1: \documentclass[aps,pre,showpacs,twocolumn,superscriptaddress]{revtex4}
2: \usepackage{epsfig}
3: \usepackage{graphicx}
4: \usepackage{amsmath}
5: \usepackage{amssymb}
6: \usepackage{verbatim} % useful for program listings
7: \usepackage{color} % use if color is used in text
8: \usepackage{subfigure} % use for side-by-side figures
9:
10: %\bibstyle{apsrev.bib}
11:
12: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
13: \def\ARTICLETYPE{{paper}}
14: \def\WIDTHA{0.46\textwidth} %% 1 FIGURE WIDTH / two-col
15: \def\WIDTHB{0.42\textwidth} %% FIGURES WIDTH / two-col
16: \def\WIDTHC{0.36\textwidth} %% FIGURES WIDTH / two-col
17: \def\WIDTHC{0.30\textwidth} %% FIGURES WIDTH / two-col
18: %\def\WIDTHB{0.8\textwidth} %% FIGURES WIDTH / PREPRINT
19: %\def\WIDTHA{0.8\textwidth} %% 1 FIGURE WIDTH / PREPRINT
20: \newcommand{\beq}{\begin{equation}}
21: \newcommand{\eeq}{\end{equation}}
22: \newcommand{\jacob}[1]{{\huge{$\bullet$}}{\em #1}}
23: \newcommand{\maya}[1]{{\large{$\clubsuit$}}{\em #1}}
24: \newcommand{\peter}[1]{{\large{$\heartsuit$}}{\em #1}}
25: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
26: \begin{document}
27: \title{Reinforced walks in two and three dimensions}
28: \date{\today}
29: \author{Jacob G. Foster} \affiliation{Complexity Science Group, Department of Physics and
30: Astronomy, University of Calgary, Calgary, Canada}
31: \author{Peter Grassberger} \affiliation{Complexity Science Group, Department of Physics and
32: Astronomy, University of Calgary, Calgary, Canada}
33: \affiliation{Institute for Biocomplexity and Bioinformatics,
34: University of Calgary, Calgary, Canada }
35: \author{Maya Paczuski} \affiliation{Complexity Science Group, Department of Physics and
36: Astronomy, University of Calgary, Calgary, Canada}
37:
38:
39: \begin{abstract}
40: In probability theory, reinforced walks are random walks on a lattice (or more generally a graph) that preferentially revisit neighboring `locations' (sites or bonds) that have been visited before. In this paper, we consider walks with one-step reinforcement, where one preferentially \emph{revisits} locations irrespective of the number of visits. Previous numerical simulations [A. Ordemann {\it et al.}, Phys. Rev. E {\bf 64}, 046117 (2001)] suggested that the site model on the lattice shows a phase transition at finite reinforcement between a random-walk like and a collapsed phase, in both 2 and 3 dimensions. The very different mathematical structure of bond and site models might also suggest different phenomenology (critical properties, etc.). We use high statistics simulations and heuristic arguments to suggest that site and bond reinforcement are in the same universality class, and that the purported phase transition in 2 dimensions actually occurs at zero coupling constant. We also show that a quasi-static approximation predicts the large time scaling of the end-to-end distance in the collapsed phase of both site and bond reinforcement models, in excellent agreement with simulation results.\end{abstract}
41:
42: \maketitle
43:
44:
45: \section{Introduction}
46:
47: \subsection{Background: Physics}
48: Random walks with memory have a large number of applications in physics and other sciences.
49: Many variants have thus been studied in different contexts. The best known example is
50: presumably the self-avoiding walk \cite{deGennes}, which models the large scale behavior of flexible
51: chain polymers in good solvents. As pointed out by Amit {\it et al.} \cite{Amit} the name
52: `self-avoiding walk' is something of a misnomer, since this model describes either static self-avoiding
53: {\it chains} or self-{\it killing} walks. When the self-avoiding walker tries to revisit
54: a site it has visited before, it is not gently turned towards another neighboring site; it is killed. In a more general version of this model, walkers carry an initial weight of unity, which decreases by a fixed factor whenever a site is revisited (the Domb-Joyce model \cite{domb}). If a site $i$ had been visited $n$ times before, the weight is diminished at the $n+1$ visit by $e^{nu}$ with $u<0$.
55:
56: When the sign of the interaction is changed to $u>0$, so that the weight is multiplied
57: by a factor $e^{nu}>1$ at each revisit, the resulting self-attracting walk degenerates in any
58: finite dimension; for large times, the walker just oscillates between two sites. This extreme
59: behavior is avoided if the weight change is independent of $n$ and one distinguishes only
60: between sites which have and have not been visited before. This model is related to the
61: Donsker-Varadhan \cite{donsker} ``Wiener sausage" problem \cite{mehra} and leads to
62: a power law scaling $R_t \sim t^{1/(d+2)}$ for the end-to-end distance after
63: $t$ time steps in $d$ dimensions of space.
64:
65: In contrast to these ``static" models, where instances are weighted and the weights are modified by
66: interactions, one can define ``dynamic" models where the walks are biased by the interaction. The oldest such model is the {\it true self-avoiding walk} (TSAW) of Amit {\it et al.} \cite{Amit}. Assume
67: that at time $t$ the walker is at site $i$, and that the number of previous visits to any of the
68: ${\cal N}$ neighbors is $n_j,\; j=1,...,{\cal N}$. Then the probability to step to neighbor $j$ at
69: the next time is
70: \beq
71: p_j = \frac{e^{n_ju}}{\sum_{j'=1}^{\cal N} e^{n_{j'}u}}\quad u < 0\;. \label{TSAW}
72: \eeq
73: This is a much milder modification than the original self-avoiding walk. Accordingly, the
74: r.m.s. end-to-end distance scales as $R^2_t \sim t$ for $d>2$, while there are logarithmic
75: corrections at the `upper critical dimension' $d=d_c=2$. In contrast, the upper critical dimension
76: for the self-avoiding walk is $d_c=4$, and $R^2_t \sim t^{2\nu_d}$ for $d<d_c$ with $\nu_d<1/2$
77: \cite{deGennes}.
78:
79: \subsection{True self-attracting walks}
80:
81: When the sign of $u$ is switched to positive, the resulting ``true self-attracting walks"
82: (TSATWs) are also closer to random walks than the ordinary self-attracting walks. It seems that the
83: behavior of the TSATW with $p_j$ given by Eq.(\ref{TSAW}) but with $u>0$ is unknown. On the other
84: hand, there are several numerical studies of the TSATW with one-step reinforcement
85: \cite{sapo,reis,prasad,lee,ordemann1,ordemann2}
86: \beq
87: p_j = \frac{e^{\kappa_ju}}{\sum_{j'=1}^{\cal N} e^{\kappa_{j'}u}}\quad u > 0\;. \label{TSAW1}
88: \eeq
89: where $\kappa_j=0$ if the site $j$ has never been visited before, and $\kappa_j=1$ otherwise. By far
90: the most extensive studies were those of \cite{ordemann1,ordemann2}, which claimed that one-step reinforcement TSATWs on the lattice showed a non-trivial phase transition in both $d=2$ and $d=3$, with $u_c(d=2) = 0.88\pm 0.05$ and $u_c(d=3) = 1.92\pm 0.03$. In both cases, the behavior of $R^2_t$ is supposed to change at $u_c$ from $R^2_t\sim t$ at $u<u_c$ to
91: \beq
92: R^2_t\sim t^{2/(d+1)} \label{R2}
93: \eeq
94: at $u>u_c$. Hence the phase transition is between a random-walk-like and a collapsed phase. At the critical point, $R_t$ scales with a new exponent $\nu_c$ which is $0.40\pm 0.01$ in $d = 2$ and $0.303\pm 0.005$ in $d = 3$ \cite{ordemann2}. These phase transitions are also seen in the average number $\langle S_t \rangle$ of sites visited up to time $t$. This scales as $\langle S_t\rangle \sim t$ for $u<u_c$ (with a logarithmic correction for $d = 2$), but as $t^k$ with
95: \beq
96: k=d/(d+1)
97: \eeq
98: for $u>u_c$. The latter was derived from a quasi-static approximation \cite{dalmaroni} in
99: Ref.~\cite{sapo}. The quasi-static approximation seems to be satisfied to high precision (see below). At $u=u_c$, Ordemann {\it et al.} found $k_c = 0.80\pm 0.01$ ($d = 2$) resp. $0.91\pm 0.01$ ($d = 3$) \cite{ordemann2}.
100:
101: \subsection{Background: Mathematics}
102:
103: In a parallel and largely independent development, these and similar random walks with
104: memory have been extensively studied in the probability theory literature. For a recent survey, see
105: \cite{pemantle}. The rigorous mathematical study of reinforced walks displays much more breadth than the rather limited study of one-step site reinforcement in the statistical physics literature. In contrast to the physics literature, which focuses on the site model, {\it bond} or ``edge" reinforced random walks (ERRW) have been studied in great detail and with multiple reinforcement as in Eq.(\ref{TSAW}) with positive $u$. Such walks (most clearly on trees) are closely related to P\'{o}lya urn processes and similar problems with reinforcement that can be solved exactly (note that walks with bond reinforcement are often called `trails' in the physics literature \cite{trails}). For the models with multiple \emph{site} reinforcement discussed above (called vertex-reinforced random walks or VRRW), the related urn process is Friedman-like and less tractable \cite{pemantle, pemantle1}, see endnote \footnote{We summarize the difference between these two urn processes following \cite{pemantle1}. In the simplest P\'{o}lya urn, one begins with a single red and black ball. In a series of draws, one removes a ball at random and returns it to the urn with another of the same type. The fraction of red balls, surprisingly, converges to a random limit uniformly distributed on the closed interval $[0,1]$. In the Friedman urn, one reinforces both the drawn color and the other color. More surprisingly, the limit here converges to {1/2} even if the reinforcements are extremely biased towards the drawn color -- as in \cite{pemantle1} $1000$ to $1$ -- although the convergence is extraordinarily slow.}.
106:
107: The mathematicians have discovered profound differences between these two models. For example, ERRW is recurrent on finite graphs \cite{dia88, kr99}, meaning that every edge is traversed infinitely often, while VRRW is not, becoming trapped e.g. on a line of five vertices \cite{pemantle, pemantle1, tar04} or more generally on ``trapping subgraphs" \cite{pemantle, pemantle1, pv99, vol1}. Many properties of the ERRW remain unknown; for example, the recurrence of ERRW on the infinite 2-d lattice is an open question. Even the one-step ERRW model (called once-reinforced in the mathematics literature) has only been successfully studied on a few special graphs, e.g. the infinite regular or Galton-Watson tree (where it is transient \cite{pemantle, dkl02, die05}) or the infinite ladder (where it is recurrent \cite{pemantle, sel06}). The recurrence of one-step ERRW on the infinite 2-d lattice remains essentially open, although Sellke showed the separate recurrence of each coordinate \cite{pemantle, sellke}. Pemantle \cite{pemantle, pemantle1} provides a more complete description of these and other results.
108:
109: The difference in mathematical tractability and underlying structure might suggest that models with bond and site reinforcement show different phenomenology. But this result would be unexpected from considerations of universality in statistical physics.
110:
111: \subsection{Overview of Results}
112:
113: In the present paper, we clarify some of these issues by means of high precision
114: simulations. Our main results are:
115: \begin{itemize}
116: \item Bond and site reinforced TSATWs with one-step reinforcement show the same critical behavior and are likely in the same universality class;
117: \item There is no finite $u$ phase transition in the 2-dimensional TSATW model with one-step
118: reinforcement. Walks are in the collapsed phase for all $u>0$ and the phase
119: transition happens at $u_c=0$;
120: \item The critical point and the critical exponents for TSATWs with one-step reinforcement in
121: $d = 3$ are markedly different from the values obtained in \cite{ordemann1,ordemann2}; and
122: \item The quasistatic approximation for the end-to-end distance seems to become exact as $t\to\infty$ for the collapsed
123: phase.
124: \end{itemize}
125:
126: \section{Numerical Methods and Results}
127:
128: \subsection{Methods}
129: \label{methods}
130:
131: Simulations of TSATWs with one-step reinforcement are straightforward. To keep track of
132: previous visits, one has to store a one-bit ``spin" variable $s_i$ for each site (bond) $i$, and
133: clear all spins after each walk. For convenience we sometimes used one byte per spin, which has the
134: added advantage that clearing is needed only after every 255th walk. This requires
135: $L^d/8$ resp. $L^d$ bytes of memory for site TSATWs and $dL^d/8$ resp $dL^d$ bytes for bond
136: TSATWs. Memory limitations were
137: more severe than CPU time so in the following we show more detailed results for site TSATWs than for bond TSATWs.
138:
139: The most serious potential source of systematic errors arises from lattices that are too small. If
140: open boundary conditions (b.c.) are used, the walk cannot go beyond the boundary, and both $R_t$ and $\langle S_t\rangle$ are underestimated. If periodic b.c. are used and the walk wraps around the lattice, it finds visited terrain in front of it and $R_t$ is overestimated, while $\langle S_t\rangle$ is still
141: underestimated.
142: We used lattices with helical boundary conditions and with up to $N = 2^{32}$ sites ($d = 2$)
143: resp. $2^{34}$ sites ($d = 3$),
144: see endnote \footnote{We always used lattice sizes such that the total number $N$ of sites was a power of
145: two, in order to implement the boundary condition $i\equiv i-N\;\mod N$ by bit masking. The
146: requirement that $N=L^d$ was not strictly enforced (unless $N$ was such that this would
147: give an integer $L$), so the lattices were only approximately of cubic shape. This is of
148: no consequence in the following.}. For each walk, the spans $x_{\rm max}-x_{\rm min}$ in all
149: $d$ directions were measured, and it was checked that the fraction of walks where any span was
150: $\ge L$ did not exceed $10^{-4}$. This restricted the number of steps per
151: walk to $t_{\rm max} \le 10^8$ for $d = 2$, and to $t_{\rm max} \approx 4\times 10^7$ for $d = 3$. The total number
152: of walks for each parameter setting was typically $\approx 2\times 10^4$ to $\approx 2\times 10^5$.
153:
154: \begin{figure}
155: \begin{center}
156: \epsfig{file=fig-R2var.ps, width=6.cm, angle=270}
157: \caption{(color online) Statistical errors (1 st.d.) of the three estimators for the average
158: squared end-to-end distance divided by $t$ (direct, $\widehat{R_t^2}/t$, and
159: $[R_t^2]_{\rm opt}/t$ from top to bottom) for 2-d site TSATWs with $u=0.34$. The number of
160: walks in this sample was $10^4$, which is only a small fraction of our total sample.}
161: \label{fig2var}
162: \end{center}
163: \end{figure}
164:
165: \subsection{Variance reduction}
166: \label{variance}
167: For small $u$, where walk-to-walk variation is significant, substantially increased accuracy is obtained by the following variance reduction procedure. Assume that the walker has already made $t$ steps and is presently at a site with cartesian coordinates ${\bf x}_t$. Given ${\bf x}_t$ and the states of the neighboring sites (i.e., visited or unvisited), one can calculate the expected increment $\widehat{\Delta \bf x}_{t+1}$ for the next step, since one knows the probability for the walker to step in each direction. From this, one obtains an estimate for the increment of $R_{t+1}^2$
168: \beq
169: \widehat{\Delta R_{t+1}^2} = \widehat{[{\bf x}_{t+1}^2 - {\bf x}_{t}^2]} = 2 {\bf x}_t \cdot \widehat{\Delta \bf x}_{t+1} +1\;.
170: \eeq
171: where we have used the fact that $\widehat{ \Delta \textbf{x}_{t+1} \cdot \Delta \textbf{x}_{t+1}} = 1$. The improved estimate is obtained by summing these increments,
172: \beq
173: \widehat{R_{t}^2} = \sum_{t'=1}^t \widehat{\Delta R_{t'}^2}\;.
174: \eeq
175: Further improvement is obtained by taking the optimal linear combination of the direct sample
176: average and this estimator,
177: \beq
178: [R_t^2]_{\rm opt} = \alpha_t R_t^2 + (1-\alpha_t) \widehat{R_t^2} \;,
179: \eeq
180: with $\alpha_t$ fixed for each $t$ such that the variance of $[R_t^2]_{\rm opt}$ is minimal. Differentiating the variance of $[R_t^2]_{\rm opt}$ with respect to $\alpha_t$ and minimizing this variance requires the estimation of the variances of $R_t^2$ and $\widehat{R_t^2}$ as well as their covariance. In Fig.~\ref{fig2var} we show the errors (single standard deviations, divided
181: by $t$) of the three estimators for 2-d site TSATWs with $u=0.34$.
182:
183:
184:
185: \subsection{Site TSATWs in d = 3}
186: \label{site3}
187: The \emph{average} r.m.s end-to-end distance divided by the number of steps, $t^{-1}R_t^2$, is shown
188: in Fig.~\ref{figR3} for site TSATWs in $d = 3$. In this and in all subsequent figures, curves are
189: not labelled by $u$ but by $w = \exp(u)$. We see clearly that there are significant corrections
190: to scaling (all curves bend upward at small $t$), but they are no worse than in other
191: nonequilibrium critical phenomena. A more careful analysis, taking these corrections into
192: account, gives $u_c = 1.831\pm 0.002$ ($w_c = \exp(u_c) = 6.24\pm 0.01$) and $\nu_c = 0.378\pm 0.004$. In particular, we can rule
193: out the possibility that $u_c > 1.85$ from the simple fact that all curves for $u > 1.85$ (i.e. for $w > 6.35)$ are clearly S-shaped and curve down at large $t$. These estimates are incompatible
194: with those of \cite{ordemann2}, $u_c = 1.92\pm 0.03$ and $\nu_c = 0.303\pm 0.005$. Possible
195: explanations for these earlier results are that corrections to scaling were neglected in \cite{ordemann2} or that the lattices used were too small.
196:
197: \begin{figure}
198: \begin{center}
199: \epsfig{file=fig-R3crit.ps, width=6.cm, angle=270}
200: \caption{(color online) Average squared end-to-end distances for site TSATWs in 3 dimensions, divided
201: by the number $t$ of steps. Each curve corresponds to a fixed value of $u$, with $u$ increasing
202: from top to bottom. Here and in all subsequent figures, the curves are labelled by $w = \exp(u)$.
203: The critical value $u=u_c$ corresponds to a straight curve in the limit
204: $t\to\infty$ whose slope is $2\nu_c - 1$. Statistical errors are comparable to the thickness
205: of the curves.}
206: \label{figR3}
207: \end{center}
208: \end{figure}
209:
210: The cross-over behavior near $u\approx u_c$ can be fitted to the usual ansatz
211: \beq
212: R_t^2 = t^{2\nu_c} F[(u-u_c)t^\phi] \;\;+\;{\rm corrections} \label{scale3}
213: \eeq
214: with $\phi=0.185\pm 0.020$, as seen from the data collapse shown in Fig.~\ref{figcoll3}.
215: The deviations from a perfect collapse seen in this figure are due to the corrections to scaling at small $t$ seen in Fig.~\ref{figR3}, which are not included in the scaling ansatz Eq.~(\ref{scale3}).
216: The apparent collapse could have improved by the widespread practice of plotting the sub- and
217: supercritical branches separately, without demanding that they join smoothly (the function $F(z)$
218: must be analytic at $z=0$). But the results obtained in this way would be spurious.
219:
220: \begin{figure}
221: \begin{center}
222: \epsfig{file=fig-R3coll.ps, width=6.cm, angle=270}
223: \caption{(color online) The same data as in Fig.~\ref{figR3}, plotted now as $R_t^2/t^{2\nu_c}$
224: versus $(u-u_c)t^\phi$, with $\nu_c = 0.378, u_c = 1.831$ $(w_c = 6.24)$, and $\phi=0.185$. If there were
225: perfect scaling, all data would fall on a single curve.}
226: \label{figcoll3}
227: \end{center}
228: \end{figure}
229:
230: Results for the average number of visited sites, $\langle S_t\rangle$, again divided by $t$, are
231: shown in Fig.~\ref{figS3}. This time the corrections to scaling are much bigger. This is not
232: unexpected, since there are also large corrections to the asymptotic law $\langle S_t\rangle \sim t$
233: for ordinary 3-d random walks. These corrections make an independent estimate of $u_c$ impossible,
234: whence we shall use the estimate obtained from $R_t$, i.e. $u_c = 1.831\pm 0.002$. The corrections to scaling also make the estimation of the exponent $k_c$ very
235: uncertain, in spite of the extremely small statistical errors (much smaller than the thickness of the curves). Our best estimate is $k_c = 0.977\pm 0.010$. This is again incompatible with the estimate $0.91\pm 0.01$ of \cite{ordemann2}. The leading correction to scaling exponent, defined as $\langle S_t\rangle = t^{k_c} [a +b/t^\Delta + o(t^{-\Delta})]$, is found to be $\Delta =0.22\pm 0.03$. This is to be compared to $\Delta =1/2$ for ordinary 3-d walks \cite{torney}.
236:
237: \begin{figure}
238: \begin{center}
239: \epsfig{file=fig-S3crit.ps, width=6.cm, angle=270}
240: \caption{(color online) Average number of visited sites for site TSATWs in 3 dimensions, divided by
241: the number $t$ of steps. Curves correspond to the same values of $u$ as in Fig.~\ref{figR3}.
242: Statistical errors are much smaller than the thickness of the curves.}
243: \label{figS3}
244: \end{center}
245: \end{figure}
246:
247: For the supercritical case, $u>u_c$, the following argument was given in \cite{sapo}: Let us assume that
248: the visited sites form, for large $t$, a compact $d-$dimensional domain $V_t$ whose volume increases as
249: $S_t \equiv |V_t| \propto R_t^d$ with $R_t \sim t^\nu$. Its surface is fuzzy but not fractal, i.e. it
250: increases as $|\partial V_t| \propto R_t^{d-1}$. If the walker is uniformly distributed inside $V_t$,
251: then the chance for it to be at the boundary is $|\partial V_t| / S_t \propto 1/R_t$. This is then also
252: proportional to the chance that the walker will make the next step outside $V_t$, i.e. $d\langle S\rangle
253: /dt \sim t^{-\nu}$. Integrating this gives $\nu = 1/(d+1)$ \cite{sapo}. The main assumption here
254: is \emph{not} that $\partial V_t$ is non-fractal (as stated in \cite{sapo,ordemann2}), but that the
255: walker is uniformly distributed inside $V_t$. This would be exact if the boundary would not grow at all
256: (i.e. in the limit $u\to\infty$), but for finite $u$ it corresponds to a {\it quasistatic
257: approximation} in the sense of \cite{dalmaroni}.
258:
259: In order to test this quasistatic approximation of the supercritical behavior, we plot in Fig.~\ref{figR3s} the ratio $R_t^2/\sqrt{t}$ for several values of $u>u_c$. We see very large corrections to scaling (the corrections to $\langle S_t\rangle$ would be even larger), but the curves do seem to become horizontal for $t\to\infty$. For $u\ge 2.5$, our best estimate is $R_t\sim t^\nu$ with $\nu =0.25\pm 0.01$, in perfect agreement with the prediction of the quasistatic approximation.
260:
261: \begin{figure}
262: \begin{center}
263: \epsfig{file=fig-R3super.ps, width=6.cm, angle=270}
264: \caption{(color online) $R_t^2 / \sqrt{t}$ for site TSATWs in 3 dimensions, for five values of
265: $u$ which are all larger than $u_c$. The curves seem to become horizontal as $t\to\infty$, with the asymptotic behavior appearing for smaller values of $t$ as $u$ increases.}
266: \label{figR3s}
267: \end{center}
268: \end{figure}
269:
270: \subsection{Site TSATWs in $d = 2$}
271: \label{site2}
272: In two dimensions the situation seems at first glance similar, except for the fact that
273: corrections to scaling are even larger. The latter is not surprising: random walks are recurrent
274: in $d = 2$, while they are not in any $d > 2$. The number of visited sites increases not as $t$
275: in $d = 2$, but as $S_t = \pi t/\ln(8t)[1+O(1/\ln t)]$ \cite{torney}. Related to this is the fact that
276: true self avoiding walks have upper critical dimension $d = 2$, leading to logarithmic corrections in
277: most observables for $d = 2$. As a consequence, one should also expect logarithmic corrections for
278: TSATWs.
279:
280: Results for the end-to-end distance are shown in Fig.~\ref{figR2}. Again we show a log-log plot
281: of $R_t^2/t$, for easy comparison with Fig.~\ref{figR3}. The main difference between these two
282: plots is that the curves fan out in Fig.~\ref{figR2} already for very small $t$, while they fan
283: out only at much later times in Fig.~\ref{figR3}. While the curves for $u<u_c$ in Fig.~\ref{figR3}
284: first seem to follow the scaling $R_t \sim t^{\nu_c}$ and cross over to $R_t \sim t$ only at
285: large $t$, no such cross-over is seen in Fig.~\ref{figR2}. Careful inspection shows that all
286: curves for $u>0.58$ (i.e. $e^u > 1.79$) bend down at large $t$, indicating that $u_c \le 0.58$ and that the estimate
287: $u_c=0.88\pm 0.05$ of \cite{ordemann2} is untenable. If we want to see a critical point with an
288: associated non-trivial power law in these data, then a possible candidate is $u_c\approx 0.54$ and
289: $\nu_c \approx 0.47$.
290:
291:
292: \begin{figure}
293: \begin{center}
294: \epsfig{file=fig-R2crit.ps, width=6.cm, angle=270}
295: \caption{(color online) Same as Fig.~\ref{figR3}, but for $d = 2$.}
296: \label{figR2}
297: \end{center}
298: \end{figure}
299:
300: An attempted data collapse for the data of Fig.~\ref{figR2}, again using Eq.(\ref{scale3}) and
301: optimized values $u_c = 0.548, \nu_c = 0.475$, and $\phi = 0.085$, is shown in Fig.~\ref{figcoll2}.
302: We might mention that the exponents proposed in \cite{ordemann2}, $\nu_c = 0.40\pm 0.01$ and
303: $\phi \approx 0.2$, seem to be ruled out. A data collapse using these exponents is shown in
304: panel (b) of Fig.~\ref{figcoll2}. Although it has an acceptable overall dispersion, this is
305: achieved mainly by fitting well the small-$t$ data, and grossly misrepresenting data for
306: large $t$.
307:
308: \begin{figure}
309: \begin{center}
310: \epsfig{file=fig-R2coll-a.ps, width=6.cm, angle=270}
311: \epsfig{file=fig-R2coll-b.ps, width=6.cm, angle=270}
312: \caption{(color online) Attempted data collapse analogous to Fig.~\ref{figcoll3}, but for the
313: data of Fig.~\ref{figR2}. Parameters in panel (a) are $\nu_c = 0.475, u_c = 0.548$ $(w_c = 1.730)$, and
314: $\phi=0.085$, while panel (b) uses the values $\nu_c = 0.40, u_c = 0.88$ $(w_c = 2.411), \phi=0.2$
315: proposed in \cite{ordemann2}.}
316: \label{figcoll2}
317: \end{center}
318: \end{figure}
319:
320: \begin{figure}
321: \begin{center}
322: \epsfig{file=fig-R2psi.ps, width=6.cm, angle=270}
323: \caption{(color online) The function $\Psi_t(u)$ [see Eq.~(\ref{psi})] plotted against $t$ for
324: both positive and negative values of $u$, including $u=0$ $(w = 1)$. The fact that this figure resembles
325: a typical cross-over plot as in Fig.~\ref{figR3} suggests that $u=0$ is a critical point.}
326: \label{figPsi}
327: \end{center}
328: \end{figure}
329:
330: \begin{figure}
331: \begin{center}
332: \epsfig{file=fig-R2psi-b.ps, width=6.cm, angle=270}
333: \caption{(color online) Part of the data shown in Fig.~\ref{figPsi}, but plotted against $1/t^{0.22}$.
334: This exponent gave the straightest curves for $w= \exp(u) <0.5$. No exponent would give straight
335: curves for $w\geq 0.8$.}
336: \label{figPsi2}
337: \end{center}
338: \end{figure}
339:
340: Although the collapse seen in Fig.~\ref{figcoll2}(a) is satisfactory, the smallness
341: of $\phi$ and the closeness of $\nu_c$ to the random walk value $\nu=1/2$ suggest a very different
342: interpretation. We propose that there is in fact \emph{no} phase transition at any $u_c>0$. Instead,
343: the TSATW is collapsed for any $u>0$, i.e. $u_c=0$. This is also consistent with the fact that
344: 2-d random walks are recurrent, i.e. the interaction should be a relevant perturbation for any
345: $u>0$. It is difficult to obtain direct numerical evidence for this scenario, due to the very
346: slow cross-over from the random walk behavior to the collapsed behavior, and due to the presence of strong corrections. In order to make any progress, we have to understand better these corrections.
347:
348: In order to analyze the behaviour for very small $u$ more closely, let us define the quantity
349: \beq
350: \Psi_t(u) = - {1\over u} \ln [R_t^2/t] \label{psi}
351: \eeq
352: It is obviously well defined for $u\neq 0$, but it can be defined also for $u=0$ using l'H\^opital's
353: rule,
354: \beq
355: \Psi_t(0) = - \lim_{u\to 0} {1\over u} \ln [R_t^2/t] = - {1\over t} {\partial R_t^2\over \partial u}.
356: \eeq
357: We used here the fact that $R_t^2 = t$ exactly for $u=0$. Numerically, $\Psi_t(0)$ can be estimated
358: by a slight generalization of the reduced variance method discussed in subsection~\ref{variance}.
359: We simulate just ordinary random walks, but keeping track of the visited sites and calculating
360: $\partial R _t^2 /\partial u$ using Eqs.~(2), (5), and (6).
361:
362: It is easily seen that $\Psi_t(u)$ is positive for all $u$. Plots of $\Psi_t(u)$ versus $t$, both for
363: positive {\it and for negative} values of $u$, are shown in Fig.~\ref{figPsi}.
364: Assume there is a collapse transition at $u=u_c$. We then expect that $\Psi_t(u)$ diverges as
365: $\ln t$ for $u>u_c$ and $t\to\infty$, while it should stay bounded for $u<u_c$. More precisely,
366: we expect that $\Psi_t(u) \sim const -a/t^\delta$ for $u<u_c$, where $\delta$ is another correction
367: to scaling exponent. Plotting $\Psi_t(u)$ versus $t^{-\delta}$ should thus give straight lines
368: converging to finite values for $t^{-\delta}\to 0$ if $u<u_c$, but upward bent curves diverging
369: for $t^{-\delta}\to 0$ if $u>u_c$. One such plot, showing $\Psi_t(u)$ versus $t^{-0.22}$, is given in
370: Fig.~\ref{figPsi2}. From this and similar plots with different exponents, we conclude that (i) the
371: data are consistent with this scenario; (ii)
372: the critical point is at $u_c\approx 0$, most likely at $u_c=0$ exactly; (iii) the correction to
373: scaling exponent is $\delta = 0.22\pm 0.05$; and (iv) at the critical point, $\Psi_t$ scales
374: either as $\Psi_t(0) \sim \ln \ln t$ or $\Psi_t(0) \sim [\ln t]^\alpha$ with $0<\alpha \ll 1$. The
375: former ($\Psi_t(0) \sim \ln \ln t$) seems preferred, but a clear distinction between these
376: alternatives is not possible.
377:
378: %positive {\it and for negative} values of $u$, are shown in Fig.~\ref{figPsi}.
379: %exists and is non-zero. Moreover, it can be estimated numerically by a slight generalization
380: %of the reduced variance method discussed in subsection~\ref{variance}. This suggests that we
381: %study the analogous quantity
382: %\beq
383: % \Psi_t(u) = - {1\over u} \ln [R_t^2/t] \label{psi}
384: %\eeq
385: %for $u = 0$ as well as $u\neq 0$. Plots of $\Psi_t(u)$ versus $t$, both for positive and for {\it negative}
386: %values of $u$, are shown in Fig.~\ref{figPsi}. A careful analysis (e.g. by plotting the data against
387: %different powers $t^{-\Delta}$ and looking for straight extrapolations for $t^{-\Delta}\to 0$)
388: %indicates that $\Psi_t(u) \to const$ for $u<0$, with $\Delta \approx 0.2-0.3$. This suggests
389: %that in two dimensions true self-{\it repelling} walks where one distinguishes only between visited and unvisited sites (in contrast to the Amit {\it et al.} TSAWs \cite{Amit}) show no logarithmic corrections. They are
390: %swollen as compared to ordinary random walks, but the swelling is by a finite factor. In
391: %contrast, for $u=0$ the curve in Fig.~\ref{figPsi} seems to increase forever, albeit very slowly.
392: %The data suggest more a behavior $\Psi_t(0) \sim \ln \ln t$ than $\Psi_t(0) \sim [\ln t]^\alpha$
393: %with $\alpha>0$, although we cannot exclude a very small value of the exponent. In any case,
394: %this suggests that there is some singularity at $u=0$ rather than at $u>0$.
395:
396: Studying $S_t$, the number of visited sites, is not very revealing. As seen from Fig.~\ref{figS2},
397: there is no value of $u$ for which the curve is straight. $u\approx 0.7,\; w = e^u \approx 2$
398: yields the straightest curve in the large $t$ range $10^5 < t < 10^8$, but this is clearly not
399: asymptotic, as the curves for larger $u$ indicate (they have not crossed over to asymptotic behavior
400: and hence curve up for large $t$, although they finally curve down, for {\it very} large $t$).
401:
402: \begin{figure}
403: \begin{center}
404: \epsfig{file=fig-S2crit.ps, width=6.cm, angle=270}
405: \caption{(color online) Same as Fig.~\ref{figS3}, but for $d = 2$. The curves correspond to $w = 0.2, 0.4, 0.6, \ldots
406: 2.6$ (from top to bottom).}
407: \label{figS2}
408: \end{center}
409: \end{figure}
410:
411: For coupling constants $u\gg 1$ one finds again that the prediction of the quasistatic
412: approximation, $R_t \sim t^{1/3}$, is in excellent agreement with the data (see Fig.~\ref{figR2s}).
413: As in the 3-d case, corrections to this prediction are very large for small values of $u$,
414: but they decrease quickly for $u\to\infty$. One would like of course to verify the
415: quasistatic approximation for smaller $u$, but this seems at present impossible without going to lattice sizes beyond the reach of our normal computational resources.
416:
417: \begin{figure}
418: \begin{center}
419: \epsfig{file=fig-R2super.ps, width=6.cm, angle=270}
420: \caption{(color online) $R_t^2 / t^{2/3}$ for site TSATWs in 2 dimensions, for seven values of
421: $u$ which are all larger than $u_c$. The curves seem to become horizontal as $t \to \infty$, with the asymptotic behavior appearing for smaller values of $t$ as $u$ increases.}
422: \label{figR2s}
423: \end{center}
424: \end{figure}
425:
426:
427: \subsection{Bond TSATWs in $d = 3$}
428:
429: We now turn to the bond reinforced random walk in $d = 3$. Results for the end-to-end distance $R_t^2$ are shown in Fig.~\ref{EFig2}. The plot is superficially quite similar to Fig.~\ref{figR3}, with substantial corrections to scaling and the critical reinforcement $u_c$ occurring at much higher $u$. This latter fact is unsurprising as bond-reinforcement is much more ``dilute" than its site-reinforced cousin; consider that the equivalent of a visited site in the bond-reinforced model must have all six bonds visited in $d = 3$. Analysis suggests $u_c = 2.475 \pm 0.003$ and $\nu_c = 0.380\pm 0.004$. Note that $\nu_c$ is within error of the estimate $\nu_c = 0.378 \pm 0.004$ for the Site TSATW in $d = 3$. This is the first piece of evidence that the bond and site models are in the same universality class.
430:
431: \begin{figure}
432: \begin{center}
433: \epsfig{file=EFig2.ps, width=6.cm, angle=270}
434: \caption{(color online) Average squared end-to-end distances for bond TSAWs in 3 dimensions, divided by
435: the number $t$ of steps. Each curve corresponds to a fixed value of $u$, with $u$ increasing
436: from top to bottom. The critical value $u=u_c = 2.475$ $(w_c \approx 11.88)$ corresponds to a straight curve in the limit
437: $t \to \infty$ whose slope is $2\nu_c - 1$. Here and in all subsequent figures, the curves are
438: labelled by $w = \exp(u)$. Statistical errors are comparable to the thickness of the curves.}
439: \label{EFig2}
440: \end{center}
441: \end{figure}
442:
443: In Fig.~\ref{EFig3} we show a data collapse with the same scaling ansatz Eq.(\ref{scale3}); $u_c = 2.475, \nu_c = 0.380$ and $\phi=0.185\pm 0.020$. The critical exponents $\nu_c$ and $\phi$ are within error and identical, resp., to those for the Site TSATW in $d = 3$, see Fig.~\ref{figcoll3}, and the data collapse is if anything even better than that of Fig.~\ref{figcoll3}. We see similar corrections to scaling at small $t$ and excellent collapse at large $t$, facilitated by our ability to simulate long walks (~$10^7$) due to the high value of $u_c$.
444:
445: \begin{figure}
446: \begin{center}
447: \epsfig{file=EFig3.ps, width=6.cm, angle=270}
448: \caption{(color online) The same data as in Fig.~\ref{EFig2}, plotted now as $R_t^2/t^{2\nu_c}$
449: versus $(u-u_c)t^\phi$, with $\nu_c = 0.380, u_c = 2.475$, and $\phi=0.185$. Note that the critical exponents are within error bars of those used in the data collapse of Fig.~\ref{figcoll3}.}
450: \label{EFig3}
451: \end{center}
452: \end{figure}
453:
454: Results for the average number of visited sites, $\langle S_t\rangle$ are not shown, but are similar to Fig.~\ref{figS3}, with substantial corrections to scaling. We thus use the estimate of $u_c$ obtained from $R_t$, i.e. $u_c = 2.475 \pm 0.002$. The best estimate of the exponent $k_c$ (again made difficult by corrections to scaling) is $k_c = 0.970\pm 0.010$. This is incompatible with the estimate $0.91\pm 0.01$ of \cite{ordemann2} but within the error of our estimate for the site-reinforced model, $k_c = 0.977 \pm 0.010$. The leading correction to scaling exponent, defined as $\langle S_t\rangle = t^{k_c} [a +b/t^\Delta + o(t^{-\Delta})]$, is found to be $\Delta =0.25\pm 0.05$. This is within error of the estimate $\Delta =0.22\pm 0.03$ in the site case.
455:
456: Hence in all cases the estimates of the critical exponents for the site-reinforced model given by \cite{ordemann2} are excluded by our results as candidate exponents for the bond-reinforced model. The estimates of all exponents for the bond-reinforced model agree (within error) with those we obtained for the site-reinforced case, see Sec.~\ref{site3}. It is thus unsurprising that Fig.~\ref{EFig5} similarly verifies the quasistatic approximation $R_t\sim t^\nu$ with $\nu =0.25$ for $u > u_c$ in the large $t$ limit. Crossover to the asymptotic behavior occurs at smaller $t$ as $u \to\infty$. It is harder to verify the large $t$ limit for small $u > u_c$; we cannot run sufficiently long walks while limiting spurious self-intersection and hence reliably estimating $R_t$.
457:
458: \begin{figure}
459: \begin{center}
460: \epsfig{file=EFig5.ps, width=6.cm, angle=270}
461: \caption{(color online) $R_t^2 / \sqrt{t}$ for bond TSAWs in 3 dimensions, for five values of
462: $u$ which are all larger than $u_c$. The curves seem to become horizontal as $t \to \infty$, with the asymptotic behavior appearing for smaller values of $t$ as $u$ increases. This verifies that the quasistatic holds for bond as well as site reinforcement.}
463: \label{EFig5}
464: \end{center}
465: \end{figure}
466:
467: \subsection{Bond TSATWs in $d = 2$}
468:
469: Our results for $d = 3$ indicate that the bond- and site-reinforced models are in the same universality class. This implies that $u_c = 0$ for bond TSATW in $d = 2$. To test this, we study small $u$ walks, which will cross over to the collapsed behavior only for large values of $t$. This regime is even more difficult to study in the bond reinforced case, due to the dilute nature of the bond reinforcement. The simulations used are as large as possible ($2^{32}$ sites and walks of $ \approx 10^8$ steps) but in many cases these walks just reach the beginning of what may be the scaling regime.
470:
471: \begin{figure}
472: \begin{center}
473: \epsfig{file=EFig6.ps, width=6.cm, angle=270}
474: \caption{(color online) Same as Fig.~\ref{EFig2}, but for $d = 2$.}
475: \label{EFig6}
476: \end{center}
477: \end{figure}
478:
479: In Fig~\ref{EFig6} we show results for the end-to-end distance $R_t^2/t$. As in the site-reinforced case for $d = 2$, the curves fan out in Fig.~\ref{EFig6} for small $t$, with no apparent cross-over of the sort seen in Fig.~\ref{figR2} or Fig.~\ref{EFig2}. An estimate of the critical reinforcement $u_c$ from these data is extremely difficult. An attempted data collapse for the data of Fig.~\ref{EFig6}, using the scaling ansatz Eq.(\ref{scale3}) and
480: optimized values $u_c = 0.73, \nu_c = 0.481$, and $\phi = 0.058$, is shown in Fig.~\ref{EFig7}(a). While the data collapse acceptably for these values, the exponents proposed in \cite{ordemann2}, $u_c = 0.88, \nu_c = 0.40\pm 0.01$ and $\phi \approx 0.2$, can be ruled out, as a data collapse using these exponents is completely unsatisfactory, Fig.~\ref{EFig7}(b) .
481:
482: \begin{figure}
483: \begin{center}
484: \epsfig{file=EFig7.ps, width=6.cm, angle=270}
485: \epsfig{file=EFig7bad.ps, width=6.cm, angle=270}
486: \caption{(color online) Attempted data collapse analogous to Fig.~\ref{EFig3}, but for the
487: data of Fig.~\ref{EFig6}. Parameters in panel (a) are $\nu_c = 0.481, u_c = 0.730$ $(w_c = 2.075)$, and
488: $\phi=0.058$, while panel (b) uses the values $\nu_c = 0.40, u_c = 0.88$ $(w_c = 2.411)$, $\phi=0.2$
489: proposed in \cite{ordemann2}. The latter collapse is completely unsatisfactory.}
490: \label{EFig7}
491: \end{center}
492: \end{figure}
493:
494: As the estimated $\phi = 0.058$ is even smaller than that obtained for the site-reinforced model in $d = 2$ (where $\phi = 0.085$) and as $\nu_c = 0.481$ is very close to the random walk value $\nu = 1/2$, we argue that there is no phase transition for $u_c > 0$ in the bond-reinforced model, either. In particular, similar heuristic arguments about the recurrence of 2-d random walks suggest again that any non-zero reinforcement is a relevant perturbation. Hence we study again the function $\Psi_t(u)$ [Eq.~(\ref{psi})] for $u > 0$, $u = 0$, and $u < 0$ (see Fig.~\ref{EFig8}). As for the site case (subsection \ref{site2}), plotting $\Psi_t(u)$ against $1/t^\delta$ with different exponents $\delta$ reveals the detailed asymptotic behavior. Such plots (not shown here) indicate that $u_c = 0$, and that the correction to scaling exponent in the uncollapsed phase $u<u_c$ is $\delta = 0.20 \pm 0.05$, well within error of the estimated $\delta = 0.22 \pm 0.05$ of the site case.
495:
496: \begin{figure}
497: \begin{center}
498: \epsfig{file=EFig8.ps, width=6.cm, angle=270}
499: \caption{(color online) The function $\Psi_t(u)$ [see Eq.~(\ref{psi})] plotted against $t$ for
500: various positive and negative values of $u$, including $u=0$. This figure again resembles a cross-over
501: plot as in Fig.~\ref{figPsi}, suggesting that $u=0$ $(w = 1)$ is a critical point for bond-reinforcement as well.}
502: \label{EFig8}
503: \end{center}
504: \end{figure}
505:
506: As was the case for the site-reinforced model in $d = 2$, it is not particularly illuminating to study the number of visited sites $S_t$. The corrections to scaling are even larger, and hence it is impossible to estimate the correct scaling exponent from these data.
507:
508: For coupling constants $u\gg 0$ one finds that the prediction of the quasistatic
509: approximation, $R_t \sim t^{1/3}$, is in excellent agreement with the data (see Fig.~\ref{EFig10}). Corrections to this prediction are very large for small values of $u$,
510: but become irrelevant as $u \to \infty$, as is already apparent at $w = e^{u}=16$. Familiar limitations to accessible lattice size make the quasistatic approximation impossible to verify for smaller $u$, where crossover to the asymptotic behavior takes place at very large $t$. Settling the validity of the quasistatic approximation in the small $u$ regime will in all likelihood require the development and application of appropriate analytical methods.
511:
512: \begin{figure}
513: \begin{center}
514: \epsfig{file=EFig10.ps, width=6.cm, angle=270}
515: \caption{(color online) $R_t^2 / t^{2/3}$ for bond TSAWs in 2 dimensions, for seven values of
516: $u$ which are all larger than $u_c$. The curves seem to become horizontal for $t \to \infty$
517: and $u \to \infty$.}
518: \label{EFig10}
519: \end{center}
520: \end{figure}
521:
522: \section{Discussion and Outlook}
523: Despite its simplicity, the once-reinforced site variant of the True Self Attracting Walk (TSATW) has generated considerable controversy since its original statement by Sapozhnikov \cite{sapo}. In this paper we have used a combination of careful high-statistics simulations and heuristic arguments to attempt a resolution of many of these disputes.
524:
525: In $d = 3$ we confirm the existence of a phase transition from random walk-like to collapsed behavior for finite reinforcement $u_c$. Our simulations provide overwhelming evidence for rejecting the proposed $u_c$ and scaling exponents of \cite{ordemann1, ordemann2}. We find $u_c = 1.831\pm 0.002$ and $\nu_c = 0.378\pm 0.004$, with the crossover exponent $\phi = 0.185 \pm 0.020$. In addition we verify the quasistatic approximation $R_t^2 \sim \sqrt{t}$ for large $t$ and $u$.
526:
527: In $d = 2$ we argue that there is no phase transition at any finite reinforcement $u_c$. For any $u>0$ the walks go to a collapsed phase, and the ``critical behavior" at $u_c = 0$ is simply that of a random walk. The quasistatic approximation $R_t^2 \sim t^{2/3}$ is also verified for large $t$ and $u$.
528:
529: In addition to the site-reinforced TSATW, we studied the bond-reinforced variant and found evidence that despite the underlying mathematical differences (and related difficulties) bond-reinforced TSATW is in the same universality class as site-reinforced TSATW. In $d = 3$ we found a phase transition at finite reinforcement $u_c = 2.475 \pm .003$ and scaling behavior extremely similar to the site-reinforced model, with similar success of the quasistatic approximation. In $d = 2$ we found evidence of a phase transition at $u_c = 0$ although the evidence here is somewhat weaker due to the long time needed to cross over to the collapsed behavior and the memory limitations imposed by the extremely large lattices needed to minimize spurious self-intersection.
530:
531: An obvious limitation of our work is the lack of an analytical understanding as to why the transition to a collapsed phase should occur at any finite reinforcement in the site- and bond- models in $d = 2$. In the mathematics literature, the once-reinforced ERRW has been studied by mapping it to a diffusion with a drift term (directed inward) at the boundary \cite{pemantle, dav96}. As far as we know this technique has only been applied in $d = 1$ and is even in this case of considerable technical difficulty. There are also techniques mapping the stochastic process to a (deterministic) dynamical system, the so-called ``stochastic approximation", which is largely unknown to the physics literature \cite{pemantle, pemantle1}. This suggests that some sensible map to a continuous process or a dynamical system might enable an analytic proof of $u_c = 0$ in one or both of the variants of once-reinforced TSATW in $d = 2$.
532:
533: More generally, the universality result we propose for site and bond TSATW in $d = 3$ and $d = 2$ suggests the possibility of a deep dialogue between the statistical physics and probability literatures. The perspective of statistical physics generates different questions (with respect to phase transitions, critical behavior, and universality) that complement the rigorous results derived within the probability community. Furthermore, the probability literature as reviewed in \cite{pemantle} contains an enormous number of unexplored models for random walks with reinforcement. It is also clear that these walk processes are specific instances of a general study of random processes with reinforcement, with many applications in the biological and social as well as physical sciences \cite{pemantle}. The statistical physics of such models remains an almost entirely open question.
534:
535: We end on a cautionary note about the use of simulation in these problems. As pointed out by Pemantle \cite{pemantle1}, the convergence times for some random processes with reinforcement can be astronomical; the Friedman urn, for example, does not reach its asymptotic behavior until a googol updates or more. This suggests that in some cases the high statistics simulations that would be applied by statistical physicists may only be probing the transient behavior of such models. While the transient behavior has its own intrinsic interest, we suggest that a dialogue between the two fields would do much to drive research in mutually beneficial directions--while avoiding pitfalls along the way.
536:
537: Acknowledgment: J.G.F. and P.G. are supported by iCORE.
538:
539: \begin{thebibliography}{99}
540: \bibitem{deGennes} P.-G. de Gennes, {\it Scaling Cpts in Polymer Physics} (Cornell University Press,
541: Ithaca, 1979).
542: \bibitem{Amit} D. Amit, G. Parisi, and L. Peliti, Phys. Rev. B {\bf 27}, 1635 (1983).
543: \bibitem{domb} C. Domb and G.S. Joyce, J. Phys. C {\bf 5}, 956 (1972).
544: \bibitem{donsker} M.D. Donsker and S.R.S Varadhan, Comm. Pure Appl. Math. {\bf 28}, 525 (1975);
545: {\bf 32}, 721 (1979).
546: \bibitem{mehra} V. Mehra and P. Grassberger, Physica D {\bf 168}, 244 (2002).
547: \bibitem{sapo} V.B. Sapozhnikov, J. Phys. A {\bf 27}, L151 (1994); {\bf 27}, 3935 (1998).
548: \bibitem{reis} F.D.A. Aaroro Reis, J. Phys. A {\bf 28}, 3851 (1995).
549: \bibitem{prasad} M.A. Prasad, D.P. Bhatta, and D. Arora, J. Phys. A {\bf 29}, 3037 (1996).
550: \bibitem{lee} J.W. Lee, J. Phys. A {\bf 31}, 3929 (1998).
551: \bibitem{ordemann1} A. Ordemann, G. Berkolaiko, S. Havlin, and A. Bunde, Phys. Rev. E {\bf 61}, R1005 (2000).
552: \bibitem{ordemann2} A. Ordemann, E. Tomer, G. Berkolaiko, S. Havlin, and A. Bunde,
553: Phys. Rev. E {\bf 64}, 046117 (2001).
554: \bibitem{dalmaroni} A. Jim{\'e}nez-Dalmaroni and H. Hinrichsen, Phys. Rev. E {\bf 68}, 036103 (2003).
555: \bibitem{pemantle} R. Pemantle, Probability Surveys {\bf 4}, 1 (2007); arXiv:math.PR/0610076v2 (2007).
556: \bibitem{pemantle1} R. Pemantle, Random processes with reinforcement, http://citeseer.ist.psu.edu/617028.html. Preprint (2001).
557: \bibitem{trails} A. Guttman, J. Phys. A {\bf 18}, 567 (1985).
558: \bibitem{dia88} P. Diaconis, In J. Bernardo, M. de Groot, D. Lindley, and A. Smith, eds., {\it Bayesian Statistics}, pp. 111-125. (Oxford Univresity Press, Oxford, 1988).
559: \bibitem{kr99} M. Keane and S. Rolles, In {\it Infinite Dimensional Stochastic Analysis} volume 52 of {\it Verhandelingen, Afdeling Natuurkunde. Eerste Reeks. Koninklijke Nederlandse Akademie van Wetenschappen}, pp. 217-234. (R. Neth. Acad. Arts Sci., Amsterdam, 1999).
560: \bibitem{tar04} P. Tarr{\`e}s, Ann. Probab. {\bf 32}, 2650 (2004).
561: \bibitem{pv99} R. Pemantle and S. Volkov, Ann. Probab. {\bf 27}, 1368 (1999).
562: \bibitem{vol1} S. Volkov, Ann. Probab. {\bf 29}, 66 (2001).
563: \bibitem{dkl02} R. Durrett, H. Kesten, and V. Limic, Prob. Thoer. Rel. Fields {\bf 122}, 567 (2002).
564: \bibitem{die05} J. Die, Statistics Probab. Lett. {\bf 73}, 115 (2005).
565: \bibitem{sel06} T. Sellke, Elec. J. Prob. {\bf 11}, 301 (2006).
566: \bibitem{sellke} T. Sellke, Reinforced random walks on the $d$-dimensional lattice. Preprint (1994).
567: \bibitem{torney} D.C. Torney, J. Stat. Phys. {\bf 44}, 49 (1986).
568: \bibitem{derkachov} S.E. Derkachov, J. Honkonen and A.N. Vasil'iev, J. Phys. A: Math. Gen. {\bf 23}, 2479 (1990).
569: \bibitem{dav96} B. Davis, Ann. Probab. {\bf 24}, 2007 (1996).
570: \end{thebibliography}
571:
572: \end{document}
573: