cond-mat0008010/ofc.tex
1:                                                                                
2: %%%%%%%%%%%%July 20, 2000 %%%%%%%%%%%%%%%
3: \documentstyle[aps,prl,multicol,epsf]{revtex}
4: %\documentstyle[aps,manuscript,epsf]{revtex}
5: \begin{document}
6: \title{\bf Self-Organized Criticality and Universality in a Nonconservative 
7: Earthquake Model}
8: 
9: \author{Stefano Lise and Maya  Paczuski}
10: \address{Department of Mathematics, Huxley Building, Imperial College
11: of Science, Technology, and Medicine, London UK SW7 2BZ \\}
12: \date{\today}
13: 
14: \maketitle %\parskip 2ex
15: 
16: \begin{abstract}
17:  
18:  We make an extensive numerical study of a two dimensional
19:  nonconservative model proposed by Olami-Feder-Christensen to describe
20:  earthquake behavior.  By analyzing the distribution of earthquake
21:  sizes using a multiscaling method, we find evidence that the model is
22:  critical, with no characteristic length scale other than the system
23:  size, in agreement with previous results.  However, in contrast to previous
24:  claims, we find convergence to universal behaviour
25:  as the system size increases, over a range of values of the dissipation
26:  parameter, $\alpha$.  We also find that both ``free'' and ``open''
27:  boundary conditions tend to the same result.
28:  Our analysis indicates that, as $L$ increases, the behaviour
29:  slowly converges toward a power law distribution of earthquake sizes
30:  $P(s) \sim s^{-\tau}$ with exponent $\tau \simeq 1.8$. The universal
31:  value of $\tau$ we find numerically agrees quantitatively with the empirical
32:  value ($\tau=B+1$) associated with the Gutenberg-Richter law.
33:  
34: 
35: 
36: \end{abstract}
37: 
38: 
39: {PACS numbers: 05.40.+j, 91.30.Px}
40: 
41: 
42: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
43: \begin{multicols}{2}
44: 
45: \section{Introduction}
46: 
47: Earthquakes may be the most dramatic example of self-organized
48: criticality (SOC) \cite{soc,review} that can be observed by humans on
49: earth.  Most of the time the crust of the earth is at rest, or
50: quiescent. These periods of stasis are punctuated by sudden, thus far
51: unpredictable bursts, or earthquakes.  According to the empirical
52: Gutenberg-Richter (GR) law \cite{gutenberg}, the distribution of
53: earthquake events is scale-free over many orders of magnitude in
54: energy.  The GR scaling extends from the smallest measurable
55: earthquakes, which are equivalent to a truck passing by, to the most
56: disastrous that have been recorded where hundreds of thousands of
57: people have perished.
58: 
59: The relevance of SOC to earthquakes was first pointed out by Bak and
60: Tang \cite{bak-tang}, Sornette and Sornette \cite{sornette}, as well
61: as Ito and Matsusaki \cite{ito}.  According to this theory, plate tectonics
62: provides energy input at a slow time scale into a spatially extended,
63: dissipative system that can exhibit breakdown events via a chain
64: reaction process of propagating instabilities in space and time.  The
65: GR law arises from the system of driven plates building up to
66: a critical state with avalanches of all sizes.  These above-mentioned
67: authors used a spatially
68: extended (but conservative) cellular automata model as a prototype
69: resembling earthquake dynamics which gave a power law distribution of
70: avalanches, or earthquakes.  This was followed by a study using
71: block-spring models \cite{burridge} by Carlson and Langer
72: \cite{carlson-langer}, who found characteristic earthquake sizes,
73: rather than asymptotic critical behavior.  Later studies of a continuous
74: ``train'' block-spring model by de Sousa Vieira \cite{vieira}
75: recovered criticality. The train model describes a driven elastic
76: chain sliding over a surface with friction.  It has been conjectured
77: to be in the same universality class as interface depinning and a
78: model of avalanches in granular piles, which agrees with numerical
79: simulation results \cite{universal}.
80: 
81: 
82:  Several groups made lattice representations of the block-spring
83: model.  These models were nonconservative \cite{takayasu,nakanishi}
84: and were driven uniformly, but did not display SOC.  Then Olami,
85: Feder, and Christensen (OFC) introduced a nonconservative model on a
86: lattice that displayed SOC \cite{ofc}.  In this simplified earthquake
87: model, sites on a lattice are continuously loaded with a force.  After
88: a threshold force is reached, the sites transfer part of their force
89: to their local neighborhood when discharging.  Each discharge event is
90: accompanied by a local loss in accumulated force from the system, when
91: the force on each element is reset to zero.  A uniform driving force
92: is slowly applied to all the elements and the model is completely
93: deterministic.  This conceptually simple and seemingly numerically
94: tractable model reproduces some of the qualitative phenomenology of
95: the statistics of earthquake events such as power law behavior over a
96: range of sizes, intermittency or clustering of large events
97: \cite{ofc}, and lack of predictability \cite{pepke}.
98: 
99: 
100: Although SOC and this type of modelling approach has been more or less
101: accepted as a reasonable description of the phenomena of earthquakes
102: (see for example Ref. ~\cite{turcotte} and references therein), the
103: OFC model itself has had a controversial existence, both on the
104: theoretical \cite{klein,lise,french,grass1} and numerical side
105: \cite{ofc,kertez,socolar,grass2,middleton,corral}.  The initial
106: numerical studies found that the distribution of earthquake sizes
107: obeyed finite size scaling (FSS) over the range of system sizes that
108: could be studied at the time \cite{ofc}.  This placed the
109: nonconservative model into the framework of standard critical
110: behavior.  However, these simulations also indicated that there was no
111: universality.  In particular the exponents characterizing the power
112: law distributions appeared to vary with both the dissipation
113: parameter, $\alpha$, and the form of boundary condition.  If this were
114: the case then the OFC model would be very different from familiar
115: critical systems where most microscopic details are irrelevant and
116: have no effect on critical coefficients.  In fact an argument was made
117: \cite{ofc} that one should not expect universal behavior in far from
118: equilibrium critical phenomena.  If this were correct, it would
119: drastically limit the application of any known theoretical tools to
120: these problems.
121: 
122: Another strange aspect was that the dimension $D$ characterizing the
123: scaling of the cutoff in the earthquake size distribution was found
124: numerically to be larger than two. This is inconsistent with the fact
125: that each site can only discharge a finite number of times in an
126: earthquake event, requiring $D \leq 2$ for the two dimensional model
127: \cite{klein}.  This last result together with the strange lack of
128: universality cast some doubt on whether the OFC model was actually
129: critical or just close to being critical, with some large as yet
130: undetermined length scale beyond which the earthquake distribution
131: would always be cut off.   Hwa and Kardar as well as
132: Grinstein and collaborators postulated that conservation of the
133: quantity being transported was required for criticality
134: \cite{hwa-kardar}, but the theoretical arguments made do not take into
135: account SOC phenomena such as avalanches and long-term memory
136: associated with the self-organization process (for more details see
137: \cite{paczuski}).  The fact that the random neighbor version of the
138: nonconservative OFC model is never critical but has an essential
139: singularity as the conservative limit is approached
140: \cite{french,grass1} has added to the mystery surrounding the behavior
141: of the lattice model.
142: 
143: 
144: In a previous large scale numerical simulation study of the model
145: discussed here, Grassberger \cite{grass2} also claimed that the model
146: was critical but found that some of the conclusions of OFC ``have to
147: be modified considerably.''  He argued that the probability
148: distribution of earthquake sizes does not show ordinary FSS over the
149: larger range of systems he was able to study, and ``conjecture(d) that
150: the cutoff of $P(s)$ becomes a step function for $L\rightarrow
151: \infty$,'' although he did not present direct numerical evidence of this.
152: 
153: There are a number of important, unresolved questions about the
154: behavior of the model, which have enormous implication for any type of
155: eventual theoretical understanding.  Is the nonconservative model on a
156: lattice (or for fixed connectivity matrix)
157: critical?  If so, is the critical behavior of the model
158: universal over a range of values of $\alpha$, or for different
159: boundary conditions?  Is it described by a
160: power-law distribution at all?  Are there any other universal quantities?
161: What type of data analysis
162: technique besides FSS would be useful to extract the large scale
163: behavior of the nonconservative model?  Our numerical study and
164: analysis will address those issues and answer those questions.
165: 
166: \subsection{Summary}
167: In the first section we review the definition of the OFC model and
168: present some numerical data for the distribution of earthquake sizes
169: using standard FSS.  For the range of lattice sizes we have simulated
170: (up to linear size $L=512$), our data confirm the lack of apparent FSS
171: in the model, particularly in the cutoff region.  In Section III, we
172: present an extensive set of results using a multiscaling method.
173: We analyze
174: the rescaled probability distribution, $\frac{\log P(s)}{\log L}$
175: \cite{note}, in
176: terms of the quantity $D_{av}\equiv \log s/\log L$, with $s$ being the
177: size of an earthquake.  We observe that there are no avalanches with
178: $D_{av}$ larger than two, consistent with the bound imposed on the
179: cutoff $s_{co}$ (see previous discussion).  By analyzing how this
180: distribution behaves for different values of the nonconservation
181: parameter, $\alpha$, and system size, $L$, we show how the multiscaled
182: probability distribution tends to converge to a universal curve as $L$
183: increases.  The direction of convergence on increasing $L$ changes  as
184: $\alpha$ varies enabling us to put fairly firm limits on the
185: asymptotic curve.  The model appears not be to described at
186: all by FSS.  However,  for $s<s_{co}$ the distribution converges
187: toward a power law with a universal exponent $\tau \simeq 1.8$ over a
188: range of $\alpha$ values.  Moreover the cutoff in this distribution
189: becomes very sharp as $L$ increases and its behavior indicates that
190: $s_{co} \rightarrow {\rm const}(\alpha) L^2$ as $L\rightarrow \infty$.
191: In Section IV we summarize our main conclusions.
192: 
193: \section{Definition of the Model}
194: We consider a two-dimensional square lattice of $L \times L $ sites.
195: At each site $i$, a force $F_i$ is assigned to be a real variable.
196: Initially, the force at each site is chosen randomly from the uniform
197: distribution between 0 and 1. The dynamics proceeds by two steps in
198: the limit of infinite time scale separation between the slow drive,
199: representing motion of the tectonic plates, and the earthquake process
200: \cite{ofc}.
201: 
202: \begin{enumerate}
203: \item{Increase  the force at all sites: Find the largest force $F_{max}$ 
204:   in the system and increase  the force at all sites by the same 
205:   amount $1- F_{max}$.}
206: 
207: \item{Relax all unstable sites, i.e. sites with $F_i \geq 1$: The force
208:   of an unstable site is reset to zero,  $F_i \rightarrow 0$, and a  
209:   fraction of it, $\alpha F_i$, is distributed  to each of its four 
210:   nearest neighbours, $F_{nn} \rightarrow F_{nn} + \alpha F_i$.   
211:   This step is repeated in a parallel update until there are no 
212:   unstable sites left.}
213: 
214: \end{enumerate}
215:   This two step rule is iterated indefinitely.  The sequence of
216:   toppling events (step 2) between application of the uniform drive (step 1)
217:   defines an avalanche or earthquake.  Since only a fraction,
218:   $4\alpha$, of the force is redistributed in each toppling, the model
219:   is nonconservative for $\alpha < 1/4$.
220: 
221: To completely define the model we need to specify the boundary
222: conditions, by defining the dissipation parameter,
223: $\alpha$, at the sites on the boundaries ($\alpha _{bc}$).
224: As in OFC, we consider both ``free'' and ``open'' boundary conditions.  
225: The sites on the boundaries of the system can be considered to be bounded by 
226: fictitious sites with $F_i= - \infty$, which can never discharge and only 
227: absorb force from the boundary sites.  In the case of open boundary 
228: conditions, the sites at the boundary have the same $\alpha$ parameter 
229: as all other sites in the bulk ($\alpha _{bc}=\alpha$).  In the case of 
230: free boundary conditions, the sites at the boundary have their $\alpha$ 
231: parameters modified in order to have the same level of dissipation per
232: reset event as 
233: for sites in the bulk.  This latter condition implies 
234: $\alpha _{bc}=\alpha / (1-\alpha)$, except at corner sites where
235: $\alpha _{bc,c}=\alpha / (1-2\alpha)$. 
236: The model with periodic boundary conditions is not critical
237: \cite{socolar,grass2,middleton},  and will not 
238: be discussed.
239: It is probably worthwhile to underline at this point that the model is 
240: completely deterministic, the only possible source of randomness coming 
241: from the initial conditions.
242: 
243:  
244: After a transient period of  many earthquakes, the model settles into a 
245: statistically stationary state.  One way to characterize this state is 
246: to measure statistical properties of the earthquakes.  The size of an 
247: earthquake, $s$, is defined as the number of resets of the local force
248: $F_i \rightarrow 0$ in the system in between applications of
249: the uniform force.  One can also measure the temporal
250: duration, $t$, in terms of the parallel update, or the radius of
251: gyration $r$ of the sites which participated in the earthquake event.
252: 
253: \subsection{Finite Size Scaling}
254: We focus on the probability distribution of earthquake sizes, $s$, in
255: a system of size $L$, $P_L(s)$.  If the model is critical then this
256: distribution will have no scale other than the physical extent $L$ and
257: the lattice constant, which is set to unity.  One ansatz that can
258: describe critical behavior is the FSS ansatz that was previously used
259: by OFC:
260: \begin{equation}
261: P_L(s) \sim L^{-\beta} G(\frac{s}{L^{D}})
262: \end{equation}
263: where $G$ is a suitable scaling function, and $\beta$ and $D$ are
264: critical exponents describing the scaling of the distribution
265: function.  As shown in Fig.~1, the model does not exhibit FSS.  In
266: this figure we chose $D=2$ as the largest possible allowed value.
267: Still the cutoff in the ``collapsed'' probability distribution moves
268: to the right as $L$ increases.  Nevertheless, for earthquake sizes
269: smaller than the cutoff, this figure shows
270: what appears to be a convergence to a
271: well-defined power law, $P_L(s) \sim s^{-\tau}$, as $L$ increases with
272: the power law exponent $\tau =\beta/D \simeq 1.8$ for both
273: $\alpha=0.18$ and $\alpha=0.21$, and possibly also for $\alpha=0.15$.
274: \begin{center}
275: \begin{figure}[hb]
276: %\protect\vspace{0.3cm}
277: \narrowtext
278:  \epsfxsize=3.65truein
279: \epsffile{fig_1a.eps}
280: %\protect\vspace{0.3cm}
281: \caption[1]{\label{fs_scal}Finite-size scaling plots of $P_L(s)$ in
282:  systems with open boundary conditions for (a) $\alpha=0.15$, (b)
283:  $\alpha =0.18$, and (c) $\alpha=0.21$. The critical exponents are
284:  $D=2$ and $\beta=3.6$, the slope of the straight line is $\tau =
285:  1.8$. Statistics are derived from at least $10^9 $ avalanches per
286:  data set. For visual clarity, curves (b) and (c) have been shifted
287:  along the horizontal axis, $x \rightarrow x+1$ and $x \rightarrow
288:  x+2$ respectively. }
289: \end{figure}
290: \end{center}
291: 
292: \section{Multiscaling Analysis}
293: The FSS ansatz is only one possible description of critical behavior.
294: As pointed out some time ago by Kadanoff and co-workers, some SOC
295: phenomena are better described by a multifractal ansatz, rather than
296: FSS \cite{kadanoff}.  This form has recently been used to clarify the
297: behavior of the Bak-Tang-Weisenfeld \cite{soc} model by Stella and
298: co-workers, who have measured different moments associated with the
299: distribution \cite{stella}.  For the OFC model, it appears to us that
300: a clearer picture can be obtained by simply examining the probability
301: distribution directly.
302: 
303: 
304: 
305: 
306: The multiscaling ansatz postulates for the probability distribution 
307: function $P_L(s)$ a form 
308: \begin{equation}
309: {\log P_L(s/s_o) \over \log (L/l_o)} = F\Bigl( {\log s/s_o \over \log
310: L/l_o}\Bigr) \quad ,
311: \end{equation}
312: where $s_o$ and $l_o$ are parameters which typically (but not always)
313: reflect phenomena at small scales associated with the lattice \cite{note}.
314: Usually, a multiscaling analysis consists of choosing these two
315: parameters in order to get the best collapse for different system
316: sizes.  This is quite different than FSS where the critical exponents
317: themselves, reflecting behavior at large scales, must be chosen in an
318: {\it ad hoc} manner in order to obtain the ``best'' collapse.  We do not
319: attempt to collapse the data using the multiscaling form of Eq.~2.
320: Instead, we define $l_o=s_o=1$ to represent the smallest earthquake
321: which occurs at only one site and involves only one
322: discharge. Moreover, we define the dimension of an earthquake of size
323: $s$ in a systems of size $L$ as
324: $$D_{av}= \log s/\log L \quad ,$$ and we denote with $D_{max}$ the
325: largest value of $D_{av}$.
326: 
327: 
328: In Fig.~2 we show the multiscaled probability distribution according
329: to Eq.~2, for different system sizes for $\alpha=0.21$.  One observes
330: immediately that all avalanches have dimension $D_{av}<2$, as required,
331: with the 
332: largest dimension $D_{max}$ approaching two as the system size increases.  
333: Also it is clear that the  shape of the cutoff function is sharpening as the
334: system size increases.  Since the largest dimension $D_{max}$ cannot be 
335: larger than two, we can infer from this that the cutoff region narrows 
336: to the region $D_{av} \rightarrow 2$ and becomes increasingly sharp.
337: \begin{figure}[hb]
338: \narrowtext
339: \protect\vspace{0.3cm}
340:  \epsfxsize=3.75in
341: \centerline{\epsffile{fig_2.eps}}
342: \protect\vspace{0.3cm}
343: \caption[2]{\label{ms_021} Multiscaling plot of $P_L(s)$ for $\alpha=0.21$ 
344:  and open boundary conditions. The slope of the straight line is 
345:  $\tau = 1.8$.}
346: \end{figure}
347: Note that the increase of $D_{max}$ as $L$ increases is totally
348: inconsistent with the notion that the OFC model is slightly off
349: criticality, because in that case one would expect that the relative
350: size of the largest dissipating
351: events with respect to the maximum total force allowed in the system
352: would decrease in larger
353: systems.  In fact what
354: happens is exactly the
355: opposite.  In larger systems a larger fraction of the
356: total energy can be dissipated in the largest event that occurs,
357: and $D_{max}$ increases with $L$.  This result
358: is completely consistent with the nonconservative model being
359: critical, rather than slightly off criticality.
360: 
361: 
362: In order to get more explicit visual information on the probability
363: distribution we try to subtract out the leading asymptotic term, which we
364: propose is a power law as suggested by Fig.~2.  This is
365: \begin{equation}
366: {\log P_L(s) \over \log (L)}= F(D_{av})= -(\tau D_{av})
367: + F_{cutoff}(D_{av}) \quad ,
368: \end{equation}
369: where $F_{cutoff}$ in the limit $L \rightarrow \infty$ should be 
370: constant up to a cutoff near $D_{max}$. 
371: \begin{center} 
372: \begin{figure}[hb]
373: \narrowtext
374: \protect\vspace{0.3cm}
375:  \epsfxsize=4.75in
376: \centerline{\epsffile{fig_3.eps}}
377: \protect\vspace{0.3cm}
378: \caption[3]{\label{mf_scal}  Plots of $F_{cutoff}$ for  
379: (a) $\alpha=0.15$, (b) $\alpha =0.18$, (c) $\alpha=0.21$, and 
380: (d) $\alpha=0.23$; we set $\tau =1.8$. Boundary conditions are open. 
381: Different curves correspond, from left to right, to $L=32,64,128,256,512$.}
382: \end{figure}
383: \end{center}
384: We get a consistent picture for a range of $\alpha$ values by choosing
385: $\tau=1.8$, as shown in Fig.~3.  Although it appears for small system
386: sizes that smaller values of $\alpha$ give a steeper distribution with
387: a larger value of $\tau$ (so the lines tend to decrease from left to
388: right rather than remaining horizontal), it is clear from this figure
389: that as the system size increases all the different values of $\alpha$
390: appear to reach the same value of $\tau=1.8$, corresponding to a
391: completely horizontal line in this figure.  The deviation for small
392: systems is more pronounced for $\alpha=0.15$, and less again for
393: $\alpha=0.18$, becoming the smallest for $\alpha=0.21$. For $\alpha$
394: closer to the conservative limit, as shown for $\alpha =0.23$, the
395: approach to the asymptotic horizontal behavior changes direction.
396: Namely smaller systems appear to have a smaller exponent $\tau$ than
397: larger systems, at least for small avalanches.  Thus rather than
398: having the slope increase as $L$ increases, for $\alpha > \sim .21$,
399: the apparent slope decreases as $L$ increases, as clearly demonstrated
400: in this figure.
401: 
402: We ascribe the change of direction to a crossover effect of the conservative
403:  fixed point.  For $\alpha$ close to 1/4, smaller avalanches behave as
404:  avalanches would in the conservative system.  It is only the larger
405:  avalanches that are affected by nonconservative dissipation.  This is
406:  associated with the fact that as $\alpha$ approaches 1/4, each site
407:  can topple more and more times in a single earthquake.  For any
408:  finite value of dissipation ($1-4\alpha$), the maximum number of
409:  times that a site can reset in an earthquake
410:  is determined by this dissipation and is
411:  finite in the limit of large $L$.  However for small systems, the largest
412:  avalanches are not large enough to be effected by dissipation, and
413:  the cutoff in the number of times that a site can topple is not
414:  determined by ($1-4\alpha$) but by $L$.  Then the  cutoff in the
415:  number of topplings at a given site is the same as in a conservative
416:  system of the same size.
417: 
418: The results we have described thus far are for the model with open boundary
419: conditions.  Fig.~4 shows that  the same
420: behavior occurs for the OFC model with free boundary conditions, with
421: the same value $\tau=1.8$.  In this case, the asymptotic behavior as $L$
422: increases is approached for decreasing apparent slope for both
423: $\alpha=0.17$ and $\alpha=0.20$, as is plainly evident in the figure.
424: Again the cutoff appears to sharpen as $L$ increases and approach $D_{max}=2$.
425: 
426: \begin{figure}[hb]
427: \narrowtext
428: \protect\vspace{0.3cm}
429:  \epsfxsize=4.5in
430: \centerline{\epsffile{fig_4.eps}}
431: \protect\vspace{0.3cm}
432: \caption[4]{\label{mfs_fbc} Plots of $F_{cutoff}$ in systems with free 
433:  boundary conditions for (a) $\alpha=0.17$ and (b) $\alpha=0.20$. 
434:  The exponent has been set to $\tau =1.8$ as before. System sizes are 
435:  $L=32,64,128,256,512$.}
436: 
437: \end{figure}
438: 
439: 
440: Our analysis indicates that the OFC model exhibits SOC with a
441: universal power law distribution of events $P(s) \sim s^{-1.8}$ for
442: $\log s < (D_{max} \log L)$.  As shown in Fig.~3, the cutoff gets sharper
443: and sharper as $L$ increases with $D_{max}$ approaching 2 from below
444: as $L$ increases for all values of $\alpha$ we have studied.  As
445: indicated above, the incorrect results obtained by OFC were due to the
446: fact that there is a strong system size dependence that varies with
447: $\alpha$ and is not described by FSS.  In addition to a systematic
448: change in apparent $\tau$ as $L$ increases, the largest dimension of
449: earthquakes, $D_{max}$, is increasing only slowly towards two, and
450: probability distribution of avalanche dimensions is getting sharper at
451: the cutoff.  This means that large avalanches are suppressed in small
452: systems relative to the total amount of force in the system as compared to a
453: larger system, and that a FSS ``fit'' for any $L$ range will always
454: give an apparent $D>2$, which is not allowed.  In this sense the
455: model appears to violate $FSS$ for all values of $L$.
456: 
457: 
458: \section{Conclusions}
459: 
460: The main results of this paper are as follows: The nonconservative
461: model on a two dimensional lattice self-organizes into a critical
462: state.  The critical state is robust and universal over a range of
463: values of the dissipation parameter, $\alpha$, and for different
464: boundary conditions.  The model does not exhibit finite-size scaling.
465: The cutoff becomes sharper as $L$ increases with largest earthquakes
466: of dimension $D_{max}$ approaching two from below.  Nevertheless, the
467: probability distribution of earthquake sizes is a power law with a
468: universal exponent $\tau \simeq 1.8$.  This value can be identified
469: with an exponent $B=\tau -1$ for the distribution of energy dissipated
470: in earthquakes.  According to the Gutenberg-Richter law this is a
471: power law with $B=0.8 {\rm \ to \ }1$, completely consistent with our
472: result.
473: 
474: \medskip
475: 
476: We thank K. Christensen and H. J. Jensen for helpful conversations,
477: and P. Bak for comments on the manuscript.  S.L. was supported
478: by the EPSRC through a post-doctoral research fellowship. 
479: 
480: 
481: \begin{references}
482: \bibitem{soc} P. Bak, C. Tang, and K. Wiesenfeld, Phys. Rev. Lett. {\bf 59},381
483:  (1987); Phys. Rev. A  {\bf 38}, 364 (1988).
484: 
485: \bibitem{review}
486: For reviews see P. Bak,
487:  {\it How Nature Works: The Science of Self-Organized Criticality} 
488:      (Copernicus, New York, 1996); 
489: H. J. Jensen, {\it Self-Organized Criticality,}
490: (Cambridge University Press, New York, 1998).
491: 
492: 
493: \bibitem{gutenberg}
494: B. Gutenberg and C.F. Richter, Ann. Geofis. {\bf 9}, 1 (1956).
495: 
496: \bibitem{bak-tang}
497: P. Bak and C. Tang,
498: J. Geophys. Res. B {\bf 94}, 15635 (1989).
499: 
500: \bibitem{sornette} A. Sornette and D. Sornette, Europhys. Lett.
501: {\bf 9}, 197 (1989).
502: 
503: \bibitem{ito}
504: K. Ito and M. Matsuzaki,
505: J. Geophys. Res. B {\bf 95}, 6853 (1990).
506: 
507: \bibitem{burridge} R. Burridge and L. Knopoff, Bull. 
508: Seismol. Soc. Am. {\bf 57},341 (1967).
509: 
510: \bibitem{carlson-langer}
511: J. M. Carlson and J. S. Langer,
512: Phys. Rev. Lett. {\bf 52}, 2632 (1989);
513: J. M. Carlson, J. S. Langer, and B. E. Shaw.
514: Rev. Mod. Phys. {\bf 66}, 657
515: (1994).
516: 
517: \bibitem{vieira} M. de Sousa Vieira, Phys. Rev. A {\bf 46}, 6288
518:  (1992); M. de Sousa Vieira {\it et al}, Phys. Rev. E {\bf 53}, 1441
519:  (1996); M. de Sousa Vieira, Phys. Rev. E {\bf 61}, 6056 (2000).
520: 
521: \bibitem{universal} M. Paczuski and S. Boettcher.  
522: Phys. Rev. Lett. {\bf 77}, 111 (1996).
523: 
524: \bibitem{takayasu} H. Takayasu and M. Matsuzaki, Phys. Lett. A {\bf 131},
525: 244 (1988).
526: 
527: \bibitem{nakanishi} H. Nakanishi, Phys. Rev. A {\bf 43}, 6613 (1991).
528: 
529: \bibitem{ofc} Z. Olami, H.J.S. Feder, and K. Christensen,
530: Phys. Rev. Lett. {\bf 68}, 1244 (1992); K. Christensen and Z. Olami,
531: Phys. Rev. A {\bf 46}, 1829 (1992); K. Christensen and Z. Olami,
532: J. Geophys. Res. {\bf 97}, 8729 (1992); K. Christensen, Z. Olami, and
533: P. Bak, Phys. Rev. Lett. {\bf 68}, 2417 (1992).
534: 
535: 
536: 
537: 
538: 
539: 
540: \bibitem{pepke} S. L. Pepke and J. M.  Carlson,
541: Phys. Rev. E {\bf 50}, 236 (1994).
542: 
543: \bibitem{turcotte} D. L. Turcotte,
544: Rep. Prog. Phys. {\bf 62} (10) 1377 (1999).
545: 
546: \bibitem{klein} W. Klein and J. Rundle, Phys. Rev. Lett. {\bf 71}, 1288 (1993);
547:          K. Christensen, Phys. Rev. Lett. {\bf 71}, 1289 (1993).
548: 
549: \bibitem{lise} S. Lise and H. J. Jensen,
550: Phys. Rev. Lett. {\bf 76}, 2326 (1996).
551: 
552: \bibitem{french}
553: M. L. Chabanol and V. Hakim, Phys. Rev. E {\bf 56}, 2343 (1997).
554: 
555: 
556: \bibitem{grass1} H. M. Broker and P. Grassberger, Phys. Rev. E {\bf 56},
557: 3944 (1997).
558:     
559: 
560: \bibitem{kertez} I.M. J\'anosi and J. Kerte\'sz, Physica A {\bf 200}, 174
561: (1993).
562: 
563: \bibitem{socolar} J.E.S. Socolar, G. Grinstein, and C. Jayaprakash,
564: Phys. Rev E, {\bf 47}, 2366 (1993).
565: 
566: \bibitem{grass2} P. Grassberger, Phys. Rev. E {\bf 49}, 2436 (1994).
567: 
568: \bibitem{note} Throughout this work, in both figures and text,
569:  $\log$ is the logarithm in base 10.
570: 
571: \bibitem{middleton}
572: A. A. Middleton and C. Tang,
573: Phys. Rev. Lett. {\bf 74}, 742 (1995).
574: 
575: \bibitem{corral} A. Corral, C. J. Perez, A. Diaz-Guilera, and A. Arenas,
576: Phys. Rev. Lett. {\bf 74}, 118 (1995).
577: 
578: \bibitem{hwa-kardar}
579: T. Hwa and M. Kardar,
580:  Phys. Rev. Lett. {\bf 62}, 1813 (1989);
581: G. Grinstein, D.-H. Lee, and S. Sachdev,
582:  Phys. Rev. Lett. {\bf 64},
583: 1927 (1990).
584: 
585: \bibitem{paczuski}
586: M. Paczuski, Phys. Rev. E {\bf 52}, 2137 (1995).
587: 
588: \bibitem{kadanoff}
589: L. P. Kadanoff, S. R. Nagel, L. Wu, and S. Zhou,
590: Phys. Rev. A {\bf 39}, 6524 (1989).
591: 
592: \bibitem{stella}
593: M. De Menech, A. L. Stella, and C. Tebaldi, Phys. Rev. E 
594: {\bf 58}, 2677 (1998);
595:  C. Tebaldi, M. De Menech, and A. L. Stella, Phys. Rev. Lett.
596: {\bf 83},  3952 (1999).
597: 
598: 
599: 
600: 
601: 
602: 
603: 
604: 
605: 
606: 
607:  
608: 
609:  
610: \end{references}
611: \end{multicols}
612: 
613: 
614: 
615: 
616: 
617: 
618: 
619: 
620: 
621: 
622: 
623: 
624: \end{document}
625: 
626: 
627: