1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2: % Parametric invariant Random Matrix Model describing
3: % the crossover to a non-perturbative line shape
4: % and the emergence of multifractality (Jan 2006)
5:
6:
7: \documentclass[aps,pre,twocolumn,floats,showpacs]{revtex4}
8: \usepackage{epsfig}
9: \usepackage{bm}
10: \usepackage{latexsym}
11:
12: \begin{document}
13: \newcommand{\hide}[1]{}
14: \newcommand{\tbox}[1]{\mbox{\tiny #1}}
15: \newcommand{\half}{\mbox{\small $\frac{1}{2}$}}
16: \newcommand{\sinc}{\mbox{sinc}}
17: \newcommand{\const}{\mbox{const}}
18: \newcommand{\trc}{\mbox{trace}}
19: \newcommand{\intt}{\int\!\!\!\!\int }
20: \newcommand{\ointt}{\int\!\!\!\!\int\!\!\!\!\!\circ\ }
21: \newcommand{\eexp}{\mbox{e}^}
22: \newcommand{\bra}{\left\langle}
23: \newcommand{\ket}{\right\rangle}
24: \newcommand{\EPS} {\mbox{\LARGE $\epsilon$}}
25: \newcommand{\ar}{\mathsf r}
26: \newcommand{\im}{\mbox{Im}}
27: \newcommand{\re}{\mbox{Re}}
28: \newcommand{\bmsf}[1]{\bm{\mathsf{#1}}}
29: \newcommand{\mpg}[2][1.0\hsize]{\begin{minipage}[b]{#1}{#2}\end{minipage}}
30:
31:
32: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
33: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
34:
35:
36: \title{Parametric invariant Random Matrix Model and the emergence
37: of multifractality}
38:
39: \author{
40: J. A. M\'endez-Berm\'udez$^{1,2,3}$, Tsampikos Kottos$^{1,4}$, Doron Cohen$^{2}$
41: }
42:
43: \affiliation{
44: $^1$Max-Planck-Institut f\"ur Dynamik und Selbstorganisation, Bunsenstra\ss e 10,
45: D-37073 G\"ottingen, Germany \\
46: $^2$Department of Physics, Ben-Gurion University,
47: Beer-Sheva 84105, Israel \\
48: $^3$Instituto de F\'{\i}sica, Universidad Aut\'onoma de Puebla, Apartado
49: Postal J-48, Puebla 72570, M\'exico \\
50: $^4$Department of Physics, Wesleyan University, Middletown, Connecticut
51: 06459-0155, USA
52: }
53:
54:
55: \begin{abstract}
56:
57: We propose a random matrix modeling
58: for the parametric evolution of eigenstates.
59: The model is inspired by a large class
60: of quantized chaotic systems. Its unique
61: feature is having parametric invariance
62: while still possessing the non-perturbative breakdown
63: that has been discussed by Wigner 50~years ago.
64: Of particular interest is the emergence
65: of an additional crossover to multifractality.
66:
67: \end{abstract}
68: \pacs{05.45.Mt, 05.45.Df, 72.15.Rn, 71.30.+h}
69: \maketitle
70:
71:
72:
73:
74: %%%%%%%%%%%%%%%%%%%%%%%%%%%% Introduction %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
75:
76:
77: \section{Introduction}
78:
79:
80: The analysis of structural changes of eigenstates as
81: a parameter $x$ is varied, has sparked a great deal
82: of research activity for many years
83: \cite{W55,BCH,MKC05,MLI04,B03,VLG02,CH2000,CK01,HKG06}.
84: Of particular importance are quantized {\em chaotic}
85: or complex systems where the change of $x$ may represent
86: the effect of some externally controlled field (like
87: electric field, gate voltage or magnetic flux). Thus,
88: these studies are relevant for diverse areas of physics
89: ranging from nuclear \cite{HZB95,W55} and atomic physics
90: \cite{FGGP99} to quantum chaos \cite{CK01,MLI04,B03}
91: and mesoscopics \cite{VLG02}.
92:
93: In all these studies, Random Matrix Theory (RMT) played
94: a dominant role as a reference theory that describes the
95: {\it universal} properties of the eigenstates of complex
96: systems. RMT was introduced 50 years ago by Wigner as a
97: tool to describe the statistical properties of the
98: eigenvalues of complex nuclei.
99: Until recently, the matrices in the frame of RMT were
100: assumed to be homogeneous, i.e., all matrix elements were
101: set to have identical statistical properties. Under this
102: simplification random matrices are rotationally invariant,
103: a property that simplifies their theoretical analysis. In
104: physical applications this implies that interactions are
105: assumed to be so strong and complex that no other parameters,
106: apart from the symmetry of the Hamiltonian matrix, are
107: relevant. As a result, such random matrices can be associated
108: to the extreme case of maximal chaos, which is known to appear
109: in various physical systems such as heavy nuclei, atoms,
110: metallic clusters, etc. Moreover, one can treat full random
111: matrices as a typical model when describing local statistical
112: properties of spectra and eigenstates in some range of the
113: energy spectrum, typically, in the semiclassical region.
114:
115: On the other hand, the conventional RMT can not describe
116: important phenomena such as localization of eigenstates,
117: neither can be directly applied to obtain spectra of
118: realistic models. For this reason, much attention has been
119: recently paid to the so-called Wigner Band Random Matrix
120: (WBRM) model which is characterized by the free parameter
121: $b$, that represents the effective bandwidth of a
122: Hamiltonian matrix.
123: Among the important applications of the WBRM model we
124: mention the study of localization in quasi-one-dimensional
125: disordered systems. Also the WBRM model has been applied
126: to the analysis of either {\em chaotic} or complex
127: conservative quantum systems that are present
128: in nuclear physics as well as in atomic and molecular physics.
129:
130:
131: Despite its success, the standard WBRM model
132: has {\em severe limitations in modeling realistic
133: systems}. We explain these limitations in section II,
134: and further motivate the introduction of
135: a new RMT ensemble to which we refer as
136: the Winger Lorentzian Random Matrix (WLRM) model.
137: %
138: In Sec. III, the WLRM model is shown
139: to have a ``parametric~invariance" property
140: which is characteristic of any realistic system
141: but is missing in the standard WBRM model.
142: The analysis of the local density of states
143: of the WLRM model is done in Sec. IV.
144: The multifractal properties of the eigenstates
145: of the WLRM ensemble are analyzed in Sec.V.
146: Our conclusions are summarized in Sec. VI.
147:
148:
149:
150:
151: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
152: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
153: \section{RMT modeling}
154:
155: The pioneering work in this field has been done
156: by Wigner \cite{W55}, who has motivated the
157: studies of RMT models of the type
158: %
159: \begin{equation}
160: {\cal H}=\bm{E}+x\bm{B} \ .
161: \label{ham}
162: \end{equation}
163: %
164: Both $\bm{E}$ and $\bm{B}$ are real symmetric matrices
165: of size $N\times N$. The elements of the diagonal
166: matrix $\bm{E}$ are the ordered energies $\{E_n\}$,
167: with mean level spacing $\Delta$,
168: while $\bm{B}$ is a banded {\em random} matrix which
169: is characterized by a band profile $C(r)$.
170: Namely, the entries of $\bm{B}$ are random numbers
171: that are drawn from a normal distribution with zero
172: mean and variance given by
173: %
174: \begin{equation}
175: \langle |B_{nm}|^2 \rangle = C(n-m) \ .
176: \end{equation}
177: %
178: For the study of {\em spectral statistics}
179: of the energy levels it turns out that full matrices ($C(r)=1$),
180: say of the Gaussian Orthogonal ensemble (GOE),
181: are enough in order to capture the universality
182: which is found in quantized chaotic system.
183: But for the study of {\em parametric evolution}
184: of the eigenstates it is essential to take into account
185: the band profile, which is dictated
186: by semiclassical considerations.
187: Namely, $C(r)$ is merely the scaled version
188: of a classical power spectrum $\tilde{C}(\omega)$
189: which is obtained via a Fourier transform
190: of the classical correlation function of the
191: generalized force ${\cal F}(t) = -\partial {\cal H}/\partial x$.
192:
193:
194:
195:
196:
197: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
198: \subsection{The standard WBRM model}
199:
200:
201: The standard WBRM model assumes a rectangular band profile:
202: %
203: \begin{equation}
204: C(r)= \left\{ \matrix{
205: 1 & \quad \mbox{$r\le b$} \cr
206: 0 & \quad \mbox{$r> b$}
207: } \right. \ .
208: \label{WBRM}
209: \end{equation}
210: %
211: For this model Wigner has found that the eigenstates undergo
212: a transition from a {\em perturbative} Lorentzian-type line
213: shape to a {\em non-perturbative} semicircle line-shape.
214: %
215: It should be clear that the Wigner Lorentzian can be regarded
216: as the outcome of perturbation theory to infinite order,
217: while the semicircle line-shape is beyond any order
218: of perturbation theory.
219:
220:
221: The {\em existence of a transition} from a perturbative
222: to a non-perturbative line shape is a generic feature
223: of {\em any} realistic (quantized) Hamiltonian.
224: In the latter case the {\em semi}-circle line-shape
225: is replaced by a {\em semi}-classical line shape.
226:
227:
228:
229:
230: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
231: \subsection{The modified WBRM model}
232:
233:
234: The WBRM model suffers from a serious drawback.
235: Unlike generic canonically quantized Hamiltonians,
236: the statistical properties of its Hamiltonian
237: matrix are not invariant under $x \rightarrow x + \const$.
238: %
239: In fact there exists two wise modified versions
240: of the WBRM model \cite{Alhassid,Wilkinson}
241: which are manifestly $x$~invariant by construction.
242: For example we cite one of them:
243: %
244: \begin{equation}
245: \mathcal{H} = \bm{E}+\cos(x)\bm{B}_1 + \sin(x)\bm{B}_2
246: \label{modWLRM}
247: \end{equation}
248: %
249: Here $\bm{B}_1$ and $\bm{B}_2$ are uncorrelated
250: banded matrices. It is quite easy to be convinced
251: that this model is $x$~invariant. One simply has
252: to set $x \rightarrow x + \const$, to expand the $\sin()$
253: and the $\cos()$, to define $\bm{B}_1'$ and $\bm{B}_2'$,
254: and to observe that they are uncorrelated
255: with the same band profile as $\bm{B}_1$ and $\bm{B}_2$.
256:
257: However there is a ``price" for using such modified model.
258: It is not difficult to prove that the parametric
259: nature of this model is essentially perturbative:
260: The associated local density of states does
261: {\em not} exhibit the non-perturbative crossover that
262: has been highlighted in the previous subsection!
263:
264:
265: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
266: \subsection{The Winger Lorentzian Random Matrix model}
267:
268:
269: In the present paper we introduce a
270: new RMT ensemble to which we refer as
271: the Winger Lorentzian Random Matrix (WLRM) model.
272: The Hamiltonian is assumed to have the standard
273: form of Eq.(\ref{ham}), and it is characterized
274: by the band profile
275: %
276: \begin{equation}
277: C(r) =
278: \frac{1}{1+\left( r/b \right)^2} \, .
279: \label{PBRM}
280: \end{equation}
281: %
282: There are several good reasons that motivate
283: the introduction and the study of this model,
284: which we are going to clarify:
285: %
286: %
287: \begin{itemize}
288: %
289: \item[{\bf (1)}]
290: There is a major class of quantized
291: chaotic systems that can be described using
292: this model.
293: %
294: \item[{\bf (2)}] Unlike the standard WBRM model it has
295: the desired $x$~invariance property that
296: characterizes quantized models.
297: %
298: \item[{\bf (3)}] Unlike the common $x$~invariant
299: version of the WBRM model it exhibits the transition
300: to a non-perturbative line shape.
301: %
302: \item[{\bf (4)}] The emergence of multifractality,
303: which is absent in the WBRM model,
304: is a fascinating issue by itself.
305: %
306: \end{itemize}
307:
308:
309: Let us expand on the first point. We recall that
310: the flat band profile of the standard WBRM model
311: is motivated by the realization that many observables
312: (say ${\cal F}(t)$) of chaotic systems exhibit ``white" power
313: spectrum: $\tilde{C}_{{\cal FF}}(\omega)\sim \mbox{const}$.
314: However, in many cases it is $G(t)=\dot{\cal F}$ that
315: has the ``white" power spectrum \cite{BCH,MKC05}.
316: In the latter case the relation
317: $\tilde{C}_{{\cal FF}}(\omega)=\tilde{C}_{GG}(\omega)/\omega^2$
318: implies Lorentzian tails.
319:
320: %%%%%%
321: \hide{
322: A major example is the Aharonov-Bohm cylindrical
323: billiard \cite{MKC05}. The motion on the two dimensional
324: surface of this system is chaotic due to the
325: collisions of the particle with a deformed or possibly rough
326: upper boundary. The control parameter $x$
327: is the magnetic flux $\Phi$ through the cylinder.
328: The corresponding Hamiltonian has the form (\ref{ham}),
329: where $\bf{B}_{nm}$ are the matrix elements of the
330: current operator. The band profile of $B$ is of
331: the Lorentzian type, which reflects the velocity-velocity
332: correlation function: Namely, the associate power spectrum
333: $\tilde{C}_{vv}(\omega)$ possess $1/omega^2$ tails.
334: It follows because the velocity ($v$) is the time derivative
335: of a random like force which is exerted on the particle
336: due to collisions with the walls.
337: }
338: %%%%%
339:
340:
341:
342: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
343: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
344: %%%%%%%%%%%%%%%%%%%%%%%%%%% X-INVARIANCE %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
345: \section{Parametric invariance}
346:
347:
348:
349:
350: %%%%%%%%%%%%%%%%%%%
351: \begin{figure}[t]
352: \begin{center}
353: \epsfxsize=8cm
354: \leavevmode
355: \epsffile{blm_fig1.eps}
356: \caption{(a) The band profile $C(r)$ of the standard Wigner model
357: using $b=500$ and $x^*=0$, 0.1, 1, 3, 7, 15, 31, and 63.
358: (b) The band profile $C(r)$ of the Lorentzian Wigner model
359: with $b=1$ and $x^*=2.5$, 27, 277, and 2777.
360: The dashed line in the inset of (b) with decay $\sim r^{-2}$
361: is plotted to guide the eye.}
362: \label{fig:fig1}
363: \end{center}
364: \end{figure}
365: %%%%%%%%%%%%%%%%%%%
366:
367:
368: %%%%%%%%%%%%%%%%%%%%
369: \begin{figure}[t]
370: \begin{center}
371: \epsfxsize=8cm
372: \leavevmode
373: \epsffile{blm_fig2.eps}
374: \caption{The band profiles for the power law matrix
375: $C(r) = [1+(r/b)^\mu]^{-1}$ with (a) $\mu=0.5$ and (b) 5.
376: $b=1$. In (a) $x^*=2.5$, 27, 277, and 2777.
377: In (b) $x^*=0.13$, 1.4, 14, and 144.
378: The dashed line in the inset of (a) [(b)] with decay
379: $\sim r^{-0.5}$ [$\sim r^{-5}$] is plotted to guide the eye.}
380: \label{fig:fig2}
381: \end{center}
382: \end{figure}
383: %%%%%%%%%%%%%%%%%%%
384:
385:
386: An important feature of a generic canonically quantized
387: Hamiltonian $\mathcal{H}(Q,P;x)$ is its parametric $x$~invariance.
388: Given $x^*$ we can represent the Hamiltonian
389: by a matrix
390: %
391: \begin{eqnarray}
392: \mathcal{H} \ \ = \ \ \bm{E} + \delta x\bm{B}
393: \end{eqnarray}
394: %
395: where $\delta x = x-x^*$.
396: If we take two different values of $x^*$
397: we get two different $\bm{B}$ matrices.
398: But if the two values of $x^*$ belong
399: to the same {\em classically small window},
400: then (by definition) the band profile $C(r)$
401: comes out the same. Still from a quantum mechanical
402: point of view a classically small range
403: of $x$~values is typically regarded as {\em huge}.
404: This means that in general, quantum perturbation theory
405: cannot be used in order to describe the parametric
406: evolution within this range.
407:
408:
409: In Fig.~\ref{fig:fig1} we present $C(r)$
410: for the standard WBRM model and for the WLRM model.
411: We see that the profile of the perturbation matrix $\bm{B}$ of
412: the WBRM model is deformed as $x$ is increased, while that of
413: the WLRM model remains the same. We have found out that
414: this $x$~invariance does not hold for other (non-Lorentzian)
415: power law profiles. As examples, in Fig.~\ref{fig:fig2}
416: we present the profiles of $\bm{B}$ for increasing $x$ in the
417: case of the power-law profile $C(r) = [1+(r/b)^\mu]^{-1}$ with
418: $\mu=0.5$ and 5. We conclude that the $x$~invariance is a
419: unique property of the WLRM ensemble.
420:
421:
422:
423:
424:
425:
426:
427:
428:
429:
430: %%%%%%%%%%%%%%%%%%%%%%%%%%%% LDOS %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
431: \section{LDOS analysis}
432:
433:
434: The local density of states (LDOS) is the major tool
435: for the characterization of the parametric evolution
436: of the eigenstates. The overlap of the eigenstates
437: $|n(x)\rangle$ for a given value of $x$ with the
438: eigenstate $|m(0)\rangle$ of the $x=0$ Hamiltonian
439: is $P(n|m) = |\langle n(x)|m(0)\rangle|^2$.
440: This can be regarded as a distribution with respect
441: to $n$. By averaging over the reference level $m$
442: we get the line shape $P(n-m)$. Up to trivial scaling
443: this is the LDOS.
444:
445:
446: The considerations that are required in order to generalize
447: the calculation of the LDOS line shape for a general band
448: profile have been introduced in \cite{CH2000,CK01}. Here, we
449: apply such methodology in order to analyze the parametric
450: evolution of the LDOS for the WLRM model.
451:
452:
453: For $x=0$ the LDOS is trivially ${P(r)=\delta_{r,0}}$ due to
454: orthogonality. As $x$ increases, perturbative tails start to appear.
455: By employing standard first-order perturbation
456: theory we get $P_{\tbox{FOPT}} (r) \approx 1$ for $r=0$, while
457: %
458: \begin{equation}
459: P_{\tbox{FOPT}}(r) = {x^2 |\bm{B}_{nm}|^2 \over (E_n{-}E_m)^2}
460: = \frac{x^2}{\Delta^2} \frac{b^2}{(b^2+r^2)} \frac{1}{r^2}
461: \label{Pst}
462: \end{equation}
463: %
464: for $r\neq 0$. In the second equality we have
465: substituted the matrix elements of $\bm{B}$
466: using Eq.~(\ref{PBRM}). The above expression applies
467: for $x<x_c$ where $x_c$ is the perturbation
468: strength needed to mix neighboring levels only.
469:
470: For $x>x_c$ Wigner had found, in the frame of the WBRM model,
471: that the LDOS line shape can
472: be calculated using perturbation theory to infinite order.
473: In case of the WBRM model one
474: obtains a Lorentzian. Assuming the validity of infinite
475: order perturbation theory we come
476: out with a Lorentzian-type approximation
477: for the LDOS of the WLRM model:
478: %
479: \begin{eqnarray}
480: P_{\tbox{PRT}}(r) &=& {x^2 |\bm{B}_{nm}|^2 \over \Gamma^2 + (E_n{-}E_m)^2}
481: \nonumber\\
482: &=& \frac{x^2}{\Delta^2} \frac{b^2}{(b^2+r^2)}
483: \frac{1}{[(\Gamma/\Delta)^2+r^2]} \ .
484: \label{Pw}
485: \end{eqnarray}
486: %
487: Eq.~(\ref{Pw}) is an approximation
488: because all orders of perturbation are treated
489: within a Markovian-like approach (by iterating
490: the first order result) and convergence of the expansion is pre-assumed.
491: The energy scale $\Gamma$ defines the region where a non-perturbative mixing of levels occurs.
492: Inside this region the perturbative profile $P_{\tbox{PRT}}(r)$ does not describe the actual LDOS
493: lineshape. $\Gamma$ is determined by imposing normalization of $P_{\tbox{PRT}}(r)$.
494: For the WBRM model it was found \cite{W55}
495: that $\Gamma= x^2 C(1)/\Delta$. For the WLRM model we get
496: %
497: \begin{equation}
498: \Gamma = \frac{b\Delta}{2} \left[ \sqrt{1+\frac{4\pi x^2}{b\Delta^2}} - 1\right].
499: \label{gamma}
500: \end{equation}
501:
502:
503:
504:
505: %%%%%%%%%%%%%%%
506: \begin{figure}[t]
507: \begin{center}
508: \epsfxsize=8cm
509: \leavevmode
510: \epsffile{blm_fig3.eps}
511: \caption{The LDOS lineshape $P(r)$ in the (a) standard perturbative,
512: (b) extended perturbative, and (c) non-perturbative regimes for the
513: WLRM model with $b=1$, $\Delta=1$, and $N=5000$.
514: For this set of parameters $x_c\approx 0.8$ and $x_{\tbox{prt}}\approx 1.5$.
515: The first-order perturbation theory profile $P_{\tbox{FOPT}}(r)$ from
516: Eq.~(\ref{Pst}) is included in (a).
517: The Wigner-type lineshape $P_{\tbox{PRT}}(r)$ from
518: Eq.~(\ref{Pw}), which is expected to be valid in the perturbative
519: regime, is also included (no fitting parameters) in (b) and (c).
520: The dashed line in the insets of (a) and (b) with decay $\sim r^{-4}$
521: is plotted to guide the eye.
522: The gray line in the inset of (c) is a semicircle fitting to $P(r)$.}
523: \label{fig:fig3}
524: \end{center}
525: \end{figure}
526: %%%%%%%%%%%%%%%
527:
528:
529:
530: Obviously, for $\Gamma\ll \Delta$ the (infinite-order)
531: LDOS profile $P_{\tbox{PRT}}(r)$ reduces
532: to the standard first-order perturbation
533: theory expression $P_{\tbox{FOPT}}(r)$. Therefore $x_c$
534: can be determined by the condition $\Gamma(x_c) \approx \Delta$,
535: leading to
536: %
537: \begin{equation}
538: x_c \approx \frac{\Delta}{\sqrt{\pi}} \sqrt{1+\frac{1}{b}} \, .
539: \label{xc}
540: \end{equation}
541:
542:
543: In Fig.~\ref{fig:fig3} we display our numerical results
544: for $P(r)$ for the WLRM model with
545: $b=1$, $\Delta=1$, $N=5000$, and various
546: perturbation strengths $x$.
547: We see that the agreement
548: with the perturbative expression (\ref{Pw})
549: persist up to some perturbation strength $x_{\tbox{prt}}$.
550: Above $x_{\tbox{prt}}$ the LDOS lineshape $P(r)$
551: becomes semicircle in complete analogy with the WBRM model scenario.
552:
553:
554: We want to find the value $x_{\tbox{prt}}$ up
555: to which the perturbative expression $P_{\tbox{PRT}}(r)$
556: describes reasonably good the LDOS lineshape.
557: To this end we compare the dispersion of $P_{\tbox{PRT}}(r)$,
558: %
559: \begin{equation}
560: \delta E_{\tbox{PRT}} = \Delta \times \sqrt{\sum_r r^2 P_{\tbox{PRT}}(r)} \ ,
561: \end{equation}
562: %
563: to the dispersion of the actual LDOS \cite{CK01}
564: %
565: \begin{equation}
566: \delta E = x\sum_{r\ne 0} C(r) \ .
567: \end{equation}
568: %
569: Expressions for $\delta E_{\tbox{PRT}}$ and $\delta E$
570: in case of our WLRM model can be obtained
571: by replacing the sums above by integrals:
572: %
573: \begin{equation}
574: \delta E_{\tbox{PRT}} \approx x\sqrt{\pi} \ b(b+\Gamma/\Delta)^{-1/2} \ ,
575: \label{dEW}
576: \end{equation}
577: %
578: \begin{equation}
579: \delta E \approx x\sqrt{2b} \left[ \pi/2 - \arctan(1/b) \right]^{1/2} .
580: \label{dE}
581: \end{equation}
582: %
583: Note that for small $x$, $\delta E_{\tbox{PRT}} = \delta E \approx x(\pi b)^{1/2}$.
584: $\delta E$ is a linear function of $x$ for all perturbation strengths
585: while for large enough $x$ $\delta E_{\tbox{PRT}}$ becomes sublinear:
586: $\delta E_{\tbox{PRT}} \propto x^{1/2}$. See Fig.~\ref{fig:fig4}.
587: The border $x_{\tbox{prt}}$ is identified as the perturbation strength for
588: which $\delta E_{\tbox{PRT}} (x_{\tbox{prt}}) \approx \gamma \delta E(x_{\tbox{prt}})$,
589: where $\gamma<1$:
590: %
591: \begin{equation}
592: x_{\tbox{prt}} \approx \Delta\sqrt{b}
593: \frac{\sqrt{\pi-2\gamma^2[\pi/2 - \arctan(1/b)]}}{2\gamma^2[\pi/2 - \arctan(1/b)]} \, .
594: \label{xprt}
595: \end{equation}
596: %
597: We typically use $\gamma = 0.8$.
598:
599:
600:
601:
602:
603: %%%%%%%%%%%%%%%%%%
604: \begin{figure}[t]
605: \begin{center}
606: \epsfxsize=8cm
607: \leavevmode
608: \epsffile{blm_fig4.eps}
609: \caption{$\delta E_{\tbox{PRT}}$ and $\delta E$ as a function of
610: $x$ for the WLRM model with $b=100$, $\Delta=1$, and $N=5000$.
611: The $50\%$ probability width of $P(r)$ ($\bigtriangleup$)
612: and of $P_{\tbox{PRT}}(r)$ (dashed line) are plotted to verify the
613: validity of
614: Eq.~(\ref{Pw}) for $x<x_{\tbox{prt}}$. $\delta E_{\tbox{PRT}}$
615: (dotted line) and $\delta E$ ($\circ$) were obtained from
616: $\delta E_{\tbox{PRT}}^2 = \Delta \sum_r r^2 P_{\tbox{PRT}}(r)$ and
617: $\delta E^2 = \Delta \sum_r r^2 P(r)$, respectively.
618: The numerically obtained $\delta E_{\tbox{PRT}}$ (dotted line)
619: is hardly visible in the plot because it lies on top of
620: $\delta E_{\tbox{PRT}}$ from Eq.~(\ref{dEW}).
621: We also show $\Gamma$ (thin full line) from Eq.~({\ref{gamma}}).
622: For this set
623: of parameters $x_c\approx 0.57$ and $x_{\tbox{prt}}\approx 5.35$.}
624: \label{fig:fig4}
625: \end{center}
626: \end{figure}
627: %%%%%%%%%%%%%%%
628:
629:
630: The validity of Eqs.~(\ref{Pw}), (\ref{dEW}), and
631: (\ref{dE}) is confirmed in Fig.~\ref{fig:fig4} for $b=100$,
632: $\Delta=1$, and $N=5000$. There, we observe excellent
633: agreement between the expressions for $\delta E_{\tbox{PRT}}$
634: and $\delta E$ and the corresponding numerics.
635: To verify the validity of Eq.~(\ref{Pw}) we compare the
636: $50\%$ probability width (defined as the energy width of the
637: central $r$ region that contains $50\%$ of the probability)
638: of $P(r)$ to that of $P_{\tbox{PRT}}(r)$ finding reasonable
639: good agreement for $x<x_{\tbox{prt}}$, as expected.
640:
641:
642:
643: Does the LDOS analysis capture all the features
644: of the eigenstates? The answer turns out to be negative.
645: In case of the WBRM model there is still another
646: regime which is not captured by the LDOS analysis.
647: Namely, for $x>x_{\infty}$ where $x_{\infty}=b^{3/2}x_c$
648: the eigenstates of the Hamiltonian become
649: exponentially localized. This is the well known
650: Anderson (strong) localization effect. For $x>x_{\infty}$
651: the Hamiltonian is essentially $\mathcal{H}=\bm{B}$
652: as if the diagonal $\bm{E}$ does not exist.
653: Do we have an analogous type of crossover in case
654: of the WLRM model? The answer must be positive because
655: we know \cite{EM00} that in case of
656: the WLRM model, the Hamiltonian $\mathcal{H}=\bm{B}$
657: has multifractal (rather than localized) eigenstates.
658:
659:
660:
661:
662: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
663: %%%%%%%%%%%%%%%%%%%%%%%%%%% Multifractality %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
664: \section{Multifractality}
665:
666:
667: The multifractal structure of the eigenstates is
668: commonly characterized by the fractal dimension $D_2$
669: which is associated with the scaling of the inverse
670: participation ratio. Given an eigenstate
671: of $\mathcal{H}$ which is represented by a column
672: vector $\Psi_n$ we define the participation ratio as
673: %
674: \begin{equation}
675: \mathcal{N}_2 = \left[ \sum_n |\Psi_n|^4 \right]^{-1} \ .
676: \end{equation}
677: %
678: The exponent $D_2$ is defined via the scaling relation
679: %
680: \begin{equation} \label{eDdef}
681: \overline{\mathcal{N}_2} \ \ \propto \ \ N^{D_2} \ .
682: \end{equation}
683: %
684: where ${\overline{\mathcal{N}_2} \equiv \exp(\langle \ln \mathcal{N}_2 \rangle)}$.
685: The measure $\overline{\mathcal{N}_2}$ constitutes an estimate
686: for the typical number of non-zero eigenfunction components
687: of the column vector. For a localized state it equals
688: a constant number and hence $D_2=0$, while for an
689: extended non-fractal state it is proportional
690: to the size of the matrix $N$ and hence $D_2=1$.
691: In general one finds $0<D_2<1$.
692: %
693: The fractal dimension $D_2$ manifests itself in a variety
694: of physical circumstances. As examples we mention the conductance
695: distribution in metals \cite{AKL91,BHMM01}, the statistical properties
696: of the spectrum \cite{BHMM01}, the anomalous spreading of a wave-packet,
697: the spatial dispersion of the diffusion coefficient \cite{HK99} and the
698: anomalous scaling of delay times \cite{MK05}.
699:
700:
701:
702:
703:
704: %%%%%%%%%%%%%%%%%%%%%%%%%%% X = infty %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
705:
706:
707: In \cite{EM00} it was shown that the fluctuations
708: of $\mathcal{N}_2/\overline{\mathcal{N}_2}$ for $\mathcal{H}=\bm{B}$
709: are characterized by a universal probability distribution.
710: A theoretical estimation \cite{MFDQS96} gives
711: \begin{equation}
712: D_2 = \left\{ \matrix{
713: 1-(\pi b)^{-1} & \mbox{$b \gg 1$} \cr
714: 2b & \mbox{$b \ll 1$}
715: } \right. \ . \nonumber
716: \end{equation}
717: We notice that $D_2=D(b)$ gives a global fit
718: for the fractal dimension, where we define
719: %
720: \begin{equation} \label{Db}
721: D(b) = \frac{1}{1+(2.34 b)^{-1}} \ .
722: \label{D2inf}
723: \end{equation}
724: %
725: %
726: Note that (\ref{D2inf}) is also in agreement
727: with the numerical found \cite{V02}
728: value ${D_2 \approx 0.7}$ for $b=1$.
729: For sake of later analysis we have found that
730: the associated proportionality
731: factor in Eq.~(\ref{eDdef}) is $\exp(-G(b))$ where
732: %
733: \begin{equation}
734: G(b) \approx \frac{1}{1+(1.23 b)^{-1}} \ .
735: \end{equation}
736:
737:
738:
739:
740: %%%%%%%%%%%%%%%%%%%%%%%%%%% finite X %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
741:
742: Thus, based on the above,
743: we may say that for $x=\infty$
744: we expect to have multifractal
745: eigenstates. We turn now to
746: discuss the more general case
747: of finite $x$. We want to see how
748: the multifractality emerges as
749: we increase $x$ form zero to infinity.
750:
751:
752:
753:
754: In the numerics we assume without loss
755: of generality that the mean level spacing
756: is $\Delta=1$. This implies that for $b>1$
757: the threshold for mixing of levels is
758: %
759: %
760: ${x_c=\pi^{-1/2}}$,
761: while for $b<1$ it is
762: ${x_c=(\pi b)^{-1/2}}$.
763: %
764: %
765: Our main interest is in the
766: non-trivial regime $x > x_c$.
767: %
768: %
769: In Fig.~\ref{fig:fig5} we plot $ \ln \overline{\mathcal{N}_2}(x)$
770: for several values of $b$ and $N$.
771: We calculate the average using $25\%$ of the
772: eigenstates at the middle of the spectra
773: from a number of realizations of $\bm{B}$ summing up
774: a total of $100,000$ data values for each $(b,N)$.
775:
776:
777:
778: %%%%%%%%%%%%%%%%
779: \begin{figure}[t]
780: \begin{center}
781: \epsfxsize=8cm
782: \leavevmode
783: \epsffile{blm_fig5.eps}
784: \caption{$\bra \ln \mathcal{N}_2(x) \ket$ as a function of $x$
785: for $N=128$, 256, 512, and 1024.
786: The bandwidth is $b=100 (\circ)$, $40 (\Box)$, $10 (\Diamond)$,
787: $4 (\bigtriangleup)$, $1 (\lhd)$, $0.4 (\bigtriangledown)$,
788: $0.1 (\rhd)$, $0.04 ({+})$, and $0.01 ({\times})$.}
789: \label{fig:fig5}
790: \end{center}
791: \end{figure}
792: %%%%%%%%%%%%%%%%
793:
794:
795:
796: Looking at Fig.~\ref{fig:fig5} we see that for ${x=0}$
797: we have ${\ln \overline{\mathcal{N}_2} = 0}$
798: because all the eigenstates have only one component.
799: We observe that this zero value persists up
800: to a point $x=x_0(b)$. In principle one can
801: argue that $x_0(b)$ should be of the order $x_c$
802: and (for $b>1$) not larger
803: than $x_{\tbox{prt}} \approx \sqrt{b}x_c$.
804: However, from Eq.~(\ref{Db}) it is clear that
805: for $b>100$ we already have $D_2 \approx D(\infty) = 1$.
806: Therefore the $b$ dependence of $x_0$ can be
807: neglected, and in practice cannot be detected.
808: We shall see below that for any practical purpose
809: one can take $x_0\approx 0.15$.
810:
811:
812: As $x$ is increased beyond $x_0$ the participation
813: ratio $\mathcal{N}_2$ becomes larger.
814: As long as $x$ is not too large the $\bm{E}$ term
815: in the Hamiltonian dominates,
816: and therefore the size of the matrix is
817: of no importance. Indeed, we see that
818: the curves in Fig.~\ref{fig:fig5} are $N$ independent
819: for small $x$~values. From this plot we find
820: that the slope of the curves is given by
821: %
822: \begin{equation}
823: F(b) \approx \left\{ \matrix{
824: 0 & \mbox{$b<0.1$} \cr
825: 0.57+0.2\ln(b) & \mbox{$b\ge 0.1$}
826: } \right. \ .
827: \end{equation}
828:
829:
830:
831: For large enough $x$, the value
832: of $\ln \overline{\mathcal{N}_2}(x)$
833: saturates to the $x=\infty$ multifractal result.
834: We call the crossover point $x_{\infty}$.
835: %
836: Using the knowledge of both the $x < x_{\infty}$
837: behavior (as described in the previous paragraph),
838: and the $x > x_{\infty}$ behavior (saturation),
839: we deduce that
840: %
841: \begin{equation}
842: x_{\infty}(N;b) = x_0 \mbox{e}^{-G(b)/F(b)} N^{D(b)/F(b)} \ .
843: \end{equation}
844:
845:
846:
847: %%%%%%%%%%%%%%%%%%%%%%%
848: \begin{figure}[t]
849: \begin{center}
850: \epsfxsize=8cm
851: \leavevmode
852: \epsffile{blm_fig6.eps}
853: \caption{The scaled version of Fig.~\ref{fig:fig5}.
854: $ \bra \ln \mathcal{N}_2(x) \ket / \bra \ln \mathcal{N}_2(\infty) \ket$
855: is plotted against
856: $F(b) \ln(x/x_0) / \bra \ln \mathcal{N}_2(\infty) \ket$.
857: The parametric crossovers at $x=x_0$ and $x=x_{\infty}$,
858: are indicated by dashed lines.}
859: \label{fig:fig6}
860: \end{center}
861: \end{figure}
862: %%%%%%%%%%%%%%%%%%%%%%
863:
864:
865: Putting together all the above findings
866: we end up with the following global
867: scaling relation for the participation ratio:
868: %
869: \begin{equation}
870: \overline{\mathcal{N}_2}(x) = \left\{
871: \begin{array}{ll}
872: \approx 1 & \quad x < x_0 \\
873: (x/x_0)^{F(b)} & \quad x_0 < x < x_{\infty}(N;b) \\
874: \eexp{-G(b)}N^{D(b)} & \quad x > x_{\infty}(N;b)
875: \end{array}
876: \right. .
877: \label{D2theo}
878: \end{equation}
879: %
880: In Fig.~\ref{fig:fig6} we demonstrate this scaling. We plot
881: $\ln \overline{\mathcal{N}}_2(x) / \ln \overline{\mathcal{N}}_2(\infty)$
882: as a function of the scaled variable
883: $F(b) \ln(x/x_0) / \ln \overline{\mathcal{N}}_2(\infty)$.
884: We see clearly the trivial crossover at $x=x_0$
885: and the non-trivial crossover to multifractality
886: at $x=x_{\infty}$.
887:
888:
889: Possibly it is more instructive to describe
890: the behavior of $\overline{\mathcal{N}_2}$ as a function of $N$
891: for a given $x$. The interesting case is to have
892: a fixed value of $x$ which is much larger than $x_0$.
893: As we increase~$N$ we have a multifractal growth
894: $\overline{\mathcal{N}_2} \propto N^{D_2}$. This
895: goes on as long as $x_{\infty}(N;b)$ remains
896: smaller than $x$. After that $\overline{\mathcal{N}_2}$
897: reaches saturation, as implied by Eq.~(\ref{D2theo}).
898: The saturation value is related to the ``width"
899: of the LDOS, and hence has an algebraic dependence
900: on the dimensional strength ($x/x_0$) of the perturbation \cite{CH2000}.
901: It is pleasing to note that for $b\sim1$
902: we observe $F(b) \approx 2/3$ which is related
903: to considerations as in \cite{CH2000}.
904:
905:
906: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
907: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
908: \section{Conclusions}
909:
910:
911: We have analyzed the parametric evolution
912: of eigenstates for the WLRM model. Both, the standard WBRM and the
913: WLRM models, exhibit a crossover from a perturbative regime
914: where the LDOS is Lorentzian-like to a non-perturbative
915: regime where the LDOS is semicircle-like. However, there
916: is in both cases an additional crossover which is not
917: captured by the conventional LDOS analysis:
918: In the case of the standard WBRM model it is the
919: well studied crossover to an Anderson localization regime,
920: where the eigenstates become exponentially localized;
921: In the case of the WLRM model it is the emergence of multifractality.
922:
923: We have also shown that the WLRM model possess the $x$~invariance
924: property, absent in the standard WBRM model and in other models
925: that do not have a Lorenzian band profile.
926: Both the non-perturbative crossover and the $x$~invariance
927: property characterize realistic quantized Hamiltonians.
928:
929:
930: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
931: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
932:
933: \ \\
934:
935: {\bf Acknowledgments.}
936: This research was supported by a grant from the GIF, the
937: German-Israeli Foundation for Scientific Research and Development,
938: and by the Israel Science Foundation (grant No.11/02).
939:
940:
941: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
942: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
943:
944: \begin{thebibliography}{99}
945:
946:
947: \bibitem{W55}
948: E. Wigner, Ann. Math. {\bf 62}, 548 (1955); {\bf 65}, 203 (1957).
949:
950: % power law
951:
952: \bibitem{BCH}
953: A. Barnett, D. Cohen and E.J. Heller, Phys. Rev. Lett. {\bf 85},
954: 1412 (2000); J. Phys. A {\bf 34}, 413 (2001).
955:
956: \bibitem{MKC05}
957: J. A. M\'endez-Berm\'udez, T. Kottos, and D. Cohen, Phys. Rev. E
958: {\bf 72}, 027201 (2005).
959:
960: % ldos
961:
962: \bibitem{MLI04}
963: J. A. M\'endez-Berm\'udez, G. A. Luna-Acosta, and F. M. Izrailev,
964: Physica E {\bf 22}, 881 (2004); Phys. Rev. E {\bf 68}, 066201 (2003).
965:
966: \bibitem{B03}
967: L. Benet, et. al., J. Phys. A: Math. Gen. {\bf 36}, 1289 (2003); L. Benet, et. al.,
968: Phys. Lett. A {\bf 277}, 87 (2000); F. Borgonovi, I. Guarneri, F. M. Izrailev,
969: Phys. Rev. E {\bf 57}, 5291 (1998).
970:
971: \bibitem{CK01}
972: D. Cohen and T. Kottos, Phys. Rev. E {\bf 63}, 036203 (2001).
973:
974: \bibitem{VLG02}
975: R. O. Vallejos, C. H. Lewenkopf, and Y. Gefen, Phys. Rev. B
976: {\bf 65}, 085309 (2002); G. Murthy, et. al., ibid, {\bf 69}, 075321 (2004);
977: L. G. G. V. Dias da Silva, et. al., ibid, {\bf 69}, 075311 (2004).
978:
979: \bibitem{CH2000}
980: D. Cohen and E. J. Heller, Phys. Rev. Lett. {\bf 84}, 2841 (2000).
981:
982: \bibitem{HKG06}
983: M. Hiller, T. Kottos, and T. Geisel, submitted (2006).
984:
985: \bibitem{HZB95}
986: V. K. B. Kota, Phys. Rep. {\bf 347}, 223 (2001); V. Zelevinsky, et. al.,
987: Phys. Rep. {\bf 276}, 85 (1996).
988:
989: \bibitem{FGGP99}
990: L. Kaplan and T. Papenbrock, Phys. Rev. Lett. {\bf 84}, 4553 (2000);
991: V. V. Flambaum, A. A. Gribakina, G. F. Gribakin, M. G. Kozlov, Phys. Rev. A {\bf 50},
992: 267 (1994).
993:
994: % invariance
995:
996: \bibitem{Wilkinson}
997: E.J. Austin and M. Wilkinson, Nonlinearity {\bf 5}, 1137 (1992).
998:
999: \bibitem{Alhassid}
1000: H. Attias and Y. Alhassid, Phys. Rev. E {\bf 52}, 4776 (1995).
1001:
1002:
1003:
1004:
1005: % multifractality
1006:
1007: \bibitem{EM00}
1008: F. Evers and A. D. Mirlin, Phys. Rev. Lett. {\bf 84}, 3690 (2000);
1009: E. Cuevas, M. Ortuno, V. Gasparian, and A. Perez-Garrido, Phys. Rev.
1010: Lett. {\bf 88}, 016401 (2002).
1011:
1012:
1013: \bibitem{AKL91}
1014: B. L. Altshuler, V. E. Kravtsov, I. V. Lerner, in {\it Mesoscopic
1015: Phenomena in Solids}, ed. B. L. Altshuler, P. A. Lee, R. A. Webb
1016: (North Holland, Amsterdam, 1991).
1017:
1018: \bibitem{BHMM01}
1019: D. Braun, E. Hofstetter, G. Montambaux, and A. MacKinnon, Phys. Rev.
1020: Lett. {\bf 64}, 155107 (2001); K. Slevin and T. Ohtsuki, ibid. {\bf 82},
1021: 382 (1999); K. A. Muttalib and P. W\"olfle, ibid. {\bf 83}, 3013 (1999);
1022: P. Markos, ibid. {\bf 83}, 588 (1999); B. Shapiro, ibid. {\bf 65}, 1510
1023: (1990).
1024:
1025: \bibitem{HK99}
1026: J. T. Chalker and G. J. Daniell, Phys. Rev. Lett. {\bf 61}, 593 (1988);
1027: B. Huckestein and R. Klesse, Phys. Rev. B {\bf 59}, 9714 (1999).
1028:
1029: \bibitem{MK05}
1030: J. A. M\'endez-Berm\'udez and T. Kottos, Phys. Rev. B {\bf 72}, 064108 (2005).
1031:
1032: \bibitem{MFDQS96}
1033: A. D. Mirlin, Y. V. Fyodorov, F.-M. Dittes, J. Quezada, and T. H.
1034: Seligman, Phys. Rev. E {\bf 54}, 3221 (1996).
1035:
1036: \bibitem{V02}
1037: I. Varga, Phys. Rev. E {\bf 66}, 094201 (2002).
1038:
1039:
1040:
1041:
1042:
1043:
1044: \end{thebibliography}
1045:
1046: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1047: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1048: \end{document}
1049: