cond-mat0104032/v2.tex
1: %%%%%%%%%%%January 5, 2001 %%%%%%%%%%%%%%%
2: \documentstyle[aps,prl,multicol,epsf]{revtex}
3: %\documentstyle[aps,manuscript,epsf]{revtex}
4: \begin{document}
5: \title{\bf Scaling in a Nonconservative Earthquake Model of Self-Organised 
6:           Criticality }
7: 
8: \author{Stefano Lise and Maya  Paczuski}
9: \address{Department of Mathematics, Huxley Building, Imperial College
10: of Science, Technology, and Medicine, London UK SW7 2BZ \\}
11: \date{\today}
12: 
13: \maketitle %\parskip 2ex
14: 
15: \begin{abstract}
16: We numerically investigate the Olami-Feder-Christensen model for earthquakes 
17: in order to characterise its scaling behaviour. We show that ordinary finite 
18: size scaling in the model is violated due to global, system wide events. 
19: Nevertheless we find that subsystems of linear dimension small compared to 
20: the overall system size obey finite (subsystem) size scaling, with universal 
21: critical coefficients, for the earthquake events localised within the 
22: subsystem.  We provide evidence, moreover, that large earthquakes responsible
23: for breaking finite size scaling are initiated predominantly near the 
24: boundary.
25: 
26: \end{abstract}
27: 
28: \vspace{0.3cm}
29: {PACS numbers: 05.65.+b, 45.70.Ht}
30: 
31: 
32: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
33: \begin{multicols}{2}
34: 
35: \section{Introduction}
36: 
37: Many dynamical phenomena in nature are intermittent. This ``bursty'' dynamics 
38: may be related to an underlying complex state, often characterised by long 
39: range correlations in space and time. For example, the crust of the earth 
40: alternates long periods of relative quiescence with burst of activity 
41: (earthquakes), which have a wide range of possible sizes.  The behaviour 
42: of earthquakes is described by the empirical Gutenberg-Richter (GR)
43: law~\cite{gutenberg}, where the distribution of energy dissipated in
44: earthquake events is a power law over many orders of magnitude in
45: energy. The GR scaling extends from the smallest measurable earthquakes, 
46: which are equivalent to a truck passing by, to the most disastrous that 
47: have been recorded.  Similar scale free behaviour of bursts is observed 
48: in vastly different kinds of physical systems such as flux motion through 
49: disordered type II superconductors placed in a magnetic field~\cite{field}, 
50: or in granular piles, under some conditions~\cite{frette}, etc. 
51: 
52: Self-organised criticality (SOC)~\cite{BTW} has been proposed as a general
53: dynamical principal behind the observed complex behaviour of many 
54: natural phenomena.  It refers to the fundamental property of slowly driven,
55: extended systems to organise, over a sufficiently long transient period, 
56: into a dynamical critical state which lacks  any characteristic time or length 
57: scale.  The amplitude of the response of the system to an external 
58: perturbation follows a power law distribution. A number of simple models have 
59: been developed to test the applicability of SOC to a variety of complex 
60: interacting dynamical systems, such as sand piles and earthquakes 
61: (for a review see e.g. ref.~\cite{bak_book,jen_book,turc_rev}).
62: 
63: One of the basic theoretical problems is to identify robust, and thus
64: physically relevant mechanisms for SOC to emerge, and to define a
65: space of parameters and dynamical processes where SOC is a stable
66: feature.  Although it has been proposed that the presence of
67: conservation laws (e.g. sand grains being transported in a sand pile)
68: or special symmetries was necessary for
69: SOC~\cite{hwa-kardar,grinstein}, many known examples of physical
70: phenomena and some models have been found where no apparent
71: conservation law or special symmetry exists. These include, besides
72: earthquakes, biological evolution, forest fires, epidemics, (possibly)
73: solar flares,  (possibly) reconnection
74: events in the magnetotail, etc.~\cite{bak_book,jen_book,turc_rev}.  
75: In contrast to conservative systems, the mechanisms for SOC in 
76: non-conservative systems are not very well established. 
77: 
78: 
79: A nonconservative SOC model that in recent years has attracted much
80: attention is the so called OFC model~\cite{ofc}. The OFC model is 
81: a simplified lattice representation of a spring-block model for 
82: earthquake dynamics which was originally introduced by Burridge and 
83: Knopoff~\cite{burridge}. The Burridge-Knopoff model can be schematized as 
84: a two dimensional network of blocks interconnected by springs. All blocks 
85: are subject to an external driving force, which pulls them, and to a static 
86: friction, which opposes their motion. In the OFC model, each  site of a 
87: lattice is associated with a continuous variable, which represent the force 
88: acting on a block.
89: A slow driving is applied to the system by increasing uniformly and 
90: simultaneously the forces of all the elements. When the force on a site 
91: exceeds some threshold value (the maximal static friction), the site relaxes 
92: and distributes part of its force to  nearest neighbour sites.   
93: Each such discharge event is accompanied by a local loss in accumulated force 
94: from the system. 
95: This conceptually simple and seemingly numerically tractable model 
96: reproduces some of the qualitative phenomenology of the statistics of 
97: earthquake events such as power law behaviour over a range of sizes, and 
98: intermittency or clustering of large events~\cite{ofc_2}.
99: Extensions of the model have been recently developed  which reproduce 
100: to some extent Omori's law and other temporal patterns associated with
101: earthquakes~\cite{hainzl}.
102: 
103: 
104:     
105: In the context of nonconservative models, the OFC model is of particular 
106: interest as it is possible to directly control the level of conservation of 
107: the dynamics through a parameter $\alpha$. Early analysis on relatively small 
108: systems indicated that the OFC model exhibited SOC, in the sense that 
109: earthquakes in the steady state obeyed finite size scaling (FSS) when the 
110: system size was varied~\cite{ofc}.  However, the critical coefficients 
111: obtained using the FSS ansatz were found to be nonuniversal. In particular 
112: the exponents characterising the power law distributions appeared to vary 
113: with both the dissipation parameter, $\alpha$, and the form of the boundary 
114: conditions.  This would have been in sharp contrast to the usual fixed
115: point picture of critical phenomena where most microscopic details are
116: irrelevant.  Moreover some apparent critical exponents obtained using FSS
117: violated physical bounds~\cite{klein}, putting some doubt on the existence of
118: criticality in the model. Recently it was shown using a multiscaling
119: analysis of large-scale simulations that, actually, the avalanche size
120: distribution has a universal power law behaviour, independent of the
121: dissipation parameter and for different boundary conditions, but that
122: the cutoff in the power-law distribution does not behave according to
123: FSS~\cite{lisepac}.  In larger systems, proportionally more of the force 
124: can be dissipated in the largest events that occur, and the cutoff function
125: becomes sharper and sharper as the system size increases.
126: 
127: 
128: Departures from standard FSS have been reported for various SOC models
129: as, for example, the sandpile model~\cite{stella}, the Drossel-Schwabl (DS)
130: forest fire model~\cite{schenk,pastor_1} and the Zhang model~\cite{pastor_2}. 
131: In this paper we address the question of the origin of the breaking of 
132: FSS and its relation to the mechanism responsible for SOC in the OFC
133: model. In particular we test the implicit assumption behind the FSS
134: hypothesis that a finite systems behaves as a subsystem of a larger 
135: system. The paper is organised as follows. 
136: In the section II we describe in some detail the model.  In section
137: III we present the results of our numerical study relative to two different 
138: type of probability distributions for earthquake sizes. The first distribution
139: concerns earthquakes which are localised within a given subsystem. We show 
140: that this subset of earthquakes exhibits ordinary FSS as long as the linear 
141: extent,  $\lambda$, of the subsystem is sufficiently small compared to the 
142: linear extent, $L$, of the entire system. The second distribution groups 
143: earthquakes according to the position of their starting site relative to the 
144: boundaries of the system. From this investigation, we deduce that FSS is 
145: violated due to large events initiated in a region near the boundary.
146: Finally, in section IV we discuss our results and draw some conclusions.
147: 
148: \section{The model}
149: 
150: We consider a two-dimensional square lattice of $L \times L$ sites.  
151: To each site $i$ of the lattice we associate a continuous variable 
152: $F_i$, which initially take some random values between zero and a 
153: threshold value $F_{th}$.  The dynamics proceeds then indefinitely. In the 
154: limit of infinite time scale separation between the slow driving and the 
155: (almost) instantaneous earthquake process, the dynamics is:  
156: \begin{enumerate}
157: \item {\em Uniform drive}: all forces  $F_i$ are increased at the same rate, 
158:   until one of them reaches the value $F_{th}$.
159: \item {\em Earthquake}: when a site becomes unstable (i.e 
160:   $F_i \geq F_{th}$),  the uniform driving is stopped and the system evolves 
161:   according to the following local relaxation rule       
162:      \begin{equation}
163:      \label{av_dyn}
164:            F_i \geq F_{th}  \Rightarrow \left\{ \begin{array}{l}
165:                                        F_i \rightarrow 0 \\
166:                          F_{nn} \rightarrow F_{nn} + \alpha F_i
167:                                       \end{array} \right.
168:       \end{equation} 
169:    until there are no more unstable sites. In eq.~(\ref{av_dyn}), the subscript
170:   ``nn'' stands for the four nearest neighbours to site $i$.  
171: \end{enumerate}
172: Since only a fraction, $4 \alpha$, of the force is redistributed in each 
173: relaxation event (toppling), the model is nonconservative for 
174: $\alpha<1/4$. In the following we concentrate on this case, i.e.  
175: $ 0< \alpha <1/4$. 
176: 
177: 
178: 
179: To completely define the model we need to specify the boundary conditions.
180: Boundaries are believed to play a crucial role for the observation of
181: critical behaviour in the OFC model. It has been suggested that they act 
182: as inhomogeneities which frustrate the natural tendency of the model to
183: order into a periodic state~\cite{grass2,socolar}.
184: Indeed, for sufficiently small values of the conservation parameter 
185: $\alpha$ ($\alpha < \alpha _c \simeq 0.18$), a system with periodic boundary 
186: conditions quickly reaches an exactly periodic state with only earthquakes 
187: of size one. For larger values of $\alpha$ the situation is slightly more 
188: complicated with multiple topplings involved in a single avalanche, but 
189: the avalanches are still localised and criticality is not 
190: observed~\cite{grass2}.
191: A system with open boundaries is prevented from reaching a periodic state 
192: because boundary sites have fewer neighbours and therefore cycle at a 
193: different frequency from bulk sites.  Middleton and Tang suggested that the 
194: inhomogeneity created by the boundaries propagates into the bulk of the 
195: system, developing, in this way, long range spatial correlations.  They named 
196: this mechanism ``marginal synchronisation'' or phase locking~\cite{middleton}. 
197: In accordance with previous studies, therefore, we consider open boundary 
198: conditions. If a boundary (corner) site topples, an extra amount $\alpha F_i$ 
199: ($2 \alpha F_i$) is simply lost by the system.
200: 
201: 
202: 
203: \section{Results: probability distributions}
204: 
205: After a sufficiently long transient time, the system reaches a stationary 
206: state. Several statistical properties can be used to characterise this state.
207: Most previous studies of the OFC model have focused on the behaviour of the 
208: probability distribution of earthquake sizes, $P_L (s)$, where $L$ is the
209: size of the system and $s$ is the total number of topplings events during an 
210: earthquake~\cite{ofc,lisepac,grass2,socolar,middleton,kertesz,corral,ceva}. 
211: We choose instead to analyse the behaviour of different distributions for 
212: avalanches sizes, which distinguish between earthquakes according to the 
213: region of the lattice involved (e.g. bulk or boundary) and 
214: the coordinates of the triggering site (see figure~\ref{scheme}). This 
215: investigation is particularly relevant for the OFC model in view of the 
216: strong inhomogeneity in the spatial distribution of 
217: avalanches~\cite{grass2,middleton,ceva}. According to ref.~\cite{grass2}, 
218: for example, large avalanches are localised near the boundary (at least for 
219: $\alpha < \alpha _c$). As a minor technical remark, we point out that we 
220: exclude from our data avalanches which involve only one site ($s=1$) as they 
221: appear to behave according to their own statistics~\cite{grass2}. As we are 
222: mainly interested in asymptotically large earthquakes, this does not alter our 
223: conclusions.   
224: 
225: We consider a subsystem of linear extent $\lambda$ centred in a system 
226: of size $L$. The first distribution we introduce, $P_{conf}(\lambda,L,s)$, 
227: is the normalised distribution of earthquake sizes restricted to earthquakes 
228: which are confined entirely within the subsystem (e.g. avalanche (a) in 
229: fig.~\ref{scheme}). The model is driven according to its usual dynamics 
230: but only those particular earthquakes are counted.  According to our 
231: definition, the case $\lambda=L$ corresponds to the distribution of
232: avalanches which do not reach the boundary of the system. 
233: As shown in fig.~\ref{fig_2}, the distribution  $P_{conf}(\lambda,L,s)$ 
234: becomes independent of $L$, if $L$ is considerably larger then  $\lambda$
235: (approximately $L \ge 2 \lambda$). When $L$ approaches $\lambda$, this is no 
236: longer the case and the cutoff in the distribution is pushed to larger sizes.
237: Although we have shown in Fig.~\ref{fig_2} only the distributions for 
238: $\alpha=0.18$ and $\lambda=32,64$, analogous considerations apply to 
239: different values of $\alpha$ and for different sizes, $\lambda$. Since for
240: a generic $L$,  $P_{conf}(\lambda,L,s) \neq P_{conf}(\lambda,\lambda,s)$, 
241: a small portion of a large system is substantially different from a finite 
242: system of the same size, contrary to what happens in equilibrium critical 
243: phenomena. A similar observation was made in ref.~\cite{schenk} for the DS 
244: forest fire model. 
245: In the following, we denote with $P_{conf}(\lambda,s)$ the distribution 
246: $P_{conf}(\lambda,L,s)$ in the limit where the distribution does not appear 
247: to depend on $L$.
248: In order to determine numerically these distributions, for each value of 
249: $\lambda$ we have simulated (for at least $2\cdot 10^9$ earthquakes) a system 
250: of size $L=2 \lambda$. The dependence on $L$ of $P_{conf}$ can in this case be 
251: safely neglected. With this choice, the accuracy of the measures and  the 
252: range of scales investigated are optimised, within our computational limits. 
253: 
254: 
255: 
256: In Fig.~\ref{fig_3} we report a FSS collapse of $P_{conf}(\lambda,s)$ for 
257: different values of $\alpha$ . Contrary to the entire distribution of 
258: earthquake sizes, $P_L(s)$, we observe that $P_{conf}(\lambda,s)$ satisfies 
259: the FSS hypothesis, i.e. 
260: $P_{conf}(\lambda,s)\simeq \lambda^{-\beta}f(s/\lambda^D)$, with universal 
261: critical coefficients. The curve corresponding to $\alpha=0.15$ and 
262: $\lambda=256$ shows some noisy behaviour, due to the difficulties in
263: collecting good statistics in this case. Indeed by decreasing $\alpha$, the 
264: relative fraction of earthquakes in the bulk of the system (with size $s>1$) 
265: diminishes. Nonetheless, there is no evident  sign that FSS is violated in 
266: this case. The critical exponents used in the fit of Fig.~\ref{fig_3} are 
267: $\beta=3.6$ and $D=2$, independent of the dissipation parameter $\alpha$. 
268: The value of the histogram exponent $\tau= \beta/D \simeq 1.8$ we obtain is 
269: the same as that found for $P_L(s)$ \cite{lisepac}.  
270: In addition, the value of $D$ we find corresponds to the largest possible
271: value for the entire distribution ($D_{max}$ in ref.~\cite{lisepac}), as it 
272: can be shown that non-conservation requires $D \le 2$ \cite{klein}.
273: 
274: 
275: The scaling behaviour of $P_{conf}$ appear to be reasonably robust with respect
276: to translating the subsystem within the entire system; in fig.~\ref{fig_4} we
277: report a FSS plot for the subsystem placed on a boundary and on a corner of 
278: the system for  $\alpha=0.18$. While the FSS collapse for the subsystem placed 
279: on the boundary is rather good, some deviations from FSS are observed in the 
280: cut-off region for the case of the subsystem placed in the corner. We believe
281: this behaviour can be ascribed to the enhanced boundary effects in the latter
282: case (two side of the subsystem are boundary sides instead of only one) and 
283: would disappear if larger (sub)systems could be studied.
284: This picture is confirmed by choosing different $\alpha$ values: for the 
285: subsystem in the corner, deviations from FSS are more pronounced for 
286: $\alpha=0.21$ and are absent for $\alpha =0.15$. For  the subsystem on the 
287: boundary, instead, the quality of FSS collapse is rather convincing in all 
288: cases.  
289: 
290: 
291: 
292: We introduce next the distributions $P_<(\lambda,L,s)$ and $P_>(\lambda,L,s)$.
293: These are the normalised distribution of earthquakes which start respectively
294: within ($P_<$) and outside ($P_>$) the subsystem of size $\lambda$,
295: irrespective of whether they stay in or go out of the subsystem (see 
296: fig.~\ref{scheme}). 
297: The only difference between these two distributions, $P_<$ and $P_>$, is the
298: location of the site that triggers the avalanche. 
299: We observe numerically that the distributions $P_<(\lambda,L,s)$ and 
300: $P_>(\lambda,L,s)$ become independent of $\lambda$ respectively in the limit 
301: $\lambda \ll L$ and $\lambda \simeq L$. As an example we report in 
302: Fig.~\ref{fig_5} the behaviour of $P_<(\lambda,L,s)$ and $P_>(\lambda,L,s)$
303: for $\alpha =0.18$, $L=256$ and for various $\lambda$. 
304: For simplicity, in the following we denote with $P_<(L,s)$ and $P_>(L,s)$ the 
305: distributions in the limit where they do not depend on $\lambda$. 
306: 
307: We consider therefore two centred subsystems of linear extent 
308: $\lambda_2>\lambda_1$, such that the above conditions are satisfied.
309: More specifically, we choose $\lambda_1 = \frac{3}{16} L$ and 
310: $\lambda_2 = \frac{7}{8} L$. In this case $P_>$ corresponds to the subset 
311: of earthquakes which are initiated in some ``boundary'' region and $P_<$ 
312: corresponds to the subset of earthquakes which are initiated within some 
313: ``bulk'' region.
314: In fig.~\ref{fig_6} we report a FSS scaling plot both for  $P_>$ and
315: $P_<$. In this figure we choose $D=2$ as the maximum allowed value.  
316: It is clear that the boundary distribution $P_>$  cannot be collapsed 
317: according to the FSS ansatz. In fact it develops a sharper and sharper 
318: cutoff which changes shape and which has an excess of large events (the
319: cutoff moves towards right for increasing $L$).
320: The bulk distribution $P_<$ instead does not develop a noticeably sharper 
321: cutoff and does not appear to change its shape. It may possibly
322: be collapsed according to the FSS ansatz. Consistent with the
323: results for $P_{conf}$ and for $P_L$, the power law exponent for  $P_<$
324: is $\tau= \beta _1/D \simeq 1.8$.
325: 
326: 
327: The above numerical analysis indicates that the large events which
328: violate FSS are triggered by sites in a boundary region. Indeed the 
329: behaviour of the cutoff for the collapsed probability distributions 
330: $P_>$ and $P_L$ is very similar (see Fig.~1 in ref.~\cite{lisepac}). 
331: Although large earthquakes are focussed mainly toward the boundary, 
332: as suggested in ref.~\cite{grass2}, they occur also in the bulk of the
333: system, even for low values of $\alpha$, as can be deduced from 
334: Fig.~\ref{fig_3} and Fig.~\ref{fig_6}. Moreover, we do not observe any
335: significant qualitative change in the behaviour of the system around 
336: $\alpha = \alpha _c \simeq 0.18$, as claimed in  ref.~\cite{grass2}. 
337: 
338: 
339: 
340: \section{Discussion and conclusions}
341: 
342: Similarly to other SOC systems~\cite{stella,schenk,pastor_1,pastor_2}, the 
343: nonconservative OFC model shows relevant deviations from simple 
344: FSS~\cite{lisepac}.
345: In this paper we have investigated the origin of this phenomenon, finding 
346: that FSS in the OFC model is violated because of large, system wide 
347: earthquakes. In fact, we have found that earthquakes localised within a given 
348: subsystem do obey ordinary FSS, with universal critical exponents,
349: independently of whether the subsystem is placed at the centre or on a 
350: boundary of the system. The value of the power law exponent, 
351: $\tau \simeq 1.8$, for the ``confined'' distribution agrees with the one for 
352: the entire distribution. We have shown, moreover, that the probability 
353: distribution for earthquakes initiated in a boundary region do not obey
354: FSS, because of an ``excess'' of large events. This would result in an 
355: apparent exponent $D>2$ which is not allowed in the nonconservative case.
356: On the other hand, the probability distribution for earthquakes starting
357: in the bulk of the system is compatible with a FSS hypothesis. In particular, 
358: the critical exponent is $D=2$ in this case, indicating that large 
359: earthquakes responsible for breaking finite size scaling are initiated 
360: predominantly near the boundary of the system.
361:  
362: 
363: Self-organised criticality in the OFC model has been ascribed to a mechanism 
364: of ``marginal synchronisation''~\cite{middleton}. A  system with open 
365: boundaries becomes almost synchronised by an invasion process where spatial 
366: correlations develop from the boundaries.
367: It was suggested that sites close to the boundaries start to organise
368: themselves first, building up long range correlations. The critical region
369: grows with time, until, in the stationary state, it invades the whole
370: lattice~. Our findings on the large events occurring at the 
371: boundary seem to indicate that the effect of synchronisation is stronger for 
372: boundary sites than for  bulk sites. This view is supported also by the 
373: ``on screen'' observation that large earthquakes tend to be triggered 
374: repetitively by the same sites over a long time scale (a result which seems 
375: to be confirmed also by the study in ref.~\cite{ceva}).
376: 
377: \medskip
378: 
379: S.L. acknowledges financial support from EPSRC (UK),  
380: Grant No. GR/M10823/01. 
381: 
382: \begin{thebibliography}{99}
383: 
384: 
385: \bibitem{gutenberg}
386: B. Gutenberg and C.F. Richter, Ann. Geofis. {\bf 9}, 1 (1956).
387: 
388: \bibitem{field} 
389: S. Field, J. Witt, F. Nori and X.S. Ling,
390: Phys. Rev. Lett. {\bf 74}, 1206 (1995);
391: C. Heiden and G.I. Rochlin,
392: Phys. Rev. Lett. {\bf 21}, 691 (1968).
393: 
394: \bibitem{frette}
395: V. Frette, {\it et. al.},
396: Nature {\bf 379}, 49 (1996). 
397: 
398: \bibitem{BTW} P. Bak, C. Tang, and K. Wiesenfeld, Phys. Rev. Lett. {\bf59},
399: 381 (1987); Phys. Rev. A. {\bf 38}, 364 (1988).
400: 
401: \bibitem{bak_book}
402: P. Bak,
403:  {\it How Nature Works: The Science of Self-Organized Criticality} 
404:      (Copernicus, New York, 1996).
405: 
406: \bibitem{jen_book}
407: H. Jensen,
408:  {\it Self-Organized Criticality}
409:       (Cambridge University Press,New York, 1998).
410: 
411: \bibitem{turc_rev} D. L. Turcotte,
412: Rep. Prog. Phys. {\bf 62} (10) 1377 (1999).
413: 
414: \bibitem{hwa-kardar}
415: T. Hwa and M. Kardar,
416:  Phys. Rev. Lett. {\bf 62}, 1813 (1989).
417: 
418: \bibitem{grinstein}
419: G. Grinstein, D.-H. Lee, and S. Sachdev,
420:  Phys. Rev. Lett. {\bf 64}, 1927 (1990).
421: 
422: \bibitem{ofc} 
423: Z. Olami, H.J.S. Feder, and K. Christensen,
424: Phys. Rev. Lett. {\bf 68}, 1244 (1992);
425:  K. Christensen and Z. Olami,
426: Phys. Rev. A {\bf 46}, 1829 (1992).
427: 
428: \bibitem{burridge} R. Burridge and L. Knopoff, Bull. 
429: Seismol. Soc. Am. {\bf 57},341 (1967).
430: 
431: \bibitem{ofc_2}
432: K. Christensen and Z. Olami,
433: J. Geophys. Res. {\bf 97}, 8729 (1992); 
434: K. Christensen, Z. Olami, and P. Bak, 
435: Phys. Rev. Lett. {\bf 68}, 2417 (1992).
436: 
437: \bibitem{hainzl} 
438: S. Hainzl, G. Zoller, and J. Kurths,
439: J. Geophys. Res. {\bf 104}, 7243 (1999); 
440: Nonlinear Proc. Geoph. {\bf 7}, 21 (2000). 
441: 
442: \bibitem{klein} 
443: W. Klein and J. Rundle, Phys. Rev. Lett. {\bf 71}, 1288 (1993).
444:        
445: \bibitem{lisepac} 
446: S. Lise and M.Paczuski, Phys. Rev. E, {\bf 63}, 036111 (2001). 
447: 
448: \bibitem{stella}
449: M. De Menech, A. L. Stella, and C. Tebaldi, 
450: Phys. Rev. E {\bf 58}, 2677 (1998);
451:  C. Tebaldi, M. De Menech, and A. L. Stella, Phys. Rev. Lett.
452: {\bf 83},  3952 (1999).
453: 
454: \bibitem{schenk}
455: K. Schenk, B. Drossel, S. Clar, and F. Schwabl,
456: Eur. Phys. J. B {\bf 15}, 177 (2000)
457: 
458: \bibitem{pastor_1}
459: R. Pastor-Satorras and A. Vespignani,
460: Phys. Rev. E {\bf 61}, 4854 (2000) 
461: 
462: \bibitem{pastor_2}
463: R. Pastor-Satorras and A. Vespignani,
464: Eur. Phys. J. B {\bf 18}, 197 (2000) 
465: 
466: \bibitem{grass2} P. Grassberger, Phys. Rev. E {\bf 49}, 2436 (1994).
467: 
468: 
469: \bibitem{socolar} J.E.S. Socolar, G. Grinstein, and C. Jayaprakash,
470: Phys. Rev E, {\bf 47}, 2366 (1993).
471: 
472: 
473: \bibitem{middleton}
474: A. A. Middleton and C. Tang,
475: Phys. Rev. Lett. {\bf 74}, 742 (1995).
476: 
477: 
478: \bibitem{kertesz} I.M. J\'anosi and J. Kerte\'sz, Physica A {\bf 200}, 174
479: (1993).
480: 
481: 
482: \bibitem{corral} \'{A}. Corral, C. J. P\'erez, A. D\'{\i}az-Guilera, 
483: and A. Arenas, 
484: Phys. Rev. Lett. {\bf 74}, 118 (1995).
485: 
486: \bibitem{ceva} H. Ceva, Phys. Lett. A, {\bf 245}, 413 (1998).  
487: 
488: 
489: \end{thebibliography}
490: 
491: \begin{figure}[hb]
492: \narrowtext
493: \epsfxsize=2.5in
494: \centerline{\epsffile{scheme.eps}}
495: \protect\vspace{0.2cm}
496: \caption[1]{\label{scheme}  
497: Schematic representation of different types of avalanches.  
498: The continuous line represents the lattice of size $L$, the dashed line 
499: the subsystem of linear extent $\lambda$. Triggering sites are denoted 
500: with a full circle, toppling sites with a cross. Avalanche (a) contributes 
501: to the distribution $P_{conf}(\lambda,L,s)$; 
502: (a) and (b) to $P_<(\lambda,L,s)$; (c) to $P_>(\lambda,L,s)$.
503: }
504: \end{figure}
505: 
506: 
507: \begin{figure}[hb]
508: \narrowtext
509: \epsfxsize=4.in
510: \centerline{\epsffile{fig_2.eps}}
511: \caption[1]{\label{fig_2}  
512: Probability distribution $P_{conf}(\lambda,L,s)$ for $\alpha=0.18$ and 
513: (a) $\lambda=32$ and (b)  $\lambda=64$.
514: }
515: \end{figure}
516: 
517: 
518: 
519: \begin{figure}[hb]
520: \narrowtext
521: \epsfxsize=4.in
522: \centerline{\epsffile{fig_3b.eps}}
523: \caption[1]{\label{fig_3}
524: Finite-size scaling plots of $P_{conf}(\lambda,s)$ (with the subsystem 
525: placed at the centre) for (a) $\alpha=0.15$,
526: (b) $\alpha=0.18$ and (c) $\alpha=0.21$. The critical exponents are 
527: $\beta=3.6$ and $D=2$; the slope of the straight line is $\tau =1.8$. 
528: For visual clarity, curves (a) and (c) have been shifted along the 
529: $x$-axis, $x \rightarrow x-1$ and  $x \rightarrow x+1$ respectively. 
530: }
531: \end{figure}
532: 
533: 
534: 
535: \begin{figure}[hb]
536: \narrowtext
537: \epsfxsize=4.in
538: \centerline{\epsffile{fig_4_18.eps}}
539: \caption[1]{\label{fig_4}
540: Finite-size scaling plots of $P_{conf}$ for the subsystem placed
541: (a) on a corner and (b) on a boundary of the system ($\alpha=0.18$). 
542: The critical exponents are $\beta=3.6$ and $D=2$; the slope of the straight 
543: line is $\tau =1.8$. Curve (a) has been shifted, $x \rightarrow x-1$. 
544: }
545: \end{figure}
546: 
547: 
548: \begin{figure}[hb]
549: \narrowtext
550: \epsfxsize=4.5in
551: \centerline{\epsffile{fig_5.eps}}
552: \caption[5]{\label{fig_5}
553: Probability distributions (a) $P_<(\lambda,L,s)$  and (b) $P_>(\lambda,L,s)$
554: for $\alpha=0.18$, for $L=256$ and for various $\lambda$.} 
555: \end{figure}
556: 
557: 
558: \begin{figure}[hb]
559: \narrowtext
560: \epsfxsize=4.5in
561: \centerline{\epsffile{fig_6.eps}}
562: \caption[5]{\label{fig_6}
563: Finite-size scaling plots of (a) $P_<(L,s)$ and (b) $P_>(L,s)$. Different
564: sets of curves corresponds, from left to right, to $\alpha=0.15$ (shifted 
565: by $x \rightarrow x-1$), $\alpha=0.18$ and $\alpha=0.21$ (shifted by 
566: $x \rightarrow x+1$). The exponents are  $\beta _1=3.6$,  $\beta _2=3.9$ 
567: and $D=2$.}
568: \end{figure}
569: 
570: \end{multicols}
571: \end{document}
572:  
573: 
574: 
575: 
576: 
577: 
578: