1: \documentclass[aps,prb,twocolumn,floats,showpacs]{revtex4}
2: \usepackage{epsfig}
3: \usepackage{bm}
4: \usepackage{latexsym}
5:
6: \begin{document}
7:
8: \newcommand{\hide}[1]{}
9: \newcommand{\tbox}[1]{\mbox{\tiny #1}}
10: \newcommand{\half}{\mbox{\small $\frac{1}{2}$}}
11: \newcommand{\sinc}{\mbox{sinc}}
12: \newcommand{\const}{\mbox{const}}
13: \newcommand{\trc}{\mbox{trace}}
14: \newcommand{\intt}{\int\!\!\!\!\int }
15: \newcommand{\ointt}{\int\!\!\!\!\int\!\!\!\!\!\circ\ }
16: \newcommand{\eexp}{\mbox{e}^}
17: \newcommand{\bra}{\left\langle}
18: \newcommand{\ket}{\right\rangle}
19: \newcommand{\EPS} {\mbox{\LARGE $\epsilon$}}
20: \newcommand{\ar}{\mathsf r}
21: \newcommand{\im}{\mbox{Im}}
22: \newcommand{\re}{\mbox{Re}}
23: \newcommand{\bmsf}[1]{\bm{\mathsf{#1}}}
24: \newcommand{\new}[1]{{\bf #1}}
25:
26: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
27:
28: \title{Resonance width distribution for high-dimensional random media }
29: \author{Matthias Weiss$^{1}$, J. A. M\'endez-Berm\'udez$^{2,3}$, and Tsampikos Kottos$^{2,4}$}
30: \affiliation{$^1$Cellular Biophysics Group (BIOMS), German Cancer Research
31: Center, Im Neuenheimer Feld 580, D-69121 Heidelberg, Germany}
32: \affiliation{$^2$Max-Planck-Institut f\"ur Dynamik und Selbstorganisation, Bunsenstrasse 10,
33: D-37073 G\"ottingen, Germany}
34: \affiliation{$^3$Department of Physics, Ben-Gurion University, Beer-Sheva 84105, Israel}
35: \affiliation{$^4$Department of Physics, Wesleyan University, Middletown, CT
36: 06459-0155, USA}
37:
38: \date{\today}
39:
40: \begin{abstract}
41: We study the distribution of resonance widths ${\cal P}(\Gamma)$ for
42: three-dimensional (3D) random scattering media and analyze how it changes
43: as a function of the randomness strength. We are able to identify in
44: ${\cal P}(\Gamma)$ the system-inherent fingerprints of the metallic,
45: localized, and critical regimes. Based on the properties of resonance
46: widths, we also suggest a new criterion for determining and analyzing
47: the metal-insulator transition. Our theoretical predictions are verified
48: numerically for the prototypical 3D tight-binding Anderson model.
49: \end{abstract}
50: \pacs{03.65.Nk, 71.30.+h, 72.20.Dp, 73.23.-b}
51:
52: \maketitle
53:
54: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
55: \section{Introduction}
56: \label{sec:introduction}
57:
58: Quantum mechanical scattering in systems with complex internal dynamics
59: has been a subject of intensive research activity for a number of years.
60: The interest was motivated by various areas of physics, ranging from nuclear
61: \cite{MW69}, atomic\cite{atomic} and molecular \cite{molecular} physics,
62: to mesoscopics \cite{B97}, quantum chaos \cite{S89,S99}, and classical
63: wave scattering \cite{DS90}. Recently, the interest in this subject was
64: renewed due to technological developments in quantum optics associated
65: with the construction of new type of lasers \cite{WAL95,NS97} and the
66: experimental investigation of atoms in optical lattices \cite{Raizen}.
67:
68: The most fundamental object which characterizes the process of quantum
69: scattering is the scattering matrix $S$, where $S$ relates the amplitudes
70: of waves that enter and leave a scattering region. Of great interest are
71: the statistical properties of the poles of the $S$ matrix. They determine
72: the conductance fluctuations of a quantum dot in the Coulomb blockade regime
73: \cite{ABG02} or the current relaxation \cite{AKL91}. The poles of the $S$
74: matrix are related to resonance states occurring at complex energies ${\cal
75: E}_n = E_n - \frac i2 \Gamma_n$, where $E_n$ is the position and $\Gamma_n$
76: the width of the resonance. Resonances correspond to ``eigenstates" of the
77: open system that decay in time due to the coupling to the ``outside world".
78:
79: For chaotic systems Random Matrix Theory (RMT) is applicable and the
80: distributions of resonance widths ${\cal P}(\Gamma)$ is known. A review can
81: be found in Ref. \onlinecite{FS97} (see also Ref. \onlinecite{KS00}).
82: As the disorder increases, the system becomes diffusive and the deviations
83: from RMT increase drastically. For low dimensional random systems
84: in the metallic regime the distribution of resonances ${\cal P} (\Gamma)$
85: was found recently \cite{BGS91,OKG02,OKG03}. For the strongly disordered
86: limit, where localization dominates, ${\cal P} (\Gamma)$ was investigated
87: by various groups \cite{TF00,SJNS00,PROT04} as well. At the same time an
88: attempt to understand systems at critical conditions was undertaken in Refs.
89: \onlinecite{SOKG00} and \onlinecite{KW02}. The latter deals with random
90: systems of higher dimensions, the most prominent of which is the
91: three-dimensional (3D) Anderson model. It undergoes a Metal-Insulator
92: Transition (MIT) with increasing strength of disorder \cite{A58}.
93:
94: In this paper we extend our previous analysis on the 3D Anderson model \cite{KW02}
95: and study the distribution of resonance widths as we change the disorder strength.
96: Based on the analysis of ${\cal P}(\Gamma)$ we propose a new method to locate
97: the MIT. The paper is organized as follows: In Section \ref{sec:basics} the 3D
98: Anderson model and the scattering formalism are
99: introduced. In Sec. \ref{sec:results} we discuss the consequences of
100: localization in the distribution of resonance widths in the diffusive and
101: localized regimes as well as at the MIT and show the numerical results
102: supporting our arguments. In Sec. \ref{sec:scale} we investigate a new method for determining
103: and analyzing the emergence of the MIT and propose a scaling theory near
104: the critical point. Finally, our conclusions are given in Sec.
105: \ref{sec:conclusions}.
106:
107: %------OKOK 09.08.2005
108: \section{The 3D Anderson model and the scattering setup}
109: \label{sec:basics}
110:
111: The Anderson model with diagonal disorder on a 3D cubic lattice
112: is described by the tight-binding Hamiltonian (TBH)
113: %
114: \begin{equation}
115: \label {tbh}
116: H_0=\sum_{\bf n} |{\bf n}\rangle W_{\bf n}\langle {\bf n}| + \sum_{{\bf (n,m)}}
117: |{\bf n}\rangle \langle {\bf m}| \ ,
118: \end{equation}
119: %
120: where ${\bf n}\equiv (n_x,n_y,n_z)$ labels all the $N=L^3$ sites of the cubic
121: lattice, while the second sum is taken over all nearest-neighbor pairs
122: ${\bf (n,m)}$ on the lattice. The on-site potential $W_{\bf n}$ for
123: $1\leq n_x,n_y,n_z\leq L$ is independently and identically distributed with
124: probability ${\cal P}(W_{\bf n})$. We use three different distributions for
125: the random potential: (a) a box distribution, i.e., the $W_{\bf n}$ are
126: uniformly distributed on the interval $[-W/2, W/2]$; (b) a Gaussian distribution
127: with zero mean and variance $W^2/12$; and (c) a Cauchy distribution
128: ${\cal P}(W_{\bf n})=W/\pi (W_{\bf n}^2+W^2)$. For the system defined by
129: Eq.~(\ref{tbh}) the MIT for $E\simeq 0$ occurs for $W=W_c$ with (a)
130: $W_c\simeq 16.5$, (b) $W_c\simeq 21.3$, and (c) $W_c\simeq 4.26$ (see
131: Ref.~\onlinecite{RMS01}). Then, for $W<W_c$ ($W>W_c$) the system is in the
132: metallic (insulating) regime.
133:
134: \begin{figure}
135: \begin{center}
136: \epsfxsize=6.4cm
137: \leavevmode
138: \epsffile{Fig1.eps}
139: \caption{Scattering setup. The sample is a cubic lattice of linear length $L$.
140: To each of the $M=L^2$ sites of the layer $n_x=1$ semi-infinite single mode leads
141: are attached.}
142: \label{fig:setup}
143: \end{center}
144: \end{figure}
145:
146: We turn the isolated system to a scattering one by attaching $M=L^2$ semi-infinite single
147: mode leads to each site of the layer $n_x=1$, as depicted in Fig. \ref{fig:setup}.
148: Each lead is described by a one-dimensional semi-infinite TBH
149: %
150: \begin{equation}
151: \label{leads}
152: H_M=\sum^{-\infty}_{n=1} (|n><n+1| + |n+1><n|) \ .
153: \end{equation}
154:
155: Using standard methods \cite{MW69} one can write the scattering matrix in the form
156: \cite{SOKG00,KW02}
157: %
158: \begin{equation}
159: \label{smatrix}
160: S(E) = {\bf 1}-2i \sin (k)\, {\cal W}^{\,T} (E-{\cal H}_{\rm eff})^{-1} {\cal W} \ ,
161: \end{equation}
162: %
163: where ${\bf 1}$ is the $M\times M$ unit matrix, $k=\arccos(E/2)$ is the wave
164: vector supported in the leads, and ${\cal H}_{\rm eff}$ is an effective
165: non-hermitian Hamiltonian given by
166: %
167: \begin{equation}
168: \label{Heff}
169: {\mathcal{H}}_{\rm eff}=H_0- e^{ik} {\cal W}{\cal W}^{\,T} \ .
170: \end{equation}
171: %
172: Here, ${\cal W}$ is a $N\times M$ matrix that specifies at which site of the sample we
173: attach the leads. Its elements are equal to zero or $\sqrt{w}$, with $0<\sqrt{w}\le 1$,
174: where $w$ is the coupling strength. Below, unless stated
175: otherwise, we will always consider the case $w=1$.
176: Moreover, since $\arccos (E/2)$ changes only slightly in the center of the band,
177: we set $E=0$ and neglect the energy dependence of ${\mathcal{H}}_{\rm eff}$.
178: The poles of the $S$ matrix are then equal to the complex zeros of
179: \begin{equation}
180: \label{poleseq}
181: \det [{\cal E}-H_{\rm eff}]=0.
182: \end{equation}
183: >From Eqs.~(\ref{smatrix}) and (\ref{poleseq}) it is clear that the formation of
184: resonances is closely related to the dynamics in the scattering region, governed
185: by $H_0$.
186:
187: In order to investigate the distributions of resonance widths
188: we used samples with $L=20$ as a maximum size. For better statistics a
189: considerable number of different disorder realizations was considered.
190: In all cases we had at least 10 000 data for statistical processing.
191:
192: %OKOK 9.8.2005
193: \section{Distribution of resonance widths: Results and discussion}
194: \label{sec:results}
195:
196: %---------------------------------------------------------------------------------------
197: \subsection {Metallic Regime}
198: \label{subsec:difres}
199:
200: When the disorder strength $W$ is smaller than $W_c$, but still large enough so that
201: the mean free path is smaller than the system size, the system is in the metallic regime.
202:
203: Recently, a lot of research activity was devoted to the understanding
204: of the statistical properties of various physical quantities (such as
205: conductance, local density of
206: states, current relaxation times) in finite-size random systems in the metallic regime.
207: The outcome of these studies indicated that the tails of these distribution functions
208: show large deviations from the universal Random Matrix Theory (RMT) results, expected
209: to be valid \cite{E83} in the limit of infinite dimensionless conductance
210: $g=\Gamma_{\rm Th}/ \Delta=D L$. Here, $\Gamma_{\rm Th} \sim D/L^2$ is the typical
211: inverse time (Thouless time) that an excitation needs to diffuse (with diffusion
212: coefficient $D$) in order to reach the boundary of a system, with linear size
213: $L$, and $\Delta\sim 1/L^3$ is the mean level spacing.
214:
215: The origin of these deviations was found to be related to the existence of
216: eigenstates which are unusually localized around a center of localization.
217: These states are precursors of the Anderson localization and were termed {\it
218: prelocalized} states \cite{M00,FE95,KOG03,UMRS00}. In 3D conductors they have
219: sharp amplitude peaks on the top of a homogeneous background \cite{M00,FE95}.
220:
221: We start our analysis by investigating the effects of prelocalized states in the
222: distribution of resonance widths. It is natural to expect that these states with
223: localization centers at the bulk of the sample are affected only weakly when opening
224: the system at the boundaries. Therefore, prelocalized states decay very slowly to the
225: continuum leading us to the conclusion that the corresponding resonance widths
226: (inverse lifetime) $\Gamma$ are smaller than the mean level spacing $\Delta$. Hence,
227: assuming the validity of standard first order perturbation theory (that can be applied
228: if the coupling of the sample to the leads is weak, $w\ll 1$) we get
229: \begin{equation}
230: \label{pertgamma}
231: {\Gamma\over 2} = \langle \Psi|{\cal W}^{\dagger}{\cal W}|\Psi\rangle \propto
232: \sum_{n\in {\rm boundary}}|\Psi(n)|^2 \sim L^2 |\Psi(L)|^2 \ ,
233: \end{equation}
234: where $|\Psi (L)|^2$ is the wavefunction intensity of a pre-localized state at
235: the boundary of the sample. At the same time, the distribution of wavefunction
236: components at the boundary was found to be \cite{FE95}
237: \begin{equation}
238: \label{theta}
239: {\cal P}(\theta) \sim \exp\left[-C_1\ln^3 \left(\theta\right)\right] \ ,
240: \end{equation}
241: with $\theta^{-1}=L\Psi(L)$ and $C_1 \propto g$. Using Eq.~(\ref{theta})
242: together with Eq.~(\ref{pertgamma}) we obtain
243: \begin{equation}
244: \label{difgammaS}
245: {\cal P}(1/\Gamma) \sim \exp\left[-C_2 \ln^3 (1/\Gamma)\right] \ ,
246: \end{equation}
247: where $C_2 \propto g$.
248:
249: As can be seen from Fig.~\ref{fig:Fig2} the prediction of Eq.~(\ref{difgammaS}),
250: obtained using perturbation theory, holds even for strong coupling. Indeed,
251: the reported data for the 3D Anderson model in the diffusive regime, plotted
252: as $\ln {\cal P}(1/\Gamma)$ vs $\ln^3(1/\Gamma)$, shows a linear behaviour.
253: This comes as a surprise since in Fig.~\ref{fig:Fig2} we have considered
254: perfect coupling, $w=1$.
255: Differences in slope correspond to different dimensionless conductances $g$
256: induced by the varying values of disorder strengths.
257:
258: Next, we turn to the analysis of ${\cal P}(\Gamma)$ for
259: $\Gamma \gtrsim \Gamma_{\rm Th}\gg \Delta$.
260: In order to go on we need to recall that
261: the inverse of $\Gamma$ represents the quantum lifetime of a particle in a resonant state
262: escaping into the leads. Moreover, we assume that the particles are uniformly distributed
263: inside the sample and spread until they reach the boundary where they are absorbed. Then,
264: we can associate the corresponding lifetimes with the time $t_R\sim 1/ \Gamma_R $ a particle
265: needs to reach the boundaries, when starting a distance $R$ away. The relative number of
266: states that require a time $t<t_R$ in order to reach the boundaries (or equivalently the
267: number of states with $\Gamma >\Gamma_R$) is
268: \begin{equation}
269: \label{igam}
270: {\cal P}_{\rm int}(\Gamma_R)=\int_{\Gamma_R}^{\infty}{\cal P}(\Gamma)d\Gamma
271: \sim\frac{V(t_R)}{L^3} \ ,
272: \end{equation}
273: where $V(t_R)\sim L^3-(L-R)^3$ is the volume populated by all particles with
274: lifetimes $t<t_R$.
275:
276: \begin{figure}
277: \begin{center}
278: \epsfxsize=8.4cm
279: \leavevmode
280: \epsffile{Fig2.eps}
281: \caption{The distribution of resonance widths, plotted as ${\cal P}(1/\Gamma)$
282: vs $\ln^3(1/\Gamma)$, for $\Gamma < \Delta$ in the diffusive regime.
283: $L=16$ and $W=10$, 12, and 14 (from left to right).
284: The dashed lines are (shifted) linear fittings to the distributions.}
285: \label{fig:Fig2}
286: \end{center}
287: \end{figure}
288:
289: Assuming now diffusive spreading,
290: \[
291: R^2=D\cdot t_R \ ,
292: \]
293: we get from Eq.~(\ref{igam}) (to leading order with respect to $\Gamma_{\rm Th}/\Gamma$)
294: \begin{equation}
295: \label{gamlargedif}
296: {\cal P}({\widetilde \Gamma}) \sim \left({{\widetilde \Gamma_{\rm Th}}\over {\widetilde \Gamma}}\right)^{3/2} \
297: =\left({g\over{\widetilde \Gamma}}\right)^{3/2}
298: \end{equation}
299: where we refer to the rescaled variable $\widetilde \Gamma = \Gamma/\Delta$.
300: Equation (\ref{gamlargedif}) is valid as long as the leads are attached
301: to the boundary of the sample.
302:
303: Here, it is interesting to point out that a different way of opening the system might
304: lead to a different power law behavior for ${\cal P}(\Gamma)$. Such a situation can
305: be realized if instead of opening the system at the boundaries we attach one lead
306: somewhere in the sample. In such a case we have
307: \[
308: {\cal P}_{\rm int}(\Gamma_R) \sim \frac{V(t_R)}{L^3} \approx \frac{R^3}{L^3} =
309: \frac{(D \cdot t_R)^{3/2}}{L^3} \sim \left( \frac{\Gamma_{\rm Th}}{\Gamma_R}
310: \right)^{3/2} \ ,
311: \]
312: leading to
313: \begin{equation}
314: \label{finalg2}
315: {\cal P}(\Gamma) \sim \Gamma^{-5/2} \ ,
316: \end{equation}
317: where we used $V(t_R)\sim R^3$. The above results are correct for any number
318: of leads $M$ such that the ratio $M/L^3$ scales as $1/L^3$.
319:
320: If, on the other hand, we attach a single lead to the boundary of the sample we obtain
321: \begin{equation}
322: \label{finalg1a}
323: {\cal P}(\Gamma) \sim \Gamma^{-2} \ .
324: \end{equation}
325: This behavior is due to the fact that the decay takes place at the surface, leading
326: to a situation similar to that of a 2D system \cite{OKG03}.
327:
328: \begin{figure}
329: \begin{center}
330: \epsfxsize=8.4cm
331: \leavevmode
332: \epsffile{Fig3.eps}
333: \caption{The resonance width distribution ${\cal P}({\widetilde \Gamma})$
334: for $L=16$ and various scattering configurations (from up to down): $M=L^2$
335: leads attached to the boundary, one lead attached to the boundary, and one
336: lead attached to a site in the bulk of the sample. The dashed lines are the
337: corresponding theoretical predictions for $\Gamma \gtrsim \Gamma_{\rm Th}$
338: given by Eqs.~(\ref{gamlargedif}-\ref{finalg1a}), see main text.}
339: \label{fig:metfig1}
340: \end{center}
341: \end{figure}
342:
343: In Fig. \ref{fig:metfig1} we present numerical data for the 3D Anderson
344: model in the metallic regime. We verify the validity of the
345: theoretical predictions given by Eqs. (\ref{gamlargedif}-\ref{finalg1a})
346: by using various configurations: $M=L^2$ leads attached to the boundary,
347: one lead attached to the boundary, and one lead attached to a site in
348: the bulk of the sample, respectively. We observe a good agreement with
349: the expected behavior in all cases.
350:
351: %-------------------------------------------------------------------------------------------------
352: \subsection {Localized Regime}
353: \label{subsec:locres}
354:
355: When the disorder strength $W$ is larger than $W_c$ the system is in
356: the localized regime. In this regime the eigenfunctions are exponentially
357: localized in space and, as a consequence, transmission is inhibited and
358: the system behaves as an insulator.
359:
360: Various groups \cite{TF00,SJNS00,PROT04} had investigated the resonance width
361: distribution of low dimensional random media in the localized regime during
362: the last years. In the region of exponentially narrow resonances $\Gamma<\Gamma_0
363: =\exp(-2L/l_{\infty})$ the distribution was found to be log-normal, i.e.,
364: \begin{equation}
365: \label{locres2}
366: {\cal P}(\widetilde \Gamma)\sim \exp\left[-\left(4 {L\over l_{\infty}}\right)^{-1}
367: \ln^2(\widetilde \Gamma)\right] , \quad \Gamma< \Gamma_0 \ .
368: \end{equation}
369: This result is analogous to the conductance distribution of localized systems.
370: Equation (\ref{locres2}) essentially relies on two assumptions: first, that
371: eigenfunction components are randomly distributed with no long-range correlations;
372: and second, that they are exponentially localized with a normal distribution of
373: localization lengths.
374:
375: It is reasonable to assume that the same arguments leading to Eq.~(\ref{locres2})
376: applies as well for high-dimensional random media like the 3D Anderson model in the
377: localized regime. Indeed, our numerical results reported in Fig.~\ref{fig:locfig1}(a)
378: show good agreement with the theoretical expectation (\ref{locres2})
379: which supports our assumption.
380:
381: \begin{figure}
382: \begin{center}
383: \epsfxsize=8.4cm
384: \leavevmode
385: \epsffile{Fig4.eps}
386: \caption{\label{loc}${\cal P}(\widetilde{\Gamma})$ in the localized regime for
387: various combinations of $W$ and $L$ in the range $\widetilde{\Gamma}\le 1$.
388: The log-normal decay is highlighted by Gaussian fits (full curves) whose maximum
389: decreases with increasing strength of disorder and also shifts towards smaller
390: values of $\widetilde{\Gamma}$.
391: When keeping the ratio $l_\infty/L\approx0.136$ fixed, similar distributions
392: were obtained for different combinations of $L$ and $W$ (filled circles and open
393: squares).
394: (b) For $\widetilde{\Gamma}\ge 1$ the power-law decay
395: ${\cal P}(\widetilde{\Gamma})\sim 1/\widetilde{\Gamma}$ is observed (dashed line)
396: which becomes more robust for increasing strength of disorder.}
397: \label{fig:locfig1}
398: \end{center}
399: \end{figure}
400:
401: On the other hand, we found [see Fig.~\ref{fig:locfig1}(b)] that the long tails
402: of the distribution behave as
403: \begin{equation}
404: \label{locres1}
405: {\cal P}({\widetilde \Gamma})\sim \left({l_{\infty}\over L}\right)
406: {1\over {\widetilde \Gamma}},\quad \Gamma_0< \Gamma \ll 1/L \ .
407: \end{equation}
408: Equation (\ref{locres1}) can be easily understood when employing Eq.~(\ref{igam}).
409: The new ingredient is that wavefunctions are exponentially localized:
410: $|\Psi(r)|\sim l_{\infty}^{-3/2} \exp(-r/l_{\infty})$. Using simple perturbation
411: arguments \cite{TF00} [see Eq.~(\ref{pertgamma})], i.e. $\Gamma \sim |\Psi(r)|^2$, we obtain
412: \[
413: R^3 \sim l_{\infty}^3 \ln^3(l_{\infty}^3\Gamma) \ .
414: \]
415: By inserting this into Eq.~(\ref{igam}) we get Eq.~(\ref{locres1}), to leading order
416: with respect to $l_{\infty}/L$.
417:
418: The region of large $\Gamma$ values is essentially determined by the
419: coupling to the continuum,
420: so it should be model-dependent. Nevertheless, it is reasonable to assume that the
421: number of resonances involved is constant, of order $l_{\infty}$, and therefore the
422: extreme tail of the distribution should subside at large $L$ at the rate
423: $\sim l_{\infty}/L$.
424:
425: Let us finally note that in the thermodynamic limit $L\rightarrow \infty$ the
426: probability of finding an eigenstate at any finite distance from the boundary
427: is equal to zero. Thus the distribution of the resonance widths in this case
428: approaches a delta function centered at zero.
429:
430: %okok 09.08.2005
431: %-------------------------------------------------------------------------------------------------
432: \subsection {Criticality}
433: \label{subsec:critres}
434:
435: The MIT, where $W=W_c$, is characterized by several {\it critical} properties:
436: the level statistics acquires a scale-independent form \cite{M00,SSSLS93,AS86,CKL96}
437: while the eigenfunctions show strong fluctuations on all length scales and
438: obey multifractal distributions \cite{M00,FE95,W80,SG91,EM00}.
439:
440: In Ref. \onlinecite{KW02} it was found that ${\cal P}(\widetilde \Gamma)$ follows
441: a new universal distribution, i.e., independent of the microscopic details of the
442: random potential and number of attached leads.
443: Specifically, it decays asymptotically with the power
444: %
445: \begin{equation}
446: \label{MITgam}
447: {\cal P}({\widetilde \Gamma})\sim g_{\rm c}^{1/3} {\widetilde \Gamma}^{-(1+1/3)} \ ,
448: \end{equation}
449: %
450: which is different from those found for chaotic, metallic, or localized systems
451: (see Ref. \onlinecite{FS97}, Eq.~(\ref{gamlargedif}), and Eq.~(\ref{locres1}),
452: respectively).
453:
454: One can relate the power-law decay (\ref{MITgam}) to the anomalous diffusion at the
455: MIT. Indeed, at the MIT the conductance of a 3D disordered sample has a
456: finite value $g_{\rm c}\sim 1$. Approaching the MIT from the metallic regime one has
457: $g\sim E_T/ \Delta$, where $E_T=D/R^2$ is the Thouless energy, $D$ is the diffusion
458: coefficient, and $\Delta \sim 1/R^3$ is the mean level spacing of a
459: sample with linear size $R$. This yields $D\sim g_{\rm c}/R$ at $W_c$.
460: Taking into account that $D=R^2/t_R$, we get for the spreading of an excitation at
461: the MIT
462: %
463: \[
464: R^3 \sim g_{\rm c} \cdot t_R \ .
465: \]
466: %
467: Then, straightforward application of Eq.~(\ref{igam}) leads to Eq.~(\ref{MITgam}).
468:
469: \begin{figure}
470: \begin{center}
471: \epsfxsize=8.4cm
472: \leavevmode
473: \epsffile{Fig5.eps}
474: \caption{Universal behavior of ${\cal P}({\widetilde \Gamma})$ at the MIT [reported
475: here as ${\cal P}(\ln({\widetilde \Gamma}))$] for various sample sizes $L$ and potential
476: distributions.}
477: \label{fig:crit1}
478: \end{center}
479: \end{figure}
480:
481: In Figs.~\ref{fig:crit1} and \ref{fig:crit2} we report some numerical results for
482: the 3D Anderson model at the MIT.
483: Figure \ref{fig:crit1} shows the distribution of the logarithm of the rescaled
484: resonance widths ${\cal P}(\ln({\widetilde \Gamma}))$ for the three
485: different distributions ${\cal P}(W_{\bf n})$ of the random potential and for
486: various sample sizes $L$. The body of the distribution function in all cases
487: coincides and does not change its shape or width. Of course, the far tail of
488: this universal distribution develops better with increasing $L$.
489: The sharp peak appearing at the right is an artifact of our choice to neglect
490: the energy dependence of ${\cal H}_{\rm eff}$. We thus confirm that at the
491: MIT the distribution of rescaled resonances is indeed scale-invariant independent
492: of the microscopic details of the potential.
493:
494: >From Fig. \ref{fig:crit2}(a) an inverse power law ${\cal P}_{ \rm int}({\widetilde
495: \Gamma}) \sim {\widetilde \Gamma}^{ -\alpha}$ is evident. The best fit to the numerical
496: data yields $\alpha= 0.333\pm 0.005$ in accordance with Eq.~(\ref{MITgam}). Here,
497: the case of perfect coupling ($w=1$) has been considered. Different coupling strengths
498: are going to affect this behavior as can be seen from Fig. \ref{fig:crit2}(b).
499:
500: \begin{figure}
501: \begin{center}
502: \epsfxsize=8.4cm
503: \leavevmode
504: \epsffile{Fig6.eps}
505: \caption{(a) The integrated distribution ${\cal P}_{\rm int}({\widetilde \Gamma})$
506: in the case of perfect coupling, $w=1$, for various sample sizes $L$ and potential
507: distributions. The dashed line is the theoretical prediction
508: ${\cal P}_{\rm int}({\widetilde \Gamma}) \sim {\widetilde \Gamma}^{-0.333}$, see
509: Eq. (\ref{MITgam}).
510: (b) ${\cal P}_{\rm int}({\widetilde \Gamma})$ for a box distribution, $L=10$, and
511: different coupling strengths: $w=0.001$, 0.01, and 0.5 from left to right.}
512: \label{fig:crit2}
513: \end{center}
514: \end{figure}
515:
516: %-----------------------------------------------------------------------------------
517: \section{ A scaling Theory for the Resonances Widths}
518: \label{sec:scale}
519:
520: In the original proposal of the scaling theory of localization, the conductance $g$
521: is the relevant parameter \cite{A58,AALR79}. A manifestation of this statement is seen
522: in Eqs.~(\ref{difgammaS},\ref{gamlargedif},\ref{locres2},\ref{locres1},\ref{MITgam}) where
523: ${\cal P}_{\rm int}^{0}\equiv {\cal P}_{\rm int}({\widetilde \Gamma}_0)$ is proportional
524: to the conductance $g$. It is therefore natural to expect that ${\cal P}_{\rm int}^{0}$
525: will follow a scaling behavior for finite $L$ (and for some ${\widetilde \Gamma}_0\sim 1$)
526: that is similar to the one obeyed by the conductance $g$.
527: The following scaling hypothesis was therefore postulated in Ref. \onlinecite{KW02}:
528: %
529: \begin{equation}
530: \label{scale}
531: {\cal P}_{\rm int}^{0}(W,L)=f(L/l_{\infty}(W)) \ .
532: \end{equation}
533: %
534: In the insulating phase
535: ($W>W_c$) the conductance of a sample with length $L$ behaves as
536: $g(L)\sim \exp(-L/l_\infty)$ due to the exponential localization of the eigenstates,
537: and therefore we have $g(L_1) < g(L_2)$ for $L_1>L_2$. Based on Eq.~(\ref{MITgam})
538: we expect the same behavior for ${\cal P}_{\rm int}^{0}$; i.e., for every finite $L_1>L_2$
539: we must have ${\cal P}_{\rm int}^{0}(W,L_1)<{\cal P}_{\rm int}^{0}(W,L_2)$. On the other
540: hand, in the metallic regime ($W<W_c$) we have that $g(L)=D L$ and therefore we expect
541: from Eq.~(\ref{MITgam}) ${\cal P}_{\rm int}^{0}(W,L_1)>{\cal P}_{\rm int}^{0}(W,L_2)$.
542: Thus, the critical point is the one at which the size effect changes its sign, or in other
543: words, the point where all curves ${\cal P}_{\rm int}^{0}(W,L)$ for various $L$ cross.
544: One can reformulate the last statement by saying that in the thermodynamic limit
545: $L\rightarrow \infty$ at $W=W_c$ the number of resonances with width larger
546: than the mean level spacing goes to a constant.
547:
548: \begin{figure}
549: \begin{center}
550: \epsfxsize=5.4cm
551: \leavevmode
552: \epsffile{Fig7.eps}
553: \caption{${\cal P}_{\rm int}^0(W,L)$ as a function of $W$ for different
554: system sizes $L$ provides a means to determine the critical point $W_c$ of the MIT
555: (vertical line at $W=16.5$).}
556: \label{fig:fig3a}
557: \end{center}
558: \end{figure}
559:
560: In Fig. \ref{fig:fig3a}, we show the evolution of ${\cal P}_{\rm int}^{0}(W)$ for
561: different values of $L$ using the box distribution. From this analysis the
562: critical disorder strength $W=W_c=16.5\pm 0.5$ was determined in agreement with other
563: calculations\cite{RMS01}. A further verification of the scaling hypothesis
564: (\ref{scale}) is shown in Fig.~\ref{fig:fig3b} where the same data are reported as a
565: function of the scaling ratio $L/l_{\infty}$. Note that all points collapse on two
566: separate branches for $W<W_c$ and $W>W_c$.
567:
568: \begin{figure}
569: \begin{center}
570: \epsfxsize=5.4cm
571: \leavevmode
572: \epsffile{Fig8.eps}
573: \caption{The one-parameter scaling of ${\cal P}_{\rm int}^0(W,L)$
574: [Eq.~(\ref{scale})] is confirmed for various system sizes $L$ and
575: disorder strengths $W$ using the box distribution.}
576: \label{fig:fig3b}
577: \end{center}
578: \end{figure}
579:
580: %=======================================================================================
581: \section{Conclusions}
582: \label{sec:conclusions}
583:
584: We have studied the properties of the resonance width distribution in
585: various regimes for the 3D Anderson model.
586:
587: In the metallic regime we obtained the forms of ${\cal P}(\Gamma)$ for small and large
588: $\Gamma$ and show that they are determined by the underlying diffusive dynamics and by
589: the existence of prelocalized states. For the localized regime we also explored the
590: limits of small and large $\Gamma$. In the first limit we found that ${\cal P}(\Gamma)$
591: shows a log-normal behavior while in the latter the distribution is power-law like.
592: At the MIT we show that ${\cal P}({\widetilde \Gamma})$, with
593: ${\widetilde \Gamma}=\Gamma/\Delta$, has a {\it universal} form, i.e.,
594: independent of the microscopic details of the random potential and number
595: of attached leads. Specifically, it decays asymptotically with a power
596: which is different from those found in the diffusive and localized regimes.
597: In addition, based on resonance widths, we suggested a new method for determining
598: and analyzing the emergence of the MIT and propose a scaling theory near the critical
599: point.
600:
601: \vspace*{0.5cm}
602: %=======================================================================================
603: \section{Acknowledgments}
604:
605: T. K. acknowledges many useful comments and discussions with Prof. B. Shapiro.
606: J.A.M.-B. and T.K. acknowledge support by a grant from the GIF, the German-Israeli
607: Foundation for Scientific Research and Development. M.W. acknowledges support from
608: the BIOMS initiative in Heidelberg.
609:
610: %-----------------------------------------------------------------------------
611:
612:
613: \begin{thebibliography}{99}
614:
615: \bibitem{MW69} C. Mahaux and H. A Weidenm\"uller,{\it Shell Model Approach in Nuclear
616: Reactions}, (North-Holland, Amsterdam), (1969); I. Rotter, Rep. Prog. Phys. {\bf 54},
617: 635 (1991); J. J. M. Verbaarschot, H. A. Weidenm\"uller, M. R. Zirnbauer, Phys. Rep. {\bf
618: 129}, 367 (1985).
619:
620: \bibitem{atomic} M. H. Nayfeh et al., {\it Atomic Spectra and Collisions in External
621: Fields}, eds. (Plenum, New York), Vol. 2 (1989).
622:
623: \bibitem{molecular} P. Gaspard, in {\it Quantum Chaos, Proceedings of E. Fermi
624: Summer School 1991}, G. Casati et al., eds. (North-Holland) 307.
625:
626: \bibitem{B97} C. Beenakker, Rev. Mod. Phys. {\bf 69}, 731 (1997).
627:
628: \bibitem{S89} U. Smilansky, in {\it Les Houches Summer School on Chaos and Quantum
629: Physics}, M.-J. Giannoni et al., eds. (North-Holland) 371-441 (1989).
630:
631: \bibitem{S99} H.-J. St\"ockmann {\it Quantum Chaos: An Introduction}, Cambridge
632: Univ. Press (1999); E. Kogan, P. A. Mello, H. Liqun, Phys. Rev. E {\bf 61},
633: R17 (2000); C.W.J. Beenakker and P.W. Brouwer, Physica E {\bf 9}, 463 (2001).
634:
635: \bibitem{DS90} E. Doron, U. Smilansky, A. Frenkel, Phys. Rev. Lett. {\bf 65}, 3072 (1990);
636: H. J. St\"ockmann, J. Stein, Phys. Rev. Lett. {\bf 64}, 2215 (1990); J. Stein, H.-J.
637: St\"ockmann, and U. Stoffregen, Phys. Rev. Lett. {\bf 75}, 53 (1995).
638:
639: \bibitem{WAL95} D. S Wiersma, M. P. Van Albada, A. Lagendijk, Nature (London) {\bf 373},
640: 203 (1995); A. Z. Genack et.al. Phys. Rev. Lett. {\bf 82}, 715 (1999).
641:
642: \bibitem{NS97} J. U. Nockel and A. D. Stone, Nature (London) {\bf 385}, 45 (1997);
643: C. Gmachl et.al., Science {\bf 280}, 1556 (1998).
644:
645: \bibitem{Raizen} M. Raizen, C. Salomon, Q. Niu, Physics Today {\bf 50} 30-34 (1997).
646:
647: \bibitem{ABG02} I. L. Aleiner, P. W. Brouwer, L. I. Glazman, Phys. Rep. {\bf 358}, 309 (2002);
648: Y. Alhassid, Rev.Mod.Phys. {\bf 72}, 895 (2000).
649:
650: \bibitem{AKL91} B. L. Altshuler, V. E. Kravtsov, I. V. Lerner, in {\it Mesoscopic
651: Phenomena in Solids}, eds. B. L. Altshuler, P. A. Lee and R. A. Webb (North Holland,
652: Amsterdam), (1991).
653:
654: \bibitem{FS97} Y. V. Fyodorov, H-J Sommers, J. Math. Phys. {\bf 38} 1918 (1997).
655:
656: \bibitem{KS00} T. Kottos and U. Smilansky, Phys. Rev. Lett. {\bf 85}, 968 (2000);
657: Journal of Physics A: Math. and General {\bf 36}, 3501 (2003).
658:
659: \bibitem{BGS91}F. Borgonovi, I. Guarneri, D. Shepelyansky, Phys. Rev. A {\bf 43},
660: 4517 (1991).
661:
662: \bibitem{OKG02} A. Ossipov, Tsampikos Kottos and T. Geisel, Phys. Rev E {\bf 65},
663: 055209(R) (2002).
664:
665: \bibitem{OKG03} A. Ossipov, Tsampikos Kottos and T. Geisel, Europhys. Lett. {\bf 62},
666: 719 (2003).
667:
668: \bibitem{TF00} M. Titov and Y. V. Fyodorov, Phys. Rev. B {\bf 61}, R2444
669: (2000); M. Terraneo, and I. Guarneri, Eur. Phys. J. B {\bf 18}, 303 (2000).
670:
671: \bibitem{SJNS00} Starykh OA, Jacquod PRJ, Narimanov EE, Stone AD, Phys. Rev. E {\bf 62},
672: 2078 (2000).
673:
674: \bibitem{PROT04} F. A. Pinheiro, M. Rusek, A. Orlowski, B. A. van Tiggelen, Phys. Rev. E
675: {\bf 69}, 026605 (2004).
676:
677: \bibitem{SOKG00} F. Steinbach, A. Ossipov, Tsampikos Kottos, and Theo Geisel,
678: Phys. Rev. Lett., {\bf 85} 4426, (2000).
679:
680: \bibitem{KW02} T. Kottos and M. Weiss, Phys. Rev. Lett., {\bf 89} 056401, (2002).
681:
682: \bibitem{A58} P. W. Anderson, Phys. Rev. {\bf 109}, 1492 (1958); A. MacKinnon and B. Kramer,
683: Rep. Prog. Phys. {\bf 56}, 1469 (1993)
684:
685: \bibitem{RMS01} A. Cohen, Y. Roth and B. Shapiro, Phys. Rev. B {\bf 38}, 12125 (1988);
686: C. M. Soukoulis, X. S. Wang, Q. M. Li, M. M. Sigalas, Phys. Rev. Lett. {\bf 82},
687: 668 (1999); K. Slevin and T. Ohtsuki, ibid. {\bf 82}, 382 (1999); K. A. Muttalib and
688: P. W\"olfle, ibid. {\bf 83}, 3013 (1999); P. Markos, ibid. {\bf 83}, 588 (1999).
689:
690: \bibitem{E83} K. B. Efetov, Adv. Phys. {\bf 32}, 53 (1983); K. B. Efetov, {\it Supersymmetry in
691: Disorder and Chaos}, (Cambridge University Press, Cambridge, 1997).
692:
693: \bibitem{M00} A. D. Mirlin, Phys. Rep. {\bf 326}, 259 (2000).
694:
695: \bibitem{FE95} V. I. Falko and K. B. Efetov, Europhys. Lett. {\bf 32}, 627 (1995);
696: Phys. Rev. B {\bf 52}, 17413 (1995); B. A. Muzykantskii and D. E. Khmelnitiskii,
697: Phys. Rev. B {\bf 51}, 5480 (1995); Y. V. Fyodorov and A. Mirlin, Int. J. Mod.
698: Phys. B {\bf 8}, 3795 (1994); I. Smolyarenko and B. L. Altshuler, Phys. Rev. B {\bf 55}, 10451 (1997).
699:
700: \bibitem{KOG03} Tsampikos Kottos, A. Ossipov and T. Geisel, Phys. Rev E {\bf 68},
701: 066215 (2003).
702:
703: \bibitem{UMRS00} V. Uski, B. Mehlig, R. A. R\"omer, and M. Schreiber, Phys. Rev.
704: B {\bf 62}, R7699 (2000); V. Uski, B. Mehlig, and M. Schreiber, ibid. {\bf 63},
705: 241101(R) (2001); B. Nikoli\'c , Phys. Rev. B {\bf 64}, 014203 (2001).
706:
707: \bibitem{SSSLS93} B. I. Shklovskii, B. Shapiro, B. R. Sears, P. Lambrianides, and
708: H. B. Shore, Phys. Rev. B {\bf 47}, 11487 (1993)
709:
710: \bibitem{AS86} B. L. Alt'shuler and B. I. Shklovskii, Zh. Eksp. Teor. Fiz. {\bf 91},
711: 220 (1986) [Sov. Phys. JETP {\bf 64}, 127 (1986)]; A. G. Aronov, A. D. Mirlin, Phys.
712: Rev. B {\bf 51}, 6131 (1995); V. E. Kravtsov and I. V. Lerner, Phys. Rev. Lett. {\bf 74},
713: 2563 (1995); I. Zharekeshev and B. Kramer, Jpn. J. Appl. Phys. {\bf 34}, 4361 (1995).
714:
715: \bibitem{CKL96} J. T. Chalker, V. E. Kravtsov and I. V. Lerner, Pis'ma Zh. Eksp. Teor.
716: Fiz. {\bf 64}, 355 (1996) [JETP Lett. {\bf 64}, 386 (1996)].
717:
718: \bibitem{W80} F. Wegner, Z. Phys. B {\bf 36}, 209 (1980).
719:
720: \bibitem{SG91} M. Schreiber and H. Grussbach, Phys. Rev. Lett. {\bf 67}, 607 (1991); D. A.
721: Parshin and H. R. Schober, ibid. {\bf 83}, 4590 (1999); A. Mildenberger, F. Evers, and
722: A. D. Mirlin, Phys. Rev. B {\bf 66}, 033109 (2002).
723:
724: \bibitem{EM00} F. Evers and A. D. Mirlin, Phys. Rev. Lett. {\bf 84}, 3690 (2000); E. Cuevas,
725: M. Ortuno, V. Gasparian, and A. Perez-Garrido, ibid. {\bf 88}, 016401 (2002);
726: I. Varga, Phys. Rev. B {\bf 66}, 094201 (2002); I. Varga and D. Braun, ibid.
727: {\bf 61}, R11859 (2000).
728:
729: \bibitem{AALR79} E. Abrahams, P. W. Anderson, D. C. Licciardello, and T. V. Ramakrishnan,
730: Phys. Rev. Lett. {\bf 42}, 673 (1979); L. P. Gorkov, A. I. Larkin and D. E. Khmelnitskii,
731: JETP Lett. {\bf 30} 228 (1979).
732:
733: \end{thebibliography}
734:
735: \end{document}
736: