1: \documentclass[12pt]{article}
2: \usepackage{graphicx}
3: \catcode`\@=11
4: \topmargin 0pt
5: \oddsidemargin 0pt
6: \headheight 0pt
7: \headsep 0pt
8: \textheight 9in
9: \textwidth 6.25in
10: \marginparwidth 0.875in
11: \def\numberbysection{\@addtoreset{equation}{section}
12: \def\theequation{\thesection.\arabic{equation}}}
13: \def\baselinestretch{1.05}
14: \numberbysection
15:
16: \newcommand{\beq}{\begin{equation}}
17: \newcommand{\beqa}{\begin{eqnarray}}
18: \newcommand{\eeq}{\end{equation}}
19: \newcommand{\eeqa}{\end{eqnarray}}
20: \newcommand{\abs}[1]{\vert#1\vert}
21: \newcommand{\bigmean}[1]{\left\langle#1\right\rangle}
22: \renewcommand{\c}{{\rm c}}
23: \newcommand{\cum}[1]{\langle\!\langle#1\rangle\!\rangle}
24: \renewcommand{\d}{{\rm d}}
25: \newcommand{\demi}{{\textstyle\frac12}}
26: \newcommand{\e}{{\rm e}}
27: \newcommand{\eps}{\varepsilon}
28: \newcommand{\frad}[2]{\displaystyle{\displaystyle#1\over\displaystyle#2}}
29: \newcommand{\g}{\gamma}
30: \renewcommand{\i}{{\rm i}}
31: \newcommand{\mean}[1]{\langle#1\rangle}
32: \newcommand{\quart}{{\textstyle\frac14}}
33: \newcommand{\rms}{{\rm rms}}
34: \newcommand{\tquart}{{\textstyle\frac34}}
35: \newcommand{\x}{{\rm cr}}
36: \newcommand{\var}{\mathop{{\rm var}}}
37: \newcommand{\C}{{\bf C}}
38: \newcommand{\D}{\Delta}
39: \newcommand{\E}{{\bf E}}
40: \newcommand{\K}{{\bf K}}
41: \renewcommand{\Re}{\mathop{{\rm Re}}}
42:
43: \begin{document}
44: \centerline{\Large\bf On the statistics of superlocalized states}
45: \vspace{.3cm}
46: \centerline{\Large\bf in self-affine disordered potentials}
47: \vspace{1.6cm}
48: \centerline{\large J.M.~Luck\footnote{luck@spht.saclay.cea.fr}}
49: \vspace{.4cm}
50: \centerline{Service de Physique Th\'eorique\footnote{URA 2306 of CNRS},
51: CEA Saclay, 91191 Gif-sur-Yvette cedex, France}
52: \vspace{1cm}
53: \begin{abstract}
54: We investigate the statistics of eigenstates
55: in a weak self-affine disordered potential in one dimension,
56: whose Gaussian fluctuations grow with distance
57: with a positive Hurst exponent $H$.
58: Typical eigenstates are superlocalized on samples much larger than
59: a well-defined crossover length, which diverges in the weak-disorder regime.
60: We present a parallel analytical investigation
61: of the statistics of these superlocalized states in
62: the discrete and the continuum formalisms.
63: For the discrete tight-binding model, the effective localization length
64: decays logarithmically with the sample size,
65: and the logarithm of the transmission is marginally self-averaging.
66: For the continuum Schr\"odinger equation,
67: the superlocalization phenomenon has more drastic effects.
68: The effective localization length decays as a power of the sample length,
69: and the logarithm of the transmission is fully non-self-averaging.
70: \end{abstract}
71: \vfill
72: \noindent To be submitted for publication to the Journal of Physics A
73:
74: \noindent P.A.C.S.: 71.23.An, 73.20.Fz, 72.15.Rn, 05.40.-a.
75:
76: \newpage
77: \section{Introduction}
78:
79: The Anderson localization of a quantum-mechanical particle
80: by a random potential is by now an old and well-understood problem~\cite{loc}.
81: This is especially so in one dimension.
82: Consider for definiteness the tight-binding equation:
83: \beq
84: \psi_{n+1}+\psi_{n-1}+V_n\psi_n=E\psi_n.
85: \label{tb}
86: \eeq
87: In the usual situation where the site potentials $V_n$ are uncorrelated,
88: all the eigenstates are known to be exponentially localized
89: (see e.g.~\cite{thou,pen}).
90:
91: The peculiar features of the localization problem
92: for various kinds of one-dimensional disordered potentials
93: with non-trivial correlations have also been investigated,
94: including dimer models~\cite{dim},
95: potentials with power-law correlations~\cite{pow},
96: potentials generated by chaotic dynamical systems~\cite{cha},
97: and potentials whose correlations are designed on purpose~\cite{adh}.
98: These examples share the common feature of {\it spatial stationarity}:
99: the statistics of the potential is invariant under translation, so that
100: its two-point correlation has a well-defined thermodynamical limit
101: $\mean{V_mV_n}=C_{m-n}$, which only depends on the distance $\abs{m-n}$.
102:
103: The more exotic situation of {\it non-stationary} random potentials,
104: whose fluctuations grow with distance,
105: has only been considered more recently~\cite{m1,m2,b1,m3,b2,b3}.
106: An example of most physical interest is that of a
107: {\it self-affine} Gaussian potential with Hurst exponent $0<H<1$, such that
108: \beq
109: \mean{(V_m-V_n)^2}=\D^2\abs{m-n}^{2H}.
110: \label{hur}
111: \eeq
112: Such a sequence of potentials can be generated
113: by fractional Brownian motion,
114: with a proper choice of stationary but correlated increments
115: \beq
116: \eps_n=V_n-V_{n-1}.
117: \label{inc}
118: \eeq
119: The particular case of the usual Brownian motion,
120: corresponding to stationary and independent increments
121: ($\mean{\eps_m\eps_n}=\D^2\delta_{mn}$), has a Hurst exponent $H=1/2$.
122:
123: An alternative way of characterizing non-stationary potentials
124: consists in considering their structure factor
125: $S(q)=\mean{\hat V(q)\hat V(-q)}$,
126: assuming a power-law divergence of the form
127: \beq
128: S(q)\sim\abs{q}^{-\alpha}
129: \eeq
130: in the long-wavelength limit ($q\to0$).
131: Only the smaller values of the scaling exponent ($\alpha<1$)
132: correspond to stationary potentials,
133: with long-range correlations falling off as $C_n\sim\abs{n}^{-(1-\alpha)}$.
134: Larger values of $\alpha$ necessarily correspond to non-stationary potentials.
135: The above self-affine potentials with stationary increments
136: are obtained in the range $1<\alpha<3$, with the correspondence $\alpha=2H+1$.
137:
138: The main novel feature induced by the non-stationarity of the potential
139: is that the effective disorder strength depends on the spatial scale.
140: The typical potential fluctuation over a distance $N$
141: indeed grows as $V_\rms(N)=\D N^H$.
142: It becomes therefore comparable with the bandwidth when the distance
143: reaches the crossover length
144: \beq
145: N_\x=\D^{-1/H}.
146: \label{nx}
147: \eeq
148: The situation of most interest is that of a weak disorder ($\D\ll1$),
149: so that $N_\x\gg1$.
150: The eigenstates are conventionally localized on scales $N\ll N_\x$,
151: with a localization length scaling as $\xi\sim1/\D^2$.
152: On larger scales ($N\gg N_\x$), however,
153: eigenstates become strongly localized or {\it superlocalized}~\cite{b2,b3},
154: because wavefunctions fall off very fast
155: in the classically forbidden zones of the potential ($\abs{E-V_n}>2$).
156: Let us mention that an alternative viewpoint~\cite{m1,m2}
157: consists in keeping fixed the effective potential width
158: $V_\rms(N)=\D N^H=\Sigma$ at the scale of the system size~$N$.
159: The microscopic disorder strength $\D=\Sigma/N^H$
160: therefore gets rescaled to smaller and smaller values.
161: The crossover length $N_\x=N/\Sigma^{1/H}$
162: grows proportionally to the sample size,
163: whereas the localization length scales as $\xi\sim N^{2H}$.
164: For a large enough non-stationarity ($H>1/2$),
165: such that $\xi$ may become much larger than the sample size $N$,
166: a crossover line in the $(\Sigma,E)$ plane
167: between extended and localized states
168: has been observed~\cite{m1,m2}, and theoretically explained~\cite{b2,b3}.
169:
170: The goal of the present work is to provide the first quantitative analysis
171: of the statistics of superlocalized eigenstates,
172: especially in the regime $N\gg N_\x$,
173: where the superlocalization phenomenon is fully developed.
174: We shall successively deal with the discrete tight-binding model (Section~2)
175: and the continuum Schr\"odinger equation (Section~3).
176:
177: The key quantity considered throughout this work
178: is the effective Lyapunov exponent
179: \beq
180: \g_N=\frac{1}{N}\ln\abs{\psi_N},
181: \label{ly}
182: \eeq
183: where $\psi_n$ is the generic (growing) solution to~(\ref{tb}).
184: The effective Lyapunov exponent provides an estimate of the global growth rate
185: of this solution over $N$ lattice sites,
186: and therefore of the effective decay rate of eigenstates over the same range.
187: In other words, the effective localization length at scale $N$ is
188: \beq
189: \xi_N=\frac{1}{\g_N}.
190: \eeq
191: The effective Lyapunov exponent is also a central quantity in
192: the theory of coherent quantum transport.
193: Indeed the two-probe Landauer formula~\cite{lan}
194: expresses the zero-temperature conductance $g_N$
195: of a sample made of $N$ lattice sites,
196: in terms of the intensity transmission $T_N$ across that sample, as
197: \beq
198: g_N=\frac{2e^2}{h}\,T_N.
199: \label{lanfor}
200: \eeq
201: Furthermore, in the insulating regime where the transmission is small,
202: it is known to scale as $T_N\sim 1/\abs{\psi_N}^2$, hence
203: \beq
204: \ln T_N\approx-2N\g_N.
205: \label{lnt}
206: \eeq
207:
208: The effective Lyapunov exponent,
209: and related physical quantities such as the effective localization length
210: and the logarithm of the transmission (conductance),
211: exhibit different kinds of scaling behavior
212: in the two situations to be investigated successively,
213: the discrete tight-binding model (Section~2)
214: and the continuum Schr\"odinger equation (Section~3).
215: This basic difference is further commented on in the Discussion (Section~4).
216:
217: \section{The discrete tight-binding model}
218:
219: \subsection{Reminder}
220:
221: We find it useful to first give a brief reminder on
222: the conventional situation of a stationary random potential.
223: The statistics of the size-dependent effective Lyapunov exponent $\g_N$
224: is then very similar to that of the free-energy density
225: of a disordered statistical-mechanical system~\cite{pen,cpv,alea}.
226: The effective Lyapunov exponent $\g_N$ has a well-defined self-averaging limit
227: $\g$ in the $N\to\infty$ limit, simply referred to as the Lyapunov exponent.
228: Its reciprocal $\xi=1/\gamma$ is interpreted as the localization length
229: of the problem.
230:
231: The fluctuations of $\g_N$ around $\g$ are Gaussian and scale as
232: $\var{\g_N}=\mean{(\g_N-\g)^2}\approx\sigma^2/N$
233: for a finite but large enough sample ($N\g\gg1$).
234: More generally, the product $N\g_N$ is extensive,
235: in the strong sense that all its cumulants
236: \beq
237: \cum{(N\g_N)^k}\approx c_k N
238: \eeq
239: grow linearly with $N$~\cite{pen}, with amplitudes $c_1=\g$, $c_2=\sigma^2$,
240: and so on.
241: It is worth noticing that $N\g_N$, which is analogous to the total free energy,
242: is also a physical observable in the present context,
243: as it is nothing but the logarithm of the transmission (conductance),
244: up to a constant factor [see~(\ref{lnt})].
245:
246: Furthermore, in the usual situation of independent site potentials
247: with $\mean{V_n}=0$ and $\mean{V_mV_n}=W^2\delta_{mn}$,
248: the Lyapunov exponent behaves as follows in the weak-disorder limit ($W^2\ll1$):
249:
250: \begin{itemize}
251:
252: \item Inside the band, i.e., for $\abs{E}<2$, setting $E=2\cos q$,
253: the celebrated result~\cite{thou}
254: \beq
255: \g\approx\frad{W^2}{8\sin^2q}=\frad{W^2}{2(4-E^2)}
256: \label{in}
257: \eeq
258: shows that the localization length diverges as $1/W^2$.
259:
260: \item Outside the band, i.e., for $\abs{E}>2$, setting $\abs{E}=2\cosh t$,
261: i.e.,
262: \beq
263: t=\ln\frad{\abs{E}+\sqrt{E^2-4}}{2}>0,
264: \eeq
265: the generic solution to~(\ref{tb}) grows exponentially as $\psi_n\sim\e^{nt}$
266: in the absence of disordered potential, hence
267: \beq
268: \g\to t\qquad(W^2\to0).
269: \label{out}
270: \eeq
271:
272: \item
273: Near band edges, i.e., for $E\to\pm2$ and $W^2\to0$ simultaneously,
274: the Lyapunov exponent obeys a scaling law of the form
275: \beq
276: \g\approx W^{2/3}\;F\!\left(\frad{\abs{E}-2}{W^{4/3}}\right),
277: \eeq
278: where the scaling function $F$ is explicitly known
279: in terms of Airy functions~\cite{dg}.
280:
281: \end{itemize}
282:
283: \subsection{General results}
284:
285: We now turn to the current problem,
286: namely the tight-binding equation~(\ref{tb}) with a weak self-affine potential
287: whose fluctuations grow with distance according to~(\ref{hur}),
288: with $\D^2\ll1$.
289: We choose the zero of energies as the site potential at the origin ($V_0=0$),
290: and assume that the origin sits in an allowed zone, i.e., $\abs{E}<2$.
291:
292: A long enough sample ($N\gg N_\x$) generically
293: consists of an alternation of allowed zones
294: ($\abs{E-V_n}<2$) and of forbidden zones ($\abs{E-V_n}>2$).
295: It can be argued from the above results
296: that the growth rate of a generic wavefunction
297: is proportional to $\D^2$ in allowed zones,
298: i.e., inside the `local band' around site $n$
299: (where energies are shifted by $V_n$)~[see~(\ref{in})],
300: whereas it is of order unity in forbidden zones,
301: i.e., outside the `local band'~[see~(\ref{out})].
302:
303: This picture is fully corroborated by the plots shown in
304: References~\cite{b2,b3},
305: where eigenstates are indeed seen to be essentially constant
306: in allowed zones, and to drop very suddenly in forbidden zones.
307: Our main goal is to turn the above picture
308: into a quantitative description of the statistics
309: of the effective Lyapunov exponent~(\ref{ly}).
310:
311: Our starting point is the approximate formula
312: \beq
313: \psi_N\approx\exp\left(\sum_{n=0}^{N-1} t_n\right),
314: \label{wd}
315: \eeq
316: with the definition
317: \beq
318: E-V_n=2\cosh t_n,
319: \eeq
320: hence
321: \beq
322: \g_N=\frac{1}{N}\sum_{n=0}^{N-1}\Re t_n,
323: \label{lyd}
324: \eeq
325: with
326: \beq
327: \Re t_n=\left\{\matrix{
328: \ln\frad{\abs{E-V_n}+\sqrt{(E-V_n)^2-4}}{2}>0\hfill
329: &\hbox{in forbidden zones}&(\abs{E-V_n}>2),\hfill\cr\cr
330: 0\hfill&\hbox{in allowed zones}\hfill&(\abs{E-V_n}<2).\hfill
331: }\right.%}
332: \label{dichod}
333: \eeq
334:
335: Let us first comment on~(\ref{wd}) for a while.
336: This expression is a full discrete analogue of the W.K.B.
337: integral~(\ref{psiwkb}), to be used in Section~3.
338: It provides a quantitative description of the growing solution to~(\ref{tb})
339: in the forbidden zones of the potential,
340: under the sole hypothesis that the sequence
341: of site potentials has small increments:
342: \beq
343: \eps_n=V_n-V_{n-1}\ll1.
344: \eeq
345: Expression~(\ref{wd}) can be easily derived by introducing
346: the Riccati variables~\cite{alea,tn,prb}
347: \beq
348: R_n=\frac{\psi_{n+1}}{\psi_n}.
349: \label{ric}
350: \eeq
351: Assuming $\psi_0=1$, we have $\psi_n=R_0\cdots R_{n-1}$, and
352: \beq
353: \g_N=\frac{1}{N}\sum_{n=0}^{N-1}\ln\abs{R_n}.
354: \label{garic}
355: \eeq
356: The variables $R_n$ obey the recursion
357: \beq
358: R_n=E-V_n-\frac{1}{R_{n-1}}.
359: \label{ricrec}
360: \eeq
361: If we now set
362: \beq
363: x_n=\frad{\e^{t_n}-R_n}{1-\e^{t_n}R_n},\qquad
364: R_n=\frac{\e^{t_n}-x_n}{1-\e^{t_n}x_n},
365: \eeq
366: the variables $x_n$ obey
367: \beq
368: x_n=\e^{-2t_n}\frac{x_{n-1}+\delta_n}{1+\delta_nx_{n-1}},
369: \label{xrec}
370: \eeq
371: with
372: \beq
373: \delta_n=\frac{\e^{t_{n-1}}-\e^{t_n}}{\e^{t_n+t_{n-1}}-1}
374: \approx\frac{\eps_n}{4\sinh^2 t_n},
375: \eeq
376: to leading order as $\eps_n\ll1$.
377: To the same order, the recursion~(\ref{xrec}) can be linearized to
378: $x_n\approx\e^{-2t_n}(x_{n-1}+\delta_n)$,
379: implying that the $x_n$'s are proportional to the $\eps_n$'s.
380: One has therefore $R_n\approx\exp(t_n)$,
381: up to terms proportional to the $\eps_n$'s.
382: The expressions~(\ref{wd}) and~(\ref{lyd}) follow at once.
383: Keeping track of higher powers of the $\eps_n$'s in the above equations
384: would be an efficient way of performing systematic
385: weak-disorder expansions~\cite{alea,prb}.
386:
387: Our analysis of the statistics of the effective Lyapunov exponent $\g_N$
388: is based on the estimate~(\ref{lyd}).
389: It turns out to be advantageous to introduce the Laplace representation
390: \beq
391: \Re t_n=\int\!\frac{\d s}{\i\pi s}\,\cosh(sE)\,K_0(2s)\,\exp(sV_n),
392: \eeq
393: where $K_0$ is the modified Bessel function,
394: and the integration contour is a vertical line with $\Re s>0$.
395: Using the Gaussian statistics of the potentials $V_n$,
396: with $V_0=0$ and~(\ref{hur}), we have
397: \beq
398: \mean{\exp(sV_n)}=\exp\left(\demi\D^2s^2n^{2H}\right),
399: \eeq
400: and similar expressions for higher-order characteristic functions
401: such as $\mean{\exp(s_1V_m+s_2V_n)}$.
402: We can therefore express the correlation functions of the variables $\Re t_n$
403: as multiple contour integrals.
404: By means of~(\ref{lyd}), the moments of the effective Lyapunov exponent $\g_N$
405: can, in turn, be expressed as multiple integrals.
406: The rescaling $x=m/N_\x$, $y=n/N_\x$ implies that, as could be expected,
407: the final results only depend on the system size $N$
408: through the dimensionless ratio
409: \beq
410: X=\frac{N}{N_\x}=\D^{1/H}N,
411: \eeq
412: where the crossover scale $N_\x$ has been introduced in~(\ref{nx}).
413: We thus obtain
414: \beqa
415: &&\mean{\g_N}=\frac{1}{X}\int_0^X\d x
416: \int\!\frac{\d s}{\i\pi s}\,\cosh(sE)\,K_0(2s)\,
417: \exp\left(\demi s^2x^{2H}\right),\label{conave}\\
418: &&\mean{\g_N^2}=\frac{1}{X^2}\int_0^X\d x\int_0^X\d y
419: \int\!\frac{\d s_1}{\i\pi s_1}\,\cosh(s_1E)\,K_0(2s_1)\,
420: \int\!\frac{\d s_2}{\i\pi s_2}\,\cosh(s_2E)\,K_0(2s_2)\,\nonumber\\
421: &&{\hskip 20truemm}\times
422: \exp\left(\demi\!\left(s_1^2x^{2H}+s_1s_2(x^{2H}+y^{2H}-\abs{x-y}^{2H})
423: +s_2^2y^{2H}\right)\right),
424: \label{convar}
425: \eeqa
426: and so on.
427:
428: The regimes of short samples ($N\ll N_\x$)
429: and of long samples ($N\gg N_\x$) deserve to be considered separately.
430:
431: \subsection{Short samples}
432:
433: We first consider relatively short samples,
434: such that $N\ll N_\x$, i.e., $X\ll1$.
435: In this regime, and for $\abs{E}<2$, the contour integral in~(\ref{conave})
436: is dominated by a saddle point at $s_c\approx(2-\abs{E})/x^{2H}\gg1$.
437: We thus obtain the following exponentially small estimate
438: \beq
439: \mean{\g_N}\sim\exp\left(-\frac{(2-\abs{E})^2}{2\,X^{2H}}\right)
440: \label{inst}
441: \eeq
442: for the mean effective Lyapunov exponent.
443:
444: The right-hand side of~(\ref{inst}) has a simple interpretation.
445: It scales indeed as the probability that $E-V_N$ has reached the closest
446: forbidden zone, i.e., $E-V_N=2$ if $E>0$ and $E-V_N=-2$ if $E<0$,
447: so that the superlocalization phenomenon just sets in.
448:
449: For a weak but finite disorder strength $\D$,
450: the mean effective Lyapunov exponent~$\mean{\g_N}$
451: is already of order $\D^2$ in the regime of usual localization,
452: i.e., for $X\ll1$.
453: The onset of superlocalization manifests itself as a crossover
454: of $\mean{\g_N}$ to the behavior~(\ref{inst}),
455: and it takes place for a system size such that
456: \beq
457: X\sim\abs{\ln\D}^{-1/(2H)}\ll1,
458: \qquad\hbox{i.e.,}\quad
459: N\sim\left(\D\abs{\ln\D}^{1/2}\right)^{-1/H}.
460: \eeq
461:
462: \subsection{Long samples}
463:
464: The superlocalization phenomenon is fully developed
465: in the converse regime of long samples ($N\gg N_\x$, i.e., $X\gg1$).
466: The contour integral in~(\ref{conave})
467: is then dominated by the branch cut of $K_0(2s)\approx-(\ln s+\C)$ at $s=0$,
468: with $\C$ being Euler's constant.
469: We thus obtain
470: \beq
471: \mean{\g_N}\approx H(\ln X-1)-\frac{\C+\ln 2}{2}.
472: \label{masy}
473: \eeq
474: The mean effective Lyapunov exponent
475: thus diverges logarithmically with the system size $N$ for $N\gg N_\x$,
476: irrespective of the energy $E$.
477:
478: As a matter of fact, the full distribution of the effective Lyapunov exponent
479: simplifies in the limit of a long sample.
480: Indeed, for $n\gg N_\x$ one has $\mean{V_n^2}\gg1$,
481: so that most often $\abs{V_n}\gg1$, hence $\Re t_n\approx\ln\abs{V_n}$, and
482: \beq
483: \g_N\approx\frac{1}{N}\sum_{n=0}^{N-1}\ln\abs{V_n}.
484: \eeq
485: The rescaling $x=n/N$ yields the following asymptotic form
486: of the effective Lyapunov exponent of long samples:
487: \beq
488: \g_N\approx H\ln X+G_H\approx H\ln(N/N_\x)+G_H\approx\ln(\D N^H)+G_H,
489: \label{gas}
490: \eeq
491: where the random additive term $G_H$ is the following non-linear functional
492: of the normalized fractional Brownian motion:
493: \beq
494: G_H=\int_0^1\d x\,\ln\abs{v(x)}.
495: \label{gfun}
496: \eeq
497: The expression~(\ref{gas}), which holds irrespectively of the energy $E$,
498: is the main result of this section.
499: The functional $G_H$ is investigated in the Appendix.
500: It has a universal distribution which only depends on the Hurst exponent $H$.
501: In the case $H=1/2$, corresponding to usual Brownian motion,
502: similar non-linear integral functionals have been met in several
503: physical problems~\cite{monmaj}.
504: To our knowledge, the distribution of $G_H$ is not known explicitly,
505: even in the Brownian case.
506: Expressions for the mean and the variance of $G_H$
507: are derived in the Appendix for all $H$,
508: as well as the full distribution of $G_1$ in the ballistic limit ($H=1$).
509: The mean $\mean{G_H}$ has a simple linear dependence~(\ref{gyave})
510: on the Hurst exponent~$H$, which can already be read off from~(\ref{masy}),
511: while the more complex expression~(\ref{gvar}) for the variance $\var{G_H}$
512: only simplifies for $H=0$, $1/2$, and~$1$.
513: Figure~\ref{figvarg} illustrates the dependence of $\var{G_H}$
514: on the exponent $H$,
515: whereas Figure~\ref{figg} shows a plot of the probability density
516: of $G_1$ (ballistic case: exact expression~(\ref{ball}))
517: and of $G_{1/2}$ (Brownian case),
518: the latter being measured from an ensemble of numerically generated,
519: and suitably rescaled, long random walks.
520: The distributions is observed to be clearly asymmetric (skew),
521: with very fast decaying tails at large values of $G_H$,
522: and at small values of $G_H$ for any $H<1$.
523:
524: \begin{figure}[htb]
525: \begin{center}
526: \includegraphics[angle=90,width=.6\linewidth]{figvarg.eps}
527: \caption{\small
528: Plot of the variance of the functional $G_H$ entering the result~(\ref{gas}),
529: evaluated from the exact expression~(\ref{gvar}),
530: against the Hurst exponent $H$.
531: Symbols: results~(\ref{zer}), (\ref{gdemi}), and~(\ref{un})
532: for the particular cases $H=0$, $1/2$, and~$1$.}
533: \label{figvarg}
534: \end{center}
535: \end{figure}
536:
537: \begin{figure}[htb]
538: \begin{center}
539: \includegraphics[angle=90,width=.6\linewidth]{figg.eps}
540: \caption{\small
541: Plot of the probability density
542: of the functional $G_H$ entering the result~(\ref{gas}).
543: Thin full line: exact result~(\ref{ball}) for $H=1$ (ballistic limit).
544: Line with symbols (showing histogram bins):
545: numerical result for $H=1/2$ (Brownian case).}
546: \label{figg}
547: \end{center}
548: \end{figure}
549:
550: \subsection{Numerical results}
551:
552: We have confronted our analytical predictions
553: with numerical results in the Brownian case ($H=1/2$).
554: This situation is indeed simpler to handle than the generic one.
555: On the one hand, the integrals over the rescaled spatial positions $x$, $y$
556: in the predictions~(\ref{conave}),~(\ref{convar}) are elementary.
557: On the other hand, sequences of Brownian potentials $V_n$ can easily
558: be generated numerically,
559: by summing independent increments $\eps_n=\pm\D$ with $\D\ll1$,
560: according to~(\ref{inc}), with $V_0=E=0$.
561:
562: \begin{figure}[htb]
563: \begin{center}
564: \includegraphics[angle=90,width=.6\linewidth]{fig1.eps}
565: \caption{\small
566: Plot of the mean effective Lyapunov exponent for a Brownian potential
567: ($H=1/2$) with $E=0$,
568: against the logarithm of the scaling variable $X=N\D^2$.
569: Full curve: analytical prediction~(\ref{conave}).
570: Dashed line: asymptotic behavior~(\ref{masy}),
571: i.e., $\mean{\g_N}\approx(\ln(N\D^2)-1-\C-\ln 2)/2$.
572: Symbols: numerical data for $\D=0.1$ and $\D=0.2$.}
573: \label{fig1}
574: \end{center}
575: \end{figure}
576:
577: The growing solution $\psi_n$ has been evaluated
578: by means of the recursion~(\ref{ricrec}) for the Riccati variables $R_n$,
579: with a random initial condition ($R_1=\tan\phi$ with a uniform angle~$\phi$).
580: The effective Lyapunov exponent has been measured by means of~(\ref{garic}).
581: The mean and the variance of the effective Lyapunov exponent
582: are respectively shown in Figures~\ref{fig1} and~\ref{fig2}.
583: The continuous curves represent
584: the analytical results~(\ref{conave}),~(\ref{convar}),
585: whereas symbols show numerical data for $10^5$ samples
586: with two strengths of disorder, $\D=0.1$ and $\D=0.2$.
587: The accuracy of the data collapse and of the quantitative agreement
588: with theoretical predictions provides a convincing check of our analysis.
589:
590: \begin{figure}[htb]
591: \begin{center}
592: \includegraphics[angle=90,width=.6\linewidth]{fig2.eps}
593: \caption{\small
594: Plot of the variance of the effective Lyapunov exponent
595: for a Brownian potential ($H=1/2$) with $E=0$,
596: against the logarithm of the scaling variable $X=N\D^2$.
597: Full curve: analytical prediction~(\ref{convar}).
598: Dashed line: limiting value~(\ref{gdemi}):
599: $\var{G_{1/2}}=(\pi^2-4)/16=0.366850$.
600: Symbols: numerical data for $\D=0.1$ and $\D=0.2$.}
601: \label{fig2}
602: \end{center}
603: \end{figure}
604:
605: \section{The continuum Schr\"odinger equation}
606:
607: We now turn to the Schr\"odinger equation
608: in a one-dimensional potential $V(x)$, which reads
609: \beq
610: -\psi''(x)+V(x)\psi(x)=E\psi(x),
611: \label{sch}
612: \eeq
613: in reduced units.
614: The potential $V(x)$ is again assumed to be a weak self-affine Gaussian
615: disordered potential with Hurst exponent $0<H<1$, such that
616: \beq
617: \mean{(V(x)-V(y))^2}=\D^2\abs{x-y}^{2H},
618: \label{chur}
619: \eeq
620: with $V(0)=0$ and $\D^2\ll1$.
621:
622: \subsection{General results}
623:
624: The present situation shares most of the characteristic features
625: of the tight-binding problem investigated in Section~2.
626: Its analysis will therefore only be described succinctly.
627:
628: A long enough sample is again generically an alternation
629: of classically allowed zones ($V(x)<E$) and of forbidden zones ($V(x)>E$).
630: We assume $E>0$, so that the origin belongs to an allowed zone.
631: The growth rate of the generic solution of~(\ref{sch})
632: up to distance~$L$ is measured by the effective Lyapunov exponent
633: \beq
634: \g(L)=\frac{1}{L}\ln\abs{\psi(L)},
635: \label{lys}
636: \eeq
637:
638: The local growth rate of wavefunctions
639: is again proportional to $\D^2$ in allowed zones,
640: whereas it is of order unity in forbidden zones.
641: It is therefore legitimate to use the celebrated W.K.B.
642: integral (see e.g.~\cite{ll,wkb})
643: \beq
644: \psi(x)\sim\exp\left(\int_0^x\d y\,\sqrt{(V(y)-E)_+}\right),
645: \label{psiwkb}
646: \eeq
647: with the definition
648: \beq
649: \sqrt{(V(y)-E)_+}=\left\{\matrix{
650: \sqrt{V(y)-E}\hfill&\hbox{in forbidden zones}&(V(y)>E),\hfill\cr\cr
651: 0\hfill&\hbox{in allowed zones}\hfill&(V(y)<E).\hfill
652: }\right.%}
653: \label{dichoc}
654: \eeq
655:
656: The integral formula~(\ref{psiwkb}) provides a quantitative description,
657: with exponential accuracy,
658: of the growing solution to~(\ref{sch}) in the forbidden zones of the potential,
659: whenever the latter is slowly varying.
660: One has therefore
661: \beq
662: \g(L)\approx\frac{1}{L}\int_0^L\d x\,\sqrt{(V(x)-E)_+}.
663: \label{lyc}
664: \eeq
665:
666: Equations~(\ref{lyc}) and~(\ref{dichoc}) are the continuum analogues
667: of~(\ref{lyd}) and~(\ref{dichod}).
668: They constitute the starting point of the subsequent
669: analysis of the statistics of the effective Lyapunov exponent.
670: It is again advantageous to introduce a Laplace representation:
671: \beq
672: \sqrt{(V(x)-E)_+}=\int\!\frac{\d s}{2\i\pi}\,\sqrt{\frac{\pi}{4s^3}}
673: \,\exp(s(V(x)-E)).
674: \eeq
675: The correlation functions of the variables $\sqrt{(V(x)-E)_+}$,
676: and therefore the moments of the effective Lyapunov exponent,
677: can again be expressed as explicit multiple integrals.
678: We will only need the expression of the mean effective Lyapunov exponent,
679: which reads
680: \beq
681: \mean{\g(L)}=\frac{1}{L}\int_0^L\d x
682: \int\!\frac{\d s}{2\i\pi}\,\sqrt{\frac{\pi}{4s^3}}
683: \exp\left(\demi\D^2s^2x^{2H}-sE\right).
684: \eeq
685: The rescalings $z=x/\ell_\x$, $u=sE$ show that the above result scales as
686: \beq
687: \mean{\g(L)}=\frac{E^{1/2}}{X}\int_0^X\d z
688: \int\!\frac{\d u}{2\i\pi}\,\sqrt{\frac{\pi}{4u^3}}
689: \exp\left(\demi u^2z^{2H}-u\right),
690: \label{schave}
691: \eeq
692: in terms of the length ratio
693: \beq
694: X=\frac{L}{\ell_\x},
695: \eeq
696: where the energy-dependent crossover length
697: \beq
698: \ell_\x=(E/\D)^{1/H}
699: \eeq
700: is the typical size of the first allowed zone containing the origin.
701:
702: \subsection{Short samples}
703:
704: For rather small samples, such that $L\ll\ell_\x$, i.e., $X\ll1$,
705: the contour integral in~(\ref{schave})
706: is dominated by a saddle point at $u_c\approx z^{-2H}\gg1$.
707: We are thus left with the exponentially small estimate
708: \beq
709: \mean{\g(L)}\sim\exp\left(-\frac{1}{2\,X^{2H}}\right).
710: \eeq
711: This expression again scales as the probability that $V(L)$ is equal to $E$,
712: so that a forbidden zone has just been entered.
713:
714: For a weak but finite disorder strength $\D$,
715: the mean effective Lyapunov exponent
716: again has a finite limit of order $\D^2$ for $X\ll1$.
717: The onset of superlocalization again takes place for a system size such that
718: \beq
719: X\sim\abs{\ln\D}^{-1/(2H)}\ll1,
720: \qquad\hbox{i.e.,}\quad
721: L\sim\left(\frac{E}{\D\abs{\ln\D}^{1/2}}\right)^{1/H}.
722: \eeq
723:
724: \subsection{Long samples}
725:
726: The superlocalization phenomenon is fully developed
727: in the converse regime of long samples ($L\gg\ell_\x$, i.e., $X\gg1$).
728: The contour integral in~(\ref{schave})
729: is dominated by the square-root branch cut at $s=0$.
730: We thus obtain a power-law growth of the mean effective Lyapunov exponent:
731: \beq
732: \mean{\g(L)}\approx\frac{2^{1/4}}{(H+2)\sqrt{\pi}}
733: \,\Gamma\!\left(\tquart\right)(\D L^H)^{1/2},
734: \label{meany}
735: \eeq
736: irrespective of the energy $E$.
737:
738: The full distribution of the effective Lyapunov exponent
739: again simplifies in the limit of a long sample.
740: Indeed, for $x\gg\ell_\x$, $E$ is most often negligible with respect to $V(x)$.
741: Equation~(\ref{lyc}) therefore simplifies to
742: \beq
743: \g(L)\approx\frac{1}{L}\int_0^L\d x\,\sqrt{(V(x))_+}.
744: \eeq
745: The rescaling of $x$ by $L$ yields the following asymptotic form
746: of the effective Lyapunov exponent of long samples:
747: \beq
748: \g(L)\approx E^{1/2}\,X^{H/2}\,Y_H\approx E^{1/2}\,(L/\ell_\x)^{H/2}\,Y_H
749: \approx(\D L^H)^{1/2}\,Y_H,
750: \label{glas}
751: \eeq
752: where the fluctuating factor $Y_H$ is another non-linear functional
753: of the normalized fractional Brownian motion:
754: \beq
755: Y_H=\int_0^1\d x\,\sqrt{(v(x))_+}.
756: \label{yfun}
757: \eeq
758: This functional is investigated in the Appendix.
759: It has a universal distribution which only depends on the Hurst exponent $H$.
760: Its first two moments are evaluated for all values of $H$
761: [see respectively~(\ref{gyave}) and~(\ref{y2})],
762: as well as the full distribution of $Y_1$ in the ballistic limit.
763: Figure~\ref{figwh} illustrates the dependence of the dimensionless ratio
764: \beq
765: W_H=\frac{\mean{Y_H^2}}{\mean{Y_H}^2}
766: \label{whdef}
767: \eeq
768: on the exponent $H$.
769: Figure~\ref{figy} shows a plot of the probability density
770: of $Y_1$ (ballistic case: exact expression~(\ref{ball}))
771: and of $Y_{1/2}$ (Brownian case),
772: the latter being measured from an ensemble of numerically generated,
773: and suitably rescaled, long random walks.
774: The delta peak at $Y=0$, which is present in the ballistic limit ($H=1$),
775: gets smeared out into a continuous density which diverges as $Y\to0$
776: for any $H<1$.
777:
778: \begin{figure}[htb]
779: \begin{center}
780: \includegraphics[angle=90,width=.6\linewidth]{figwh.eps}
781: \caption{\small
782: Plot of the moment ratio $W_H$, defined in~(\ref{whdef}),
783: of the quantity $Y_H$ entering the result~(\ref{glas}),
784: evaluated from the exact expression~(\ref{yw}),
785: against the Hurst exponent $H$.
786: Symbols: results~(\ref{zer}), (\ref{wdemi}), and~(\ref{un})
787: for the particular cases $H=0$, $1/2$, and $1$.}
788: \label{figwh}
789: \end{center}
790: \end{figure}
791:
792: \begin{figure}[htb]
793: \begin{center}
794: \includegraphics[angle=90,width=.6\linewidth]{figy.eps}
795: \caption{\small
796: Plot of the probability density
797: of the random variable $Y_H$ entering the result~(\ref{glas}).
798: Thin full line: continuous component (delta peak at $Y=0$ omitted)
799: of the exact result~(\ref{ball}) for $H=1$ (ballistic limit).
800: Line with symbols (showing histogram bins):
801: numerical result for $H=1/2$ (Brownian case).}
802: \label{figy}
803: \end{center}
804: \end{figure}
805:
806: \section{Discussion}
807:
808: The present work is devoted to the specific features of localization
809: in a weak self-affine disordered potential in one dimension,
810: whose fluctuations grow with distance with a positive Hurst exponent $H$.
811: The most salient feature of such non-stationary potentials
812: is that the effective disorder strength depends on the spatial scale.
813: Even in the regime of most physical interest,
814: where the fluctuations of the disordered potential are small
815: at the microscopic scale ($\D\ll1$), the effective disorder becomes strong
816: beyond a well-defined crossover length $N_\x$ or $\ell_\x$,
817: which diverges as $\D^{-1/H}$.
818: Samples much longer than this crossover length
819: typically consist of an alternation of classically allowed zones,
820: where the growth rate of a generic wavefunction is small
821: and proportional to the square disorder strength~$\D^2$,
822: and of forbidden zones, where this growth rate is large.
823: Typical eigenstates on such samples turn out to be superlocalized~\cite{b2,b3}.
824:
825: In this paper we present a parallel analytical investigation
826: of the statistics of superlocalized eigenstates,
827: for both the discrete tight-binding model
828: and the continuum Schr\"odinger equation.
829: First of all, the clear separation of the growth rates of wavefunctions
830: in allowed and in forbidden zones in the weak-disorder regime
831: is shown to allow for a semi-classical treatment
832: of the superlocalization phenomenon, by means of the W.K.B. formalism.
833: The key quantity is the size-dependent effective Lyapunov exponent,
834: which is closely related to the transmission across the sample,
835: and therefore to its zero-temperature conductance by the Landauer formula.
836:
837: For the discrete tight-binding model, investigated in Section~2,
838: the effective Lyapunov exponent behaves as
839: \beq
840: \g_N\approx H\ln(N/N_\x)+G_H\approx\ln(\D N^H)+G_H
841: \label{dgas}
842: \eeq
843: in the fully superlocalized regime of long samples
844: ($N\gg N_\x$) [see~(\ref{gas})].
845: The fluctuating part~$G_H$ is an {\it additive term} of order unity.
846: It is given by the functional~(\ref{gfun}),
847: whose limit distribution only depends on the Hurst exponent $H$.
848: These results translate as follows in terms of physical quantities.
849: One of the most characteristic features of superlocalization
850: is the fall-off of the effective localization length $\xi_N$
851: as a function of $N$.
852: This decay is however only logarithmic, as $\xi_N\approx1/[H\ln(N/N_\x)]$.
853: As far as the transmission $T_N$ is concerned
854: (or equivalently the conductance $g_N$),
855: the relevant quantity is its logarithm, as in most localization problems.
856: Its mean grows as $\mean{\ln T_N}\approx-2HN\ln(N/N_\x)$,
857: whereas its variance grows as $\var{\ln T_N}\approx4\var G_H N^2$.
858: The logarithm of the transmission is therefore
859: {\it marginally self-averaging},
860: as its relative variance $\var{\ln T_N}/\mean{\ln T_N}^2\sim1/[\ln(N/N_\x)]^2$
861: falls off logarithmically.
862:
863: For the continuum Schr\"odinger equation, investigated in Section~3,
864: the effective Lyapunov exponent behaves as
865: \beq
866: \g(L)\approx E^{1/2}\,(L/\ell_\x)^{H/2}\,Y_H\approx(\D L^H)^{1/2}\,Y_H
867: \label{dglas}
868: \eeq
869: [see~(\ref{glas})] in the fully superlocalized regime.
870: The fluctuating part $Y_H$ of this result, given by the functional~(\ref{yfun}),
871: is now involved {\it as a multiplicative factor.}
872: This multiplicative law generates more pronounced superlocalization effects,
873: and especially stronger fluctuations in physical quantities.
874: The effective localization length $\xi(L)\approx1/(Y_H\D^{1/2}L^{H/2})$
875: now decays as a power of the sample length, with a fluctuating amplitude.
876: As far as the transmission is concerned,
877: its mean grows as $\mean{\ln T(L)}\sim-\D^{1/2}L^{1+H/2}$,
878: i.e., more rapidly than linearly with the sample length.
879: The logarithm of the transmission is now {\it fully non-self-averaging},
880: as its relative variance $\var{\ln T(L)}/\mean{\ln T(L)}^2\to W_H-1$
881: has a non-trivial limit for $L\gg\ell_\x$.
882:
883: The effective Lyapunov exponent and related physical quantities
884: therefore exhibit two different kinds of statistics
885: in the discrete and in the continuum formalism.
886: The basic difference between both situations
887: is already present in the case of a constant potential.
888: It is indeed deeply rooted in the dispersion relations of the models.
889: For the tight-binding model with a constant site potential~$V$ outside the band,
890: the growth rate~$t$ of wavefunctions
891: is such that $E-V=2\cosh t$, and therefore only diverges logarithmically
892: as $t\approx\ln\abs{V}$ for large $V$.
893: For the continuum Schr\"odinger equation with a constant potential~$V$,
894: the growth rate now obeys $V-E=t^2$, and diverges as $t\approx\sqrt{V}$.
895: Replacing in the above dispersion estimates $t$ by $\g$,
896: and $V$ by the product of a random number of order unity by
897: $V_\rms\sim\D N^H$ or $\D L^H$, leads to the correct qualitative forms
898: of the scaling equations~(\ref{dgas}) and~(\ref{dglas}).
899: Finally, this line of reasoning also shows
900: that the results for the Schr\"odinger equation cannot be recovered
901: from those of the tight-binding model by going to the continuum limit,
902: because this limit corresponds to $t\to0$,
903: whereas superlocalization is essentially due to large values of $t$.
904: Table~\ref{one} summarizes the discussion.
905:
906: \begin{table}[ht]
907: \begin{center}
908: \begin{tabular}{|c|c|c|c|c|}
909: \hline
910: Quantity&tight-binding model&Schr\"odinger equation&Anderson\\
911: &$N\gg N_\x$ sites&length $L\gg\ell_\x$&localization\\
912: \hline
913: $\mean{\xi}$&$1/\ln(N/N_\x)$&$(\D L^H)^{-1/2}$&$1/\D^2$\\
914: $-\mean{\ln T}$&$N\ln(N/N_\x)$&$\D^{1/2}L^{1+H/2}$&$\D^2 L$\\
915: $\var{\ln T}/\mean{\ln T}^2$&$1/[\ln(N/N_\x)]^2$&constant&$1/(\D^2L)$\\
916: \hline
917: \end{tabular}
918: \caption{\small Scaling behavior of various quantities
919: in the regime of fully developed superlocalization:
920: mean effective localization length, mean and relative variance
921: of the logarithm of the transmission (conductance).
922: Only the qualitative scaling behavior is given
923: (dependence on the sample size $N$ or $L$ and on the disorder strength $\D$).
924: The emphasis is put on the difference between
925: the discrete tight-binding model
926: (additive fluctuation in the effective Lyapunov exponent)
927: and the continuum Schr\"odinger equation (multiplicative fluctuation).
928: The well-known results for conventional Anderson localization
929: are recalled for comparison.}
930: \label{one}
931: \end{center}
932: \end{table}
933:
934: Let us close up with a word on the generality of our results.
935: For any weak self-affine (not necessarily Gaussian)
936: random potential obeying power laws
937: of the form~(\ref{hur}) or~(\ref{chur}), with any positive Hurst exponent~$H$,
938: both the functional form of the results~(\ref{dgas}),~(\ref{dglas})
939: and the scaling laws recalled in Table~\ref{one}
940: can be shown to still hold true.
941: The fluctuating parts $G_H$ and~$Y_H$ are, however,
942: only given by~(\ref{gfun}) and~(\ref{yfun}) for the Gaussian
943: self-affine potentials associated with fractional Brownian motion.
944: Their distribution involve in general further details
945: of the statistics of the potentials.
946:
947: \subsubsection*{Acknowledgements}
948:
949: It is a pleasure to thank Dominique Boos\'e for very stimulating discussions.
950:
951: \appendix
952: \setcounter{equation}{0}
953: \def\theequation{A.\arabic{equation}}
954: \section*{Appendix. The functionals $G_H$ and $Y_H$}
955:
956: This Appendix is devoted to the random quantities $G_H$ and $Y_H$,
957: which respectively enter our key results~(\ref{gas}) and~(\ref{glas}).
958: They are defined as the following non-linear functionals
959: \beq
960: G_H=\int_0^1\d x\,\ln\abs{v(x)},\qquad
961: Y_H=\int_0^1\d x\,\sqrt{(v(x))_+},
962: \eeq
963: of the normalized fractional Brownian motion $v(x)$
964: with Hurst exponent $0<H<1$ on $0\le x\le1$, so that $v(0)=0$ and
965: \beq
966: \mean{(v(x)-v(y))^2}=\abs{x-y}^{2H}.
967: \eeq
968:
969: \subsubsection*{First moments}
970:
971: The means (first moments)
972: \beq
973: \mean{G_H}=\int_0^1\d x\,\mean{\ln\abs{v(x)}},\qquad
974: \mean{Y_H}=\int_0^1\d x\,\bigmean{\sqrt{(v(x))_+}},
975: \eeq
976: can be readily evaluated.
977: Indeed, $v(x)$ is a Gaussian with variance
978: $\mean{v(x)^2}=x^{2H}$, so that its probability density reads
979: \beq
980: P(v(x))=\frac{1}{x^H\sqrt{2\pi}}\,\exp\left(-\frac{v(x)^2}{2\,x^{2H}}\right).
981: \eeq
982: Elementary integrals yield
983: \beq
984: \mean{\ln\abs{v(x)}}=H\ln x-\frac{\C+\ln 2}{2},\qquad
985: \bigmean{\sqrt{(v(x))_+}}
986: =\frac{1}{2^{3/4}\sqrt{\pi}}\,\Gamma\!\left(\tquart\right)x^{H/2},
987: \eeq
988: so that
989: \beq
990: \mean{G_H}=-H-\frac{\C+\ln 2}{2},
991: \qquad
992: \mean{Y_H}=\frac{2^{1/4}}{(H+2)\sqrt{\pi}}\,\Gamma\!\left(\tquart\right).
993: \label{gyave}
994: \eeq
995:
996: \subsubsection*{Second moments}
997:
998: The second moments
999: \beq
1000: \mean{G_H^2}=2\int_0^1\d x\int_x^1\d y\,\mean{\ln\abs{v(x)}\,\ln\abs{v(y)}},
1001: \quad
1002: \mean{Y_H^2}=2\int_0^1\d x\int_x^1\d y\,\bigmean{\sqrt{(v(x))_+(v(y))_+}},
1003: \eeq
1004: now involve two-point observables,
1005: expressed in terms of the correlated Gaussian variables $v(x)$ and $v(y)$,
1006: such that
1007: \beq
1008: \mean{v(x)^2}=x^{2H},\qquad\mean{v(y)^2}=y^{2H},\qquad
1009: \mean{v(x)v(y)}=\demi\left(x^{2H}+y^{2H}-(y-x)^{2H}\right).
1010: \label{vxy}
1011: \eeq
1012: It is convenient to use polar co-ordinates in the $v(x)$, $v(y)$ plane.
1013: For $x\le y$, we set
1014: \beq
1015: z=\frac{x}{y}\le1,
1016: \eeq
1017: and
1018: \beq
1019: v(x)=r\cos\theta,\qquad v(y)=\frac{r}{z^H}\,\sin(\theta+\alpha),
1020: \eeq
1021: with
1022: \beq
1023: \sin\alpha
1024: =\frac{\mean{v(x)v(y)}}{\sqrt{\mean{v(x)^2}\,\mean{v(y)^2}}}
1025: =\frac{1+z^{2H}-(1-z)^{2H}}{2z^H}\qquad(0\le\alpha\le\pi/2).
1026: \label{aldef}
1027: \eeq
1028: This parametrization ensures that the angle $\theta$ is
1029: uniformly distributed between $0$ and $2\pi$,
1030: whereas the radial variable~$r$ has a probability density
1031: \beq
1032: P_r(r)=\frac{r}{x^{2H}}\,\exp\!\left(-\frac{r^2}{2\,x^{2H}}\right).
1033: \eeq
1034:
1035: The calculation of $\mean{\ln\abs{v(x)}\,\ln\abs{v(y)}}$
1036: splits into the radial integrals
1037: \beq
1038: \mean{\ln r}=\frac{\ln(2x^{2H})-\C}{2},\qquad
1039: \mean{(\ln r)^2}_\c=\frac{\pi^2}{24},
1040: \label{rad}
1041: \eeq
1042: where $\mean{\cdots}_\c$ denotes the connected part, and the angular integrals
1043: \beq
1044: \mean{\ln\abs{\!\cos\theta}}=\mean{\ln\abs{\!\sin\theta}}=-\ln2,\qquad
1045: \mean{\ln\abs{\!\cos(\theta)}\,\ln\abs{\!\sin(\theta+\alpha)}}_\c=
1046: \frac{\alpha^2}{2}-\frac{\pi^2}{24}.
1047: \label{ang}
1048: \eeq
1049: The derivation of the second result requires the use of the Fourier series
1050: \beq
1051: \sum_{n=1}^\infty(-1)^n\,\frac{\cos(2n\theta)}{n}=-\ln(2\abs{\!\cos\theta}),
1052: \qquad
1053: \sum_{n=1}^\infty(-1)^n\,\frac{\cos(2n\theta)}{n^2}=\theta^2-\frac{\pi^2}{12},
1054: \eeq
1055: where the latter equality holds for $\abs{\theta}\le\pi$.
1056: We thus obtain the following simple expression
1057: \beq
1058: \mean{\ln\abs{v(x)}\,\ln\abs{v(y)}}_\c=\frac{\alpha^2}{2}
1059: \eeq
1060: for the connected correlation function,
1061: which leads us to the result
1062: \beq
1063: \var{G_H}=\mean{G_H^2}_\c=\demi\int_0^1\alpha^2\,\d z,
1064: \label{gvar}
1065: \eeq
1066: where $\alpha$ is defined in terms of $z$ by~(\ref{aldef}).
1067:
1068: Similarly, the calculation of $\bigmean{\sqrt{(v(x))_+(v(y))_+}}$
1069: splits into the radial integral
1070: \beq
1071: \mean{r}=\sqrt{\frac{\pi}{2}}\,x^H,
1072: \eeq
1073: and the angular integral
1074: \beq
1075: \int_{-\alpha}^{\pi/2}
1076: \frac{\d\theta}{2\pi}\sqrt{\cos\theta\,\sin(\theta+\alpha)}
1077: =\frac{k^2}{\pi}\int_0^{\pi/2}\frac{\cos^2\phi\,\d\phi}{\sqrt{1-k^2\sin^2\phi}}
1078: =\frac{1}{\pi}\left(\E(k)-(1-k^2)\K(k)\right),
1079: \label{angul}
1080: \eeq
1081: where we have set
1082: \beq
1083: \sin\left(\theta+\frac{\alpha}{2}-\frac{\pi}{4}\right)=k\sin\phi,
1084: \eeq
1085: with
1086: \beq
1087: k^2=\sin^2\left(\frac{\pi}{4}+\frac{\alpha}{2}\right)
1088: =\frac{1+\sin\alpha}{2}=\frac{(1+z^H)^2-(1-z)^{2H}}{4z^H},
1089: \label{kdef}
1090: \eeq
1091: and where $\E(k)$ and $\K(k)$ are complete elliptic integrals.
1092: Finally,
1093: \beq
1094: \mean{Y_H^2}=\frac{1}{H+2}\sqrt{\frac{2}{\pi}}
1095: \int_0^1\left(\E(k)-(1-k^2)\K(k)\right)z^{H/2}\,\d z,
1096: \label{y2}
1097: \eeq
1098: so that
1099: \beq
1100: W_H=\frac{\mean{Y_H^2}}{\mean{Y_H}^2}
1101: =\frac{H+2}{2\,\pi^{3/2}}\,\Gamma\!\left(\quart\right)^2
1102: \int_0^1\left(\E(k)-(1-k^2)\K(k)\right)z^{H/2}\,\d z,
1103: \label{yw}
1104: \eeq
1105: where $k$ is defined in terms of $z$ by~(\ref{kdef}).
1106:
1107: The general expressions~(\ref{gvar}) and~(\ref{yw})
1108: simplify in the following particular cases.
1109:
1110: \subsubsection*{Ultraslow limit ($H=0$)}
1111:
1112: \noindent In this limit we have
1113: \beq
1114: \alpha=\frac{\pi}{6},\qquad k=\frac{\sqrt{3}}{2},
1115: \eeq
1116: irrespective of $z$, hence
1117: \beq
1118: \var{G_0}=\frac{\pi^2}{72}=0.137078,\qquad
1119: W_0=\frac{1}{\pi^{3/2}}\,\Gamma\!\left(\quart\right)^2
1120: \left(\E\left(\frac{\sqrt{3}}{2}\right)
1121: -\quart\,\K\left(\frac{\sqrt{3}}{2}\right)\right)=1.586206.
1122: \label{zer}
1123: \eeq
1124:
1125: \subsubsection*{Brownian case ($H=1/2$)}
1126:
1127: \noindent In this case we have
1128: \beq
1129: \sin\alpha=2k^2-1=\sqrt{z},
1130: \eeq
1131: so that a direct integration yields
1132: \beq
1133: \var{G_{1/2}}=\frac{\pi^2-4}{16}=0.366850.
1134: \label{gdemi}
1135: \eeq
1136: In order to evaluate $W_{1/2}$,
1137: it is more convenient to go back to the middle expression of~(\ref{angul}),
1138: and to integrate first over $k$, then over $\phi$.
1139: We thus obtain after some algebra
1140: \beq
1141: W_{1/2}=\frac{5}{288\pi^{3/2}}\,\Gamma\!\left(\quart\right)^2
1142: \left(50-3\sqrt{2}\,\ln\left(1+\sqrt{2}\right)\right)=1.895949.
1143: \label{wdemi}
1144: \eeq
1145:
1146: \subsubsection*{Ballistic limit ($H=1$)}
1147:
1148: \noindent In this case we have
1149: \beq
1150: \alpha=\frac{\pi}{2},\qquad k=1,
1151: \eeq
1152: irrespective of $z$, hence
1153: \beq
1154: \var{G_1}=\frac{\pi^2}{8}=1.233701,\qquad
1155: W_1=\frac{1}{\pi^{3/2}}\,\Gamma\!\left(\quart\right)^2=2.360681.
1156: \label{un}
1157: \eeq
1158: The constancy of $\alpha=\pi/2$
1159: means that $v(x)$ and $v(y)$ are fully correlated.
1160: Their correlation functions~(\ref{vxy})
1161: indeed saturate the Cauchy-Schwarz inequality.
1162:
1163: The whole fractional Brownian process degenerates into
1164: a one-variable problem in the ballistic limit.
1165: We have indeed
1166: \beq
1167: v(x)=xw,
1168: \eeq
1169: where $w\equiv v(1)$ is a Gaussian variable such that $\mean{w^2}=1$, hence
1170: \beq
1171: G_1=\ln\abs{w}-1,\qquad Y_1=\frac{2}{3}\sqrt{w_+}.
1172: \eeq
1173: By performing the appropriate changes of variables on the Gaussian law of $w$,
1174: we get the following explicit expressions for the probability densities:
1175: \beq
1176: P(G_1)=\sqrt{\frac{2}{\pi}}\,\exp\left(G_1+1-\demi\,\e^{2(G_1+1)}\right),
1177: \qquad
1178: P(Y_1)=\demi\delta(Y_1)
1179: +\frac{9\,Y_1}{2\sqrt{2\pi}}\,\exp\left(-\frac{81\,Y_1^4}{32}\right).
1180: \label{ball}
1181: \eeq
1182:
1183: %\newpage
1184: \begin{thebibliography}{99}
1185:
1186: \bibitem{loc}
1187: See e.g. I.M. Lifshitz, S.A. Gredeskul, and L.A. Pastur, {\it Introduction to
1188: the Theory of Disordered Systems} (Wiley, New-York, 1988);
1189: B. Kramer and A. MacKinnon, Rep. Prog. Phys. {\bf 56}, 1469 (1993);
1190: Y. Imry, {\it Introduction to Mesoscopic Physics} (Oxford University Press,
1191: Oxford, 1997);
1192: K. Efetov, {\it Supersymmetry in Disorder and Chaos} (Cambridge University
1193: Press, Cambridge, 1997).
1194:
1195: \bibitem{thou}
1196: D.J. Thouless, in {\it La mati\`ere mal condens\'ee -- Ill-Condensed Matter},
1197: Proceedings of Les Houches Summer School, edited by R. Balian, R. Maynard,
1198: and G. Toulouse (North-Holland, Amsterdam, 1979).
1199:
1200: \bibitem{pen}
1201: J.B. Pendry, Adv. Phys. {\bf 43}, 461 (1994).
1202:
1203: \bibitem{dim}
1204: D.H. Dunlap, H.L. Wu, and P.W. Phillips, Phys. Rev. Lett. {\bf 65}, 88 (1990).
1205:
1206: \bibitem{pow}
1207: S. Russ, S. Havlin, and I. Webman, Phil. Mag. B {\bf 77}, 1449 (1998);
1208: S. Russ, Phys. Rev. B {\bf 66}, 012204 (2002);
1209: F.A.B.F. de Moura, M.D. Coutinho-Filho, E.P. Raposo, and M.L. Lyra, Phys. Rev.
1210: B {\bf 68}, 012202 (2003).
1211:
1212: \bibitem{cha}
1213: H. Yamada, M. Goda, and Y. Aizawa, J. Phys. Cond. Matt. {\bf 3}, 10043 (1991);
1214: J.~Phys. Soc. Japan {\bf 60}, 3501 (1991);
1215: R.A. Pinto, M. Rodriguez, J.A. Gonzalez, and E. Medina,
1216: preprint cond-mat/0402364.
1217:
1218: \bibitem{adh}
1219: F.M. Izrailev and A.A. Krokhin, Phys. Rev. Lett. {\bf 82}, 4062 (1999);
1220: F.M. Izrailev, A.A. Krokhin, and S.E. Ulloa, Phys. Rev. B {\bf 63}, 041102
1221: (2001);
1222: L. Tessieri, J. Phys. A {\bf 35}, 9585 (2002).
1223:
1224: \bibitem{m1}
1225: F.A.B.F. de Moura and M.L. Lyra, Phys. Rev. Lett. {\bf 81}, 3735 (1998).
1226:
1227: \bibitem{m2}
1228: F.A.B.F. de Moura and M.L. Lyra, Physica A {\bf 266}, 465 (1999).
1229:
1230: \bibitem{b1}
1231: J.W. Kantelhardt, S. Russ, A. Bunde, S. Havlin, and I. Webman, Phys. Rev. Lett.
1232: {\bf 84}, 198 (2000).
1233:
1234: \bibitem{m3}
1235: F.A.B.F. de Moura and M.L. Lyra, Phys. Rev. Lett. {\bf 84}, 199 (2000).
1236:
1237: \bibitem{b2}
1238: A. Bunde, S. Havlin, J.W. Kantelhardt, S. Russ, and I. Webman, J. Mol. Liq.
1239: {\bf 86}, 151(2000).
1240:
1241: \bibitem{b3}
1242: S. Russ, J.W. Kantelhardt, A. Bunde, and S. Havlin, Phys. Rev. B {\bf 64},
1243: 134209 (2001).
1244:
1245: \bibitem{lan}
1246: Y. Imry and R. Landauer, Rev. Mod. Phys. {\bf 71}, S306 (1999).
1247:
1248: \bibitem{cpv}
1249: A. Crisanti, G. Paladin, and A. Vulpiani, {\it Products of Random Matrices
1250: in Statistical Physics} (Springer, Berlin, 1992).
1251:
1252: \bibitem{alea}
1253: J.M. Luck, {\it Syst\`emes d\'esordonn\'es unidimensionnels} (in French)
1254: (Collection Al\'ea-Saclay, 1992).
1255:
1256: \bibitem{dg}
1257: B. Derrida and E. Gardner, J. Phys. (France) {\bf 45}, 1283 (1984).
1258:
1259: \bibitem{tn}
1260: Th.M. Nieuwenhuizen, Physica A {\bf 113}, 173 (1982).
1261:
1262: \bibitem{prb}
1263: J.M. Luck, Phys. Rev. B {\bf 39}, 5834 (1989).
1264:
1265: \bibitem{monmaj}
1266: C. Monthus and A. Comtet, J. Phys. I (France) {\bf 4}, 635 (1994);
1267: C. Monthus, Ann. Phys. (Paris) {\bf 20}, 341 (1995);
1268: A. Comtet, C.~Monthus, and M. Yor, J. Appl. Prob. {\bf 35}, 255 (1998);
1269: S.N. Majumdar and A. Comtet, Phys. Rev. Lett. {\bf 92}, 225501 (2004).
1270:
1271: \bibitem{ll}
1272: L.D. Landau and E.M. Lifshitz, {\it Quantum Mechanics:
1273: Non-Relativistic Theory} (Pergamon, Oxford, 1959).
1274:
1275: \bibitem{wkb}
1276: N. Fr\"oman and P.O. Fr\"oman, {\it Physical Problems Solved
1277: by the Phase-Integral Method} (Cambridge University Press, Cambridge, 2002).
1278:
1279: \end{thebibliography}
1280: \end{document}
1281: