1: \documentclass[twocolumn, showpacs, preprintnumbers, amsmath, amssymb]{revtex4}
2: \usepackage{graphicx}% Include figure files
3: \usepackage{dcolumn}% Align table columns on decimal point
4: \usepackage{bm}% bold math
5: \usepackage{color}
6: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
7: \begin{document}
8: %--------------------------------------------------------------------
9: \title{Self-Affinity in the Gradient Percolation Problem}
10: %--------------------------------------------------------------------
11:
12: \author{Alex Hansen}
13: \email[Alex.Hansen@ntnu.no]{}
14: \affiliation{Department of Physics, Norwegian University of Science and
15: Technology, N--7491 Trondheim, Norway}
16:
17: \author{G.\ George Batrouni} \email[George.Batrouni@inln.cnrs.fr]{}
18: \altaffiliation{INLN, UMR CNRS 6618, Universit{\'e} de Nice-Sophia
19: Antipolis, 1361 route des Lucioles, F--06560 Valbonne, France}
20: \affiliation{Department of Physics, Norwegian University of Science
21: and Technology, N--7491 Trondheim, Norway}
22:
23: \author{Thomas Ramstad}
24: \email[Thomas.Ramstad@phys.ntnu.no]{}
25: \affiliation{Department of Physics, Norwegian University of Science and
26: Technology, N--7491 Trondheim, Norway}
27:
28: \author{Jean Schmittbuhl}
29: \email[Jean.Schmittbuhl@eost.u-strasbg.fr]{} \affiliation{Institut de
30: Physique du Globe de Strasbourg, UMR CNRS 7516, 5, rue Ren{\'e}
31: Descartes, F--67084 Strasbourg, France}
32:
33: %--------------------------------------------------------------------
34: \date{\today}
35: %--------------------------------------------------------------------
36: \begin{abstract}
37: We study the scaling properties of the solid-on-solid front of the
38: infinite cluster in two-dimensional gradient percolation. We show that
39: such an object is self affine with a Hurst exponent equal to 2/3 up to
40: a cutoff-length $\sim g^{-4/7}$, where $g$ is the gradient. Beyond
41: this length scale, the front position has the character of
42: uncorrelated noise. Importantly, the self-affine behavior is robust
43: even after removing local jumps of the front. The previously observed
44: multi affinity, is due to the dominance of overhangs at small
45: distances in the structure function. This is a crossover effect.
46: \end{abstract}
47: %--------------------------------------------------------------------
48: \pacs{05.40.+j, 02.50.-r, 47.55.Mb, 64.60.Ak}
49: %--------------------------------------------------------------------
50: \maketitle
51: %--------------------------------------------------------------------
52: Rough surfaces showing non-trivial scaling properties have been
53: extensively studied theoretically, numerically and experimentally over
54: the last couple of decades. Examples of such surfaces are those
55: appearing during brittle fracture \cite{b97} which were first
56: characterized as being fractal \cite{mpp84-bs85} but it was then
57: realized that the concept of self affinity was more appropriate
58: \cite{blp90-mhhr92}. The question of self affinity versus fractality
59: has also been the focus of intense research on invasion fronts in
60: porous media and on the dynamics of magnetic domain walls
61: \cite{cr88-cr90-mcr91-jr92}. It was recently reported that the
62: displacement fronts in self-affine fractures are self affine
63: \cite{dakh04}. More recently, a possible explanation for the observed
64: self affinity of fracture surfaces has been proposed and hinges on a
65: clear understanding of the distinction between fractality and self
66: affinity \cite{hs03a-hs03b,az04}. It has also been suggested that
67: brittle fracture surfaces are multi affine rather than simply self
68: affine \cite{sss95-bpsv06}. Whether this is so remains an open
69: question \cite{smmhsvdbr06}.
70:
71: %--------------------------------------------------------------------
72: \begin{figure}[b]
73: \includegraphics[width=7cm,clip]{fig1.eps}
74: \caption{Top-side and bottom-side SOS fronts based on
75: the perimeter of the cluster connected to the $p=1$ edge (shown in
76: the insert as dots). We also show the filtered $j_0(i)$ front
77: (see Eq.\ (\ref{transform})).}
78: \label{fig1}
79: \end{figure}
80: %--------------------------------------------------------------------
81:
82: It is the aim of this Letter to study the question of fractality, self
83: affinity and multi affinity of a front in a system which is simple
84: enough to be tractable, namely that of the gradient percolation
85: \cite{srg85}. There are already in the literature studies of this
86: system in the present context. Furuberg {\it et al.\/} \cite{fhhfj91} study
87: the jumps in the position of the solid-on-solid (SOS) front of the
88: infinite cluster, whereas Asikainen {\it et al.\/} \cite{amda02} conclude that
89: this front is multi affine. We will in this Letter show that up to a
90: given scale, the SOS front is {\it self affine\/} with a well-defined
91: Hurst exponent, whereas on larger scales its position becomes
92: uncorrelated. The self affinity is {\it not\/} caused by the jumps in
93: the position of the front due to overhangs, but related to its fractal
94: structure. The multi affinity seen by Asikainen {\it et al.\/} has its
95: origin in the overhangs resulting from the definition of the SOS
96: fronts and shows up in the structure function on small scales.
97:
98: In {\it gradient\/} percolation, a spatial gradient in the occupation
99: probability $p$ is introduced. A cartesian coordinate system $(i,j)$
100: is oriented with respect to the finite lattice of size $L_i\times L_j$
101: (assuming for the rest of this paper that the lattice is two
102: dimensional), so that the $i$ axis runs perpendicular to the gradient
103: (i.e.\ along the lower edge) and the $j$ axis along the gradient
104: (i.e.\ the left edge). The gradient is introduced in the $j$ direction
105: so that $p(j)=gj$, where the gradient $g=1/L_j$. However, the cluster
106: connected to the lower edge will reach some average value, $j=j_g$,
107: with an associated occupation probability $p_g=gj_g$. The region
108: around $j_g$ is critical and has a width $\xi$, spanning between
109: $j_{\pm}=j_g\pm\xi/2$, where $\xi$ is the correlation length
110: associated with the critical region in the direction of the gradient.
111: Defining $p_{\pm}=gj_{\pm}$ and setting
112: $\xi=|p_{\pm}-p_g|^{-\nu}=|g(j_{\pm}-j_g)|^{-\nu}$ where $\nu$ is the
113: correlation length exponent, Sapoval {\it et al.\/} \cite{srg85} found that
114: $\xi\sim g^{-\nu/(1+\nu)}=g^{-4/7}$.
115:
116: %--------------------------------------------------------------------
117: \begin{figure}[t]
118: \includegraphics[width=7cm,clip]{fig2.eps}
119: \caption{Data collapse of the averaged wavelet coefficients for the
120: bottom side front based on lattice size $L_j=64$ to 8192, while
121: $L_i=2048$. We have that $g=1/L_j$. The straight line has a slope of
122: $\zeta+1/2$, see Eq.\ (\ref{awc}).}
123: \label{fig2}
124: \end{figure}
125: %--------------------------------------------------------------------
126:
127: The infinite cluster has a fractal structure with an upper cutoff in
128: length scale set by the width of the critical region, $\xi$. We now
129: focus on the front of this infinite cluster and define precisely what
130: we mean by this front $j(i)$ in the gradient percolation problem. Our
131: starting point is the perimeter of the cluster of occupied sites that
132: is attached to the $p=1$ edge of the lattice. Since this perimeter
133: contains overhangs and therefore is multivalued when interpreted as a
134: function $j(i)$, we use the SOS method to extract a single-valued
135: function for its position, see Fig.\ \ref{fig1}. For each $i$, we use
136: either the $j$ value that is closest to the $p=0$ edge (top side) or
137: the $j$ value which is closest to the $p=1$ side (bottom side) or the
138: average over all the $j$ values attached to a given $i$ value (average
139: front).
140:
141: A trace $j(i)$ is statistically self affine if the probability
142: density, $\pi(i,j)$, for it to have a value $j$ at $i$, given that
143: $j=0$ at $i=0$, has the invariance
144: \begin{equation}
145: \label{selfaffine}
146: \lambda^\zeta \pi (\lambda i,\lambda^\zeta j)=\pi(i,j)\;,
147: \end{equation}
148: where $\zeta$ is the Hurst exponent. This invariance must be caused
149: by {\it spatial correlations} in $j$ along the $i$-axis. We note that
150: a L{\'e}vy flight, which is an uncorrelated random walk whose step
151: size $h$ is drawn from a power law distribution $N(h)\sim
152: h^{-\beta-1}$, will satisfy Eq.\ (\ref{selfaffine}) with an apparent
153: Hurst exponent $\zeta=1/\beta$. However, in this example, satisfying
154: Eq.\ (\ref{selfaffine}) is due to the step size distribution and not
155: to spatial correlations \cite{hm06}.
156:
157: We have used the Average Wavelet Coefficient (AWC) method
158: \cite{mrs97,shn98} to analyse the structure of the SOS fronts. The
159: AWC method consists of wavelet transforming $j(i)$, and averaging
160: the wavelet coefficients $w(b,a)$ at each length scale $a$ over
161: position $b$, $W(a)=\langle w(b,a)\rangle_b$. If $j(i)$ is self
162: affine, the averaged wavelet coefficients will scale as
163: \begin{equation}
164: \label{awc}
165: W(a) \sim a^{\zeta+1/2}\;.
166: \end{equation}
167: We show in Fig.\ \ref{fig2}, the averaged wavelet coefficients based
168: on the Daubechies-4 wavelets for the bottom-side fronts. The plots
169: for the top and average fronts are comparable.
170: The data are based on averages over
171: 2201 samples for $L_j$ in the range 64 to 2048 and 200 samples for
172: $L_j=4096$ and 8192. $L_i$ was set to 2048 for all the different
173: $L_j$. The gradient $g$ was set to $1/L_j$. There is a clear
174: crossover between two regimes in these plots. At smaller length
175: scales, one does indeed find the behavior of Eq.\ (\ref{awc})
176: indicating self affinity. On larger scales, the slope of the log-log
177: plots are zero indicating $\zeta=-1/2$, which corresponds to
178: uncorrelated or white noise \cite{hsb01}. Furthermore, we observe
179: excellent data collapse when $W$ is scaled by $g^{-\beta}$ and the length
180: scale, $a$, is scaled by $g^{-\alpha}$. We will show below that
181: \begin{equation}
182: \label{zeta}
183: \zeta=2-D_e =\frac{2}{3}\;,
184: \end{equation}
185: \begin{equation}
186: \label{alpha}
187: \alpha=\frac{\nu}{1+\nu}=\frac{4}{7}\;,
188: \end{equation}
189: and
190: \begin{equation}
191: \label{beta}
192: \beta=\frac{3}{2}\ \alpha=\frac{6}{7}\;.
193: \end{equation}
194: where $D_e=4/3$ is the fractal dimension of the external perimeter of the
195: front \cite{ga86-ga87}.
196:
197: %--------------------------------------------------------------------
198: \begin{figure}[t]
199: \includegraphics[width=7cm,clip]{fig3.eps}
200: \caption{Data collapse of the averaged wavelet coefficients for the
201: smoothed $j_0(i)$ based on the bottom side front. The lattice sizes
202: and gradients are as in Fig.\ \ref{fig2}. The long-dashed line has a
203: slope of 7/6, see Eq.\ (\ref{awc}), whereas the dotted line has a
204: slope of 1=1/2+1/2, consistent with uncorrelated random walks.}
205: %Note the change of scaling along the $j$ axis.}
206: \label{fig9}
207: \end{figure}
208: %--------------------------------------------------------------------
209:
210: The main goal of this Letter is to derive Eq.\ (\ref{zeta}) and thus
211: demonstrate that $\zeta$ is a proper Hurst exponent and $j(i)$ a
212: self-affine function. To this end, we need to demonstrate two things:
213: First, $j(i)$ satisfies the scaling relation (\ref{selfaffine}) and,
214: second,that this is not due to a power law tail in the step size
215: distribution. We note that since the average wavelet coefficients
216: obey Eq.\ (\ref{awc}), $j(i)$ automatically satisfies Eq.\
217: (\ref{selfaffine}). Therefore, we now need only to identify the
218: mechanism behind this scaling.
219:
220: In order to derive Eq.\ (\ref{zeta}), we start by noting that the
221: distribution of distances $m$ between crossing points between a planar
222: fractal curve with dimension $D_e$ --- e.g., the percolation perimeter
223: --- and a straight line follows the power law $\pi(m)\sim m^{-D_e}$
224: \cite{hme94}. Introducing a gradient in the $j$ direction and placing
225: the straight line in the critical region interval, $[j_-,j_+]$, and
226: parallel to the $i$ axis, the distribution of crossing point distances
227: $m$ remains the same. A self-affine curve characterized by a Hurst
228: exponent $\zeta$, leads to a distribution of crossing point
229: distances given by $\pi(m)\sim m^{-(2-\zeta)}$ \cite{hme94}. By comparing
230: this expression to $\pi(m)\sim m^{-D_e}$,
231: Eq.\ (\ref{zeta}) immediately follows. However, we still need to show
232: that $j(i)$ is indeed self affine, in other words the scaling relation
233: Eq.\ (\ref{selfaffine}) is not caused by jumps.
234:
235: First, we turn to deriving Eqs.\ (\ref{alpha}) and (\ref{beta}). The
236: correlation length in the direction of the gradient, the $j$
237: direction, is $\xi\sim g^{-\nu/(1+\nu)}$. Since the perimeter is
238: locally isotropic, this is also the correlation length in the $i$
239: direction. The crossover length scale from self affinity to
240: uncorrelated noise is the correlation length $\xi$. Hence, rescaling
241: $a \to a/\xi\sim a/g^{-\nu/(1+\nu)}$ gives data collapse along this
242: axis which demonstrates Eq.\ (\ref{alpha}). Likewise, the crossover
243: length scale in the $j$ direction is $\xi$. This implies that the
244: normalized wavelet coefficient at this scale, $W(\xi)/\xi^{1/2}$ is
245: equal to $\xi$. Hence, $W(\xi)\sim \xi^{3/2} \sim g^{-(3/2)\alpha}
246: \sim g^{-\beta}$, and $\beta=(3/2)\alpha= 6/7$, as stated in Eq.\
247: (\ref{beta}).
248:
249: In order to show that $j(i)$ is a self-affine function, we need to
250: demonstrate that $\zeta$ is not caused by the step size distribution.
251: To this end, we define the following transformation of the function
252: $j(i)\to j_k(i)$ where we factorize the function in such away that we
253: can distinguish the respective roles of persistency and step sizes,
254: \begin{equation}
255: \label{transform}
256: j_k(i)=\sum_{m=0}^i {\rm sign}[j(m+1)-j(m)]\ |j(m+1)-j(m)|^k\;,
257: \end{equation}
258: where $|j(m+1)-j(m)|=h(m)$ is the step size at position $m$.
259: We have in particular that $j_1(i)=j(i)$.
260: It was shown in \cite{fhhfj91}, that $h$ is distributed according to
261: \begin{equation}
262: \label{overhang}
263: N(h,g)=h^{-D_e-1}f(hg^\alpha)\;,
264: \end{equation}
265: where $D_e=4/3$ and $f(z)$ approaches a constant as $z\ll 1$ and falls
266: off faster than any power law as $z\to\infty$. The step size
267: distribution comes from the appearance of overhangs in the perimeter.
268: An overhang is defined as the jump made by the front from one position
269: along the $i$ axis to the next due to a backwards turn
270: \cite{hahw90,fhhfj91,bh92}. In order to confirm that the overhangs do
271: not generate the Hurst exponent $\zeta=2/3$, we analyse the filtered
272: front $j_0(i)$, defined in Eq.\ (\ref{transform}). With $k=0$, we
273: eliminate the overhangs all together \cite{bmr05}. Fig.\
274: \ref{fig9} shows the data collapse based on $j_0(i)$ corresponding to
275: the bottom side $j(i)$ shown in Fig.\ \ref{fig2}. The scaling
276: along the $i$ axis
277: is unchanged as no change in the system has been made in that
278: direction. However, since all step sizes have been reset to unity in
279: the transformation $j(i)\to j_0(i)$, the rescaling in the $j$
280: direction is no longer controlled by $j_c$. In order to regain data
281: collapse for different $g=1/L_j$, we need to rescale the lattice units
282: in this direction by the Hurst exponent, $\zeta=2/3$. The straight
283: line matching the small-$a$ region of the figure has a slope
284: $2/3+1/2$, while the straight line matching the large-$a$ portion has
285: a slope of $1/2+1/2$ corresponding to an uncorrelated random walk.
286: This shows that, for small scales, the $\zeta=2/3$ is indeed a Hurst
287: exponent. On the other hand, for longer length scales, we expect
288: random walk behavior since white noise gives precisely the exponent
289: $1/2$ in the transformation $j(i)\to j_0(i)$.
290:
291: In order to analyse the multi affinity that has been reported in this
292: problem \cite{amda02}, we construct the structure function
293: $C_k(n,g)=\langle |j(m+n)-j(m)|^k\rangle$. Multi affinity occurs when
294: $C_k(n,g)^{1/k}$ does {\it not\/} scale with a single $k$-independent
295: exponent with respect to $n$. Using the overhang distribution
296: (\ref{overhang}), we find $C_k(1,g)\sim g^{s(k)}$, where $s(k)
297: =\min[0,\alpha(D_e-k)]=\min[0,(16/21-4 k/7)]$. The self-affine
298: character of $j(i)$ cannot be visible in the structure function for
299: $n=1$ but will appear only gradually as $n$ is increased. We may
300: therefore analyse the structure function based solely on the L{\'e}vy
301: character induced by the overhangs in the small-$n$ limit. We will
302: call this the L{\'e}vy regime, whereas for larger $n$ where the self
303: affinity dominates, we will refer to as the self-affine regime. The
304: scaling with respect to $g$ for $C_k(1,g)$ persists for $n>1$ in the
305: L{\'e}vy regime since $j(i+n)-j(i)$ follows a L{\'e}vy distribution
306: whose power law tail does not change with increasing $n$. Hence, we
307: expect $C_k(n,g)\sim g^{s(k)}$ in this regime. In order to derive its
308: dependence on $n$ in the L{\'e}vy regime, we note that the
309: distribution of distances $l$ between overhangs follows the
310: same power law as the overhangs themselves. This can be seen as
311: follows. When there is a gradient present in the $j$ direction, the
312: length of the perimeter scales as $L_i^{D_e}$, when the gradient is
313: kept fixed. Making a cut through the perimeter with a straight line
314: parallel to the $i$ axis, the crossing points of the perimeter with
315: the line form a fractal set with dimension $D_e-1$. Hence, there are,
316: in a given interval $l$, $N_l \sim l^{D_e-1}$ overhangs \cite{hme94}.
317: These overhangs give rise to an effective Hurst exponent $1/D_e=3/4$
318: on the fractal set, seen e.g.\ in the width of the trace, $\Delta j
319: \sim N_l^{1/D_e}\sim l^{(D_e-1)/D_e}$. Since the overhangs form a
320: fractal set, we will need $N_b \sim l^{-(D_e-1)}$ boxes of size $l$ to
321: cover it. Due to the averaging over position $i$, there will be yet
322: another factor $l$, see \cite{m05}. We may now assemble these pieces
323: to form the scaling of the structure function in the L{\'e}vy regime,
324: $C_k(l,g)\sim N_l^k N_b l \sim l^{k\zeta_k^L}$ where
325: \begin{equation}
326: \label{affen}
327: \zeta_k^L= \left[1-\frac{1}{D_e}\right]+\frac{2-D_e}{k} = \frac{1}{4}
328: +\frac{2}{3k}\;.
329: \end{equation}
330: Therefore, in the L{\'e}vy regime, i.e.\ for small $n$, there is multi
331: affinity. A similar analysis in the self-affine regime, i.e.\ at
332: larger $n$, yields
333: \begin{equation}
334: \label{chimps}
335: \zeta_k^{SA} = \zeta = \frac{2}{3}\;.
336: \end{equation}
337: Therefore there is no multi affinity in this regime.
338: The $n$ for which there is the
339: crossover between the L{\'e}vy and the self-affine regime will depend on
340: $k$ and is governed by prefactors that the scaling analysis presented here
341: cannot access. For $n$ beyond $\xi$, the front decorrelates
342: and the structure function becomes independent of $n$.
343: We show in Fig.\ \ref{fig10}, the $k=1$, 2 and 3 structure functions. Their
344: behavior is in accordance with our predictions. However, note that for $k=2$,
345: $\zeta_2^L=7/12=0.58$ which is close to $\zeta=2/3$. Furthermore, the
346: self-affine regime is close to the decorrelated flat regime. Hence, it is
347: hard to distinguish between the L{\'e}vy and the self-affine regime for this
348: value of $k$. As $k$ increases, the L{\'e}vy regime grows, as the overhangs
349: are emphasized for larger $k$.
350:
351: %--------------------------------------------------------------------
352: \begin{figure}[t]
353: \includegraphics[width=7cm,clip]{fig4.eps}
354: \caption{$C_k(n,g)$ as a function of $n$ for $k=1$, 2 and 3. The
355: three leftmost straight lines have slopes according to Eq.\ (\ref{affen}),
356: while the bold middle line has a slope equal to 2/3 in accordance with
357: Eq.\ (\ref{chimps}). For large $n$ the structure functions become flat
358: indicating that one has reached the decorrelated regime.}
359: \label{fig10}
360: \end{figure}
361: %--------------------------------------------------------------------
362:
363: To conclude, we have shown that the structure of the interface in a
364: gradient percolation problem combines fractal and self-affine
365: properties. The perimeter that includes numerous overhangs has the
366: classical fractal structure \cite{srg85}. However, Solid-on-Solid
367: fronts that are extracted from the perimeter, have a clear self-affine
368: property up to a crossover length scale $\xi$ even if local jumps,
369: inherited from overhangs, are removed. On larger scales it shows an
370: uncorrelated noise behavior. The structure function is, however,
371: sensitive to the overhangs on smaller scales and this implies a
372: multi-affine scaling behavior in this regime. Implications of our
373: results for physical interpretations of analogical and numerical
374: experiments are important.
375:
376: We thank M.\ K.\ Alava and S.\ Zapperi for stimulating comments leading
377: to this work.
378: G.\ G.\ B.\ thanks NTNU and Norsk Hydro for his appointment as the Lars
379: Onsager Professor for 2004.
380: %---------------------------------------------------------------------
381: \begin{thebibliography}{99}
382:
383: \bibitem{b97} E.\ Bouchaud, J.\ Phys.\ Condens.\ Matt.\ {\bf 9},
384: 4319 (1997).
385: {\bf 55}, 349 (2006).
386:
387: \bibitem{mpp84-bs85} B.\ B.\ Mandelbrot, D.\ E.\ Passoja, and A.\ J.\
388: Paullay, Nature, {\bf 308}, 721 (1984); S.\ R.\ Brown and C.\ H.\
389: Scholz, J.\ Geophys.\ Res.\ {\bf 90}, 12575 (1985).
390:
391: \bibitem{blp90-mhhr92} E.\ Bouchaud, G.\ Lapasset,
392: and J.\ Plan{\'e}s, Europhys.\ Lett.\ {\bf 13}, 73 (1990); K.\ J.\
393: M{\aa}l{\o}y, A.\ Hansen, E.\ L.\ Hinrichsen, and S.\ Roux, Phys.\
394: Rev.\ Lett.\ {\bf 68}, 213 (1992).
395: (1995).
396:
397: \bibitem{cr88-cr90-mcr91-jr92} M.\ Cieplak and M.\ O.\ Robbins, Phys.\ Rev.\
398: Lett.\ {\bf 60}, 2042 (1988); M.\ Cieplak and M.\ O.\ Robbins, Phys.\ Rev.\ B,
399: {\bf 41}, 11508 (1990); N.\ Martys, M.\ O.\ Robbins and M.\ Cieplak,
400: Phys.\ Rev.\ B {\bf 44}, 12294 (1991); H.\ Ji and M.\ O.\ Robbins, Phys.\
401: Rev.\ B,{\bf 46}, 14519 (1992).
402:
403: \bibitem{dakh04} G.\ Drazer, H.\ Auradou, J.\ Koplik and J.\ P.\ Hulin,
404: Phys.\ Rev.\ Lett.\ {\bf 92}, 014501 (2004).
405:
406: \bibitem{hs03a-hs03b}
407: A.\ Hansen and J.\ Schmittbuhl, Phys.\ Rev.\ Lett.\ {\bf 90},
408: 045504 (2003); J.\ Schmittbuhl, A.\ Hansen and G.\ G.\ Batrouni, Phys.\
409: Rev.\ Lett.\ {\bf 90}, 045505 (2003).
410:
411: \bibitem{az04} M.\ J.\ Alava and S.\ Zapperi, Phys.\ Rev.\ Lett.\ {\bf 92},
412: 049601 (2004); J.\ Schmittbuhl, A.\ Hansen and G.\ G.\ Batrouni, Phys.\ Rev.\
413: Lett.\ {\bf 92}, 049602 (2004).
414:
415: \bibitem{sss95-bpsv06} J.\ Schmittbuhl, F.\ Schmitt and C.\ Scholz, J.\
416: Geophys.\ Res.\ {\bf 100}, 5953 (1995); E.\ Bouchbinder, I.\ Procaccia,
417: S.\ Santucci and L.\ Vanel, Phys.\ Rev.\ Lett.\ {\bf 96}, 055509 (2006).
418:
419: \bibitem{smmhsvdbr06} S.\ Santucci, J. Mathiesen, K.\ J.\ M{\aa}l{\o}y,
420: A.\ Hansen, J.\ Schmittbuhl, L.\ Vanel, A.\ Delaplace, J.\ {\O}.\ H.\ Bakke
421: and P.\ Ray, Cond-mat/0607385.
422:
423: \bibitem{srg85} B.\ Sapoval, M.\ Rosso and J.\ F.\ Gouyet, J.\ Phys.\ Lett.\
424: (France) {\bf 46}, L149 (1985).
425:
426: \bibitem{fhhfj91} L.\ Furuberg, A.\ Hansen, E.\ L.\ Hinrichsen, J.\ Feder and
427: T.\ J{\o}ssang, Phys.\ Script.\ T {\bf 38}, 91 (1991).
428:
429: \bibitem{amda02} J.\ Asikainen, S.\ Majaniemi, M.\ Dub{\'e} and T.\
430: Ala-Nissil{\"a}, Phys.\ Rev.\ E {\bf 65}, 052104 (2002).
431:
432: \bibitem{hahw90} A.\ Hansen, T.\ Aukrust, J.\ M.\ Houlrik and I.\ Webman,
433: J.\ Phys.\ A {\bf 23}, L145 (1990).
434:
435: \bibitem{bh92} G.\ G.\ Batrouni and A.\ Hansen, J.\ Phys.\ A {\bf 25},
436: L1059 (1992).
437:
438: \bibitem{hm06} A.\ Hansen and J.\ Mathiesen in {\it Modelling critical and
439: catas\-trophic phenomena in geoscience: A statistical phy\-sics approach,\/}
440: edited by P.\ Bhattacharyya and B.\ K.\ Chakrabarti (Springer Verlag,
441: Berlin, 2006).
442:
443: \bibitem{shb03} J.\ Schmittbuhl, A.\ Hansen and G.\ G.\ Batrouni, Phys.\
444: Rev.\ Lett.\ {\bf 90}, 045505 (2003).
445:
446: \bibitem{mrs97} A.\ R.\ Mehrabi, H.\ Rassamdana and M.\ Sahimi,
447: Phys.\ Rev.\ E, {\bf 56}, 712 (1997).
448:
449: \bibitem{shn98} I.\ Simonsen, A.\ Hansen and O.\ M.\ Nes, Phys.\ Rev.\ E
450: {\bf 58}, 2779 (1998).
451:
452: \bibitem{hsb01} A.\ Hansen, J.\ Schmittbuhl and G.\ G.\ Batrouni, Phys.\
453: Rev.\ E {\bf 63}, 062102 (2001).
454:
455: \bibitem{hme94} A.\ Hansen, K.\ J.\ M{\aa}l{\o}y and T.\ Eng{\o}y,
456: Fractals, {\bf 2}, 527 (1994).
457:
458: \bibitem{ga86-ga87} T.\ Grossmann and A.\ Aharony, J.\ Phys.\ A {\bf 19}, L745
459: (1986); T.\ Grossmann and A.\ Aharony, J.\ Phys.\ A {\bf 20}, L1193
460: (1987).
461:
462: \bibitem{bmr05} G.\ M.\ Buend{\'\i}a, S.\ J.\ Mitchell and P.\ A.\ Rikvold,
463: Microelectronics J.\ {\bf 36}, 913 (2005).
464:
465: \bibitem{m05} S.\ J.\ Mitchell, Phys.\ Rev.\ E, {\bf 72}, 065103(R) (2005).
466:
467: \end{thebibliography}
468: %---------------------------------------------------------------------
469: \end{document}
470:
471:
472:
473: