1: %\documentclass[aps,prl,twocolumn]{revtex4}
2: \documentstyle[aps,prl,multicol,epsf]{revtex}
3: %\documentstyle[aps,prl,multicol,epsf,graphics]{revtex}
4: %\documentstyle[aps,prl,twocolumn,epsf,graphics]{revtex}
5: %\documentstyle[pra,aps,eqsecnum,epsf]{revtex}
6: %\usepackage{graphicx}
7: \begin{document}
8: \title{Percolation in random environment}
9:
10: \author{R\'obert Juh\'asz$^{1,2}$ and Ferenc Igl\'oi$^{2,1,3}$}
11:
12: %\affiliation{
13: \address{
14: $^1$ Institute of Theoretical Physics,
15: Szeged University, H-6720 Szeged, Hungary\\
16: $^2$ Research Institute for Solid State Physics and Optics,
17: H-1525 Budapest, P.O.Box 49, Hungary\\
18: $^3$Laboratoire de Physique des Mat\'eriaux, Universit\'e Henri Poincar\'e (Nancy 1),
19: F-54506 Vand\oe uvre l\`es Nancy, France
20: }
21:
22: \date{\today}
23:
24: \maketitle
25:
26: \begin{abstract}
27: We consider bond percolation on the square lattice with perfectly correlated
28: random probabilities. According to scaling considerations, mapping to a random
29: walk problem and the results of Monte Carlo simulations the critical behavior
30: of the system with varying degree of disorder is governed by new, random fixed points
31: with anisotropic scaling properties. For weaker disorder both the magnetization and
32: the anisotropy exponents are non-universal, whereas for strong enough disorder the system
33: scales into an {\it infinite randomness fixed point} in which the critical
34: exponents are exactly known.
35: \end{abstract}
36:
37: %\maketitle
38:
39: \newcommand{\bc}{\begin{center}}
40: \newcommand{\ec}{\end{center}}
41: \newcommand{\be}{\begin{equation}}
42: \newcommand{\ee}{\end{equation}}
43: \newcommand{\beqn}{\begin{eqnarray}}
44: \newcommand{\eeqn}{\end{eqnarray}}
45:
46: \begin{multicols}{2}
47: \narrowtext
48:
49: \section{Introduction}
50:
51: Percolation is a paradigm for random processes\cite{staufferaharony},
52: in which the $i$-th bond (or site) of a
53: regular lattice is occupied with a probability, $p_i$, which is generally taken
54: independent of its position, $p_i=p$. In percolation theory one is interested in
55: the properties of clusters, in particular in the vicinity of the percolation transition
56: point $p=p_c$, when clusters with diverging size are formed. Using a close analogy
57: with thermal phase transitions, which is based on the $Q \to 1$ limit of the ferromagnetic
58: $Q$-state Potts model\cite{kasteleyn}, a scaling theory has been developed and in two dimensions
59: many, conjecturedly exact results have been obtained by conformal field theory\cite{cardy96}
60: and by Coulomb-gas methods\cite{cardyziff}.
61:
62: In real systems, however, the
63: occupation probabilities are generally inhomogeneous,
64: i.e., position, direction or neighborhood dependent, and there are some correlations
65: between them. The effect of quenched disorder, i.e. when the occupation probabilities
66: are position dependent random variables, can be studied by scaling considerations.
67: According to the Harris criterion\cite{harris} the relevance or irrelevance of the effect
68: of quenched disorder on the percolation transition depends on the sign of the specific
69: heat exponent, $\alpha$, of the corresponding pure Potts model in the $Q \to 1$ limit.
70: Since in any dimension
71: $\alpha<0$\cite{staufferaharony}, the critical
72: properties of ordinary and ``random'' percolation are equivalent.
73: Another form of perturbations, e.g. long-range correlations between occupation
74: probabilities\cite{weinrieb} or anisotropy, such as in directed percolation\cite{kinzel}, however,
75: leads to modified critical properties.
76:
77: In the present paper we consider the combined effect of
78: disorder, anisotropy and correlations, when the occupation probabilities
79: are random variables, which are perfectly correlated in a $d_d$ dimensional subspace.
80: This type of behavior could be relevant to describe the properties of oil or gas inside
81: porous rocks in oil reservoirs, when the rock has a layered structure\cite{staufferaharony}.
82:
83: Models with perfectly correlated disorder play an important r\^ ole in
84: statistical physics and in the theory of (quantum) phase transitions. Among the early work
85: we mention the partially exact solution of the
86: McCoy-Wu model\cite{mccoywu} (which is the two-dimensional Ising model with layered
87: randomness) and the field-theoretical investigations by Boyanovski and
88: Cardy\cite{boyanovskicardy}. As a matter of fact in random quantum systems disorder is
89: perfectly correlated along the (imaginary) time direction, i.e. here $d_d=1$. For these
90: systems, in particular for random quantum spin chains, i.e. in $(1+1)$ dimension, many
91: new, presumably exact results have been obtained recently by a strong disorder renormalization
92: group (SDRG) method\cite{MDH}. It was found that for strong enough initial disorder
93: the critical behavior of several systems is governed by a so-called infinite randomness
94: fixed point (IRFP)\cite{fisher99}, with unusual scaling properties. Here we mention
95: recent calculations on the random
96: transverse-field Ising model (RTIM)\cite{fisher92,igloi02}, random quantum Potts and
97: clock models\cite{senthil}, random
98: antiferromagnetic Heisenberg spin chains\cite{fisher94,S=1} and ladders\cite{mllri} and
99: also non-equilibrium phase transitions in the presence of quenched disorder\cite{hiv02}. In many
100: cases a cross-over between
101: weak and strong disorder regimes has been observed and a general scaling scenario has
102: been proposed\cite{cli01}.
103:
104: In the present paper we study percolation on a square lattice with strip
105: random occupation probabilities. We investigate the critical behavior of the system with
106: varying strength of disorder by scaling considerations, by random walk mappings and by
107: Monte Carlo (MC) simulations.
108: The structure of the paper is the following. The model and the relevant physical
109: quantities are introduced in Sec. II. Investigations in the weak and strong disorder
110: limits are given in Sec. III., MC simulations for intermediate disorder
111: are presented in Sec. IV. The paper is closed by a discussion in Sec. V.
112:
113: \section{The model}
114:
115: We consider bond percolation on a square lattice with sites $\{i,j\}$,
116: $1 \le i \le L$ and $1 \le j \le K$, where the
117: occupation probabilities, $0<p<1$, are random variables, which are perfectly correlated
118: along vertical lines. as indicated in Fig. \ref{fig1}. If the average value of the occupation
119: probabilities exceeds a critical value, $\langle p \rangle>p_c$, there is a percolation
120: transition in the system. The value of $p_c$ can be determined by noticing that
121: under a duality transformation,
122: which maps the ordered and the disordered phases of the system into each other,
123: the layered structure of the system is preserved and the dual value of the local
124: probability is transformed as: $\tilde{p}=1-p$\cite{kinzeldomany}.
125: Consequently the probability distribution, $P(p)$,
126: is transformed into $\tilde{P}(\tilde{p})=P(1-p)$ and the random system is self-dual, if
127: the probability distribution is symmetric: $P(p)=P(1-p)$ and thus the average
128: value of $p$ is given by $\langle p \rangle=p_c=1/2$.
129: Since there is one phase transition in the system, the self-duality
130: point corresponds to the critical point and the distance of the critical point,
131: $t$ is defined as:
132: %
133: \be
134: t=\langle p \rangle-p_c\;.
135: \label{delta}
136: \ee
137: %
138:
139: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
140: \begin{figure}[tbh]
141: \epsfxsize=7truecm
142: \begin{center}
143: \mbox{\epsfbox{fig1.eps}}
144: \end{center}
145: \caption{\label{fig1} Percolation on the square lattice with random bond occupation
146: probabilities, which are perfectly correlated along the vertical direction. The portion of
147: the lattice having the same occupation probability, $p$, is denoted by bold
148: lines, whereas the corresponding part of the dual lattice with $\tilde{p}=1-p$ is
149: shown by dashed lines.
150: }
151: \end{figure}
152: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
153:
154: In the presence of quenched disorder the mean value of a physical
155: observable, $\Phi$, is calculated as $[ \langle \Phi \rangle ]_{\rm av}$,
156: where $\langle \dots \rangle$ denotes thermal averaging for a given realization of the
157: disorder and $[ \dots ]_{\rm av}$ stands for disorder averaging.
158: In percolation the basic quantities of interest are the fractal and connectivity properties
159: of the largest clusters. In the following we use the concept of anisotropic
160: scaling\cite{binderwang},
161: when the correlation lengths in the two directions, (which correspond to the extensions of the
162: largest clusters) involve different critical exponents: $\xi_{\perp} \sim |t|^{-\nu_{\perp}}$
163: and $(\xi_{\parallel} \sim t^{-\nu_{\parallel}}$. Thus the anisotropy exponent
164: %
165: \be
166: z=\frac{\nu_{\parallel}}{\nu_{\perp}}\;
167: \label{z}
168: \ee
169: %
170: is generally different from one.
171: In the ordered phase, $t>0$, the number of points belonging to
172: the infinite cluster, $N_0$, scales around the transition point as:
173: %
174: \be
175: N_0=LK t^{\beta} \tilde{N}(L t^{\nu_{\perp}},K t^{\nu_{\parallel}})\;,
176: \label{N_0}
177: \ee
178: %
179: where $\beta$ is the critical exponent of the order parameter. At the critical point, $t=0$,
180: fixing the ratio $K/L^z=O(1)$ we obtain:
181: %
182: \be
183: N_0 \sim L^{d_{\perp}} \sim K^{d_{\parallel}}\;,
184: \label{D}
185: \ee
186: %
187: where the two fractal dimensions of the infinite cluster are given by:
188: %
189: \be
190: d_{\perp}=1+z-\beta/\nu_{\perp}\;,
191: \label{d_f}
192: \ee
193: %
194: and $d_{\parallel}=d_{\perp}/z$.
195: The distribution of cluster sizes, $R(N)$, at the critical point asymptotically
196: behaves as:
197: %
198: \be
199: R(N) {\rm d} N= N^{-\tau} \tilde{R}(N/L^{d_{\perp}}) {\rm d} N\;,
200: \label{PN}
201: \ee
202: %
203: where $\tau=2+\beta/(\nu_{\perp} d_{\perp})$. This relation can be obtained by generalizing the
204: similar result for ordinary percolation\cite{staufferaharony}.
205:
206: Correlation between two sites with coordinates, $\{i_1,j_1\}$ and $\{i_2,j_2\}$,
207: is defined as the expectation value of the connectivity,
208: $\delta(\{i_1,j_1\},\{i_2,j_2\})$, which is $1$, if the two sites belong to the
209: same cluster and zero otherwise. Here we mainly consider correlations in the
210: perpendicular direction
211: %
212: \be
213: C_{\perp}(i_1,i_2)=\frac{1}{K} \sum_{j=1}^{K}
214: [\langle \delta(\{i_1,j\},\{i_2,j\}) \rangle]_{\rm av} \;,
215: \label{Cperp}
216: \ee
217: %
218: where an average over the vertical coordinate, $j_1=j_2=j$ is also performed.
219: When correlations in the bulk are calculated we use periodic boundary
220: conditions (b.c.), (thus $i=L+1\equiv 1$), take maximal
221: distance between the sites, $i_2=i_1+L/2$, and
222: average over the position $i_1$.
223: The average bulk correlations, calculated in this way,
224: scale at the critical point as:
225: %
226: \be
227: C^b_{\perp}(L) \sim L^{-\eta_{\perp}}\;,
228: \label{Cb}
229: \ee
230: %
231: where $\eta_{\perp}=2 \beta/\nu_{\perp}$. We also considered the system with free
232: boundaries at $i=1$ and $i=L$ and calculated the correlations between two
233: surface sites. This end-to-end correlation function at the critical point
234: asymptotically behaves as:
235: %
236: \be
237: C_{\perp}(1,L) \equiv C^s_{\perp}(L) \sim L^{-\eta^s_{\perp}}\;,
238: \label{Cs}
239: \ee
240: %
241: where the decay exponent, $\eta^s_{\perp}$, is related to the surface fractal
242: properties of the infinite cluster.
243: Closing this section we quote the values of the critical exponents for two-dimensional
244: ordinary percolation\cite{staufferaharony}:
245: %
246: \be
247: \nu^{(0)}=4/3,\quad \eta^{(0)}=5/24,
248: \quad \eta^{s(0)}=2/3\;.
249: \label{exp0}
250: \ee
251: %
252:
253: \section{Strength of disorder: Limiting cases}
254:
255: The strength of disorder, $\Delta$, is related to the broadness of the probability
256: distribution, $P(p)$. In terms of the integrated probability distribution,
257: $\Pi(p)=\int_0^p P(p') {\rm d} p'$, we introduce the probabilities, $p_{1/4}$
258: and $p_{3/4}$ with the definitions: $\Pi(p_{1/4})=1/4$ and
259: $\Pi(p_{3/4})=3/4$. Since the central half of the distribution is located
260: in the region: $p_{1/4} \le p \le p_{3/4}$ its relative width is measured by:
261: %
262: \be
263: \Delta=\frac{p_{3/4}-p_{1/4}}{1-p_{3/4}+p_{1/4}}\;,
264: \label{Delta}
265: \ee
266: %
267: what we can identify with the strength of disorder.
268:
269: In this paper we used two specific forms of the distribution.
270: For the bimodal distribution ($0 \le q \le 1/2, \overline{q}=1-q$):
271: %
272: \be
273: P_{bin}(p)=\frac{1-t}{2} \delta(p-q)+\frac{1+t}{2} \delta(p-\overline{q})\;,
274: \label{bimodal}
275: \ee
276: %
277: the critical point is located at $t=0$ and the strength of disorder is
278: given by $\Delta_{bin}=(1-2q)/2q$. Thus, as
279: expected the bimodal disorder is weak for $q \approx 1/2$ and strong for
280: $q \ll 1/2$.
281:
282: The other distribution we use has a power-law form:
283: %
284: \be
285: P_{pow}(p)=\frac{1}{D 2 \overline{p}}\left( \frac{p}{\overline{p}}\right)^{-1+1/D} \qquad 0 < p <
286: \overline{p} < 1\;,
287: \label{prob}
288: \ee
289: %
290: and $P_{pow}(1-p)=P_{pow}(p)\overline{p}/(1-\overline{p})$, for $\overline{p}<p<1$. The distance
291: from the critical point is measured by $t=(\overline{p}-1/2)/(D+1)$,
292: and for $\overline{p}=1/2$, i.e. for $t=0$ the distribution is indeed symmetric.
293: In this case the strength of disorder is given by: $\Delta_{pow}=2^D-1$. Thus
294: for $D=0$ we recover the ordinary percolation and the strength of disorder is
295: monotonically increasing with $D$. Therefore $D$ will be often
296: called as the disorder parameter of the distribution.
297:
298: \subsection{Weak disorder}
299:
300: In the limit of weak disorder one usually decides about the relevance-irrelevance of the
301: perturbation by performing a stability analysis at the ordinary percolation fixed
302: point. Generalizing the method by Harris\cite{harris} the
303: cross-over exponent due to correlated disorder is calculated as:
304: %
305: \be
306: \phi=2-\nu^{(0)}=2/3\;,
307: \label{harris}
308: \ee
309: %
310: where we used $\nu^{(0)}=4/3$ in Eq.(\ref{exp0}). Since $\phi>0$, even weak
311: correlated disorder is a relevant perturbation, thus a new random fixed point is
312: expected to control the critical behavior of the model.
313:
314: \subsection{Strong disorder: Mapping to random walks}
315:
316: Next we turn to study the behavior of the system for extremely strong disorder using
317: the bimodal distribution in Eq.(\ref{bimodal}) in the limit $q \to 0$.
318: In this limiting case the percolation in a given layer with a probability $p_i$
319: has a simple, anisotropic structure (for an illustration see Fig. \ref{fig2}).
320: If this probability is extremely
321: large, $p_i=\overline{q}$, then here almost all bonds are occupied, except of a very
322: small fraction of $q$. Since the typical distance between two non-occupied bonds is
323: $l \sim 1/q$, the cluster in the $i$-th layer is composed
324: of long connected units of typical size $l$. On the other hand, if the probability
325: is extremely small, $p_j=q$, then almost all bonds in this layer are unoccupied,
326: except of a very small fraction of $q$. Since the typical distance between two
327: occupied bonds is $l \sim 1/q$, the cluster in the $j$-th
328: column is composed from long empty units of typical size $l$. Notice the duality in
329: the structure of the two types of column.
330:
331: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
332: \begin{figure}[tbh]
333: \epsfxsize=7truecm
334: \begin{center}
335: \mbox{\epsfbox{fig2.eps}}
336: \end{center}
337: \caption{\label{fig2} Structure of the percolation cluster in the extreme bimodal
338: distribution with $p_1=p_2=\overline{q}$ and $p_3=p_4=p_5=q$ (here with $q \approx 1/4$).
339: In a layer with extremely large (small) probability there are connected (empty) units of
340: typical length $l \sim 1/q$. The number of sites of the connected cluster at the other
341: surface of a strip of width, $k$,
342: $n_k$ is given by: $n_1 \sim 1/q$, $n_2 \sim 1/q^2$, $n_3 \sim 1/q$ and $n_4=O(1)$ (see text).
343: In the limit $q \to 0$ the cluster ends at $k=4$, thus $n_5=0$.}
344: \end{figure}
345: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
346:
347: With this prerequisite we consider the
348: order-parameter in the surface column, $m_s(L)$, which is the fraction of surface sites
349: belonging to a cluster of horizontal extent $L$. In order to make a statement about the
350: value of $m_s(L)$ we consider parallel strips of width $k \le L$ and introduce the
351: quantity, $n_k$, as the typical number of bonds at the $k$-th (i.e. surface) column of a cluster,
352: which is connected to the other surface of the strip. Starting with $k=1$ we have two
353: possibilities. For extremely small probability, $p_1=q$, there is no surface cluster
354: in the system, thus we have $n_1=0$.
355: Otherwise, for $p_1=\overline{q}$, a surface site is connected to all sites of a
356: ``connected unit'' of length $l$, thus we have $n_1 \sim l \sim 1/q$.
357: For $k=2$, if the probability is extremely large in the second layer, too,
358: $p_1=p_2=\overline{q}$, then a surface cluster extends up to the second layer and
359: its vertical size, which is given by $n_2$, can be estimated as follows (see Fig. \ref{fig2}).
360: The end of a cluster is signalled by the fact that in both columns unoccupied
361: bonds are in neighboring positions, which happens with a probability $q^2$,
362: from which the typical size of a cluster $n_2 \sim 1/q^2 \sim n_1/q$, follows.
363: Repeating this argument for $p_i=\overline{q}$, $i=1,2,\dots,k$ we obtain
364: $n_k \sim 1/q^k \sim n_{k-1}/q$. Now having a small probability at the following
365: layer, $p_i=\overline{q}$, $i=1,2,\dots,k$ and $p_{k+1}=q$, then only a fraction of
366: $q$ of the sites $n_k$ have a further connection, thus $n_k$ will be reduced by a
367: factor $q$ giving $n_{k+1} \sim n_k q$. Inclusion of any further layer with an
368: extremely small probability will reduce $n_j$ by a factor of $q$, until we arrive
369: at $n_{j'}=O(1)$, when for the next small probability layer we have $n_{j'+1}=0$, thus
370: the surface cluster ends at this distance.
371:
372: From this example we can read that the $n_k$ numbers are either integer powers of
373: $1/q$, $n_k \sim 1/q^{X_k}$, for $X_k=0,1,\dots$, or $n_k=0$, if formally $X_k<0$. Furthermore,
374: we have the transformation rules:
375: %
376: \be
377: n_{k+1} \sim \left\{ \begin{array}{ll}
378: n_k/q, & p_{k+1}=\overline{q} \cr
379: n_k q, & p_{k+1}=q \cr
380: \end{array}
381: \right.
382: \label{n_k}
383: \ee
384: %
385: where in the second case $n_{k+1}=0$, if $n_k=O(1)$.
386: At this point we can formulate the condition that the surface magnetization in a given
387: sample (in a rare realization) is $m_s(L)=O(1)$, if $n_k \ge O(1)$, for all
388: $k=1,2, \dots L$. For all other cases $m_s(L)=0$. Consequently to calculate
389: the {\it average value} of $m_s(L)$ it is enough to find the fraction of rare realizations,
390: $\rho_L^s$, for which $m_s(L)=O(1)$, since $[m_s]_{\rm av}\sim \rho_L^s$.
391: To calculate $\rho_L^s$ we use a random walk (RW) mapping (see an illustration in Fig. \ref{fig3}),
392: in which
393: to each disorder realization we assign a one-dimensional RW, which starts at $X_0=0$
394: and takes its $k$-th step upwards, $x_k=1$ (downwards, $x_k=-1$ ) if the corresponding bond
395: occupation probability is extremely large, $p_k=\overline{q}$ (extremely small, $p_k=q$).
396: The position of the walker at the
397: $k$-th step, $X_k=\sum_{i=1}^k x_k$ is related to $n_k$ as
398: $n_k \approx q^{-X_k}$. Then, as argued before, the surface cluster extends up to
399: a vertical distance, $L$, if $X_k \ge 0$, for every $k=1,2, \dots L$, i.e. the
400: RW has a surviving character.
401:
402: At the critical point of the percolation problem, $t=0$, the corresponding RW
403: is unbiased, and the fraction of surviving $L$-step RW-s
404: scales as $\rho_L^s \sim L^{-1/2}$. Now the fraction of clusters which
405: connect the two free boundaries of the strip over a distance $L$, and thus
406: contribute to the average end-to-end correlations in Eq.(\ref{Cs}),
407: is given by $(\rho^s_{L/2})^2$, since at each site there should be an independent
408: percolating surface cluster, which meet in the middle of the system.
409: Consequently the average end-to-end correlations at the critical point scale as
410: $C_{\perp}^s(L) \sim L^{-1}$ thus the corresponding decay exponent in the strong
411: disorder limit is given by:
412: %
413: \be
414: \eta^{s,(\infty)}_{\perp}=1\;.
415: \label{etas_inf}
416: \ee
417: %
418:
419: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
420: \begin{figure}[tbh]
421: \epsfxsize=7truecm
422: \begin{center}
423: \mbox{\epsfbox{fig3.eps}}
424: \end{center}
425: \caption{\label{fig3} Illustration of the RW mapping of percolation for a given realization of the
426: extreme binary distribution. Layers with high, $\overline{q}$, (low, $q$) probability are drawn
427: by thick (thin) lines and the corresponding RW makes a step of unit length upwards
428: (downwards). The position of the RW, in the $k$-th step, $X_k$, is related to, $n_k$, the number
429: of typical sites in the $k$-th layer of percolation, which are connected to a
430: given surface site as $n_k \sim q^{-X_k}$. The surface cluster extends to a distance, $L$,
431: if $X_k \ge 0$, for all $k=1,2, \dots L$, thus the RW has a surviving character.}
432: \end{figure}
433: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
434:
435: Using the RW mapping one can easily estimate the perpendicular size of the
436: percolating clusters, which is given by
437: $\xi_{\parallel}(L) \sim n_{L/2} \sim q^{X_{L/2}}$. Since the transverse fluctuations
438: of unbiased surviving RW-s scale as $X_{L/2} \sim L^{1/2}$ we obtain in the strong
439: disorder limit
440: %
441: \be
442: \ln \xi_{\parallel} \sim \xi_{\perp}^{1/2} \;.
443: \label{z_inf}
444: \ee
445: %
446: Consequently the anisotropy exponent, $z$, in Eq.(\ref{z}) is formally infinite
447: for strong disorder.
448:
449: Another results can be simply obtained by noticing that the same type of RW mapping
450: applies to the one-dimensional RTIM\cite{igloirieger98}, too,
451: so that we can simply borrow the results obtained in this case.
452:
453: For bulk correlations one should consider the fraction of realizations, $\rho_L$,
454: for which a given bulk site belongs to a connected cluster of vertical size, $L$.
455: As was shown in\cite{riegerigloi99} for these realizations the {\it thermal average} of
456: the position of the RW has a surviving character. The fraction of these walks is
457: given by\cite{fdm,riegerigloi99}: $\rho_L \sim L^{-(3-\sqrt{5})/4}$, consequently
458: the critical average bulk correlations being $C^b_{\perp}(L) \sim (\rho_L)^2$
459: have a decay exponent
460: %
461: \be
462: \eta_{\perp}^{(\infty)}=\frac{3-\sqrt{5}}{2}\;,
463: \label{eta_inf}
464: \ee
465: %
466: in the strong disorder limit.
467:
468: Finally, outside the critical point the mapping is related to a biased RW, with a finite
469: drift velocity, which is proportional to $t$. From the surviving probability of
470: biased RW-s one obtains for the correlation length critical exponent\cite{igloirieger98}:
471: %
472: \be
473: \nu_{\perp}^{(\infty)}=2\;,
474: \label{nu_inf}
475: \ee
476: %
477: The scaling exponents and relations in Eqs.(\ref{etas_inf}), (\ref{z_inf}),
478: (\ref{eta_inf}) and (\ref{nu_inf}) are identical with those of the IRFP of
479: the one-dimensional RTIM\cite{fisher92}, which is known to control the critical
480: behavior of several other random quantum spin chains\cite{senthil,cli01} and non-equilibrium
481: phase transitions in the presence of quenched disorder\cite{hiv02}.
482: At this point our next question is about the region of attraction of the IRFP.
483: For the RTIM, where the RW
484: mapping can be generalized for weaker disorder, any small amount of randomness seems to
485: bring the system into the IRFP\cite{fisher92}, which claim is checked by intensive
486: numerical calculations\cite{youngrieger,igloirieger98,riegerigloi99}.
487: There are, however, several other models (random quantum clock-model,
488: Ashkin-Teller model\cite{cli01}, directed percolation\cite{hiv02}, $S=1$ random
489: antiferromagnetic spin chains\cite{S=1}, etc.) where
490: weak disorder is not sufficient to bring the system into the IRFP.
491: In these cases either the pure systems fixed point stays stable against weak
492: disorder perturbations or the competition between (quantum)
493: fluctuations and weak quenched disorder leads to conventional random scaling behavior.
494: For the random percolation problem the latter scenario is likely to happen, since
495: the RW mapping can not be extended for small disorder. (The transformation law
496: for the connected sites, $n_k/n_{k-1} \approx q$ or $1/q$, does not hold around $q \approx 1/2$.)
497: We are going to study this issue numerically by MC
498: simulations in the next Section.
499:
500:
501: \section{Numerical results}
502:
503: For intermediate strength of disorder we studied the percolation by MC simulations.
504: Since the critical properties of the problem are related to the connectivity
505: properties of clusters for this purpose we implemented the standard Hoshen-Koopelman
506: labelling algorithm\cite{hk}. To decide about the shape of the lattice
507: one should take into account the expected anisotropic scaling properties of the
508: system, since the scaling functions, as in Eq.(\ref{N_0}) depend on the ratio
509: $r=L^z/K$, where $z$ is an unknown parameter. To overcome this difficulties
510: we used a strip-like geometry, when $K \gg L$, thus $r \approx 0$ for
511: all strip widths. In practice we had $K=10^5$, went up to $L=64$ and imposed
512: periodic b.c. in the vertical direction.
513: For the distribution of the disorder we used the power-law form in
514: Eq.(\ref{prob}), which has already been turned out successful in similar investigations for
515: random quantum spin chains\cite{cli01}. Since averaging in the vertical
516: direction in Eq.(\ref{Cperp}) (and also in the horizontal direction for bulk
517: correlations) is equivalent to a partial average over quenched disorder it was enough to
518: consider only a limited number ($\sim 10-20$) realizations.
519:
520: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
521: \begin{figure}[tbh]
522: \epsfxsize=8truecm
523: \begin{center}
524: \mbox{\epsfbox{fig4.eps}}
525: \end{center}
526: \caption{\label{fig4} Estimates for the anisotropy exponent for different
527: strength of disorder. The straight lines connecting the points are guide
528: to the eye, for $D>D_{\infty}\approx 1.2-1.5$ the anisotropy exponent
529: is possibly divergent. In the inset extrapolation of the size-dependent
530: effective anisotropy exponents is shown, for $D=0.25, 0.5, 0.75$ and $1.0$,
531: up to dawn.}
532: \end{figure}
533: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
534:
535: First, we determine the anisotropy exponent, $z$, by calculating the probability distribution
536: of clusters in Eq.(\ref{PN}). While the decay exponent, $\tau$ in Eq.(\ref{PN})
537: has only a weak anisotropy dependence, the scaling function $\tilde{R}(y)$
538: turned out to be sensitive of the value of $z$. As we noticed in the numerical
539: calculations $\tilde{R}(y)$ has two different regimes. For smaller values of
540: the parameter, $y=N/L^z<y^*$, the finite size effects are negligible and
541: the scaling function is approximately constant. For $y>y^*$, when the largest
542: clusters touch the boundaries, the scaling function has a characteristic variation.
543: Measuring the position
544: of $y^*$ for different widths, $L$, we obtained a series of effective anisotropy
545: exponents, which are then extrapolated to $L \to \infty$, as shown in the inset to
546: Fig. \ref{fig4}. This procedure is repeated for several disorder parameters and the extrapolated
547: anisotropy exponents are plotted in Fig. \ref{fig4}, Unfortunately, with this method we could not
548: go to very strong disorder, while the cross-over region can not be clearly located
549: for $D>1$. However, it is clear from the available data that $z$ is monotonically
550: increasing with the strength of disorder and it is likely
551: that $z$ will be divergent for $D>D_{\infty}\approx 1.2-1.5$.
552:
553: In order to obtain more information about the critical behavior of the system we have
554: calculated the bulk and the end-to-end average correlation functions at the
555: critical point, as defined in Eqs.(\ref{Cb}) and (\ref{Cs}), respectively.
556: In Fig. \ref{fig5} the average bulk correlations, $C_{\perp}^b(L)$,
557: vs. $L$ is drawn in a log-log plot. The slope of the curves, which is related
558: to the decay exponent, $\eta_{\perp}$, has a disorder dependence.
559:
560: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
561: \begin{figure}[tbh]
562: \epsfxsize=8truecm
563: \begin{center}
564: \mbox{\epsfbox{fig5.eps}}
565: \end{center}
566: \caption{\label{fig5} Average bulk correlations vs. the width of the strip for
567: different strength of disorder, from $D=0$ to $D=1.75$ in units of $0.25$ from up
568: to dawn. The typical error is generally smaller than the size of the symbols for small $D$,
569: whereas for larger $D$ it is at most twice of the size of symbols.
570: The straight lines are least-square fits.}
571: \end{figure}
572: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
573:
574: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
575: \begin{figure}[tbh]
576: \epsfxsize=7truecm
577: \begin{center}
578: \mbox{\epsfbox{fig6.eps}}
579: \end{center}
580: \caption{\label{fig6} Bulk ($\eta_{\perp}$) and surface ($\eta_{\perp}^s$)
581: decay exponents versus the strength of disorder.
582: Values at the IRFP, as given in Eqs.(\ref{eta_inf}) and
583: (\ref{etas_inf}) are denoted by dashed lines. Two typical error
584: bars are also indicated. }
585: \end{figure}
586: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
587:
588: The exponents, calculated in this way together with the decay exponent of the
589: end-to-end correlations, $\eta_{\perp}^s$, are plotted in Fig. \ref{fig6}.
590: As seen in Fig. \ref{fig6} both exponents are monotonously increasing with the strength of
591: disorder and tend to saturate at the respective IRFP values, given in Eqs.(\ref{eta_inf})
592: and (\ref{etas_inf}). The value of disorder strength, where the saturation takes
593: place, within the error of the calculation, is the same for the two exponents and it
594: is compatible with the estimate, $D_{\infty}$, as
595: calculated from the divergence of the dynamical exponent in Fig. \ref{fig4}.
596:
597: We can thus conclude that the critical behavior of the random percolation process
598: has a weak-to-strong disorder cross-over. For weaker disorder, $D<D_{\infty}$,
599: what we call the {\it intermediate disorder regime}, the critical behavior of the
600: system is controlled by a line of conventional fixed points. Here the anisotropy
601: exponent is finite, and together with the order-parameter exponents, $\eta_{\perp}$
602: and $\eta_{\perp}^s$, monotonously increasing with the strength of disorder.
603: In the {\it strong disorder regime}, $D>D_{\infty}$, the critical behavior of
604: the system is controlled by the IRFP. Here the anisotropy exponent is formally
605: infinity and the other critical exponents have no disorder dependence.
606:
607: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
608: \begin{figure}[tbh]
609: \epsfxsize=8truecm
610: \begin{center}
611: \mbox{\epsfbox{fig7.eps}}
612: \end{center}
613: \caption{\label{fig7} Scaling plot of the bulk correlation function with
614: $\nu_{\perp}=2$ at a disorder strength $D=0.75$.}
615: \end{figure}
616: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
617:
618: The non-universal nature of the critical behavior in the intermediate disorder regime is
619: possibly connected to the presence of a marginal operator, which should have vanishing
620: anomalous dimension, $x_e=0$, in the entire disorder range, $0<D<D_{\infty}$.
621: In our case the disorder perturbation is connected to the local energy-density
622: operator, for which the marginality condition, according to the Harris criterion
623: in Eq.(\ref{harris}) requires the condition $\phi=0$, thus
624: $\nu_{\perp}=2$. To verify this scenario we have calculated
625: the average bulk correlation function, $C_{\perp}^b(L,t)$, outside the critical
626: point, at a disorder strength, $D=0.75$, which is in the middle of the intermediate
627: disorder regime.
628: According to scaling considerations
629: %
630: \be
631: C_{\perp}^b(L,t)=L^{-\eta_{\perp}} \tilde{C}(tL^{1/\nu_{\perp}})\;,
632: \label{Cb_scal}
633: \ee
634: %
635: thus from an optimal scaling collapse $\nu_{\perp}$ can be determined. As shown
636: in Fig. \ref{fig7} the scaling behavior of $C_{\perp}^b(L,t)$ is compatible with
637: the conjectured value of $\nu_{\perp}=2$ and thus with the marginality
638: condition.
639:
640: \section{Discussion}
641:
642: In this paper bond percolation is studied on the square lattice with strictly
643: correlated, layered randomness. The phase diagram of the problem as a function
644: of the strength of disorder contains two regions. For strong enough disorder
645: the critical properties of the model are controlled by an IRFP, the
646: properties of which are exactly known by a RW mapping. For weaker disorder,
647: in the intermediate disorder regime the critical behavior is found to be
648: controlled by a line of conventional random fixed points, where both the
649: anisotropy exponent and the order-parameter exponents are disorder dependent.
650: The correlation length exponent, however, stays constant at its marginal value.
651:
652: This type of critical behavior is very similar to that obtained in a class
653: of random quantum spin chains\cite{cli01,hiv02}. This close similarity
654: can be understood by noting the relation between percolation and the
655: $Q \to 1$ limit of the $Q$-state ferromagnetic Potts model.
656: With layered randomness the two-dimensional Potts
657: model in the Hamiltonian limit\cite{kogut} is equivalent to a quantum
658: Potts chain, with random couplings, $J_i$, and transverse fields, $h_i$.\cite{turbanigloi02},
659: the critical behavior of which can be studied by the SDRG method\cite{senthil}.
660: In this procedure the couplings and transverse fields are put in descending order
661: and the strongest terms are successively decimated out, whereas neighboring terms
662: are replaced by renormalized values. Decimating the
663: strongest coupling, say $J_2$, yields a new effective spin cluster in a renormalized
664: transverse field of strength:
665: %
666: \be
667: \tilde h=\frac{2}{Q}{h_1 h_2 \over J_2}\;,
668: \label{recursion}
669: \ee
670: %
671: where $h_1$ and $h_2$ are the original transverse fields acting at the two end-spins
672: of $J_2$.
673: Similarly, if, the spin in the strongest transverse field, $h_2$, is decimated out, then
674: a new renormalized coupling is generated between remaining spins, which is of the form in
675: Eq.(\ref{recursion}), by interchanging $h_i \leftrightarrow J_i$, which is due
676: to duality.
677:
678: If the disorder is strong enough, so that the system under renormalization is
679: in the attractive region of the IRFP, the model specific prefactor $2/Q$
680: in Eq.(\ref{recursion}) does not matter and the critical properties are
681: universal. The region of strong attraction of the IRFP, however, is limited
682: by $Q=2$, i.e. for the RTIM. For smaller values of $Q$, like in percolation, when
683: the prefactor in Eq.(\ref{recursion}) is larger than one,
684: for weak disorder some renormalized couplings
685: and transverse fields are larger than the decimated ones.
686: If this happens frequently, i.e. when the disorder is too weak, then the SDRG method
687: is no longer valid and the critical behavior of the model is expected to be controlled
688: by a conventional random fixed point. This is exactly what we obtained by MC
689: simulations.
690:
691: We close our paper with two remarks. First, for strong enough disorder the critical
692: behavior of both ordinary and directed percolation\cite{hiv02} is controlled by
693: the same IRFP, thus the original anisotropy between the two pure problems does not
694: make any influence about the (strongly) random critical behavior. Our second remark
695: concerns possible Griffiths effects in the random percolation problem. Using the analogy
696: with random quantum spin chains for strong disorder some dynamical quantities of
697: the random percolation problem are singular also outside the critical point.
698: For example the susceptibility in a uniform field, $H$ diverges as $\chi \sim H^{-1+1/z'}$,
699: and the vertical correlation function decays algebraically as $C_{\parallel}(l) \sim l^{-1/z'}$,
700: where $z'$ is a finite dynamical exponent, which depends on the distance of the
701: critical point.
702:
703: We are indebted to Jae Dong Noh, Heiko Rieger and Lo\"{\i}c Turban for stimulating
704: discussions. This work has been supported by the Hungarian National
705: Research Fund under grant No OTKA TO34183, TO37323,
706: MO28418 and M36803, by the Ministry of Education under grant No FKFP 87/2001,
707: by the EC Centre of Excellence (No. ICA1-CT-2000-70029) and the numerical
708: calculations by NIIF 1030. The Laboratoire de Physique des Mat\'eriaux
709: is Unit\'e Mixte de Recherche No 7556.
710:
711: \begin{references}
712:
713: \bibitem{staufferaharony}
714: For a review see, Stauffer and A. Aharony, {\it Introduction to Percolation Theory},
715: (Taylor and Francis, London) (1992).
716:
717: \bibitem{kasteleyn}
718: P.W. Kasteleyn and C.M. Fortuin, J. Phys. Soc. Jpn. {\bf 46} (suppl.),
719: 11 (1969).
720:
721: \bibitem{cardy96}
722: J.L. Cardy, in {\it Fluctuating Geometries in Statistical Mechanics and Field Theory},
723: Les Houches Summer School 1994 (North-Holland, 1996).
724:
725: \bibitem{cardyziff}
726: J.L. Cardy and R.M. Ziff, cond-mat/0205404.
727:
728: %\bibitem{BG}
729: % J.P. Bouchaud and A. Georges, Phys. Rep. {\bf 195}, 127 (1990).
730:
731: \bibitem{harris}
732: A.B. Harris, J. Phys. C{\bf 7}, 1671 (1974).
733:
734: \bibitem{weinrieb}
735: A. Weinrieb and B. I. Halperin, Phys. Rev. B {\bf 27}, 413 (1983).
736:
737: \bibitem{kinzel}
738: W. Kinzel, in {\it Percolation structures and processes}, ed. G. Deutscher,
739: R. Zallen and J. Adler, Ann. Isr. Phys. Soc. {\bf 5}, (Adam Hilger, Bristol, 1983),
740: p. 425.
741:
742: \bibitem{mccoywu}
743: B. M. McCoy and T. T. Wu, Phys. Rev. Lett. {\bf 23}, 383 (1968).
744:
745: \bibitem{boyanovskicardy}
746: D. Boyanovski and J.L. Cardy, Phys. Rev. B {\bf 26}, 154 (1982).
747:
748: \bibitem{MDH}
749: S. K. Ma, C. Dasgupta, and C.-K. Hu, Phys. Rev. Lett. {\bf 43}, 1434 (1979);
750: C. Dasgupta and S. K. Ma, Phys. Rev. B {\bf 22},1305 (1980).
751:
752: \bibitem{fisher99}
753: D.S. Fisher, Physica A {\bf 263}, 222 (1999).
754:
755: \bibitem{fisher92}
756: D.S. Fisher, Phys. Rev. Lett. {\bf 69}, 534 (1992);
757: Phys. Rev. B {\bf 51} 6411 (1995).
758:
759: \bibitem{igloi02}
760: F. Igl\'oi, Phys. Rev. B {\bf 65}, 064416 (2002).
761:
762: \bibitem{senthil}
763: T. Senthil and S.N. Majumdar, Phys. Rev. Lett. {\bf 76}, 3001 (1996).
764:
765: \bibitem{fisher94}
766: D.S. Fisher, Phys. Rev. B {\bf 56}, 3799 (1994).
767:
768: \bibitem{S=1}
769: R.A. Hyman and K. Yang, Phys. Rev. Lett. {\bf 78}, 1783 (1997);
770: C. Monthus, O. Golinelli, and Th. Jolicoeur, {\it ibid.} {\bf 79}, 3254 (1997).
771:
772: \bibitem{mllri}
773: R. Melin, Y.-C. Lin, P. Lajk\'o, H. Rieger and F. Igl\'oi, Phys. Rev. B {\bf 65}, 104415 (2002).
774:
775: \bibitem{hiv02}
776: J. Hooyberghs, F. Igl\'oi and C. Vanderzande, preprint cond-mat/0203610.
777:
778: \bibitem{cli01}
779: E. Carlon, P. Lajk\'o, and F. Igl\'oi, Phys. Rev. Lett. {\bf 87}, 277201 (2001).
780:
781: \bibitem{kinzeldomany}
782: W. Kinzel and E. Domany, Phys. Rev. B{\bf 23}, 3421 (1981).
783:
784: \bibitem{binderwang}
785: K. Binder and L.S. Wang, J. Stat. Phys. {\bf 55}, 87 (1989).
786:
787: \bibitem{igloirieger98}
788: F. Igl\'oi and H. Rieger, Phys. Rev. B{\bf 57}, 11404 (1998).
789:
790: \bibitem{riegerigloi99}
791: H. Rieger and F. Igl\'oi, Europhys. Lett. {\bf 45}, 673 (1999).
792:
793: \bibitem{fdm}
794: D.S. Fisher, P. Le Doussal and C. Monthus, Phys. Rev. Lett. {\bf 80},
795: 3539 (1998).
796:
797: \bibitem{youngrieger}
798: A.P. Young and H. Rieger, Phys. Rev. B{\bf 53}, 8486 (1996).
799:
800: \bibitem{hk}
801: J. Hoshen and R. Kopelman, Phys. Rev. B{\bf 14}, 3428 (1976).
802:
803: \bibitem{kogut}
804: J. Kogut, Rev. Mod. Phys. {\bf 51}, 659 (1979).
805:
806: \bibitem{turbanigloi02}
807: See e.g. in L. Turban and F. Igl\'oi, preprint cond-mat/0110399.
808:
809: \end{references}
810: \end{multicols}
811: \end{document}