%%%%%%%%%%%%%
1067: \end{figure}
1068:
1069:
1070:
1071: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1072: \section{The standard perturbative regime}
1073:
1074: Standard perturbation theory gives the following
1075: first order expression for the LDOS
1076: %
1077: \begin{eqnarray} \label{e19}
1078: P(n|m) \ \ \approx \ \ \delta_{nm} \ + \
1079: \delta x^2
1080: \frac{|({\partial {\cal H}}/{\partial x})_{nm}|^2}
1081: {(E_n-E_m)^2}
1082: \end{eqnarray}
1083: %
1084: This expression is most straightforwardly obtained
1085: by inspecting Eqs.(\ref{e3}) and (\ref{e4}).
1086: We can define the (total) transition probability as
1087: %
1088: \begin{eqnarray} \label{e20}
1089: p(\delta x) \ \ = \ \ \sum_{r\ne0} P(r)
1090: \end{eqnarray}
1091: %
1092: Using Eq.(\ref{e19}) combined with (\ref{e13})
1093: we get the following estimate:
1094: %
1095: \begin{eqnarray} \label{e21}
1096: p(\delta x) \ \ \approx \ \
1097: \delta x^2 \times \frac{1}{\hbar^2}
1098: \int_{|\omega|>\frac{\Delta}{\hbar}}\frac{d\omega}{2\pi}
1099: \ \frac{\tilde{C}(\omega)}{\omega^2}
1100: \end{eqnarray}
1101: %
1102: Standard perturbation theory is applicable as
1103: long as $p(\delta x) \ll 1$. This can be converted
1104: into an equivalent inequality $\delta x \ll \delta x_c^{\tbox{qm}}$.
1105: By this definition $\delta x_c^{\tbox{qm}}$ is the
1106: parametric deformation which is needed in order
1107: to mix the initial level $m$ with other levels
1108: $n \ne m$.
1109:
1110:
1111: If we use Eq.(\ref{e21}) for a special deformation,
1112: then we have $\gamma>1$, and consequently
1113: the integral is not sensitive to the
1114: exclusion of the $|r|<\Delta$ region.
1115: As a result we have $\delta x_c^{\tbox{qm}} \propto \hbar$.
1116: Using Eq.(\ref{e15}) with (\ref{e14}) we get
1117: %
1118: \begin{eqnarray} \label{e22}
1119: \delta x_c^{\tbox{qm}}|_{\mbox{special}} \ \ = \ \
1120: \left(\ell_{\tbox{bl}}\frac{1}{{\mathsf V}}
1121: \oint[D(\mbf{s})]^2 d\mbf{s}
1122: \right)^{-\half} \times \lambda_{\tbox{B}}
1123: \end{eqnarray}
1124: %
1125: In case of generic deformation Eq.(\ref{e22}) is not
1126: valid because the value of the integral
1127: in (\ref{e21}) is predominantly determined
1128: by the $\omega \sim \Delta/\hbar$ lower cutoff rather
1129: than by an effective $\omega \sim 1/\tau_{\tbox{bl}}$
1130: lower cutoff. As a result one obtains
1131: %
1132: \begin{eqnarray} \label{e23}
1133: \delta x_c^{\tbox{qm}}|_{\mbox{generic}} \ \ = \ \
1134: \left(\ell_{\tbox{H}}\frac{1}{{\mathsf V}}
1135: \oint[D(\mbf{s})]^2 d\mbf{s}
1136: \right)^{-\half} \times \lambda_{\tbox{B}}
1137: \end{eqnarray}
1138: %
1139: where $\ell_{\tbox{H}}=vt_{\tbox{H}}$ is the length which is
1140: associated with the Heisenberg time
1141: $t_{\tbox{H}}=2\pi\hbar/\Delta$.
1142:
1143:
1144:
1145: It is illuminating to use the convention $D(\mbf{s})\sim 1$,
1146: such that $\delta x$ measures the typical
1147: displacement of a wall element. With this convention
1148: we get from Eqs.(\ref{e22})-(\ref{e23}) the following:
1149: %
1150: \begin{eqnarray}
1151: \label{e24_s}
1152: \delta x_c^{\tbox{qm}} &
1153: |_{\mbox{special \hspace*{3.5mm}}} \ \ & \approx \ \
1154: \lambda_{\tbox{B}}
1155: %
1156: \\ \label{e24}
1157: \delta x_c^{\tbox{qm}} &
1158: |_{\mbox{generic \hspace*{3.5mm}}} \ \ & = \ \
1159: \left(
1160: \frac{\lambda_{\tbox{B}}^{d{-}1}}{{\mathsf A}_{\tbox{eff}}}
1161: \right)^{\half}
1162: \times \lambda_{\tbox{B}}
1163: %
1164: \\ \label{e24_d}
1165: \delta x_c^{\tbox{qm}} &
1166: |_{\mbox{diffractive}} \ \ & \sim \ \
1167: \lambda_{\tbox{B}}
1168: \end{eqnarray}
1169: %
1170: In the generic case the effective area of the deformed
1171: boundary ${\mathsf A}_{\tbox{eff}}$ may be smaller than
1172: the total area ${\mathsf A}$ of the boundary. The
1173: effective area ${\mathsf A}_{\tbox{eff}}$ can be
1174: formally defined by comparing Eq.(\ref{e24}) with Eq.(\ref{e23}).
1175: %
1176: Eq.(\ref{e24_d}) has been added for sake of completeness of
1177: our presentation. It correspond to the diffractive limit
1178: ${\mathsf A}_{\tbox{eff}} \rightarrow \lambda_{\tbox{B}}^{d{-}1}$.
1179: Note that this limit is beyond the scope of the present study.
1180: %
1181: Thus in the generic case, the wall displacement needed to mix levels
1182: is much smaller than $\lambda_{\tbox{B}}$.
1183: In the generic case $\delta x_c^{\tbox{qm}} \propto \hbar^{(d{+}1)/2}$
1184: rather than $\delta x_c^{\tbox{qm}} \propto \hbar$.
1185: What happens with perturbation theory beyond $\delta x_c^{\tbox{qm}}$?
1186: This is the subject of the next section.
1187:
1188:
1189: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1190: \section{The core-tail regime}
1191:
1192: For $\delta x > \delta x_c^{\tbox{qm}}$ standard perturbation
1193: theory diverges due to the non-perturbative
1194: mixing of neighboring levels on small scale:
1195: Once $\delta x$ becomes of the order of $\delta x_c^{\tbox{qm}}$
1196: several levels are mixed, and as $\delta x$ becomes larger,
1197: more levels are being mixed non-perturbatively.
1198: Consequently it is natural to distinguish between
1199: {\em core} and {\em tail} regions \cite{vrn,frc,lds}.
1200: %
1201: Most of the spreading probability is contained within
1202: the core region, which implies a natural extension
1203: of first-order perturbation theory (FOPT):
1204: The first step is to transform Eq.(\ref{e3})
1205: into a new basis where transitions within the core are
1206: eliminated; The second step is to use FOPT (in the new basis)
1207: in order to analyze the core-to-tail transitions.
1208: Details of this procedure, which is in the spirit of
1209: degenerate perturbation theory, were discussed in \cite{frc}.
1210: The most important (and non-trivial) consequence of this
1211: procedure is the observation that mixing on small scales
1212: does not affect the transitions on large-scales.
1213: Therefore we have in the tail region $P(n|m)\propto \delta x^2$
1214: rather than $P(n|m)\propto \delta x$.
1215: The validity of this observation has been numerically
1216: illustrated in \cite{lds}.
1217:
1218:
1219:
1220: Following the above reasoning we define
1221: the {\em tail} region as consists of those levels
1222: whose `occupation' can be calculated using perturbation
1223: theory, while the {\em core} is the non-perturbative
1224: component in the vicinity of $r=0$.
1225: Assuming that only one scale characterize the core width,
1226: one arrives at the following practical approximation:
1227: %
1228: \begin{eqnarray} \label{e25}
1229: P_{\tbox{prt}}(r) =
1230: \frac{\Delta}{2\pi\hbar} \
1231: \tilde{C}\left(\frac{E_n{-}E_m}{\hbar}\right) \
1232: \frac{\delta x^2}
1233: {(\Gamma(\delta x))^2 + (E_n{-}E_m)^2}
1234: \end{eqnarray}
1235: %
1236: It is implicit in this definition that $(E_n{-}E_m)$
1237: should be regarded as a function of $r=(n-m)$.
1238: The parameter $\Gamma(\delta x)$ is determined
1239: (for a given $\delta x$) such that $P_{\tbox{prt}}(r)$
1240: has a unit normalization. One may say that
1241: $\Gamma(\delta x)$ regularizes the behavior
1242: around $r=0$. For generic deformation we get
1243: %
1244: \begin{eqnarray} \label{e26}
1245: \Gamma(\delta x) \ \equiv \ b_0\Delta \ \approx \
1246: \left(\frac{\delta x}{\delta x_c^{\tbox{qm}}}\right)^2 \times \Delta
1247: \end{eqnarray}
1248: %
1249: The core is defined as the region $|r|<b_0$,
1250: and the outer ($|r|>b_0$) regions are the tails.
1251: %
1252: For $\delta x \ll \delta x_c^{\tbox{qm}}$ we
1253: get $b_0 \ll 1$ and the core-tail structure Eq.(\ref{e25})
1254: becomes equivalent to the standard
1255: perturbative result Eq.(\ref{e19}).
1256: %
1257: It should be clear that the core-tail structure
1258: is a generalization of Wigner's Lorentzian.
1259: It is indeed a Lorentzian in the
1260: special case of a `flat' band profile.
1261:
1262:
1263:
1264: \begin{figure}
1265: %%%%%%%%%%%%%%%%%%%%%%%%%%%
1266: \centerline{\epsfig{figure=prt_rota.eps,width=\hsize}}
1267: \centerline{\epsfig{figure=prt_defo.eps,width=\hsize}}
1268: \centerline{\epsfig{figure=prt_defo_core.eps,width=\hsize}}
1269: %
1270: \vspace{.1in}
1271: \caption{
1272: Comparison between the perturbative calculation
1273: $P_{\tbox{prt}}(r)$ and the actual $P(r)$. The upper
1274: subfigure is for the rotation and the lower subfigures
1275: are for the generic deformation. The lower subfigure
1276: gives a zoom over the the `birth' of the
1277: non-perturbative semiclassical saturation profile.}
1278: %%%%%%%%%%%%%%%%%%%%%%%%%%%
1279: \end{figure}
1280:
1281:
1282:
1283:
1284: \begin{figure}
1285: %%%%%%%%%%%%%%%%%%%%%%%%%%%
1286: \centerline{\epsfig{figure=probs_rota.eps,width=\hsize}}
1287: \centerline{\epsfig{figure=probs_defo.eps,width=\hsize}}
1288: \centerline{\epsfig{figure=regions_defo.eps,width=\hsize}}
1289: %
1290: \vspace{.1in}
1291: \caption{
1292: The upper and the middle subfigures are $p(\delta x)$ for
1293: the rotation and for the generic deformation respectively.
1294: The lower subfigure is the perturbative calculation
1295: of the core width $b_0$ for the generic deformation.
1296: Also displayed is the calculated $b_c$ for $P_{\tbox{prt}}(r)$,
1297: and the actual $b_c$ for $P(r)$. }
1298: %%%%%%%%%%%%%%%%%%%%%%%%%%%
1299: \end{figure}
1300:
1301:
1302:
1303: We turn now to analyze our numerical results.
1304: We have verified (see eg Fig.7)
1305: that for $\delta x < 0.2$ we have
1306: good agreement with perturbation theory
1307: irrespective of whether we have a core component
1308: (which is the case for the generic deformation)
1309: or not (which is the case for the rotation).
1310: As we come closer to $\delta x \sim 0.2$
1311: the agreement becomes worse, and for $\delta x > 0.3$
1312: we have a total collapse of perturbation theory.
1313: %
1314: In Fig.8 we display the total transition
1315: probability $p(\delta x)$ as a function
1316: of $\delta x$. This plot should be used in order
1317: to numerically determine the value of $\delta x_c^{\tbox{qm}}$,
1318: say as the value where $p(\delta x)=1/2$.
1319:
1320:
1321:
1322: Fig.8 also displays comparison with the corresponding
1323: perturbative calculation
1324: (using Eq.(\ref{e20}) with(\ref{e25})).
1325: The agreement (for $\delta x < \delta x_c^{\tbox{qm}}$)
1326: in case of the rotation is remarkable.
1327: %
1328: For larger values of $\delta x$ one observes
1329: a linear drop of the total probability due to the non-unitary
1330: nature of the evolution. (This drop is clearly linear
1331: in a linear-linear scale which is not displayed).
1332: %
1333: The eventual rise of the total transition probability
1334: (beyond $1$) in case of the generic deformation,
1335: reflects numerical errors in the determination
1336: of the small $\kappa$ overlaps.
1337: See \cite{alexthesis} for further details.
1338:
1339:
1340:
1341: The lower subfigure in Fig.8 displays the
1342: calculated core width $b_0$ for the
1343: generic deformation. Recall that $b_0$ is determined,
1344: given $\delta x$, such that $P_{\tbox{prt}}(r)$ of Eq.(\ref{e25})
1345: is normalized. Also calculated is the width $b_c$
1346: of the region that contains $50\%$ of the probability.
1347: The width $b_c$ is calculated for both $P_{\tbox{prt}}(r)$
1348: and $P(r)$. Note that $b_c= 1$ in the standard perturbative
1349: regime. The determination of $\delta x_c^{\tbox{qm}}$ for the
1350: generic deformation becomes more convenient by
1351: using this plot. Also the crossover
1352: (at $\delta x_{\tbox{NU}} \sim 0.2$)
1353: to the non-universal regime is most pronounced.
1354:
1355:
1356:
1357:
1358: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1359: \section{The non-universal regime}
1360:
1361:
1362: The validity of Eq.(\ref{e25}) as a global approximation
1363: relies on the assumption that the core is
1364: characterized only by one scale (which is $b_0$).
1365: But this assumption ceases to be true if
1366: $\delta x$ is large enough. We have explained in
1367: Ref.\cite{wls} that the width of the {\em core}
1368: defines a `window' through which we can view
1369: the `landscape' of the semiclassical analysis.
1370: As $\delta x$ becomes larger, this `window' becomes wider,
1371: and eventually some of semiclassical structure is exposed.
1372: This is marked by the non-universal parametric scale $\delta x_{\tbox{NU}}$.
1373: For $\delta x$ larger than $\delta x_{\tbox{NU}}$,
1374: the non-universal structure of the core is exposed.
1375:
1376:
1377: What is the semiclassical structure of the LDOS?
1378: Time-domain semiclassical considerations
1379: (See Eq.(10) of \cite{wls} and related discussion there)
1380: imply that the non-universal structure of the core is
1381: %
1382: \begin{eqnarray} \label{e27}
1383: P(n-m) \ \ \approx \ \ \frac{1}{\pi} \
1384: \frac{\delta E_{\tbox{SC}}}{\delta E_{\tbox{SC}}^2+(E_n{-}E_m)^2}
1385: \end{eqnarray}
1386: %
1387: with $\delta E_{\tbox{SC}}=\hbar/\tau_{\tbox{col}}$.
1388: The definition of the collision rate $1/\tau_{\tbox{col}}$
1389: is similar to that of $1/\tau_{\tbox{bl}}$.
1390: The former is the collision rate with the deformed area
1391: ${\mathsf A}_{\tbox{dfr}}$, and therefore it may be
1392: smaller than $1/\tau_{\tbox{bl}}$, because $\tau_{\tbox{bl}}$
1393: is related to the total area ${\mathsf A}$.
1394:
1395:
1396:
1397: If, by mistake, we identified $\tau_{\tbox{col}}$
1398: with $\tau_{\tbox{bl}}$, then we would get the
1399: random wave result which is derived in Appendix A.
1400: This would be an over simplification. If it were true,
1401: it would imply that for $\delta x > \delta x_{\tbox{NU}}$
1402: we should have got the same
1403: Lorentzian-distribution in both cases
1404: (the rotation and the generic deformation).
1405: What we see, as a matter of fact, is that for the rotation
1406: (Fig.5) we have a reasonably good agreement with Lorentzian
1407: whose width is $\delta k_{\tbox{SC}}=1.5$, whereas
1408: for the generic deformation (Fig.6) there is rough
1409: agreement with Lorentzian whose width is $\delta k_{\tbox{SC}}=0.26$.
1410: The smaller width in the latter case clearly reflects
1411: having larger $\tau_{\tbox{col}}$.
1412: %
1413: For the generic deformation we also have pronounced
1414: non-Lorentzian features. Actually the global fitting
1415: to Lorentzian is quite bad. Our understanding
1416: is that these features are due to the bouncing-ball trapping:
1417: It leads to non-exponential decay of the time-dependent
1418: survival probability (see \cite{wls} for definition
1419: of the latter term), and hence to the observed
1420: non Lorentzian features of the spreading profile.
1421:
1422:
1423:
1424: We turn now to explain how $\delta x_{\tbox{NU}}$
1425: is determined. By definition it is the
1426: deformation which is required in order to expose
1427: features of the semiclassical landscape.
1428: These features start to be exposed once
1429: $\Gamma(\delta x) \sim \delta E_{\tbox{SC}}$ which
1430: should be converted into an equivalent
1431: expression $\delta x \sim \delta x_{\tbox{NU}}$.
1432: Thus we get
1433: %
1434: \begin{eqnarray} \label{e28}
1435: \delta x_{\tbox{NU}} \ \ = \ \
1436: \left(
1437: \frac{{\mathsf A}_{\tbox{dfr}}}{{\mathsf A}_{\tbox{eff}}}
1438: \right)^{\half}
1439: \times \lambda_{\tbox{B}} \ \ \sim \ \ \lambda_{\tbox{B}}
1440: \end{eqnarray}
1441: %
1442: Here ${\mathsf A}_{\tbox{dfr}}$ is the geometric area of the
1443: deformation (in the sense of scattering cross-section),
1444: while ${\mathsf A}_{\tbox{eff}}$ is the effective area of the
1445: deformation. The definition of the latter is implied
1446: by comparing Eq.(\ref{e24}) with Eq.(\ref{e23}).
1447: By rescaling $D(\mbf{s}) \mapsto \alpha D(\mbf{s})$
1448: and $\delta x \mapsto \delta x / \alpha$,
1449: we can make ${\mathsf A}_{\tbox{eff}} = {\mathsf A}_{\tbox{dfr}}$
1450: by convention.
1451: %
1452: For special deformations $\delta x_{\tbox{NU}}$
1453: coincides with $\delta x_c^{\tbox{qm}}$ implying
1454: that we get into the non-universal regime as soon
1455: as we have a breakdown of standard perturbation theory.
1456:
1457:
1458:
1459: Our theoretical consideration so far do not imply
1460: a total collapse of perturbation theory.
1461: We may have in principle a co-existence of
1462: non-universal core component and perturbative tails.
1463: Actually we see such co-existence in the lower
1464: subfigures of Fig.7, mainly for $\delta x = 0.2512$.
1465: The right peak around $\kappa=0$ is clearly non-perturbative,
1466: while the rest of the profile is in reasonable
1467: (though not very good) agreement with the perturbative
1468: calculation. We are going to explain in the next section
1469: that the total collapse of perturbation theory
1470: for $\delta x > 0.3$ is actually not related
1471: at all to the appearance of non-universal features
1472: in the core structure. It is only circumstantial
1473: that in the hard wall limit this collapse happens
1474: as soon as we enter the non-universal regime.
1475:
1476:
1477:
1478:
1479:
1480: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1481: \section{The collapse of perturbation theory}
1482:
1483:
1484:
1485: A good starting for the following discussion
1486: is to consider the classical approximation for $P(r)$.
1487: Namely, in Eq.(\ref{e1}) one approximates
1488: $\rho_n(Q,P)$ and $\rho_m(Q,P)$ by microcanonical
1489: distributions. A phase-space illustration of the
1490: energy surfaces which support $\rho_n(Q,P)$ and $\rho_m(Q,P)$
1491: can be found in Fig.1 of Ref.\cite{wls}.
1492: The classical $P(n|m)$ equals to the overlap of these surfaces.
1493:
1494: If we were dealing with a generic system
1495: we could introduce a linearized version of the Hamiltonian
1496: ${\cal H}(Q,P;x) = {\cal H}_0(Q,P) + \delta x {\cal F}(Q,P;\delta x{=}0)$,
1497: where ${\cal F}(Q,P;x) = {\partial {\cal H}}/ {\partial x}$.
1498: By definition this linearization is a good approximation
1499: provided $\delta x \ll \delta x_c^{\tbox{cl}}$, where
1500: $\delta x_c^{\tbox{cl}}$ is the classical correlation
1501: scale of ${\cal F}(Q,P;x)$ with respect to $x$.
1502: In the classical linear regime the classical $P(r)$ has
1503: the scaling property $P(r) = 1/\delta x \hat{P}(r /\delta x)$.
1504:
1505:
1506: An equivalent definition of $\delta x_c^{\tbox{cl}}$
1507: in the quantum-mechanical case is obtained by looking
1508: on the $x$ dependence of the matrix elements
1509: of ${\cal F}(Q,P;x)$ in some {\em fixed~basis}.
1510: Again we define $\delta x_c^{\tbox{cl}}$ as the
1511: respective correlation scale. It is quite clear
1512: that for cavity with soft walls we have
1513: %
1514: \begin{eqnarray}
1515: \delta x_c^{\tbox{cl}} \ \ = \ \ \ell
1516: \ \ \ \ \mbox{[soft walls]}
1517: \end{eqnarray}
1518: %
1519: where $\ell=E/f$ has been defined as the penetration
1520: distance upon collision.
1521: %
1522: %
1523: From purely classical point of view the hard wall limit
1524: $\ell\rightarrow0$ is a {\em non-linear} limit.
1525: But this is not true quantum-mechanically.
1526: Here we have
1527: %
1528: \begin{eqnarray}
1529: \delta x_c^{\tbox{cl}} \ \ \sim \ \ \lambda_{\tbox{B}}
1530: \ \ \ \ \mbox{[hard walls]}
1531: \end{eqnarray}
1532: %
1533: for $\ell < \lambda_{\tbox{B}}$. The terminology `classical
1534: correlation scale' while referring to $\delta x_c^{\tbox{cl}}$
1535: becomes misleading here, but we shall keep using it anyway.
1536:
1537:
1538:
1539: The theory of the core-tail structure \cite{frc}
1540: is valid only in the linear regime
1541: $\delta x \ll \delta x_c^{\tbox{cl}}$.
1542: Let us assume for a moment soft walls. We can ask:
1543: Is $\delta x \ll \delta x_c^{\tbox{cl}}$
1544: a sufficient condition for having a core-tail structure?
1545: The answer is definitely not.
1546: Perturbation theory has a final collapse once
1547: the core width $b_0(\delta x)$ becomes of
1548: the order of the band width $b$. This defines
1549: a parametric scale $\delta x_{\tbox{prt}}$.
1550: For $\delta x > \delta x_{\tbox{prt}}$ the LDOS
1551: becomes purely non-perturbative. In Wigner's
1552: theory of random banded matrices this corresponds
1553: to the crossover from Lorentzian to semicircle
1554: line shape.
1555:
1556:
1557: However, in the limit of hard walls the above
1558: mechanism of collapse becomes irrelevant
1559: because the band width is infinite ($b=\infty$).
1560: On the other hand we still have to satisfy
1561: the inequality $\delta x \ll \delta x_c^{\tbox{cl}}$.
1562: Thus, for $\delta x > \lambda_{\tbox{B}}$ we expect
1563: a total collapse of perturbation theory,
1564: as indeed observed in the numerical study.
1565:
1566:
1567:
1568: \appendix
1569: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1570:
1571: \section{Overlap of uncorrelated random waves}
1572:
1573: It is possible to estimate the overlap
1574: $|\langle n|m \rangle|^2$ if we assume that
1575: $|n\rangle$ and $|m\rangle$ are {\em uncorrelated}
1576: random-superpositions of plane-waves:
1577: %
1578: A random-superpositions of plane-waves is
1579: characterized by the correlation function
1580: %
1581: \begin{eqnarray} \nonumber
1582: \langle \psi_{\tbox{R}}(\mbf{x}_1)\psi_{\tbox{R}}(\mbf{x}_2)
1583: \rangle \ = \ \frac{1}{{\mathsf V}}
1584: \ \mbox{Cos}(k|\mbf{x}_2-\mbf{x}_1|)
1585: \end{eqnarray}
1586: %
1587: where $\mbox{Cos}(kr) \equiv
1588: \langle \exp(i\mbf{k}\cdot\mbf{r})\rangle_{\Omega}$
1589: is a generalized Bessel function
1590: (for further details see Appendix D of \cite{frc}).
1591: Assuming that the wavefunction of $|n\rangle$
1592: is uncorrelated with the wavefunction of $|m\rangle$
1593: one obtains
1594: %
1595: \begin{eqnarray} \nonumber
1596: \langle|\langle n|m \rangle|^2 \rangle \ = \
1597: \left(\frac{1}{{\mathsf V}}\right)^2
1598: \times \hspace*{4cm} \\ \nonumber
1599: \int\!\int
1600: \mbox{Cos}(k_n|\mbf{x}_2{-}\mbf{x}_1|) \
1601: \mbox{Cos}(k_m|\mbf{x}_2{-}\mbf{x}_1|) \
1602: \ d\mbf{x}_1 d\mbf{x}_2
1603: \end{eqnarray}
1604: %
1605: The integration is over the whole volume
1606: of the cavity. Using the definition of the
1607: $\mbox{Cos}$ function we can cast this
1608: expression into the form
1609: %
1610: $\langle|\langle n|m \rangle|^2 \rangle
1611: =\left\langle f(\mbf{q}) \right\rangle_{q}$
1612: %
1613: where the average is over
1614: the difference $\mbf{q}=(\mbf{k}_2-\mbf{k}_1)$,
1615: with all possible orientations for
1616: $|\mbf{k}_1|=k_m$ and for $|\mbf{k}_2|=k_n$.
1617: The function $f(\mbf{q})$ is defined as follows:
1618: %
1619: \begin{eqnarray} \nonumber
1620: f(\mbf{q}) \ = \ \left| \int
1621: \mbox{e}^{i\mbf{q}{\cdot}\mbf{x}}
1622: d\mbf{x} \right|^2
1623: \end{eqnarray}
1624: %
1625: The function $f(\mbf{q})$ depends mainly
1626: on $q=|\mbf{q}|$. We can obtain an estimate
1627: for $f(q)$ by considering a spherical
1628: cavity $|\mbf{x}|<R$ with the same volume.
1629: Using spherical coordinates it is straightforward
1630: to obtain the following:
1631: %
1632: \begin{eqnarray} \nonumber
1633: f(q) \ \approx \
1634: \left( {\cal FT} \left[
1635: \frac{1}{d{-}1}(R^2-x^2)^{(d{-}1)/2}
1636: \right]\right)^2
1637: \end{eqnarray}
1638: %
1639: above ${\cal FT}$ is a Fourier transform
1640: from $x$ to $q$. For $q \ll 1/R$ we
1641: have simply $f(q)=({\mathsf V})^2$. For $q \gg 1/R$
1642: we have $f(q)=({\mathsf V})^2/(Rq)^{d{+}1}$.
1643: This is because of the singularity at $x=\pm R$.
1644: The average over $q$ can be done using again
1645: spherical coordinate. We have
1646: $q \approx ((\Delta k)^2 + 2k^2(1-\cos\theta) )^{\tbox{1/2}}$,
1647: where $\Delta k = |k_n-k_m|$, and we can transform
1648: the $d\theta$ integration into $dq$ integration:
1649: %
1650: \begin{eqnarray} \nonumber
1651: \left\langle f(\mbf{q}) \right\rangle_{q} \ \ = \ \
1652: \frac{\Omega_{d{-}2}}{\Omega_{d{-}1}}
1653: \times \hspace*{4cm} \\ \nonumber
1654: \int \ \frac{qdq}{k^2}\left(
1655: \frac{\sqrt{(q^2-(\Delta k)^2)((2k)^2-q^2)}}{2k^2}
1656: \right)^{d{-}3} f(q)
1657: \end{eqnarray}
1658: %
1659: where $\Omega_d$ is the solid angle in $d$ dimensions.
1660: There is no point in trying to carry an exact
1661: integration. Rather, it is important
1662: to observe that a practical approximation for
1663: the overlap is:
1664: %
1665: \begin{eqnarray} \nonumber
1666: \langle|\langle n|m \rangle|^2 \rangle \ \approx \
1667: \left( \frac{1}{kR} \right)^{d{-}1}
1668: \frac{1}{1+(R{\cdot}(k_n{-}k_m))^2}
1669: \end{eqnarray}
1670: %
1671: In the latter expression we have neglected
1672: the $d$-dependent normalization prefactor.
1673: It is a `practical' approximation since it
1674: gives the correct behavior for both small
1675: and large values of $\Delta k$.
1676: The interpolation around $\Delta k \sim 1/R$
1677: cannot be trusted.
1678:
1679:
1680: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1681:
1682: \ \\ \ \\ \ \\
1683:
1684: \noindent
1685: {\bf Acknowledgments:}
1686:
1687: \noindent
1688: This work was funded by ITAMP and the National Science Foundation.
1689:
1690: \ \\ \ \\ \ \\
1691:
1692: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1693: \begin{thebibliography}{99}
1694:
1695:
1696: \bibitem{rmrk1}
1697: The classical approximation for $P(n|m)$ works well
1698: for ergodic eigenstates. As explained later in the
1699: text we actually consider the averaged profile.
1700: Namely, $P(n|m)$ is regarded as a function of $r=n-m$,
1701: and it is averaged over the reference state $m$.
1702: With this procedure the effect of the minority of non-ergodic
1703: eigenstates can be neglected in the classical limit.
1704:
1705:
1706: \bibitem{rmrk2}
1707: The two parametric scales of Wigner's RMT model
1708: correspond to $\delta x_c^{\tbox{qm}}$ and
1709: $\delta x_{\tbox{prt}}$ of Table 1. Consequently
1710: there are three parametric regimes in Wigner's theory.
1711: These are the standard perturbative regime
1712: ($\delta x \ll \delta x_c^{\tbox{qm}}$),
1713: the Lorentzian regime
1714: ($\delta x_c^{\tbox{qm}} \ll \delta x \ll \delta x_{\tbox{prt}}$),
1715: and the semi-circle regime ($\delta x \gg \delta x_{\tbox{prt}}$).
1716: As an artifact of this RMT model, there actually exists
1717: a forth regime (Anderson strong localization regime)
1718: that in the present context does not have a semiclassical analog.
1719:
1720:
1721: \bibitem{rmrk3}
1722: As explained in \cite{wls} the generic hierarchy is
1723: $\delta x_{\tbox{NU}} \ll \delta x_{\tbox{prt}} \ll
1724: \delta x_{\tbox{SC}} \ll \delta x_c^{\tbox{cl}}$.
1725: This hierarchy is realized in the classical limit
1726: (small $\hbar$) where we have soft walls (see section II).
1727: The hard wall limit is non-generic, and these four
1728: parametric scales coincide. In the present paper
1729: we assume hard walls unless stated otherwise.
1730:
1731:
1732: \bibitem{casati}
1733: G. Casati, B.V. Chirikov, I. Guarneri, F.M. Izrailev,
1734: Phys. Rev. E {\bf 48}, R1613 (1993);
1735: \ Phys. Lett. A {\bf 223}, 430 (1996).
1736:
1737: \bibitem{flamb}
1738: V.V. Flambaum, A.A. Gribakina, G.F. Griobakin
1739: and M.G. Kozlov, Phys. Rev. A {\bf 50} 267 (1994).
1740:
1741: \bibitem{felix}
1742: G. Casati, B.V. Chirikov, I. Guarneri and F.M. Izrailev,
1743: Physics Lett. A {\bf 223}, 430 (1996).
1744: \ F. Borgonovi, I. Guarneri and F.M. Izrailev
1745: Phys. Rev. E {\bf 57}, 5291 (1998).
1746:
1747: \bibitem{heller} E. Heller in
1748: {\em Chaos and quantum Physics},
1749: Proc. Session LII of the Les-Houches Summer School,
1750: Edited by A. Voros and M-J Giannoni
1751: (Amsterdam: North Holland 1990).
1752:
1753: \bibitem{wls}
1754: D. Cohen and E.J. Heller,
1755: Phys. Rev. Lett. {\bf 84}, 2841 (2000).
1756:
1757: \bibitem{ct}
1758: {\em Atom-photon interactions: basic processes and applications},
1759: C. Cohen-Tannoudji, J. Dupont-Roc and G. Grynberg
1760: (New York: J. Wiley 1992).
1761:
1762: \bibitem{wigner}
1763: E. Wigner, Ann. Math {\bf 62}, 548 (1955);
1764: {\bf 65}, 203 (1957).
1765:
1766: \bibitem{mario}
1767: M. Feingold and A. Peres, Phys. Rev. A {\bf 34} 591, (1986).
1768: M. Feingold, D. Leitner, M. Wilkinson, Phys. Rev. Lett. {\bf 66}, 986 (1991).
1769: M. Wilkinson, M. Feingold, D. Leitner, J. Phys. A {\bf 24}, 175 (1991).
1770: M. Feingold, A. Gioletta, F. M. Izrailev, L. Molinari, Phys. Rev. Lett.
1771: {\bf 70}, 2936 (1993).
1772:
1773: \bibitem{crs}
1774: D. Cohen, Phys. Rev. Lett. {\bf 82}, 4951 (1999).
1775:
1776: \bibitem{vrn}
1777: D. Cohen, {\em in} Proceedings of the International
1778: School of Physics `Enrico Fermi' Course CXLIII
1779: ``New Directions in Quantum Chaos'',
1780: Edited by G. Casati, I. Guarneri and U. Smilansky,
1781: IOS Press, Amsterdam, 2000.
1782:
1783: \bibitem{frc}
1784: D. Cohen, Annals of Physics {\bf 283}, 175 (2000).
1785:
1786: \bibitem{lds}
1787: D. Cohen and T. Kottos, nlin.CD/0001026,
1788: Phys. Rev. E (2001, in press).
1789:
1790: \bibitem{berry}
1791: M. V. Berry and M. Wilkinson,
1792: Proc. R. Soc. London A {\bf 392}, 15 (1984).
1793:
1794: \bibitem{dil}
1795: A. Barnett, D. Cohen, and E.J. Heller,
1796: Phys. Rev. Lett. {\bf 85}, 1412 (2000).
1797:
1798: \bibitem{wlf}
1799: A. Barnett, D. Cohen, and E.J. Heller, nlin.CD/0006041,
1800: J. Phys. A (2001, in press).
1801:
1802: \bibitem{vergini}
1803: E. Vergini and M. Saraceno,
1804: Phys. Rev. E, {\bf 52}, 2204 (1995).
1805: E. Vergini, Ph.~D.\ thesis,
1806: Universidad de Buenos Aires, 1995.
1807:
1808: \bibitem{alexthesis}
1809: A. Barnett, PhD thesis (Harvard, Sept. 2000).
1810:
1811: \bibitem{note}
1812: In order to find the average profile
1813: we divide the $\kappa$ axis into small bins.
1814: We find the average $P(n|m)$ value
1815: in each bin. In order to get $P(r)$
1816: we re-sample the average profile. The
1817: distance between the sampling points
1818: is equal to the mean level spacing.
1819: We are not interested in features on
1820: scale of the mean level spacing. Therefore
1821: the transformation from the continuous
1822: variable $\kappa$ to the integer variable
1823: $r$ should not be considered as problematic.
1824:
1825:
1826: \end{thebibliography}
1827: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1828:
1829: \end{document}
1830: