0807.0772/lfp.tex
1: \documentclass[11pt]{article}
2: % Projet d'article avec Tom Bienaime & Alain
3: % Christophe Texier. 
4: %
5: % 7 Avril 2008
6: %
7: % update: 04/07/2008
8: %
9: % 04 Jul 2008: cond-mat arXiv:0807.0772
10: %%%%%%%%%%%%%%%%%%% Soumission à J.Phys.A %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
11: %Article ID:	    A/285657/PAP
12: %Submission: 04 Jul 2008
13: %Editorial receipt: 06 Jul 2008
14: %First Referee Report: 11 August
15: %Revised version sent the 9 Sept.	
16: %Acceptation: 18 Sept.		
17: %
18: % Published : 20/10/2008
19: % J. Phys. A: Math. Theor. 41 (2008) 475001 (9pp)   
20: %             doi: 10.1088/1751-8113/41/47/475001
21: %
22: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
23: %
24: % corrections post-referee: 
25: % update 09/09/08 
26: 
27: \usepackage{amssymb}
28: \usepackage{amsmath,amsthm}  
29: \usepackage{graphicx}
30: %\usepackage{lmodern}
31: \usepackage{microtype}
32: %\usepackage{showlabels,pdfsync}
33: \usepackage[english,francais]{babel}
34: 
35: %% Le package de Tom %%%%
36: %\usepackage{psfrag}
37: %%%%%%%%%%%%%%%%%%%%%%%%%
38: 
39: %%%%%%%%%%%%%%%%%%%%%%%%%% Format de la page %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
40: 
41: \textwidth= 16cm
42: \oddsidemargin= 0.5cm
43: \evensidemargin=-0.5cm
44: \topmargin=-2cm
45: \textheight= 24cm
46: 
47: %\renewcommand{\baselinestretch}{2.} %pour envoyer en double interligne
48: 
49: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Fonts %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
50: 
51: %                       Fonte de l'AMS
52: \newfont{\ensmathquatorze}{msbm10 scaled 1400}
53: \newfont{\ensmathonze}{msbm10 scaled 1100}
54: \newfont{\ensmathdix}{msbm10}
55: \newfont{\ensmathneuf}{msbm10 scaled 833}
56: \newfont{\ensmathhuit}{msbm10 scaled 694}
57: \newfam\ensmathfam                        
58: \textfont\ensmathfam=\ensmathonze        
59: \scriptfont\ensmathfam=\ensmathdix       
60: \scriptscriptfont\ensmathfam=\ensmathhuit
61: \def\ensmf{\fam\ensmathfam\ensmathonze}         % Usage: {\ensmf texte...}
62: 
63: %%%%%%%%%%%%%%%%%%%%%%%%%%% Newcommand, Def %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
64: 
65: \def\typ{\mbox{\tiny typ}}
66: 
67: \renewcommand{\leq}{\leqslant}
68: \renewcommand{\geq}{\geqslant}
69: 
70: \def\eqdef{\stackrel{\mbox{\tiny def}}{=}}     % egal a ... par definition
71: \def\eqloi{\stackrel{\mbox{\tiny (loi)}}{=}}     % egal en loi
72: \def\eqlaw{\stackrel{\mbox{\tiny (law)}}{=}}     % egal en loi
73: \newcommand{\ket}[1]{|\kern.3ex#1\kern.3ex\rangle}
74: \newcommand{\bra}[1]{\langle\kern.3ex #1 \kern.3ex|}
75: \newcommand{\APPROX}[1]{                % approximativement egal a .. si ..
76:    {{\raisebox{-.3cm}{$\textstyle\simeq$}} \atop {\scriptstyle{#1}}}}
77: \newcommand{\APPROXs}[1]{               % similaire a ... si ...
78:    {{\raisebox{-.3cm}{$\textstyle\sim$}} \atop {\scriptstyle{#1}}}}
79: \newcommand{\leadto}[1]{                % approximativement egal a .. si ..
80:    {{\raisebox{-.3cm}{$\textstyle\longrightarrow$}} \atop {\scriptstyle{#1}}}}
81: \newcommand{\mean}[1]{\left\langle #1 \right\rangle} % moyenne statistique
82: \newcommand{\smean}[1]{\langle #1 \rangle} % moyenne statistique
83: \newcommand{\cum}[1]{\langle\langle #1 \rangle\rangle} % cumulant
84: 
85: \newcommand{\EXP}[1]{{\mbox{\large e}}^{#1}}         % exponentielle
86: \newcommand{\argcosh}{\mathop{\mathrm{argcosh}}\nolimits}
87: \newcommand{\argsinh}{\mathop{\mathrm{argsinh}}\nolimits}
88: 
89: \newcommand{\re}{\mathop{\mathrm{Re}}\nolimits}      % partie reelle
90: \newcommand{\im}{\mathop{\mathrm{Im}}\nolimits}      % partie imaginaire
91: \newcommand{\tr}[1]{\mathop{\mathrm{Tr}}\nolimits\left\{ #1 \right\}}  % Trace
92: \newcommand{\cotg}{\mathop{\mathrm{cotg}}\nolimits}  % cotangente
93: \newcommand{\sign}{\mathop{\mathrm{sign}}\nolimits}  % signe
94: \renewcommand{\min}[2]{\mathop{\mathrm{min}}\nolimits\left( #1 , #2\right)}  
95: \renewcommand{\max}[2]{\mathop{\mathrm{max}}\nolimits\left( #1 , #2\right)}  
96: 
97: \newcommand{\proba}{\mathop{\mathrm{Prob}}\nolimits}  
98: \newcommand{\heav}{\mathop{\mathrm{Y}}\nolimits}  
99: 
100: \newcommand\Dslash{D\hspace{-0.25cm}/}               % D slash de Dirac
101: 
102: \def\NN{{\ensmf N}}                 % entiers 
103: \def\ZZ{{\ensmf Z}}                 % entiers relatifs
104: \def\QQ{{\ensmf Q}}                 % rationnels 
105: \def\RR{{\ensmf R}}                 % ensemble des reels
106: \def\CC{{\ensmf C}}                 % ensemble des complexes
107: 
108: \def\I{{\rm i}}                  % le i mathematique 
109: \def\D{{\rm d}}                  % la differenciation
110: \def\Dc{{\rm D}}                 % dérivée covariante
111: 
112: \newcommand{\dz}{\partial_{z}}
113: \newcommand{\dzb}{\partial_{\bar z}}
114: \newcommand{\drond}[2]{\frac{\partial #1}{\partial #2}} % derivee partielle
115: 
116: \newcommand\dphi{{\cal D}\phi{\cal D}\bar\phi\:} % mesure pour l'integrale de 
117:                                                  % chemin
118: \newcommand\ide{1}%\hspace{-0.1cm}{\rm I}} % matrice identite.
119: 
120: %%%% Pour ajuster la hauteur des diagrammes:
121: %%%% =======================================
122: \newcommand{\diagram}[3]{\raisebox{#3}{\includegraphics[scale=#2]{#1}}}
123: 
124: %%%% COmmande \ddot antidiagonale (trouvee sur un forum %%%%
125: \newcommand\antiddots{\mathinner{\mkern2mu\raise1pt\hbox{.}\mkern2mu
126: \newline \raise4pt\hbox{.}\mkern2mu\raise7pt\hbox{.}\mkern1mu}}
127: 
128: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
129: %%%%%%%%%%%%%%%%% Christian %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
130: \newcommand{\diff}{\mathrm{d}}
131: \newcommand{\red}[1]{\textcolor{red}{#1}}
132: 
133: %\renewcommand{\showlabelsetlabel}[1]
134: %  {
135: %    {\color{red} 
136: %      \showlabelfont{\sf{#1}}
137: %    }
138: %  }
139:   
140: %\let\MarginPar\marginpar
141: %\renewcommand\marginpar[1]{\MarginPar{\sf \textcolor{red}{#1}}}
142: 
143: 
144: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
145: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
146: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
147: 
148: \begin{document}
149: 
150: \selectlanguage{english}
151: 
152: \title{Localization for one-dimensional \\ random  
153:   potentials with large local fluctuations}
154: 
155: \author{ Tom Bienaim\'e and Christophe Texier }
156: 
157: 
158: %\date{\today}
159: %\date{July 4, 2008}
160: \date{September 9, 2008}
161: 
162: \maketitle
163: 
164: \hspace{1cm}
165: \begin{minipage}[t]{13cm}
166: {\small
167: 
168: Laboratoire de Physique Th\'eorique et Mod\`eles Statistiques,
169: UMR 8626 du CNRS, 
170: 
171: Universit\'e Paris-Sud, B\^at. 100, F-91405 Orsay Cedex, France.
172: 
173: 
174: }
175: \end{minipage}
176: 
177: 
178: \begin{abstract}
179:   We study the localization of wave functions for one-dimensional
180:   Schr\"odinger Hamiltonians with random potentials $V(x)$ with short
181:   range correlations and large local fluctuations such that
182:   $\int\D{x}\,\smean{V(x)V(0)}=\infty$. A random supersymmetric
183:   Hamiltonian is also considered.  Depending on how large the
184:   fluctuations of $V(x)$ are, we find either new energy dependences of
185:   the localization length, $\ell_\mathrm{loc}\propto{}E/\ln{E}$,
186:   $\ell_\mathrm{loc}\propto{}E^{\mu/2}$ with $0<\mu<2$ or
187:   $\ell_\mathrm{loc}\propto\ln^{\mu-1}E$ for $\mu>1$, or
188:   superlocalization (decay of the wave functions faster than a simple
189:   exponential).
190: \end{abstract}
191: 
192: \noindent
193: PACS numbers~: 72.15.Rn~; 73.20.Fz~; 02.50.-r.
194: 
195: %72.   Electronic transport in condensed matter
196: %72.10.-d   Theory of electronic transport; scattering mechanisms
197: %72.10.Bg   General formulation of transport theory 
198: %72.15.Rn Localization effects (Anderson or weak localization)
199: 
200: %73.   Electronic structure and electrical properties of surfaces, interfaces,
201: %      thin films, and low-dimensional structures 
202: %73.23.-b     Electronic transport in mesoscopic systems 
203: %73.20.Fz Weak or Anderson localization
204: 
205: %02.50.-r Probability theory, stochastic processes, and statistics
206: %05.40.Jc Brownian motion
207: 
208: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
209: 
210: \vspace{0.25cm}
211: 
212: %\section{Introduction}
213: \noindent{\bf Introduction.--}
214: The phenomenon of Anderson localization \cite{And58} in one dimension
215: has been widely studied since the pioneering work of Mott \& Twose
216: arguing that all states are localized in one-dimension
217: (1d)~\cite{MotTwo61}. This statement was rigorously proven in
218: Refs.~\cite{GolMolPas77,PasFig78}. A general method to study the
219: spectral and localization properties of 1d random Hamiltonian was
220: proposed in Refs.~\cite{AntPasSly81,LifGrePas88}~: let us consider the
221: one-dimensional Schr\"odinger Hamiltonian
222: $H=-\frac{\D^2}{\D{}x^2}+V(x)$ where $V(x)$ is a random potential with
223: short range correlations. We first study the solution of the
224: stationary Schr\"odinger equation $H\psi(x;E)=E\psi(x;E)$ satisfying
225: $\psi(0;E)=0$ and $\psi'(0;E)=1$ (differenciation with respect to $x$
226: is denoted by $'$).  We define the Lyapunov exponent (inverse
227: localization length $\ell_\mathrm{loc}\equiv1/\gamma$) as the increase
228: rate~\cite{AntPasSly81,Luc92}
229: \begin{equation}
230: %  \frac1{\ell_\mathrm{loc}(E)} = 
231:   \gamma(E) \eqdef \lim_{x\to\infty}\frac{\D}{\D{x}}
232:   \bigg\langle \ln\sqrt{\psi(x;E)^2+\frac1{E}\,\psi'(x;E)^2} \bigg\rangle
233:   \:.
234: \end{equation}
235: Averaging $\mean{\cdots}$ is taken over realizations of
236: the random potential. This definition becomes more clear if the wave
237: function is parametrized in terms of an envelope and an oscillating
238: part. We substitute to the couple of functions $(\psi,\psi')$ the
239: variables $(\theta,\xi)$ according to~:
240: \begin{align}
241:   \label{pf1}
242:   \psi(x;E)  &= \phantom{k\,}\EXP{\xi(x)}\, \sin\theta(x) \\
243:   \label{pf2}
244:   \psi'(x;E) &= k\,          \EXP{\xi(x)}\, \cos\theta(x) 
245:   \:,
246: \end{align}
247: where $E=k^2$. We can rewrite the definition of the Lyapunov exponent as
248: $\gamma(E)=\frac{\D}{\D{x}}\smean{\xi(x)}$.  Therefore the Lyapunov
249: exponent gives the rate of the exponential increase of the envelope of
250: the wave function.
251: 
252: At high energy (compared to disorder), oscillations of the wave
253: function occur on the typical scale $k^{-1}$ and the Lyapunov exponent
254: is given by~\cite{AntPasSly81,LifGrePas88}\footnote{
255: %
256:   The solution $\psi(x;E)$ of the Cauchy problem exists for any value
257:   of the energy~; it is used to construct the normalized wave
258:   functions $\varphi(x)$ of the stationary Schr\"odinger equation on a
259:   finite interval $[0,L]$ satisfying boundary conditions
260:   $\varphi(0)=\varphi(L)=0$, what can only be achieved for a discrete
261:   set of energies (Sturm-Liouville problem). From this scheme we
262:   expect that the normalized wave functions present the structure
263:   $\varphi(x)\sim\sin(kx+\theta_0)\,\EXP{-|x-x_0|/\ell_\mathrm{loc}}$.
264:   Note however that this simple picture neglects the important fact
265:   that, in the exponential, $\xi(x)$ has large absolute fluctuations
266:   despite it presents negligible fluctuations relatively to its
267:   average when $\int\D{x}\,\smean{V(x)V(0)}<\infty$. These
268:   fluctuations play a very important role since they induce large
269:   fluctuations of the normalization of the wave function (see \S13.3
270:   of Ref.~\cite{LifGrePas88})~\cite{TexCom99}.  
271:   %When $V(x)$ is a white
272:   %noise it can be proven~\cite{LifGrePas88} that
273:   %$\smean{\xi(x)^2}-\smean{\xi(x)}^2\simeq\smean{\xi(x)}$.
274:   %This relation, only valid at high energy compared to disorder, is
275:   %sometimes refered to as ``{\it single parameter scaling}''.
276: %
277: }
278: \begin{equation}
279:   \label{Pastur}
280:   \gamma(E\to\infty) \simeq \frac{1}{8k^2}\int\D{x}\,\smean{V(x)V(0)}\,\cos2kx
281:   \:.
282: \end{equation} 
283: 
284: The question of the present article is to discuss the situation where
285: the potential presents large fluctuations such that
286: $\int\D{x}\,\smean{V(x)V(0)}=\infty$, which makes (\ref{Pastur}) 
287: inapplicable.
288: %
289: The integral of the correlation function may diverge for different
290: reasons.  First, it may diverge due to long range correlations. This
291: corresponds to nonstationary potentials. The localization for self
292: affine potentials such that $\smean{[V(x)-V(0)]^2}\propto|x|^{2h}$,
293: with the Hurst exponent $h>0$, was studied in
294: Refs.~\cite{DeMLyr98,RusKanBunHav01,Luc05} (the case $h=1/2$
295: corresponds to the Brownian case, partly studied in
296: Ref.~\cite{JosJay98}). This will not be the question of interest in
297: the present article.  Another reason for the divergence
298: $\int\D{x}\,\smean{V(x)V(0)}=\infty$ is for a potential with {\it
299:   short range} correlations and {\it large local} fluctuations.  This
300: is the case on which we will focus here.
301: 
302: For simplicity we consider a random potential $V(x)$ uncorrelated at
303: different positions (vanishing correlation length).  A model that
304: realizes these conditions is the following random potential
305: \begin{equation}
306:   \label{potential}
307:   V(x) = \sum_n v_n  \, \delta(x-x_n)
308:   \:,
309: \end{equation}
310: where the weights $v_n$ are chosen to be independent and identical random
311: variables distributed according to a distribution with power law tail
312: \begin{equation}
313:   \label{powerlaw}
314:   p_1(v) \propto \frac1w\left|\frac{w}{v}\right|^{1+\mu}
315:   \hspace{0.5cm}\mbox{for } v \to \pm\infty
316: \end{equation}
317: with $\mu>0$. Here $w$ is a scale for the weights to make the
318: argument of the tail dimensionless.  The positions
319: of impurities $x_n$ are also chosen to be independent random variables
320: uniformly distributed with a finite density $\rho$.
321: % with $\mean{v_n}=0$ to ensure $\mean{V(x)}=0$. 
322: When $\mu\leq2$ the second moment diverges $\smean{v_n^2}=\infty$, as
323: well as the correlation function of the potential (\ref{potential}) since
324: $\smean{V(x)V(x')}=\rho\,\smean{v_n^2}\,\delta(x-x')$.
325: 
326: Some exact results have been obtained for a tight binding Hamiltonian
327: with random on-site potential (Anderson model) distributed according
328: to a Cauchy law \cite{Llo69,Tho72,Ish73} (\S10.3 of
329: Ref.~\cite{LifGrePas88} or Ref.~\cite{Luc92}).  
330: Note also that
331: fluctuations of the envelope of the wave function and conductance
332: statistics for this discrete model were studied for power law disorder
333: in the recent works \cite{DeyLisAlt01,TitSch03}.
334: %
335: The Anderson model (AM) can be mapped \cite{Luc92} onto the problem we are
336: interested in here for {\it fixed} impurity positions and with
337: $\mu=1$. Potentials for fixed and random impurity positions share some
338: features, however, when impurities of random weights form a lattice,
339: $x_n=n/\rho$, the trace of the lattice remains for arbitrary large
340: energies (band edges remain at $k_n=n\pi\rho$). This makes the
341: definition of a high energy regime less convenient.
342: 
343: \vspace{0.25cm}
344: 
345: %\section{Ricatti variable}
346: \noindent{\bf Ricatti variable.--}
347: We follow the ideas introduced in Ref.~\cite{FriLlo60} in order to
348: study the spectrum, and apply them to the localization problem.  Let
349: us introduce the Ricatti variable
350: $z(x)\eqdef\frac{\psi'(x)}{\psi(x)}$.  From the Schr\"odinger equation
351: we see that it obeys a Langevin equation $z'=-E-z^2+V(x)$ for initial
352: condition $z(0)=\infty$. The distribution $T(z;x)$ of the Ricatti
353: variable obeys the integro-differential equation
354: \begin{equation}
355:   \frac{\partial}{\partial x}T(z;x)
356:   = \frac{\partial}{\partial z}\left[(E+z^2)T(z;x)\right]
357:   + \rho\int\D v\,p(v)\,\left[T(z-v;x)-T(z;x)\right]
358:   \:.
359: \end{equation}
360: The first term in the right hand side is the drift term coming from
361: the force $-(E+z^2)$ and the second a jump term originating from the
362: random potential (\ref{potential}). For $x$ sufficiently large, the
363: distribution reaches a limiting distribution $T(z)$ for a steady
364: current~\cite{LifGrePas88,Luc92}. Current of the Ricatti variable
365: through $\RR$ gives the number of zeros of the wave function per unit
366: length. This is also the integrated density of states (IDoS) per unit
367: length $N(E)$, therefore
368: \begin{equation}
369:   \label{eqInteg}
370:   N(E) = (E+z^2)T(z)  - \rho \int\D v\,p(v)
371:   \int_{z-v}^z\D z'\,T(z')
372:   \:.
373: \end{equation}
374: Imposing normalization of the solution of this integral equation gives
375: the IDoS. Knowing the limiting distribution $T(z)$, the Lyapunov
376: exponent can be obtained from~\cite{LifGrePas88} $\gamma=\smean{z}$.
377: Since $T(z\to\pm\infty)\simeq{}N(E)/z^2$, in order to deal with well
378: defined integral it is understood that calculation of the Lyapunov
379: exponent involves the antisymmetric part of the
380: distribution~:~$\gamma=\int\D{z}\,z\,\frac12[T(z)-T(-z)]$).
381: 
382: Let us study the high energy Lyapunov exponent. For that purpose we
383: solve the integral equation (\ref{eqInteg}) by perturbation starting
384: from the solution for $V(x)=0$. In the absence of disorder
385: ($p(v)=\delta(v)$) we have $T_0(z)=\frac1\pi\frac{k}{z^2+k^2}$. We
386: expand the distribution $T(z)=T_0(z)+T_1(z)+\cdots$ in powers of the
387: density $\rho$, as well as the IDoS.  Then eq.~(\ref{eqInteg}) is
388: solved recursively order by order.  We easily obtain $T_1(z)$ from
389: which we deduce
390: \begin{equation}
391:   \gamma(E=k^2\to\infty) \simeq \frac\rho\pi \int\D z\, \frac{z}{z^2+k^2}
392:   \int\D v\,p(v)\,
393:   \left[ \arctan\frac{z}{k} - \arctan\frac{z-v}{k} \right] 
394:   \:.
395: \end{equation}
396: This gives the general formula
397: \begin{equation}
398:   \label{RES1}
399:   \gamma(k^2) \simeq \frac\rho2
400:   \mean{\ln\left[1+\left(\frac{v}{2k}\right)^2\right]}_v
401:   \:,
402: \end{equation}
403: where the averaging is now taken over the $\delta$-peak weights~$v_n$.  
404: This is the first term of a ``concentration expansion'' that can be
405: systematically developed~\cite{LifGrePas88} (eq.~(\ref{RES1}) was
406: derived in the \S10.4 of this latter reference for non random weights
407: $v_n$. Additional averaging in eq.~(\ref{RES1}) follows from the property
408: of additivity of the variable $\xi(x)$).
409: 
410: 
411: \vspace{0.25cm}
412: 
413: %\section{New energy dependences}
414: \noindent{\bf New energy dependences.--}
415: We first consider the high energy Lyapunov exponent, $\sqrt{E}=k\gg\rho,\,w$,
416: when the weights are distributed according to (\ref{powerlaw}).  We
417: write $p_1(v)=\frac1wf(v/w)$ where $f(y)$ is a dimensionless symmetric
418: function such that $f(y\to\pm\infty)\simeq{}C\,|y|^{-1-\mu}$.  We
419: divide 
420: the integral
421: $\gamma\simeq\rho\int_0^\infty\D{y}\,f(y)\,\ln[1+(\frac{w}{2k}y)^2]$
422: into three parts~: $\gamma\sim\rho\big[
423: (\frac{w}{2k})^2\int_0^1\D{y}\,f(y)\,y^2
424: +(\frac{w}{2k})^2C\int_1^{2k/w}\D{y}\,y^{1-\mu}
425: +2C\int_{2k/w}^\infty\D{y}\,y^{-1-\mu}\ln(\frac{w}{2k}y)\big]$.
426: 
427: \noindent$\bullet$ For $\mu>2$ the Lyapunov exponent is dominated by
428: smallest $y$ ($\lesssim1$). We obtain $\gamma\propto\rho(\frac{w}{k})^2$ that
429: corresponds to expand the logarithm of
430: eq.~(\ref{RES1}) for small $v$. This is the result of
431: eq.~(\ref{Pastur})~:
432: $\gamma(k^2)\simeq\frac1{8k^2}\rho\,\smean{v^2}$.
433: 
434: \noindent$\bullet$ For $\mu=2$, eq.~(\ref{Pastur}) cannot be used
435: since $\smean{v^2}=\infty$. The integral giving the Lyapunov exponent is
436: dominated by the intermediate scale $1\lesssim{}y\lesssim{}k/w$~:
437: \begin{eqnarray}
438:   \label{lyap2}
439:     \gamma(k^2) \propto  \rho\left(\frac{w}{k}\right)^2\ln\left(\frac{2k}{w}\right)
440: %    \hspace{1cm}\mbox{for } \mu =2 
441:   \:.
442: \end{eqnarray}
443: 
444: \noindent$\bullet$ For $0<\mu<2$, the Lyapunov exponent is
445: dominated by largest $y$ ($\gtrsim{}k/w$). We obtain
446: \begin{eqnarray}
447:   \label{lyap1}
448:   \gamma(k^2) 
449:     \propto  \rho\left(\frac{w}{k}\right)^\mu  
450: %    \hspace{1cm}\mbox{for }0<\mu<2
451:   \:.
452: \end{eqnarray}
453: %
454: %For $\mu>2$,
455: %the limit $k\gg{}w$ selects the small-$v$ part of the logarithm in
456: %eq.~(\ref{RES1})~: we recover the result
457: %$\gamma(k^2)\simeq\frac1{8k^2}\rho\,\smean{v^2}\propto\rho\,(\frac{w}{k})^2$
458: %corresponding to eq.~(\ref{Pastur}).
459: %For $0<\mu<2$, $\smean{v^2}=\infty$ and the computation of (\ref{RES1})
460: %is dominated by the large-$v$ part of the logarithm, consequently we find~:
461: %\begin{eqnarray}
462: %  \label{lyap1}
463: %  \gamma(k^2) 
464: %    \propto  \rho\left(\frac{w}{k}\right)^\mu  
465: %    \hspace{1cm}\mbox{for }0<\mu<2
466: %\end{eqnarray}
467: %The marginal $\mu=2$ case gives~: 
468: %\begin{eqnarray}
469: %  \label{lyap2}
470: %    \gamma(k^2) \propto  \rho\left(\frac{w}{k}\right)^2\ln\left(\frac{2k}{w}\right)
471: %    \hspace{1cm}\mbox{for } \mu =2 
472: %\end{eqnarray}
473: The numerical dimensionless prefactors depend on the precise form of the
474: distribution and not only on its tail. 
475: 
476: 
477: The fluctuations of the random weights can be further increased by considering
478: a distribution with tail~:
479: \begin{equation}
480:   \label{loglaw}
481:   p_2(v) \propto \frac1{|v|\ln^{1+\mu}\left|\frac{v}{w}\right|}
482:   \hspace{0.5cm}\mbox{for }  v \to \pm\infty
483: \end{equation}
484: for $\mu>0$.  When $\mu>1$ we find that the Lyapunov exponent decays logarithmically
485: with energy
486: \begin{equation}
487:   \gamma(k^2) \sim \frac{\rho}{\ln^{\mu-1}(\frac{k}{w})}
488:   \:.
489: \end{equation}
490: The case $0<\mu\leq1$ is discussed in the next section.
491: 
492: 
493: \vspace{0.25cm}
494: 
495: %\section{Superlocalization}
496: \noindent{\bf Superlocalization.--}
497: On the other hand, for distribution (\ref{loglaw}) with $\mu\leq1$,
498: not only the second moment diverges $\smean{v^2}=\infty$, but the
499: expression (\ref{RES1}) shows that the Lyapunov exponent diverges as
500: well~: $\gamma=\infty$. This indicates that the logarithm of the
501: envelope of the wave function, $\xi(x)$, presents different scaling
502: properties with $x$. In order to analyze this, we remark that the
503: variable $\xi(x)$ is constant between two impurities and makes a jump
504: $\Delta\xi_n\eqdef\xi(x_n^+)-\xi(x_n^-)\sim\ln|v_n|$ across the
505: impurity $n$ (see below, the section on numerics). Therefore $\xi(x)$
506: behaves as the sum of $N\sim\rho{}x$ independent variables, each
507: distributed according to a power law distribution
508: $p(\Delta\xi)\propto\Delta\xi^{-1-\mu}$.  Using well known results
509: (recalled in appendix A) we obtain
510: \begin{eqnarray}
511:   \label{unconv1}
512:   \xi(x) &\sim& (\rho x)^{1/\mu}
513:     \hspace{1.5cm}\mbox{for }0<\mu<1 \\
514:   \label{unconv2}
515:   &\sim& (\rho x)\ln(\rho x)
516:     \hspace{1cm}\mbox{for }\mu=1
517:   \:.
518: \end{eqnarray}
519: The envelope of the wave function presents a decay faster than a
520: simple exponential.  This phenomenon is called {\it superlocalization}
521: and has been recently studied for a discrete model in
522: Ref.~\cite{BooLuc07}\footnote{ Note also that such superlocalization
523:   $\xi(x)\sim{x}^{1+h/2}$ occurs for self affine random potentials
524:   characterized by long-range correlations
525:   $\smean{[V(x)-V(0)]^2}\propto|x|^{2h}$ with $h>0$ \cite{Luc05}.  }.
526: Characterization of the localization properties cannot be limited to
527: the typical behaviours (\ref{unconv1},\ref{unconv2}) since the
528: variable $\xi(x)$ presents large fluctuations. Its distribution is
529: characterized by a power law tail
530: \begin{equation}
531:   \label{PdeXi}
532:   \mathcal{P}(\xi;x)\propto1/\xi^{1+\mu}
533: \end{equation}
534: with the same exponent as the one involved in the distribution of the weights.
535: 
536: 
537: \vspace{0.15cm}
538: 
539: %\subsection*{Conductance}
540: \noindent{\it Conductance.--}
541: We give another interpretation of the previous result in terms of the
542: conductance of a finite disordered interval of length $L$. The
543: dimensionless conductance is equal to the transmission probability
544: (Landauer formula), and presents the same exponential decay as the
545: square of the wave function modulus.  Therefore we can write
546: $g\sim\EXP{-2\xi(L)}$ (a more precise definition of the reflection
547: coefficient within the phase formalism can be found in
548: Ref.~\cite{AntPasSly81}).
549: 
550: Let us first recall some well known results valid for a potential with
551: finite local fluctuations. At high energy, when (\ref{Pastur}) holds,
552: $\xi(x)$ behaves like a Brownian motion with drift\footnote{The fact
553:   that drift and the Gaussian fluctuations involve the same parameter
554:   is refered to as ``{\it single parameter scaling}''
555:   \cite{AbrAndLicRam79} (see also \cite{DeyLisAlt01})~; it holds only
556:   at high energy since it relies on the decoupling between a fast
557:   variable (the phase $\theta(x)$ introduced above) and the slow
558:   variable $\xi(x)$.}  \cite{AntPasSly81}~:
559: $\xi(x)\eqlaw\gamma\,x+\sqrt{\gamma}\,W(x)$, where $W(x)$ is a Wiener
560: process\footnote{a normalized free Brownian motion such that
561:   $\mean{W(x)}=0$ and $\mean{W(x)W(x')}=\min{x}{x'}$.}.  It follows
562: that the distribution of the logarithm of the conductance is Gaussian
563: $\Pi(\ln{g})\simeq\frac1{\sqrt{8\pi\gamma\,L}}\exp-\frac1{8\gamma\,L}(\ln{g}+2\gamma{}L)^2$
564: \cite{Pen94}. The typical value of the conductance is
565: $g_\mathrm{typ}\sim\EXP{-6\gamma{}L}$ (while
566: $(\ln{}g)_\mathrm{typ}\sim-2\gamma{}L$) however fluctuations of the
567: logarithm are associated to a much larger scale
568: $g_\mathrm{fluct}\sim\EXP{-2\sqrt{\gamma{}L}}$.
569: 
570: Distribution of the conductance in the Anderson model with power law
571: disorder has been studied in Ref.~\cite{TitSch03} where some power law
572: distribution of the conductance was obtained for~$g\to0$.
573: 
574: %The Anderson model with on-site energies distributed according to a
575: %power law was studied in Refs.~\cite{DeyLisAlt01,TitSch03}~: the first
576: %of these references showed that for Cauchy distribution and in the
577: %weak disorder limit, the ratio of the two cumulants presents an
578: %additional factor two compared to the conventional case (with our
579: %notation $\smean{\xi(x)^2}-\smean{\xi(x)}^2\simeq2\smean{\xi(x)}$).
580: %Some non trivial distribution was obtained in this case in Ref.~\cite{TitSch03}.
581: 
582: In the superlocalization regime, the behaviour (\ref{unconv1}) is
583: associated with a decay of the conductance $g\sim\EXP{-L^{1/\mu}}$.
584: The distribution (\ref{PdeXi}) can be related to the conductance
585: distribution~:
586: \begin{equation}
587:   \Pi( \ln g) \underset{g\to0}{\sim} \frac1{|\ln g|^{\mu+1}}
588: \end{equation}
589: for $0<\mu\leq1$.
590: 
591: \vspace{0.25cm}
592: 
593: %\section{Localization for supersymmetric Hamiltonian}
594: \noindent{\bf Localization for supersymmetric Hamiltonian.--}
595: We consider another class of random Hamiltonians with the so-called
596: supersymmetric structure
597: \begin{equation}
598:   \label{Hsusy}
599:   H = -\frac{\D^2}{\D x^2} + \phi(x)^2 + \phi'(x)
600:   \hspace{0.5cm}\mbox{with }   \phi(x) = \sum_n \eta_n  \, \delta(x-x_n)
601:   \:,
602: \end{equation}
603: where $\eta_n$ are dimensionless uncorrelated weights, each
604: distributed according to a distribution $p(\eta)$.  This Hamiltonian
605: is interesting since it presents rather different spectral and
606: localization properties (in particular it leads to a delocalization
607: transition as $E\to0$). It is related to several other problems as
608: well. For example it is the square of a Dirac Hamiltonian with a
609: random mass $\phi(x)$, introduced in various contexts of condensed
610: matter physics~; the problem can also be related to
611: classical diffusion in a random force field (see
612: Refs.~\cite{LifGrePas88,BouComGeoLeD90,ComTex98} for a review).  
613: %$H$ possesses a positive spectrum.  
614: We can follow the same strategy~: the
615: Ricatti variable $z=\frac{\psi'}{\psi}-\phi$ obeys the Langevin type
616: equation $z'=-E-z^2-2z\phi(x)$ with multiplicative noise. Limiting
617: distribution of the Ricatti variable for a steady current $-N(E)$ obeys the
618: integral equation
619: \begin{equation}
620:   \label{eqIntegSusy}
621:   N(E) = (E+z^2)T(z)  + \rho \int\D\eta\,p(\eta)
622:   \int_{z}^{z\mathrm{e}^{2\eta}}\D z'\,T(z')
623:   \:.
624: \end{equation}
625: The Lyapunov exponent is now given by $\gamma=\mean{z}+\mean{\phi}$.
626: Following the same perturbative approach as before, we obtain
627: \begin{equation}
628:    \label{RESsusy}
629:   \gamma(E\to\infty) \simeq \rho
630:   \mean{\ln\cosh\eta}_\eta
631:   \:.
632: \end{equation}
633: For the supersymmetric Hamiltonian, when the average exists, the
634: Lyapunov exponent reaches a finite value at high energy (in contrast
635: with the decrease of the Lyapunov exponent for the Schr\"odinger
636: Hamiltonian)~: for $|\eta_n|\ll1$ it takes the form
637: $\gamma(E\to\infty)\simeq\frac12\int\D{x}\,\smean{\phi(x)\phi(0)}$.
638: 
639: Let us consider a power law distribution of weights
640: $p(\eta)\propto1/|\eta|^{1+\mu}$. We see from (\ref{RESsusy}) that we
641: have $\gamma=\infty$ in this case for $\mu\leq1$. The reason is
642: similar to the one discussed in the previous paragraph for the
643: Schr\"odinger equation with weights distributed according to
644: (\ref{loglaw}). Here the variable $\xi(x)$ jumps by
645: $\Delta\xi_n\sim|\eta_n|$ across the impurity (see below). Therefore,
646: for the supersymmetric case, a power law distribution of the weights
647: leads to the superlocalization,
648: eqs.~(\ref{unconv1},\ref{unconv2},\ref{PdeXi}).
649: 
650: \vspace{0.25cm}
651: 
652: %\section{Numerical calculations}
653: \noindent{\bf Numerical calculations.--} 
654: We can easily study the evolution of the phase and envelope variables
655: (\ref{pf1},\ref{pf2}) numerically.  We denote by
656: $\theta_n^\pm\eqdef\theta(x_n^\pm)$ and $\xi_n^\pm\eqdef\xi(x_n^\pm)$
657: the value of the phase and the envelope just before and right after
658: the $n$-th $\delta$-peak.  Between two impurities we have
659: $\theta_{n+1}^--\theta_n^+=k\ell_n$ and $\xi_{n+1}^--\xi_n^+=0$. The
660: length $\ell_n=x_{n+1}-x_n$ denotes the distance between consecutive
661: impurities. It is distributed according to a Poisson
662: law~$p(\ell)=\rho\,\EXP{-\rho\ell}$.  The evolution of the random
663: variables across an impurity depends on the form of the random
664: potential. We introduce the
665: notation~$\Delta\xi_n=\xi_n^+-\xi_n^-=\xi_{n+1}^--\xi_n^-$.
666: 
667: \noindent$\bullet$
668: For the {\it Schr\"odinger Hamiltonian} with potential (\ref{potential})
669: phase evolution is given by 
670: $\cotg\theta_n^+-\cotg\theta_n^-=\frac{v_n}{k}$ and evolution of
671: the envelope by
672: $
673: \Delta\xi_n=\ln\frac{\sin\theta_n^-}{\sin\theta_n^+}
674: =\frac12\ln[1+\frac{v_n}{k}\sin2\theta_n^-+\frac{v_n^2}{k^2}\sin^2\theta_n^-]
675: $.
676: 
677: \noindent$\bullet$
678: For the {\it supersymmetric Hamiltonian} (\ref{Hsusy}) we have 
679: $\tan\theta_n^+=\EXP{2\eta_n}\tan\theta_n^-$ and 
680: $
681: \Delta\xi_n=\frac12\ln\frac{\sin2\theta_n^-}{\sin2\theta_n^+}
682: =\frac12\ln[\EXP{2\eta_n}\sin^2\theta_n^-+\EXP{-2\eta_n}\cos^2\theta_n^-] 
683: $.
684: 
685: \noindent
686: IDoS is given by $N(E)=\lim_{L\to\infty}\frac{\theta(L)}{L\pi}$ and the
687: Lyapunov exponent by $\gamma(E)=\lim_{L\to\infty}\frac{\xi(L)}{L}$.
688: 
689: To be precise we consider a specific distribution with power law tail~:
690: \begin{equation}
691:   \label{ppl}
692:   p_1(v) =
693:   \frac{\mu\,\left|\frac{v}{w}\right|^{\mu-1}}
694:        {\pi w\,\big(1+\left|\frac{v}{w}\right|^{2\mu}\big)}
695:   \:.
696: \end{equation}
697: This choice has the advantage that it is very easy to simulate since
698: the cumulative distribution is straightforwardly obtained.  Using
699: eq.~(\ref{RES1}), we get the high energy Lyapunov exponent~:
700: $\gamma\simeq\frac1{\sin(\frac{\pi\mu}2)}\rho(\frac{w}{2k})^\mu$ for
701: $0<\mu<2$ and
702: $\gamma\simeq\frac2\pi\rho(\frac{w}{2k})^2\ln(\frac{2k}{w})$ for
703: $\mu=2$ (both expressions are valid for $k\gg\rho,\,w$).  The case
704: $\mu=1$ corresponds to a Cauchy law. The right hand side of
705: eq.~(\ref{RES1}) can be computed easily in this case and we obtain the
706: expression $\gamma\simeq\rho\ln(1+\frac{w}{2k})$ valid in a broader
707: range of energy (for $k\gg\rho$ but $w$ arbitrary)~; at high energy we
708: recover the known energy dependence $\gamma\simeq\rho\frac{w}{2k}$ (it
709: can be recovered from discrete models~\cite{Tho72,Luc92}).  These
710: expressions are compared to the numerical results on
711: figure~\ref{fig:GamE} and work perfectly well.
712: 
713: \begin{figure}[!ht]
714:   \centering
715:   \includegraphics[scale=0.45]{lyapunov.eps}
716:   \caption{\it For the Schr\"odinger Hamiltonian~: Lyapunov exponent
717:     as a function of the energy for weights $v_n$ distributed
718:     according to (\ref{ppl}). Numerical results (continuous lines) are
719:     compared with high energy expressions (dashed lines) derived in
720:     the text (no fit) for $\mu=0.34$, $1$ and~$2$.  Other parameters
721:     are $w=1$, $\rho=0.01$ and number of impurities $N=10^6$.  Inset~:
722:     $E\gamma(E)$ is plotted in semilog scale for $\mu=2$ in order to
723:     check  its logarithmic behaviour (dashed line corresponds to
724:     eq.~(\ref{lyap2})). }
725:   \label{fig:GamE}
726: \end{figure}
727: 
728: Next we analyze the superlocalization regime~: we consider
729: the supersymmetric Hamiltonian for weights $\eta_n$ distributed
730: according to a law similar to (\ref{ppl}) for $0<\mu<1$. The
731: distribution $\mathcal{P}(\xi;x)$ is plotted for different values of
732: $x$ on figure~\ref{fig:PdeXi}. In the inset the axes are rescaled in
733: order to check that, according to (\ref{unconv1}), the distribution
734: has the form
735: \begin{equation}
736:   \mathcal{P}(\xi;x) \simeq \frac1{(\rho{}x)^{1/\mu}}\:
737:   \varpi\left(\frac\xi{(\rho{}x)^{1/\mu}}\right)
738:   \:,
739: \end{equation}
740: %$\mathcal{P}(\xi;x)={(\rho{}x)^{-1/\mu}}\,\varpi({(\rho{}x)^{-1/\mu}}\xi)$,
741: where $\varpi(\zeta)$ is a dimensionless function. After rescaling we
742: see that the four curves corresponding to different values of $x$
743: perfectly collapse onto each other, apart for small deviations
744: corresponding to the smallest values of $\xi$ and $x$.  Finally we
745: check that the tail of the distribution is indeed a power law,
746: eq.~(\ref{PdeXi})~: in the inset of figure~\ref{fig:PdeXi} rescaled
747: distributions are plotted on a log-log scale with
748: $\varpi(\zeta)\propto\zeta^{-1-\mu}$. The agreement seems excellent.
749: %Rescaled curves are plotted in log-log scale in the inset in order to check the
750: %power low tail (16) and scaling (14). The curves perfectly overlap and their
751: %slope coincides with our prediction.
752: 
753: \begin{figure}[!ht]
754:   \centering
755: %  \psfrag{xi}{$\xi$}
756: %  \psfrag{P}{$\mathcal{P}(\xi;x)$}
757: %  \psfrag{xibis}{$\xi/(\rho x)^{1/\mu}$}
758: %  \psfrag{Pbis}{$(\rho x)^{1/\mu}\mathcal{P}(\xi;x)$}
759:   \includegraphics[scale=0.45]{distributionXi.eps}
760:   \caption{\it Superlocalization for the supersymmetric Hamiltonian.
761:     Distribution of the variable $\xi(x)$ for
762:     different values of $x=L/4$, $L/2$, $3L/4$ and $L$.
763:     Parameters are~: $k=10$, $\rho=1$, $\Lambda=0.1$ (typical scale
764:     for weights $\eta_n$) and $\mu=0.5$. 
765:     Number of impurities is $N=10^6$ and $L=1000$. 
766:     In the inset, straight line corresponds to eq.~(\ref{PdeXi}).}
767:   \label{fig:PdeXi}
768: \end{figure}
769: 
770: 
771: 
772: \vspace{0.25cm}
773: 
774: %\section{Conclusion}
775: \noindent{\bf Conclusion.--} 
776: We have analyzed the high energy localization length for random
777: potentials with short range correlations and large local fluctuations
778: such that the well-known result (\ref{Pastur}), leading to
779: $\ell_\mathrm{loc}\propto{}E$, is not valid.  
780: %
781: We have studied localization for potentials made of superposition of
782: $\delta$-peaks. Performing a concentration expansion, we have obtained
783: two general high energy formulae for the Lyapunov exponent~:
784: $\gamma(E)\simeq\rho\smean{\ln[1+\frac{v^2}{4E}]}_v$
785: \cite{LifGrePas88} for the Schr\"odinger Hamiltonian and
786: $\gamma(E)\simeq\rho\smean{\ln\cosh\eta}_\eta$ for the supersymmetric
787: Hamiltonian.  These formulae have been used to analyze the case of
788: potential with large local fluctuations.  
789: 
790: For the Schr\"odinger case, we have shown the relation between the
791: distribution of the weights of the $\delta$-peaks and the energy decay
792: of the Lyapunov exponent (inverse localization length
793: $\ell_\mathrm{loc}$).  Sufficiently large fluctuations of the weights,
794: such that $\smean{v_n^2}=\infty$, lead to a stronger localization
795: effect characterized by an increase of $\ell_\mathrm{loc}$ with energy
796: slower than linear.  These results are summarized in 
797: table~\ref{tab:summary}.
798: %
799: 
800: \begin{table}[!ht]
801:   \centering
802:   \begin{tabular}{lll}
803:      Potential distribution & & Localization length \\
804:      \hline\hline
805:      $\smean{v_n^2}<\infty$ & & $\ell_\mathrm{loc}\propto{}E$
806:      \ \hspace{1cm}\cite{LifGrePas88}\\
807:      \hline
808:      $p(v\to\pm\infty)\propto1/|v|^{\mu+1}$ 
809:          & $\mu=2$  & $\ell_\mathrm{loc}\propto{}E/\ln{}E$   \\
810:          & $0<\mu<2$ & $\ell_\mathrm{loc}\propto{}E^{\mu/2}$ \\
811:      \hline
812:      $p(v\to\pm\infty)\propto\frac1{|v|}\ln^{-1-\mu}\left|\frac{v}{w}\right|$
813:         & $\mu>1$ & $\ell_\mathrm{loc}\propto\ln^{\mu-1}E$ \\
814:      \ \hspace{2cm}superlocalization 
815:         &
816:           $\left\{\begin{array}{l} \mu=1 \\ 0<\mu<1 \end{array}\right.$
817:         & $\begin{array}{l} \xi(x)\sim x\ln x \\
818:           \xi(x)\sim{}x^{1/\mu} \end{array}$
819:         \\
820:     \hline\hline
821:   \end{tabular}
822:   \caption{\it Energy dependence of the localization length for the
823:     Schr\"odinger Hamiltonian with random potential
824:     $V(x)=\sum_nv_n\,\delta(x-x_n)$ for different broad distributions of the
825:     weights $v_n$.} 
826:   \label{tab:summary}
827: \end{table}
828: 
829: 
830: 
831: 
832: %In the Schr\"odinger case for weights distributed according to a
833: %distribution with a power law tail
834: %$p_1(v\to\pm\infty)\propto1/|v|^{\mu+1}$, this has led to a stronger
835: %localization effect characterized by a slower increase of
836: %$\ell_\mathrm{loc}$ with energy~:
837: %new energy dependences for the localization length~:
838: %$\ell_\mathrm{loc}\propto{}E/\ln{}E$ for $\mu=2$ and
839: %$\ell_\mathrm{loc}\propto{}E^{\mu/2}$ for $0<\mu<2$.   
840: %
841: %When fluctuations are further increased, by considering a distribution
842: %with tail
843: %$p_2(v\to\pm\infty)\propto\frac1{|v|}\ln^{-1-\mu}\left|\frac{v}{w}\right|$,
844: %we have obtained $\ell_\mathrm{loc}\propto\ln^{\mu-1}E$ for $\mu>1$.
845: %For $0<\mu\leq1$ the Lyapunov exponent diverges, what is the sign of the
846: %superlocalization of wave functions.
847: 
848: The understanding of fluctuations of the variable $\xi(x)$ ({\it i.e.}
849: of the localization length) plays a major role to analyze universal
850: statistical properties of Wigner time delay \cite{TexCom99}.  It would
851: be an interesting issue to study how the statistics of Wigner time
852: delay are affected by the unconventional localization properties
853: analyzed here.
854: 
855: 
856: \vspace{0.25cm}
857: 
858: %\section*{Acknowlegments}
859: \noindent{\bf Acknowlegments.--} 
860: We thank Alain Comtet and Satya Majumdar for interesting remarks and
861: Jean-Marc Luck for bringing to our attention
862: Refs.~\cite{Luc05,BooLuc07}.
863: 
864: 
865: \vspace{0.25cm}
866: 
867: %\section*{Appendix A}
868: \noindent{\bf Appendix A.--}
869: We recall well known results on the distribution of the sum of
870: independent and identically distributed ({\it i.i.d}) random
871: variables. Let us consider $N$ {\it i.i.d.} positive  variables $y_n$ and their
872: sum $Y_N=\sum_{n=1}^Ny_n$. If $\smean{y_n^2}<\infty$ the statistical
873: properties of $Y_N$ are given by central limit theorem for
874: $N\to\infty$~: Gaussian distribution centered on
875: $\smean{Y_N}=N\smean{y}$ of variance
876: $\smean{Y_N^2}_c=\smean{Y_N^2}-\smean{Y_N}^2=N\smean{y^2}_c$. If the
877: distribution of the $y_n$'s presents a power law tail
878: $p(y)\propto1/y^{\mu+1}$ with $0<\mu\leq2$ such that
879: $\smean{y_n^2}=\infty$, the situation is different~:
880: 
881: \noindent$\bullet$ For $0<\mu<1$ all moments of $y_n$ diverge. Let us
882: consider the characteristic function $g(p)=\smean{\EXP{-py}}$. We can
883: write $g(p)=1-\int_0^\infty\D{}y\,(1-\EXP{-py})p(y)$ from which we see
884: that $g(p\to0)\simeq1-C\,p^\mu$ where $C$ is some constant related to
885: the prefactor of the power law tail of $p(y)$. The characteristic
886: function for $Y_N$ reads $G_N(p\to0)\simeq\EXP{-NC\,p^\mu}$. This
887: shows that the related distribution $P_N(Y)$ presents a similar power
888: law tail and involves the typical scale $Y_N\sim{}N^{1/\mu}$.
889: 
890: \noindent$\bullet$ For $\mu=1$. A similar analysis gives
891: $g(p\to0)\simeq1-C\,p\ln1/p$ and therefore $Y_N\sim{}N\ln{N}$.
892: 
893: \noindent$\bullet$ For $1<\mu<2$ the first moment is finite
894: $\smean{Y_N}=N\mean{y}$ however fluctuations are larger than in the normal case
895: $\smean{Y_N^2}_c\sim{}N^{2/\mu}$.
896: 
897: \noindent$\bullet$ For $\mu=2$ fluctuations are 
898: $\smean{Y_N^2}_c\sim{}N\ln{N}$.
899: 
900: \noindent$\bullet$ For $\mu>2$, central limit theorem applies.
901: 
902: 
903: %\bibliographystyle{phreport}
904: %\bibliography{biblio}
905: 
906: 
907: \begin{thebibliography}{10}
908: 
909: \bibitem{And58}
910: P.~W. Anderson,
911:  Absence of diffusion in certain random lattices,
912:  Phys. Rev. {\bf 109}, 1492 (1958).
913: 
914: \bibitem{MotTwo61}
915: N.~F. Mott and W.~O. Twose,
916:  The theory of impurity conduction,
917:  Adv. Phys. {\bf 10}, 107 (1961).
918: 
919: \bibitem{GolMolPas77}
920: I.~{\relax{Ya}}. Gol'dsheid, S.~A. Molchanov, and L.~A. Pastur,
921:  A pure point spectrum of the stochastic one-dimensional Schr\"odinger
922:   operator,
923:  Funct. Anal. and App. {\bf 11}, 1 (1977).
924: 
925: \bibitem{PasFig78}
926: L.~A. Pastur and A.~L. Figotin,
927:  Teor. Math. Fiz. {\bf 35}, 193 (1978).
928: 
929: \bibitem{AntPasSly81}
930: T.~N. Antsygina, L.~A. Pastur, and V.~A. Slyusarev,
931:  Localization of states and kinetic properties of one-dimensional
932:   disordered systems,
933:  Sov. J. Low Temp. Phys. {\bf 7}(1), 1--21 (1981).
934: 
935: \bibitem{LifGrePas88}
936: I.~M. Lifshits, S.~A. Gredeskul, and L.~A. Pastur,
937:  {\em Introduction to the theory of disordered systems},
938:  John Wiley \& Sons, 1988.
939: 
940: \bibitem{Luc92}
941: J.-M. Luck,
942:  {\em Syst\`emes d\'esordonn\'es unidimensionnels},
943:  CEA, collection Al\'ea Saclay, Saclay, 1992.
944: 
945: \bibitem{TexCom99}
946: C.~Texier and A.~Comtet,
947:  Universality of the Wigner time delay distribution for
948:   one-dimensional random potentials,
949:  Phys. Rev. Lett. {\bf 82}(21), 4220--4223 (1999).
950: 
951: \bibitem{DeMLyr98}
952: F.~A. B.~F. {de~Moura} and M.~L. Lyra,
953:  Delocalization in the 1D Anderson Model with Long-Range Correlated
954:   Disorder,
955:  Phys. Rev. Lett. {\bf 81}(17), 3735 (1998).
956: 
957: \bibitem{RusKanBunHav01}
958: S.~Russ, J.~W. Kantelhardt, A.~Bunde, and S.~Havlin,
959:  Localization in self-affine energy landscapes,
960:  Phys. Rev.~B {\bf 64}, 134209 (2001).
961: 
962: \bibitem{Luc05}
963: J.-M. Luck,
964:  On the statistics of superlocalized states in self-affine disordered
965:   potentials,
966:  J.~Phys.~A: Math. Gen. {\bf 38}, 987 (2005).
967: 
968: \bibitem{JosJay98}
969: S.~K. Joshi and A.~M. Jayannavar,
970:  Distribution of Wigner delay time from single channel disordered
971:   systems,
972:  Solid State Commun. {\bf 106}(6), 363 (1998).
973: 
974: \bibitem{Llo69}
975: P.~Lloyd,
976:  Exactly solvable model of electronic states in a three-dimensional
977:   disordered Hamiltonian: non-existence of localized states,
978:  J.~Phys.~C: Solid State Phys. {\bf 2}(10), 1717 (1969).
979: 
980: \bibitem{Tho72}
981: D.~J. Thouless,
982:  A relation between the density of states and range of localization
983:   for one-dimensional random systems,
984:  J.~Phys.~C: Solid State Phys. {\bf 5}, 77 (1972).
985: 
986: \bibitem{Ish73}
987: K.~Ishii,
988:  Localization of eigenstates and transport phenomena in the one
989:   dimensional disordered system,
990:  Prog. Theor. Phys. (Suppl.) {\bf 53}, 77--138 (1973).
991: 
992: \bibitem{DeyLisAlt01}
993: L.~I. Deych, A.~A. Lisyansky, and B.~L. Altshuler,
994:  Single Parameter Scaling in 1-D Anderson localization. Exact
995:   analytical solution,
996:  Phys. Rev.~B {\bf 64}(22), 224202 (2001).
997: 
998: \bibitem{TitSch03}
999: M.~Titov and H.~Schomerus,
1000:  Anomalous Wave Function Statistics on a One-Dimensional Lattice with
1001:   Power-Law Disorder,
1002:  Phys. Rev. Lett. {\bf 91}(17), 176601 (2003).
1003: 
1004: \bibitem{FriLlo60}
1005: H.~L. Frisch and S.~P. Lloyd,
1006:  Electron levels in a one-dimensional random lattice,
1007:  Phys. Rev. {\bf 120}(4), 1175 (1960).
1008: 
1009: \bibitem{BooLuc07}
1010: D.~Boos\'e and J.-M. Luck,
1011:  Statistics of quantum transmission in one dimension with broad
1012:   disorder,
1013:  J.~Phys.~A: Math. Theor. {\bf 40}, 14045 (2007).
1014: 
1015: \bibitem{AbrAndLicRam79}
1016: E.~Abrahams, P.~W. Anderson, D.~C. Licciardello, and T.~V. Ramakrishnan,
1017:  Scaling theory of localization: absence of quantum diffusion in two
1018:   dimensions,
1019:  Phys. Rev. Lett. {\bf 42}(10), 673 (1979).
1020: 
1021: \bibitem{Pen94}
1022: J.~B. Pendry,
1023:  Symmetry and transport of waves in one-dimensional disordered
1024:   systems,
1025:  Adv. Phys. {\bf 43}(4), 461 (1994).
1026: 
1027: \bibitem{BouComGeoLeD90}
1028: J.-P. Bouchaud, A.~Comtet, A.~Georges, and P.~{Le~Doussal},
1029:  Classical diffusion of a particle in a one-dimensional random force
1030:   field,
1031:  Ann. Phys. (N.Y.) {\bf 201}, 285--341 (1990).
1032: 
1033: \bibitem{ComTex98}
1034: A.~Comtet and C.~Texier,
1035:  One-dimensional disordered supersymmetric quantum mechanics: a brief
1036:   survey,
1037:  in {\em Supersymmetry and Integrable Models}, edited by H.~Aratyn,
1038:   T.~D. Imbo, W.-Y. Keung, and U.~Sukhatme, Lecture Notes in Physics, Vol. 502,
1039:   pages 313--328, Springer, 1998,
1040:  Proceedings of a workshop held at Chicago, IL, USA, 12-14 June 1997
1041:   (also available as cond-mat/97\,07\,313).
1042: 
1043: \end{thebibliography}
1044: 
1045: 
1046: 
1047: \end{document}
1048: