1: \documentclass[11pt]{article}
2: \usepackage{amsmath}
3: \usepackage{amsfonts}
4: \usepackage{amssymb}
5: \usepackage{graphicx}
6:
7: \topmargin 0.25truein
8: \oddsidemargin 0.0truein
9: \evensidemargin 0.0truein
10: \textheight 8.5truein
11: \textwidth 6.5truein
12: \footskip 0.3truein
13: \headsep 0.6truein
14: \parskip 0.3cm
15: \setlength{\leftmargini}{.5\leftmargini}
16:
17: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
18: \begin{document}
19:
20: \newcommand{\ls}[1]
21: {\dimen0=\fontdimen6\the\font \lineskip=#1\dimen0
22: \advance\lineskip.5\fontdimen5\the\font \advance\lineskip-\dimen0
23: \lineskiplimit=.9\lineskip \baselineskip=\lineskip
24: \advance\baselineskip\dimen0 \normallineskip\lineskip
25: \normallineskiplimit\lineskiplimit \normalbaselineskip\baselineskip
26: \ignorespaces }
27: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
28: \renewcommand{\theequation}{\thesection.\arabic{equation}}
29:
30: %ENVIRONMENTS
31: \newtheorem{definition}[equation]{Definition}
32: \newtheorem{lemma}[equation]{Lemma}
33: \newtheorem{proposition}[equation]{Proposition}
34: \newtheorem{corollary}[equation]{Corollary}
35: \newtheorem{example}{Example}[section]
36: \newtheorem{conjecture}{Conjecture}[section]
37: \newtheorem{algorithm}{Algorithm}[section]
38: \newtheorem{theorem}[equation]{Theorem}
39: \newtheorem{exercise}[equation]{Exercise}
40: \newtheorem{assumption}[equation]{Assumption}
41:
42:
43: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
44: %GITTA'S DEFINITIONS
45: \def\le{\leq}
46: \def\ge{\geq}
47: \def\lt{<}
48: \def\gt{>}
49:
50: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
51: \newcommand{\Exercise}{{\bf Exercise}}
52: \newcommand{\req}[1]{(\ref{#1})}
53: \newcommand{\beps}{{\scriptscriptstyle{{ \cal E}}}}
54: \newcommand{\lip}{\langle}
55: \newcommand{\rip}{\rangle}
56: \newcommand{\lf}{\lfloor}
57: \newcommand{\lc}{\lceil}
58: \newcommand{\rc}{\rceil}
59: \newcommand{\rf}{\rfloor}
60: \newcommand{\supp}{{\rm supp\,}}
61: \newcommand{\ARn}
62: {\begin{array}[t]{c}
63: \longrightarrow \\[-0.3cm]
64: \scriptstyle {n\rightarrow \infty}
65: \end{array}}
66:
67:
68: \newcommand{\uu}{\underline}
69: \newcommand{\oo}{\overline}
70: \newcommand{\dfr}{\displaystyle\frac}
71: \newcommand{\La}{\Lambda}
72: \newcommand{\la}{\lambda}
73: \newcommand{\eps}{\varepsilon}
74: \newcommand{\om}{\omega}
75: \newcommand{\Om}{\Omega}
76: \newcommand{\Inv}{{\text{\rm Inv}\,}}
77: \newcommand{\crit}{{\rm crit}}
78:
79: %%%%%%%%%%%%%%%%%%% fonts %%%%%%%%%%%%%%%%%%%%%%%
80: \newcommand{\EE}{{\mathbb E}}
81: \newcommand{\NN}{{\mathbb N}}
82: \newcommand{\PP}{{\mathbb P}}
83: \newcommand{\QQ}{{\mathbb Q}}
84: \newcommand{\reals}{{\mathbb R}}
85: \newcommand{\ZZ}{{\mathbb Z}}
86:
87: \newcommand{\calA}{{\mathcal A}}
88: \newcommand{\calB}{{\mathcal B}}
89: \newcommand{\calF}{{\mathcal F}}
90: \newcommand{\calG}{{\mathcal G}}
91: \newcommand{\calI}{{\mathcal I}}
92: \newcommand{\calO}{{\mathcal O}}
93: \newcommand{\calQ}{{\mathcal Q}}
94: \newcommand{\calS}{{\mathcal S}}
95: \newcommand{\calV}{{\mathcal V}}
96: \newcommand{\calW}{{\mathcal W}}
97: \newcommand{\calY}{{\mathcal Y}}
98: \newcommand{\calX}{{\mathcal X}}
99:
100: \newcommand{\bfcdot}{{\boldsymbol \cdot}}
101:
102: \newcommand{\won}{{\boldsymbol 1}}
103: \newcommand{\hilbert}{\bigcirc\kern -0.8em
104: {\rm\scriptstyle {H}\;}}
105:
106: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
107: \newcommand{\limn}{\lim_{n \rightarrow \infty}}
108: \newcommand{\limk}{\lim_{k \rightarrow \infty}}
109: \newcommand{\limi}{\lim_{i \rightarrow \infty}}
110: \newcommand{\liml}{\lim_{\ell \rightarrow \infty}}
111: \newcommand{\limv}{\lim_{v \rightarrow \infty}}
112: \newcommand{\limm}{\lim_{m \rightarrow \infty}}
113: \newcommand{\limd}{\lim_{\delta \rightarrow \infty}}
114: \newcommand{\limsupn}{\limsup_{n \rightarrow \infty}}
115: \newcommand{\liminfn}{\liminf_{n \rightarrow \infty}}
116:
117: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
118: \newcommand{\proof}{\noindent \textbf{Proof:\ }}
119: \newcommand{\remark}{\noindent \textbf{Remark:\ }}
120: \newcommand{\remarks}{{\bf Remarks:}}
121:
122: \newcommand{\eff}{{\operatorname{eff}}}
123:
124: \newcommand{\dsum}{\displaystyle\sum}
125: \newcommand{\dprod}{\displaystyle\prod}
126:
127:
128: %fraction with round brackets
129: \newcommand{\ffrac}[2]
130: {\left( \frac{#1}{#2} \right)}
131:
132: \newcommand{\one}{\frac{1}{n}\:}
133: \newcommand{\half}{\frac{1}{2}\:}
134:
135:
136: %qed
137: \def\squarebox#1{\hbox to #1{\hfill\vbox to #1{\vfill}}}
138: \newcommand{\qed}{\hspace*{\fill}
139: \vbox{\hrule\hbox{\vrule\squarebox{.667em}\vrule}\hrule}\smallskip}
140: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
141:
142: %\frontmatter work in book class
143: %\mainmatter
144: %\backmatter
145:
146: \title{{Aging properties of Sinai's model of random walk in random environment}}
147: \author{Amir Dembo\\
148: Department of Statistics and\\
149: Department of Mathematics\\
150: Stanford University\\
151: Stanford, CA 94305, USA \and Alice Guionnet
152: \\
153: Ecole Normale Superieure de Lyon\\
154: Unite de Mathematiques pures et appliquees\\
155: UMR 5669, 46 Allee d'Italie \\
156: 69364 Lyon Cedex 07, France
157: \and
158: Ofer Zeitouni\\
159: Department of Electrical Engineering\\
160: Technion\\
161: Haifa 32000, Israel}
162:
163: \date{{\textit{May 26, 2001}}}
164: \maketitle
165:
166: %\newpage
167:
168: %\ls{1.5}
169: \ls{1}
170:
171: %\input{models}
172: %\input{ergodic}
173: %\input{ergodic-212}
174: %\input{CLT}
175: %\input{large}
176: \abstract{We study in this short note aging properties of Sinai's
177: (nearest neighbour) random walk in random environment.
178: With $\PP^o$ denoting the annealed law of the
179: RWRE $X_n$, our main result is a full proof of the
180: following statement due to
181: P. Le Doussal, C. Monthus and D. S. Fisher:
182: $$\lim_{\eta\to0} \lim_{n\to\infty} \PP^o
183: \left(\frac{|X_{n^h} - X_n|}{(\log n)^2} < \eta\right)
184: = \frac{1}{h^2} \left[ \frac{5}{3} - \frac{2}{3} e^{-(h-1)} \right].
185: $$
186: }
187: \section{Introduction: Sinai's model and non standard limit laws}
188: \setcounter{equation}{0}
189: \label{sec-sinai}
190:
191: We shall here study Sinai's random walk in random
192: environment, and focus on its aging properties. We begin
193: by recalling Sinai's model and it's long time behaviour.
194: This review is not meant to be exhaustive: see the recent
195: paper \cite{shi} for an updated account of Sinai's model which includes
196: many topics not covered here.
197:
198: Let
199: $N_z = \{\omega_z^-,\omega_z^0,\omega_z^+ \geq 0 :
200: \omega_z^- + \omega_z^0 +\omega_z^+ =1\}$ for $z \in \ZZ$,
201: and $\Omega = \prod_{z \in \ZZ} N_z$ equipped with the
202: product topology and the corresponding Borel $\sigma$-field
203: and shift operator. The {\it random
204: environment} is $\omega \in \Omega$ of
205: law $P \in M_1(\Omega)$, a Borel probability measure on $\Omega$.
206: For each $\omega \in \Omega$ we define (Sinai's) {\it random walk in
207: random environment} as the time-homogeneous Markov
208: chain $\{ X_n \}$ taking values in $\ZZ$ with transition
209: probabilities
210: $$
211: P_\omega(X_{n+1}=z+1|X_n=z) = \omega_z^+, \quad
212: P_\omega(X_{n+1}=z|X_n=z) = \omega_z^0, \quad
213: P_\omega(X_{n+1}=z-1|X_n=z) = \omega_z^- \;.
214: $$
215: We use $P_\omega^v$ to denote the law
216: induced on $\ZZ^\NN$ when $P_\omega^v(X_0=v)=1$,
217: refering to it as the {\it quenched} law of $\{X_n\}$.
218: Noting that $\omega \mapsto P_\omega^v(G)$ is Borel measurable
219: for any fixed Borel set $G \subset \ZZ^\NN$, we
220: define the {\it annealed} law of $\{ X_n \}$ as
221: $P^v(G)=\int_\Omega P_\omega^v(G) dP(\omega)$. Note that
222: under $P^v$ the random walk in random environment $\{ X_n \}$
223: is not a Markov chain!
224:
225: With $\rho_z=\omega_z^-/\omega_z^+$ and
226: $\oo{R}_k= k^{-1} \sum_{i=0}^{k-1} \log \rho_i$, we
227: assume throughout that the following holds.
228: \begin{assumption}
229: \label{ass-sinai}
230: The probability measure $P$ is stationary,
231: strongly mixing on $\Omega$ and such that
232: $P(\om_0^+ + \om_0^- > 0) = 1$,
233: $E_P \log \rho_0 = 0$, and there exists an $\varepsilon > 0 $ such that
234: $E_P|\log \rho_0|^{2+\varepsilon} < \infty$.
235: Further,
236: $\{\sqrt{k}\, \oo R_{[kt]}/\sigma_P\}_{t\in \reals}$ converges weakly to
237: a Brownian motion for some
238: $\sigma_P> 0$.
239: %(sufficient conditions for such convergence can be found e.g. in \cite{DU}).
240: %are as in Lemma~\ref{lem-Durret}).
241: \end{assumption}
242: (In the i.i.d. case, $\sigma_P^2=E_P (\log \rho_0)^2$).
243: Define
244: $$
245: W^n (t) = \frac{1}{\log\,n} \sum_{i=0}^{\lfloor (\log n)^2 t \rfloor}
246: \log \rho_i \cdot (\text{sign\ } t)
247: $$
248: with $t\in\reals$. By Assumption~\ref{ass-sinai},
249: $\{W^n(t)\}_{t\in\reals} $ converges weakly to $\{\sigma_P B_t\}$, where
250: $\{B_t\}$ is a two sided Brownian motion.
251:
252: Next, we call a triple $(a,b,c)$ with $a < b < c$ a valley of the path
253: $\{W^n (\cdot) \}$ if
254: \begin{align*}
255: W^n (b) & = \min_{a \le t \le c} W^n (t)\,,\\
256: W^n (a) & = \max_{a \le t \le b} W^n (t)\,,\\
257: W^n (c) & = \max_{b \le t \le c} W^n (t)\,.
258: \end{align*}
259: The \textsl{depth} of the valley is defined as
260: $$
261: d_{(a,b,c)} = \min (W^n(a) - W^n(b), W^n(c) - W^n(b))
262: \,.
263: $$
264: If $(a,b,c)$ is a valley, and $a<d< e < b$
265: are such that
266: $$
267: W^n (e) - W^n(d) = \max_{a\le x < y\le b} W^n(y) - W^n(x)
268: $$
269: then $(a,d,e)$ and $(e,b,c)$ are again valleys, which are obtained from
270: $(a,b,c)$ by a \textsl{left refinement}.
271: One defines similarly a
272: \textsl{right refinement}.
273: Define
274: \begin{align*}
275: c_0^n & = \min \{t\ge 0:\quad W^n(t) \ge 1\}\\
276: a_0^n & = \max \{t \le 0:\quad W^n (t) \ge 1\}\\
277: W^n(b_0^n) & = \min_{a_0^n\le t \le c_0^n} W^n(t)
278: \,.
279: \end{align*}
280: ($b_0^n$ is not uniquely defined, however, due to Assumption \ref{ass-sinai},
281: with $P$-probability approaching 1 as $n\to\infty$,
282: all candidates for $b_0^n$
283: are within distance converging to 0 as $n\to\infty$; we define $b_0^n$ then as
284: the smallest one in absolute value.)
285:
286: \begin{figure}[h]
287: \begin{center}
288: \includegraphics[scale=.6]{fig0.eps}
289: \caption{Left refinement of $(a,b,c)$}
290: \end{center}
291: \end{figure}
292:
293: One may now apply a (finite) sequence of refinements to find the
294: \textit{smallest} valley $(\oo{a}^n, \oo{b}^n, \oo{c}^n)$
295: with $\oo{a}^n < 0 < \oo{c}^n$, while
296: $d_{(\oo{a}^n, \oo{b}^n, \oo{c}^n)} \geq 1$.
297: We define similarly the smallest valey
298: $(\oo{a}^n_\delta, \oo{b}^n_\delta, \oo{c}^n_\delta)$
299: such that
300: $d_{(\oo{a}^n_\delta, \oo{b}^n_\delta, \oo{c}^n_\delta)} \geq 1+\delta$.
301: Let
302: \begin{equation*}
303: A_n^{J,\delta} =
304: \Biggl\{ \om\in \Omega:
305: \begin{array}[t]{c}\oo b^n=\oo b^n_\delta, \,
306: %\# \text{\ of needed refinements\ } \le J\,,
307: \text{any refinement $(a,b,c)$ of $(\oo{a}^n_\delta, \oo{b}^n_\delta,
308: \oo{c}^n_\delta)$ with $b\neq \oo{b}^n$
309: has depth } < 1-\delta\,,\\
310: \min_{t\in [\oo{a}^n, \oo{c}^n]\backslash [\oo{b}^n - \delta, \oo{b}^n +
311: \delta]}
312: W^n(t) - W^n (\oo b^n) > \delta^3\,,
313: |\oo{a}^n_\delta| + |\oo{c}^n_\delta| \le J
314: \end{array}
315: \Biggr\}
316: \end{equation*}
317: then it is easy to check by the properties of Brownian motion that
318: \begin{equation}
319: \label{sinai1}
320: \lim_{\delta \to 0}
321: \lim_{J\to\infty}
322: \lim_{n\to\infty}
323: P(A_n^{J,\delta}) = 1\,.
324: \end{equation}
325: The following theorem is due to
326: \cite{sinai}. For completeness, we include
327: a proof, which follows the approach
328: of \cite{golosov}, who dealt with a RWRE reflected at $0$,
329: i.e. with state space $\ZZ_+$.
330: \begin{theorem}
331: \label{the-sinai}
332: Assume $P(\min(\om_0, \om_0^+) < \eps) = 0$ and Assumption~\ref{ass-sinai}.
333: For any $\eta > 0$,
334: $$
335: \PP^o \left( \left| \frac{X_n}{(\log n)^2} - \oo{b}^n \right| > \eta
336: \right) \underset{n\to\infty}{\to}0
337: $$
338: \end{theorem}
339: \proof
340: Fix $\delta < \eta/2, J$ and $n_0$ large enough such that for all $n>n_0$,
341: $\om\in A_n^{J,\delta}$.
342: For simplicity of notations, assume in the sequel that
343: $\om$ is such that $\oo{b}^n > 0$.
344: Write $a^n=\oo a^n (\log n)^2, b^n=\oo b^n (\log n)^2,
345: c^n=\oo c^n (\log n)^2$, with similar notations for
346: $a^n_\delta,b^n_\delta,c^n_\delta$. Define
347: \begin{align*}
348: %\oo{B} & = \Bigl[\oo{a}^n (\log n)^2, \oo{c}^n (\log n)^2\Bigr]\\
349: \oo{T}_{b,n} & = \min \{ t \ge 0:
350: X_t = {b}^n \,
351: \text{or\ } X_t= {a}^n_\delta \}
352: \,.
353: \end{align*}
354:
355: We next recall that in dimension one, harmonic functions are readily
356: evaluated.
357: For $z \in [-m_-, m_+]$, define
358: $$
359: \calV_{m_-, m_+, \om} (z) :=
360: P_\om^z (\{X_n\} \text{\ hits\ $-m_-$ before hitting $m_+$)}
361: \,.
362: $$
363: %Note that due to the assumption $|\log \rho_z | < \infty$,
364: %or each $z$, it holds that
365: %$\calV_{m_-, m_+, \om} (z)$ is well defined as
366: %$$
367: %P_\om^z (\{X_m\} \text{\ never hits\ } [-m_-, m_+]^c) = 0\,.
368: %$$
369: The Markov property implies that
370: $\calV_{m_-, m_+, \om} (\cdot)$ is harmonic, that is it satisfies
371: \begin{equation}
372: \label{1}
373: \left\{
374: \begin{aligned}
375: (\om_z^++\om_z^-)\calV_{m_-, m_+, \om}(z) & = \om_z^- \calV_{m_-,m_+, \om} (z-1)\\
376: & \quad +\om_z^+ \calV_{m_-, m_+, \om} (z+1), \quad
377: z\in (-m_-, m_+) \\
378: \calV_{m_-,m_+, \om}(-m_-) & =1, \quad \calV_{m_-,m_+, \om}(m_+) =0
379: \end{aligned}
380: \right.
381: \end{equation}
382: Solving \req{1} (noting that the solution is unique due to
383: the maximum principle), we find
384: \begin{equation}
385: \label{2}
386: \calV_{m_-, m_+, \om} (z) =
387: \frac{\dsum_{i=z+1}^{m_+}\;
388: \dprod_{j=z+1}^{i-1} \rho_j}
389: {\dsum_{i=z+1}^{m_+} \; \dprod_{j=z+1}^{i-1} \rho_j + \dsum_{i=-m_-+1}^z
390: \left(\dprod_{j=i}^z \rho_j^{-1}\right)}
391: \end{equation}
392: Hence, for $\om\in A_n^{J,\delta}$,
393: %A:Just a precision
394: \begin{equation}
395: \label{sinai2}
396: P_\om^o \Bigl(X_{\oo{T}_{b,n}}
397: = \calV_{a_\delta^n, b^n_\delta, \om}(0)={a}^n_\delta \Bigr)
398: \le \frac{1}{1+ \frac{\exp\{
399: (\log n)(W^n(\oo a^n_\delta) - W^n(\oo b^n))\}}{J(\log n)^2}}
400: \le \frac{J(\log n)^2}{n^{1+\delta}}
401: \end{equation}
402: On the other hand, let $\tilde T_{b,n}$ have the law of
403: $\oo{T}_{b,n}$ except that the walk $\{X_\cdot\}$ is reflected at
404: ${a}^n_\delta $, and define similarly $\tilde \tau_1$.
405: %Using the same recursions as in (\ref{ergodic-oferadd1}),
406: Using the Markov property, c.f. \cite{GZrev} for a similar computation,
407: we have that
408: $$
409: E_\om^o (\tilde
410: \tau_1) = \frac{1}{\om_0^+} + \frac{\rho_0}{\om_{(-1)}^+} +
411: \cdots + \frac{ \prod_{i=0}^{ {a}^n_\delta + 2} \rho_{-i}}
412: {\om_{{a}^n_\delta-1}^+} + \prod_{i=0}^{{a}^n_\delta + 1} \rho_{-i}\,.
413: $$
414: Hence, with
415: $\tilde\om_i = \om_i$ for
416: $i\not={a}^n_\delta$ and $\tilde\om_{{a}^n_\delta}^+ =
417: 1$, for all $n$ large enough,
418: \begin{align*}
419: E_\om^o(\oo{T}_{b,n}) & \le E_\om^o (\tilde{T}_{b,n}) =
420: \sum_{i=1}^{{b}^n} \sum_{j=0}^{i-1-{a}^n_\delta}
421: \frac{\prod_{k=1}^j \rho_{i-k}}{\om_{(i-j-1)}^+} \\
422: & \le \frac{1}{\eps} \sum_{i=1}^{{b}^n} \sum_{j=0}^{i-1-{a}^n_\delta}
423: e^{(\log n) (W^n(i) - W^n(i-j))}
424: \le \frac{2J^2}{\eps} e^{\log n (1-\delta)} \le
425: n^{1- \frac{\delta}{2}}\,.
426: \end{align*}
427: We thus conclude with (\ref{sinai2})
428: %A:just to be more precise
429: that
430: $$
431: P_\om^o \Bigl( \oo{T}_{b,n} < n, \quad X_{\oo{T}_{b,n}} =
432: {b}^n\Bigr) \underset{n\to\infty}{\longrightarrow} 1
433: $$
434: implying that
435: \begin{equation}
436: \label{sinai3}
437: P_\om^o \Bigl( T_{{b}^n} < n \Bigr) \underset{n\to\infty}{\to} 1\,.
438: \end{equation}
439: Next note that another application of \req{2} yields
440: \begin{align}
441: \label{sinai4}
442: P_\om^{{b}^{n}{-1}} (X_\cdot\, \text{
443: hits ${b}^n$ before ${a}^n_\delta$})
444: & \ge 1- n^{-(1+\frac{\delta}{2})}\notag \\
445: P_\om^{{b}^{n}{+1}} (X_\cdot\, \text{
446: hits ${b}^n$ before ${c}^n_\delta$})
447: & \ge 1- n^{-(1+\frac{\delta}{2})}
448: \end{align}
449: On the same probability space,
450: construct a RWRE $\{\tilde{X}_t\}$ with the same transition mechanism
451: as
452: $\{X_t\}$ except that it is reflected at
453: ${a}^n_\delta$, i.e.\ replace
454: $\om$ by $\tilde \om$.
455: Then, using (\ref{sinai4}),
456: \begin{align*}
457: P_\om^o \left( \left| \frac{X_n}{(\log n)^2} -
458: \oo{b}^n \right| > \delta
459: \right)
460: & \le P_\om^o \Bigl(T_{{b}^n} > n \Bigr)
461: + \max_{t\le n} P_\om^{{b}^n} \left(\left| \frac{X_t}{(\log n)^2} -
462: \oo{b}^n \right| > \delta \right) \\
463: &\le P_\om^o \Bigl(T_{{b}^n} > n\Bigr) + \Bigl[
464: 1-(1-n^{-(1+\frac{\delta}{2})})^n\Bigr]
465: + \max_{t\le n} P_\om^{{b}^n} \left(\left| \frac{\tilde{X}_t}{(\log n)^2}
466: - \oo{b}^n \right| > \delta\right)
467: \end{align*}
468: Hence, in view of \req{sinai1} and \req{sinai3},
469: the theorem holds as soon as
470: we show that for $\om$ as considered here,
471: \begin{equation}
472: \label{sinai5}
473: \max_{t\le n} P_\om^{\oo{b}^n} \left(\left| \frac{\tilde{X}_t}{(\log n)^2}
474: - \oo{b}^n \right| > \delta\right)
475: \underset{n \to \infty}{\longrightarrow} 0
476: \end{equation}
477: To see (\ref{sinai5}), define
478: $$
479: f(z) = \frac{\prod_{{a}^n_\delta +1 \le i < z} \om_i^+}
480: {\prod_{{a}^n_\delta +1 \le i < z} \om_{i+1}^-} ,\quad
481: \oo{f}(z) = \frac{f(z)}{f({b}^n)}
482: $$
483: (as usual, the product over an empty set of indices is taken as 1.
484: $\oo{f}(\cdot)$
485: corresponds to the invariant measure for the resistor network
486: corresponding to $\tilde{X}_.$). Next, define the operator
487: \begin{equation}
488: \label{sinai6}
489: (Ag) (z) = \oo{\om}_{z-1}^{\,+} g(z-1) + \oo\om_{z+1}^{\,-}
490: g(z+1) + \oo\om_z^{\,0}g(z)
491: \end{equation}
492: where $\oo{\om}_z = \om_z$ for
493: $z> {a}^n_\delta$,
494: $\oo{\om}_{{a}^n_\delta}^+ = 1, \oo\om_{{a}^n_\delta-1}^+ = 0$.
495: Note that
496: $A\oo{f} = \oo{f}$, and further that
497: $$
498: P_\om^{{b}^n} (\tilde X_t = z) = A^t {\bf 1}_{b^n}(z)\,.
499: $$
500: Since $\oo{f}(z) \ge {\bf 1}_{b^n}(z)$ and $A$ is a positive operator,
501: we conclude that
502: $$P_\om^{{b}^n} (\tilde{X}_t = z) \le \oo{f}(z)\,.
503: $$
504: But, for $z$ with $|z/(\log n)^2 - \oo{b}^n | > \delta$
505: and $\om\in A^{J,\delta}_n$, it holds that
506: $\oo{f} (z) \le e^{-\delta^3 \log n}$, and hence
507: $$
508: P_\om^{{b}^n} (\tilde{X}_t=z) \le n^{-\delta^3}\,.
509: $$
510: Thus,
511: $$
512: \max_{t\le n} P_\om^{\oo{b}^n}
513: \left(\left| \frac{\tilde{X}_t}{(\log n)^2} - \oo{b}^n
514: \right| > \delta \right) \le n^{-\delta^3}\,,
515: $$
516: yielding (\ref{sinai5}) and completing the proof of the theorem.
517: \qed
518:
519: We next turn to a somewhat more detailed study of the random variable
520: $\oo{b}^n$. By replacing $1$ with $t$ in the definition of $\oo{b}^n$, one
521: obtains a process
522: $\{\oo{b}^n(t)\}_{t\ge 0}$. Further, due to Assumption~\ref{ass-sinai},
523: the process $\{\oo{b}^n (t/\sigma_P)\}_{t\ge 0}$
524: converges weakly to a process
525: $\{\oo{b}(t)\}_{t\ge 0}$, defined in terms of the Brownian motion
526: $\{B_t\}_{t\ge 0}$; Indeed, $\oo{b}(t)$ is the location of the bottom of the
527: smallest valley of
528: $\{B_t\}_{t\ge 0}$, which surrounds $0$ and has depth $t$. Throughout
529: this section we denote by $\calQ$ the law of the
530: Brownian motion $B_\cdot$. Our next goal is
531: to characterize the process
532: $\{\oo{b}(t)\}_{t\ge0}$. Toward this end,
533: define
534: \begin{align*}
535: m_+ (t) &= \min\{B_s: 0 \le s \le t\}\,,
536: \quad
537: m_- (t) = \min\{B_{-s}: 0 \le s \le t\} \\
538: T_+ (a) & = \inf \{s\ge 0: B_s-m_+(s) = a\}\,,
539: \quad
540: T_- (a) = \inf \{s\ge 0: B_{-s}-m_-(s) = a\}\\
541: s_\pm (a) & = \inf\{s\ge 0: m_\pm (T_\pm(a)) = B_{\pm s} \} \,,
542: \quad
543: M_\pm (a) =
544: \sup \{ B_{\pm \eta}: \quad 0\le \eta \le s_\pm (a) \} \,.
545: \end{align*}
546: Next, define $W_\pm(a) = B_{s_\pm (a)} $.
547: It is not hard to check that the pairs
548: $(M_+(\cdot), W_+ (\cdot))$ and $(M_-(\cdot), W_-(\cdot))$ form independent
549: Markov processes.
550: Define finally
551: $$
552: H_\pm(a) = (W_\pm (a) + a) \vee M_\pm (a)\,.
553: $$
554:
555: \begin{figure}[h]
556: \begin{center}
557: \includegraphics[scale=.6]{fig1.eps}
558: \caption{The random variables $(M_+(a), W_+ (a), s_+(a))$}
559: \end{center}
560: \end{figure}
561:
562: The following characterization of the law of $\oo{b}(a)$
563: is due to \cite{kesten1}. This characterization can be found
564: also in \cite{golosov2}.
565: \begin{theorem}
566: \label{the-sinai2}
567: For each $a>0$, $\calQ(\oo{b}(a) \in \{s_+(a), -s_-(a)\})=1$.
568: Further, $\oo{b}(a) = s_+(a)$ iff
569: $H_+(a) < H_- (a)$.
570: \end{theorem}
571:
572: \proof
573: Note that $\calQ(H_+ (a) = H_-(a))=0$.
574: That $\oo{b}(a) \in \{s_+(a), -s_-(a)\}$ is a direct consequence of the
575: definitions, i.e.\ assuming $\oo{b}(a) > 0$ and $\oo{b} (a)
576: \neq s_+(a)$ it is easy to show
577: that one may refine from the right
578: the valley defining $\oo{b}(a)$, contradicting
579: minimality. We begin by showing, after Kesten~\cite{kesten1}, that
580: $\oo{b}(a) = s_+(a)$ iff either
581: \begin{equation}
582: \label{sinai7}
583: W_-(a) > W_+ (a), \quad M_+(a) < (W_-(a) + a) \vee M_-(a)
584: \end{equation}
585: or
586: \begin{equation}
587: \label{sinai8}
588: W_-(a) < W_+ (a), \quad M_-(a) > (W_+(a) + a) \vee M_+(a)
589: \,.
590: \end{equation}
591: Indeed, assume $\oo{b} (a) = s_+(a)$, and
592: $W_-(a) > W_+ (a)$.
593: Let $(\alpha, \oo{b}(a),\gamma)$ denote the minimal valley defining
594: $\oo{b}(a)$.
595: If $-s_-(a) \le \alpha$, then
596: \begin{equation}
597: \label{sinai9}
598: M_-(a) = \max \{B_{-s}: \; s\in (0,s_-(a))\} \ge B_{-\alpha}
599: = \max \{B_{s}: -\alpha \le s \le \oo{b} (a) \} \ge M_+(a)
600: \end{equation}
601: implying (\ref{sinai7}). On the other hand, if
602: $-s_-(a) > \alpha$, refine $(\alpha, \oo{b} (a), \gamma)$ on the left
603: (find $\alpha', \beta'$ with $\alpha<\alpha' < \beta' < \oo{b} (a)$),
604: such that
605: $$
606: B_{\beta'} - B_{\alpha'} = \max_{\alpha < x< y < \oo{b} (\alpha)}
607: (B_y - B_x) \ge M_+(a) - W_-(a)
608: $$
609: and thus minimality of $(\alpha, \oo{b}(a), \gamma)$ implies that
610: $M_+(a) - W_-(a) < a$, implying (\ref{sinai7}).
611:
612: \begin{figure}[h]
613: \begin{center}
614: \includegraphics[scale=.5]{fig2.eps}
615: \qquad \quad
616: \includegraphics[scale=.5]{fig3.eps}
617: \caption{$\oo{b}(a) = s_+(a)$}
618: \end{center}
619: \end{figure}
620:
621:
622: We thus showed that if $\oo{b}(a) = s_+(a)$ and
623: $W_-(a) > W_+(a)$ then (\ref{sinai7}) holds. On the other hand, if
624: (\ref{sinai7}) holds, we show that $\oo{b}(a)=s_+(a)$ by considering the cases
625: $\alpha \le -s_-(a)$ and $-s_-(a) < \alpha$ separately.
626: In the former case,
627: necessarily
628: $\gamma > s_+$, for otherwise $M_- (\alpha) \le B_\gamma \le M_+ (a) \le
629: W_-(a)+a$
630: which together with $\oo{b}(a)= -s_-(a)$ would imply that the depth of
631: $(\alpha, \oo{b} (a), \gamma)$ is smaller than $a$.
632: Thus, under (\ref{sinai7}) if
633: $\alpha \le -s_-(a)$ then $\gamma > s_+$, and in this case
634: $\oo{b}(a) = s_+(a)$ since $B_{s_+(a)} < B_{-s_-(a)}$. Finally, if
635: $\alpha> - s_-(a)$ then $\oo{b}(a) \not= -s_-(a)$
636: and hence $\oo{b}(a) = s_+(a)$.
637:
638: Hence, we showed that if $W_-(a) > W_+(a)$
639: then (\ref{sinai7}) is equivalent to
640: $\oo{b}(a) = s_+(a)$. Interchanging the positive and negative axis, we
641: conclude that if
642: $W_-(a) < W_+(a)$, then $\oo{b}(a)= -s_-(a)$ iff
643: $M_+(a) < (W_+(a) + 1) \vee M_+(a)$. This completes the proof that
644: $\oo{b}(a) = s_+(a)$ is equivalent to (\ref{sinai7}) or (\ref{sinai8}).
645:
646: To complete the proof of the theorem, assume first
647: $W_-(a) > W_+(a)$. Then, $\oo{b}(a) = s_+(a)$ iff (\ref{sinai7}) holds,
648: i.e.\ $M_+ (a) < (W_- (a) + a) \vee M_- (a) = H_-(a)$.
649: But $H_-(a) \ge W_- (a) + a \ge W_+ (a) + a$, and hence
650: $M_+(a) < H_-(a)$ is equivalent to $M_+(a) \vee (W_+(a) + a) < H_- (a)$,
651: i.e.\ $H_+(a) < H_- (a)$.
652: The case $W_+(a) < W_-(a)$ is handled similarly by using (\ref{sinai8}).
653: \qed
654:
655: \section{Aging properties for Sinai's model}
656: One may use the representation in Theorem~\ref{the-sinai2} in order to
657: evaluate explicitly the law of $\oo{b}(a)$ (note that $\oo{b}(a)
658: \overset{{\cal L}}{=} a \oo{b} (1)$ by Brownian scaling).
659: This is done in \cite{kesten1},
660: and we do not repeat the construction here.
661: Our goal is to use Theorem~\ref{the-sinai2} to show that Sinai's model
662: exhibits \textsl{aging} properties. More precisely, we claim that
663: \begin{theorem}
664: \label{the-sinai3}
665: Assume $P(\min(\om_0, \om_0^+) < \eps) = 0$
666: and Assumption~\ref{ass-sinai}.
667: Then, for $h>1$,
668: \begin{equation}
669: \label{sinai10}
670: \lim_{\eta\to0} \lim_{n\to\infty} \PP^o
671: \left(\frac{|X_{n^h} - X_n|}{(\log n)^2} < \eta\right)
672: = \frac{1}{h^2} \left[ \frac{5}{3} - \frac{2}{3} e^{-(h-1)} \right]
673: \end{equation}
674: \end{theorem}
675: Theorem \ref{the-sinai3} was derived heuristically
676: in \cite{fisher}, to which we refer for additional aging properties and
677: discussion. The right hand side of
678: formula \req{sinai10} appears also in \cite{golosov2}, in a slightly
679: different context.
680:
681:
682:
683: \proof
684: Applying Theorem~\ref{the-sinai}, the limit in the left hand side of
685: (\ref{sinai10}) equals
686: $$
687: \calQ\Bigl(\oo{b} (h) = \oo{b}(1)\Bigr)
688: = 2\calQ \Bigl(\oo{b} (h) = \oo{b}(1) = s_+ (1) = s_+
689: (h)\Bigr)
690: $$
691: Note that
692: $$
693: \calQ(s_+ (h) = s_+(1)) = \calQ \left(
694: \begin{array}[t]{c}
695: \text{Brownian motion, started at height $1$,}\\
696: \text{hits $h$ before hitting $0$}
697: \end{array}
698: \right) = \frac{1}{h}\,.
699: $$
700: Hence, using that on $s_+(1) = s_+(h)$ one has
701: $W_+(1) = W_+(h), M_+(1) = M_+(h)$,
702: and using that the event $\{s_+(h)=s_+(1)\}$ depends only on increments
703: of the path of the Brownian motion after time $T_+(1)$,
704: one gets
705: \begin{equation}
706: \label{sinai10a}
707: \calQ\Bigl(\oo{b}(h) = \oo{b} (1) \Bigr)
708: = \frac{2}{h} \calQ\Bigl( H_+ (1) < H_-(1), (W_+ (1) + h) \vee
709: M_+ (1) < H_-(h)\Bigr)\,.
710: \end{equation}
711: Next, let
712: \begin{align*}
713: \tau_0 & = \min\{t>s_-(1): B_{-t} = W_-(1) + 1\} \\
714: \tau_h & = \min \{t>\tau_0: \; B_{-t}= W_-(1) + h \,\text{\ or \ }
715: B_t = W_-(1)\}\,.
716: \end{align*}
717: Note that $\tau_h-\tau_0$ has the same law as that of the hitting time of
718: $\{0,h\}$ by a Brownian motion $Z_t$ with $Z_0=1$. (Here,
719: $Z_t = B_{-(\tau_0+t)} - W_-(1) !$). Further, letting
720: $I_h = \won_{\{B_{\tau_h} = W_-(1)\}}
721: (= \won_{\{Z_{\tau_h-\tau_0}=0\}})$,
722: it holds that
723: \begin{align*}
724: W_-(h) & = W_-(1) + I_h \tilde{W}_- (h) \\
725: M_-(h) & = \begin{cases}
726: M_-(1), & I_h=0\\
727: M_-(1) \vee (\oo{M}_- (h) + W_-(1) + 1) \vee (\tilde M_-(h) + W_-(1)),
728: & I_h=1
729: \end{cases}
730: \end{align*}
731: where $(\tilde{W}_-(h), \tilde{M}_-(h))$ are independent of
732: $(W_-(1), M_-(1))$ and
733: possess the same law as $(W_-(h),$
734: $M_-(h))$, while $\oo{M}_-(h)$
735: is independent of
736: both $(W_- (1), M_-(1))$ and $(\tilde{W}_-(h), \tilde{M}_-(h))$
737: and has the law of the maximum of a Brownian motion, started
738: at $0$, killed at hitting $-1$ and conditioned not to hit $h-1$.
739: %$M_-(h)$.
740: (See figure 4 %A : problem with the numbering ~\ref{sinai-fig4}
741: for a graphical description of these random variables.)
742:
743: \begin{figure}[h]
744: \label{sinai-fig4}
745: \begin{minipage}[b]{0.5\linewidth}
746: \centering
747: \includegraphics[scale=.5]{fig4a.eps}
748:
749: {$I_h=0$}
750: \end{minipage}
751: \begin{minipage}[b]{0.5\linewidth}
752: \centering
753: \includegraphics[scale=.5]{fig4b.eps}
754:
755: {$I_h=1$}
756: \end{minipage}
757: \caption{Definition of auxiliary variables}
758: \end{figure}
759:
760: Set now
761: $$\hat{M}_-(h) =
762: \begin{cases}
763: h, & I_h = 0\\
764: 1+\oo{M}_-(h), & I_h=1\,,
765: \end{cases}
766: $$
767: $\tilde H_-(h)=(\tilde W_-(h)+h)\vee \tilde M_-(h)$
768: and $\Gamma(h) = \max (\tilde H_-(h), \hat M_-(h))$.
769: Note that $\tilde H_-(h)$ has the same law as $H_-(h)$ but is
770: independent of $\oo{M}_-(h)$.
771: Further, it is easy to check that
772: $(W_-(h)+h) \vee M_- (h) = (W_-(1) + \Gamma_h) \vee M_-(1)$
773: (note that either $M_-(h) = M_-(1)$ or
774: $M_-(h) > M_-(1)$ but in the latter case,
775: $M_-(h) \le W_-(1) + \Gamma(h)$.)
776: We have the following lemma, whose proof is deferred:
777: \begin{lemma}
778: \label{lem-sinai}
779: The law of
780: $\Gamma_h$ is $\frac{1}{h} \delta_h + \frac{h-1}{h} U [1,h]$, where
781: $U[1,h]$ denotes the uniform law on $[1,h]$.
782: \end{lemma}
783: Substituting in (\ref{sinai10a}), we get that
784: \begin{equation}
785: \label{sinai11}
786: \calQ
787: (\oo{b}(h) = \oo{b}(1)) =
788: E_{\calQ}\left(E_{\calQ}(\oo{b}(h)=\oo{b}(1)|\Gamma(h))\right)=
789: \frac{2}{h^2} \left[ \int_1^h Q(t) dt + Q(h)\right]
790: \end{equation}
791: where
792: $$
793: Q(t) = \calQ(H_+(1) < H_-(1), H_+ (h) < H_-(t))\,.
794: $$
795: In order to evaluate the integral in (\ref{sinai11}),
796: we need to evaluate the
797: joint law of $(H_+(1), H_+(t))$
798: (the joint law of $(H_-(1),H_-(t))$ being identical).
799: %being denoted by $B_1$ the segment $\{0\le x=y\le 1\}$ and by
800: %$B_\alpha$ the segment
801: %$\{t-1 \le y= x+t -1 \le t\}$.
802: Since $0\le H_+(1) \le 1$ and $H_+(1) \le H_+(t) \le H_+(1) + t-1$,
803: the support of the law of $(H_+(1), H_+(t))$ is the domain A defined by
804: $0 \le x \le 1$, $x \le y \le x+t-1$. Note that for
805: $(z,w) \in A$,
806: \begin{align*}
807: \calQ
808: (H_+(1) \le z, H_+(t) \le w) &=
809: \calQ(M_+(1) \le z \wedge w, W_+ (1) \le - [(1-z) \vee (t-w)]\\
810: & = \calQ\Bigl(M_+(1) \le z, W_+(1) \le - (t-w)\Bigr)
811: \,.
812: \end{align*}
813: We now have the following well know lemma. For completeness,
814: the proof is given at the end of this section:
815: \begin{lemma}
816: \label{lem-sinai2}
817: For $z+y \ge 1$, $0 \le z \le 1$, $y \ge 0$,
818: $$
819: \calQ(M_+(1) \le z, W_+(1) \le -y) = z e^{-(z+y-1)}\,.
820: $$
821: \end{lemma}
822: Lemma~\ref{lem-sinai2} implies that, for $(z,w) \in A$, $t>1$,
823: \begin{equation}
824: \label{sinai12}
825: \calQ(H_+(1) \le z, H_+(t) \le w) = z e^{-(z+t-w - 1)}\,.
826: \end{equation}
827: Denote by $B_1$ the segment $\{0\le x=y\le 1\}$ and by
828: $B_2$ the segment
829: $\{t-1 \le y= x+t -1 \le t\}$.
830: We conclude, after some tedious computations,
831: that the law of $(H_+(1), H_+(t))$:
832: \begin{itemize}
833: \item
834: possesses the density
835: $
836: f(z,\om) = (1-z) e^{-z} e^{-w-(t-1)},
837: \quad(z, w) \in A \backslash (B_1\cup B_2)
838: $
839: \item
840: possesses the density
841: $\tilde f(z,y) = (1-z) e^{-(t-1)}, \quad z=w \in B_1$
842: \item
843: possesses the density
844: $\oo{f} (z,z+t-1) = z, \quad w=z+t-1 \in B_2$.
845: \end{itemize}
846: Substituting in the expression for $Q(t)$, we find that
847: $$
848: Q(t) = \frac{5}{12} e^{-(h-t)} + \frac{1}{12} e^{-(h+t-2)}
849: \,.
850: $$
851: Substituting in (\ref{sinai12}), the theorem follows.
852: \qed
853:
854: \noindent
855: {\bf Proof of Lemma \ref{lem-sinai}:}
856: Note that $\calQ(I_h=0)=1/h$, and in this case $\Gamma_h=h$.
857: Thus, we only need to consider the case where $I_h=1$ and show that
858: under this conditioning,
859: $\max(H_-(h), 1+\oo{M}_-(h))$ possesses the law $U[1,h]$.
860: Note that by standard properties of Brownian motion,
861: $$\calQ(\hat M_-(h)\leq \xi|I_h=1)=\frac{\frac{\xi-1}{\xi}}{
862: \frac{h-1}{h}}.$$
863: We show below that the law of $\tilde H_-(h)$, which is
864: identical to the law
865: of $H_-(h)$, is uniform on $[0,h]$. Thus, using independence,
866: for $\xi\in [1,h]$,
867: $$ \calQ(\Gamma_h<\xi|I_h=1)=
868: \frac{h(\xi-1) \xi}{\xi (h-1) h}=
869: \frac{\xi-1}{h-1}\,,$$
870: i.e. the law of $\Gamma_h$ conditioned on $I_h=1$ is indeed $U[1,h]$.
871:
872: It thus only remains to evaluate the law of $H_-(h)$. By
873: Brownian scaling, the law of $H_-(h)$ is identical to the law of $hH_+(1)$,
874: so we only need to show that the law of $H_+(1)$ is uniform on $[0,1]$.
875: This in fact is a direct consequence of Lemma \ref{lem-sinai2}.
876: \qed
877:
878: \noindent
879: {\bf Proof of Lemma \ref{lem-sinai2}: \ }
880: Let $\calQ^x$ denote the law of a Brownian motion
881: $\{Z_t\}$ starting at time $0$
882: at $x$. The Markov property now yields
883: \begin{eqnarray}
884: \label{sinaifinal}
885: \calQ(M_+(1) \le z, W_+(1) \le -y) &=&
886: \calQ^o(\{Z_\cdot\} \,
887: \mbox{\rm hits $z-1$ before hitting $z$})
888: \calQ^{z-1}(M_+(1) \le z, W_+(1) \le -y)
889: \nonumber \\
890: &=&
891: z \calQ^o(M_+(1) \le 1, W_+(1) \le -y-z+1)\nonumber \\
892: & =& z \calQ^o(W_+(1)\leq -(y+z-1))\,.
893: \end{eqnarray}
894: For $x\geq 0$, let $f(x):=\calQ(W_+(1)\leq -x)$. The Markov
895: property now implies
896: $$f(x+\epsilon)=f(x) \calQ^{-x}(W_+(1)\leq -(x+\epsilon))=
897: f(x) f(\epsilon)\,.$$
898: %It follows that $d(\log f(x))/dx=-1$, implying together with
899: Since $f(0)=1$ and $f(\epsilon)=1-\epsilon+o(\epsilon)$,
900: it follows that $f(x)=e^{-x}$. Substituting in
901: \req{sinaifinal}, the lemma follows.
902: \qed
903:
904:
905:
906:
907:
908: %\appendix
909: %\input{reminder}
910:
911: %\input{bibliography}
912: %version of April 16
913: \begin{thebibliography}{99}
914: %\bibitem{De-Ze} A. Dembo and O. Zeitouni,
915: %{\it Large deviation techniques and applications}, 2nd edition,
916: %Springer, New-York (1998).
917: %\bibitem{Deu-St} J. D. Deuschel and D. W. Stroock,
918: %{\it Large deviations},
919: %Academic Press, Boston (1989).
920: \bibitem{fisher}
921: P. Le Doussal, C. Monthus and D. S. Fisher,
922: Random walkers in one-dimensional random environments:
923: exact renormalization group analysis.
924: {\it Phys. Rev. E} {\bf 59} (1999) pp. 4795--4840.
925: %\bibitem{DS} P. G. Doyle and L. Snell, {\it Random walks and electric
926: %networks}, Carus Math. Monographs {\bf 22}, MAA, Washington (1984).
927: %\bibitem{DU}
928: %R. Durrett, {\it Probability: theory and examples}, 2nd ed.,
929: %Duxbury Press, Belmont (1996).
930: \bibitem{GZrev} N. Gantert and O. Zeitouni, Large deviations for one
931: dimensional random walk in a random environment - a survey, {\it
932: Bolyai Society Math Studies} {\bf 9} (1999) pp. 127--165.
933: \bibitem{golosov} A. O. Golosov,
934: Limit distributions for random walks in random environments,
935: {\it Soviet Math. Dokl.} {\bf 28} (1983) pp. 18--22.
936: % \bibitem{golosov1} A. O. Golosov,
937: % Localization of
938: %random walks in one-dimensional random environments,
939: % {\it Comm. Math. Phys.} {\bf 92} (1984) pp. 491--506.
940: \bibitem{golosov2} A. O. Golosov,
941: {\it Comm. Moscow Math. Soc.} {\bf 199} (1985) pp. 199--200.
942: \bibitem{kesten1}
943: H. Kesten,
944: The limit distribution of Sinai's random walk
945: in random environment,
946: {\it Physica } {\bf 138A}
947: (1986) pp.~299--309.
948: %\bibitem{KKZ}
949: %H. Kesten, M. V. Kozlov and F. Spitzer, A limit law for
950: %random walk in a random environment, {\it Comp. Math.} {\bf 30} (1975)
951: %pp. 145--168.
952: \bibitem{shi} Z. Shi, Sinai's walk via stochastic calculus,
953: preprint (2001).
954: \bibitem{sinai} Ya. G. Sinai, The limiting behavior of a one-dimensional
955: random walk in random environment, {\it Theor. Prob. and Appl.}
956: {\bf 27} (1982) pp. 256--268.
957: %\bibitem{solomon}
958: %F. Solomon, Random walks in random environments, {\it Annals Probab.}
959: %{\bf 3} (1975) pp. 1--31.
960: \end{thebibliography}
961: \end{document}
962:
963:
964:
965:
966:
967: