1: %file rchrtex_n2c.tex V.19.4.2000
2: %\documentstyle[aps,preprint,epsf]{revtex}
3: \documentstyle[aps,prl,multicol,epsf]{revtex}
4: %\documentstyle[aps,prl,twocolumn,epsf]{revtex}
5: %\tighten
6:
7: \begin{document}
8:
9: \title{Scaling of avalanche queues in directed dissipative sandpiles}
10:
11: \author{Bosiljka Tadi\'c$^{1,*}$ and Vyatcheslav Priezzhev$^{2,**}$
12: }
13:
14: \address{$^1$Jo\v{z}ef Stefan Institute,
15: P.O. Box 3000, 1001 Ljubljana, Slovenia \\
16: $^2$Bogolubov Laboratory of Theoretical Physics,
17: Joint Institute of Nuclear Research, 141980 Dubna, Russia }
18:
19: %\date\today
20:
21: \maketitle
22: \begin{abstract}
23: \newline
24: Using numerical simulations and analytical
25: methods we study a two-dimensional directed sandpile automaton with
26: nonconservative random defects (concentration $c$) and varying
27: driving rate $r$. The automaton is driven
28: only at the top row and driving rate is measured by the number of
29: added particles per time step of avalanche evolution. The probability
30: distribution of duration of elementary avalanches at zero driving rate
31: is exactly given by $P_1(t,c) = t^{-3/2}\exp{[t\ln(1-c)]}$.
32: For driving rates in the interval $0< r \le 1$
33: the avalanches are queuing one after another, making increase
34: the periods of non-interrupted activity of the automaton.
35: Recognizing the probability $P_1$ as a distribution of service time of
36: jobs arriving at a server with frequency $r$, the model
37: represents a new example of the $\langle E,1,GI/\infty/1\rangle $ server
38: queue in the queue theory. We study scaling properties of the
39: busy period and dissipated energy of sequences of non-interrupted
40: activity. In the limit $c \to 0$ and varying linear system size
41: $L \ll 1/c$ we find that at driving rates $r\le L^{-1/2}$
42: the distributions of duration and energy of the avalanche queues are
43: characterized by a multifractal scaling and we determine the
44: corresponding spectral functions. For $L\gg 1/c$
45: increasing of the driving rate somewhat compensates the energy losses
46: at defects above the line $r \sim \sqrt{c}$. The scaling exponents of
47: the distributions in this region of phase diagram vary approximately
48: linearly with the driving rate. Using properties of recurrent states
49: and the probability theory we determine analytically the exact upper
50: bound of the probability distribution of busy periods.
51: In the case of conservative dynamics $c=0$ the probability
52: of a continuous flow increases as $F(\infty ) \sim r^2$ for
53: small driving rates.
54:
55:
56:
57:
58: \end{abstract}
59: \pacs{PACS numbers: 05.65.+b, 64.60.Ht, 45.70.Ht, 02.50.Hb}
60:
61: %\narrowtext
62: \begin{multicols}{2}
63:
64:
65:
66: \newpage
67:
68: \section{Introduction}
69: In the past decade the sandpile type of cellular automata played
70: a special role in understanding of self-organized criticality in
71: nonlinear dynamical systems (for a recent review see Ref.\ \cite{DD-rev}).
72: In sandpile automata the properties of the dynamics
73: which are essential for the occurrence of self-organized critical states
74: can be monitored in a direct manner. Apart from the relaxation rules,
75: these are the following properties: type of driving and time-scale
76: separation; conservation law of the dynamics; direction of mass flow;
77: role of boundaries. In addition, the Abelian nature of the toppling
78: rules in some sandpile automata enabled derivation of certain exact
79: results \cite{DR,exact}, in contrast to other dynamical systems where
80: such calculations are not available.
81: Numerous sandpile models, both deterministic and stochastic \cite{DD-rev}
82: are shown to exhibit dynamic critical states in the limit of
83: ``infinitely slow'' driving (i.e., at zero driving rate $r=0$). In this
84: limit a new avalanche is initiated
85: only after the previous one has stopped, thus the time scale
86: separation is exactly observed. On the time scale of perturbations,
87: the avalanche evolution is seen as occurring instantly.
88: The existence of the critical states in the case of directed Abelian
89: sandpile automaton at zero driving rate has been proved exactly by Dhar
90: and Ramaswamy \cite{DR}. At this point it is interesting to mention that the
91: model studied in Ref.\ \cite{DR} is characterized by local driving and
92: deterministic conservative dynamics. The automaton with conservative
93: stochastic dynamics, on the other hand, has been shown
94: to belong to another universality class \cite{TD}. The presence
95: of nonconservative defects in Dhar-Ramaswamy automaton leads to a
96: subcritical behavior \cite{Tetal}.
97:
98:
99: The behavior of driven dynamical systems at finite driving rates
100: ($r>0$) represents an important subject both for theoretical and
101: practical reasons. A finite driving rate may appear either as a
102: control parameter set from outside, or as a
103: probability distribution originating from another coupled stochastic
104: process.
105: In practice, the systems are driven by an external field, which oscillates
106: with a finite frequency. Examples are Barkhausen noise \cite{BN-r},
107: integrate and fire oscillators \cite{IFO}, granular material
108: in rotating drums \cite{drums}, etc. Queuing jobs at a server
109: \cite{queue-books}, e.g., in teletraffic, is an example where the
110: frequency of arriving jobs is given by a random process.
111:
112: Theoretically at finite driving rates $r>0$ the probability that
113: a new avalanche starts before previous one has stopped increases with
114: increasing $r$. This obviously
115: leads to different statistics of avalanches, where an avalanche is
116: understood to represent a non-interrupted activity of the system.
117: For large driving rates
118: a continuous flow (an avalanche which never stops) is expected in sandpiles.
119: Similarly, a single spanning cluster may occur in driven disordered systems.
120: Therefore, a time scale separation becomes
121: less and less apparent with increasing $r$. In addition, by
122: increasing driving rates, the local driving looses its strict sense.
123: Thus fast driven sandpiles are placed between strictly local driving,
124: where the system is driven at a single (random) site, and global driving,
125: where the same perturbation applies to all sites in the system.
126: The role of the conservation law (conservation of number of grains in the
127: interior of the system) is also expected to be changed at finite
128: driving rates. In $r\to 0$ limit, locally driven nonconservative systems
129: appear to be subcritical \cite{noncons-local}, whereas when the driving
130: is global the critical states may appear even if the dynamics is
131: dissipative \cite{noncons-global,IFO}.
132: So far neither an exact theory nor a renormalization-group analysis of
133: fast driven critical systems has been done. A general questions as the
134: existence of critical states at finite driving rates and universal
135: scaling properties of the system in the limit of large distances and
136: long times remain yet to be understood.
137:
138: Recently two numerical simulations elucidated certain important
139: properties of sandpiles at finite driving rates.
140: In a $1$-dimensional ricepile model Corral and Paczuski \cite{Al-Maya}
141: have first introduced avalanches for a finite driving rate as
142: non-interrupted periods of activity and have
143: shown that these avalanches diverge at rates $r\ge r_c(L)
144: \sim L^{-0.20}$ for a given $L$.
145: The rational behind this conclusion is that an ever-running avalanche
146: occurs for driving rates $r\sim 1/\langle t\rangle _0$, where
147: $\langle t\rangle _0 \sim L^{z(2-\tau _t)}$ is the average duration of
148: avalanches in zero driving rate \cite{Al-Maya}.
149: In another example Barrat {\it et al.} \cite{BTW-r} have shown
150: that in a $2$-dimensional symmetric
151: Abelian sandpile model mixing of time scales at finite driving rates
152: leads to correlations which appear to violate the fluctuation-dissipation
153: theorem.
154:
155:
156: In this work we study a simple $2$-dimensional model with strictly
157: directed flow of grains and deterministic toppling rules.
158: We add particles {\it only at the top row}
159: with driving rate $r$. The driving rate $r$ is defined as a
160: number of added particles per time step of avalanche evolution.
161: We consider both conservative and nonconservative dynamics.
162: A fraction of sites $c$ are considered as annealed nonconservative
163: defects. By toppling at a defect site two grains are lost, thus
164: affecting the propagation of avalanche below that site \cite{comment-sb}.
165: When $c=0$ the dynamics is strictly conservative.
166: In the $r=0$ limit and $c=0$ the model
167: has been exactly solved by Dhar and Ramaswamy \cite{DR}. The critical
168: states where shown to consist of heights
169: $h=0$ and $h=1$ occurring with equal probability 1/2 at each lattice site.
170: The duration of an avalanche is given by the probability distribution
171: for large $t$
172: \begin{equation}
173: P_1(t) \sim t^{-3/2} \ .
174: \label{p1_0}
175: \end{equation}
176: In the presence of nonconservative defects it has been shown \cite{Tetal,JT}
177: that the screening of the power-law distribution in Eq.\ (\ref{p1_0}) occurs
178: as
179: \begin{equation}
180: P_1(t) \sim t^{-3/2}\exp({-t/\xi }) \ ;\ \xi \sim 1/c \ .
181: \label{p1_c}
182: \end{equation}
183: In this paper we will refer to the distributions in Eqs.\ (\ref{p1_0})
184: and (\ref{p1_c}) as
185: probability distributions of {\it elementary} avalanches, to be
186: distinguished from the {\it combined} avalanches, which
187: occur at finite driving rates and which consist of series of elementary
188: avalanches.
189: Up to relatively high driving rates $r=1$ the model has the property
190: that successive elementary avalanches are running one after the other,
191: in contrast to cases studied in Refs.\ \cite{Al-Maya,BTW-r}, where merging
192: of avalanches may occur at any finite $r$.
193: After an elementary avalanche is over the system is characterized
194: by statistically unchanged distribution of heights, owing to a weak
195: correlation between the avalanches in the recurrent states.
196: A finite probability of avalanche collision, which accelerates flow
197: of grains, occurs in this model only for $r>1$. Here we restrict the
198: study to the case
199: $r\leq 1$, where the formation of avalanche queues is a dominant
200: phenomenon which determines the scaling properties of the system.
201:
202:
203:
204: The problem of avalanche queue in our model can be regarded as an
205: example of the server queue \cite{queue-books} of the class
206: $\langle E,1,GI/\infty,1 \rangle$, which is studied as a model in the
207: analysis of various applied problems, e.g., in telecommunications,
208: insurance, etc. These analogies are made clear by
209: noticing that the avalanches of a directed sandpile model have
210: the corresponding terms in the language of the queue theory as
211: follows: (i) elementary avalanche --- customer;
212: (ii) duration of elementary avalanche --- service time;
213: (iii) driving rate --- frequency of arrivals of customers;
214: (iv) number of elementary avalanches coexisting at a given moment
215: of time --- number of working servers;
216: (v) duration of a combined avalanche --- busy period.
217: The notation $\langle E,1,GI/\infty,1 \rangle$ means that we
218: deal with customers arriving by one and being served by one. The letter
219: $E$ means that the arrival times are generated by a Bernoulli process
220: with the distribution
221: $Prob(t=k) = p^k(1-p); k=0,1,2,... ; p>0$\ .
222: The symbol $GI/\infty$ means that the service times are
223: identical independently distributed (i.i.d.) random variables and the
224: infinite number of
225: servers provides a non-restricted number of customers which can be served
226: simultaneously. Despite a huge literature devoted to this subject
227: \cite{queue-books}, most of papers
228: focus on the distribution of $q_n$, the number of working servers at a
229: given moment of time. The importance of the tail behavior of the busy
230: period distribution for fluid queues in telecommunications, generalized
231: processor sharing and other applications was recently pointed out
232: in Ref.\ \cite{Zwart}.
233: Here we concentrate on some other properties of
234: the queue: the scaling behavior of the busy period and dissipated energy
235: distribution.
236:
237: Given the duration of elementary avalanches in Eq.\ (\ref{p1_0}), we may
238: conclude that the directed sandpile model for $r\leq 1$
239: represents a special case of the queue theory with the power-law
240: distribution of the service time with the exponent
241: $\nu =3/2 -1 =1/2$. This implies
242: that the average service time per customer diverges $\langle t\rangle \to
243: \infty $. In practice, service times are restricted to finite values,
244: which corresponds to the power-law distribution with $1 <\nu <2$
245: \cite{Zwart} . This may explain why the queue
246: with the distribution of the type given in Eq.\ (\ref{p1_0}) has not
247: been studied so far.
248: In our model a finite average duration of elementary avalanches
249: $\langle t\rangle <\infty$ is achieved in two cases: (a) In the case
250: of distribution in Eq.\ (\ref{p1_0}) when the system size $L$ is finite,
251: hence the distribution is truncated at $t=L$; (b) In the case of
252: finite dissipation $c>0$, where the distribution in Eq.\
253: (\ref{p1_c}) has a characteristic duration $\xi <\infty $ for all
254: finite $c$ values (see Refs.\ \cite{Tetal,JT} and below).
255:
256:
257:
258: For a finite $L$ we find a continuous flow phase (F) for low
259: dissipation and large driving rates, and
260: three regions with intermittent behavior of avalanche queues.
261: These are regions with subcritical (S),
262: nonuniversal (N), and multifractal (M) behavior, shown schematically
263: on the phase diagram in Fig.\ 1.
264: When the length separation $L\gg \xi $ holds, we find a
265: line in the $(r,c)$-plane where loss of particles on defects in the
266: interior of the pile becomes ``compensated'' by fast adding of particles
267: from outside. In the region above
268: the compensation line the avalanche queues exhibit a scaling behavior
269: with the scaling exponents depending on the driving rate: The slopes
270: decrease whereas the fractal dimensions increase with driving rate.
271: Cut-offs with a stretch-exponential behavior appear.
272: In the limit $c\to 0$ and when the system size $L\ll \xi $ is varied
273: multifractal scaling properties describe the scaled distributions,
274: rather than a simple finite-size scaling.
275:
276:
277: \narrowtext
278: \begin{figure}
279: \epsfxsize=82mm\epsffile[56 75 392 298]{rchr_fig1.ps}
280: \caption{\label{fig1}Schematic phase diagram for a finite system size
281: $L$. Cross-hatched area represents a crossover region between
282: $\xi \ll L$ (right) and $\xi \gg L$ (left). Regions with distinct
283: behavior of the avalanche queues are shown: subcritical (S),
284: nonuniversal (N), and multifractal scaling region (M), and flow phase (F).
285: Solid line represents the compensation line. Transition to the flow
286: phase is marked by $\diamond$ for $c=0$ and by dotted line for small
287: $c>0$. In the origin only single Dhar-Ramaswamy avalanches occur.}
288: \end{figure}
289:
290:
291:
292: In general, the distributions of combined avalanches are characterized
293: by a scaling function of two arguments of the form
294: \begin{equation}
295: P(X,r,L_c) \sim X^{-\tau _X}{\cal{P}}(XL_c^{-D_X},rL_c^{1/2}) \ ,
296: \label{scal-general}
297: \end{equation}
298: where $X$ represents either
299: duration, $t$, or number of topplings \cite{comment-n}, $n$, and $L_c
300: \equiv \min{(\xi ,L)}$.
301: The corresponding fractal dimensions $D_X$ are defined by
302: \begin{equation}
303: \langle X\rangle _{\ell} \sim \ell ^{D_X} \ ,
304: \label{fract-dim}
305: \end{equation}
306: where the average is taken over all combined avalanches of a selected length
307: $\ell $ measured along the direction of transport.
308:
309:
310:
311: Using analogy to the queue theory and the properties of the
312: recurrent states we were able to derive an exact upper limit of the
313: distributions of busy periods and to discuss the limit $L\to
314: \infty$. We also derived the expression for the probability of continuous
315: flow in the conservative limit.
316:
317:
318:
319: The paper is organized as follows: In Sec.\ II we define the model
320: and consider the case of conservative dynamics by numerical
321: simulations on finite lattice.
322: In Sec.\ III we present results of simulations in the case of finite
323: concentration of nonconservative defects. In Sec.\ IV we present details
324: of the analytical results. The paper contains a short summary of the
325: results and discussion in
326: Sec.\ V.
327:
328:
329:
330:
331:
332:
333:
334:
335: \section{Multifractal queues of Dhar-Ramaswamy avalanches}
336:
337:
338: The sandpile automaton model introduced by Dhar and Ramaswamy represents
339: an example of a self-organized criticality with exact solution in the
340: limit of zero driving rate \cite{DR}. In this Section we consider the
341: same model at finite driving rates $0 <r\leq 1$. The dynamic rules of
342: the model are summarized as follows \cite{DR}: We consider a $2$-dimensional
343: square lattice oriented downwards, with a dynamic variable---height
344: $h(i,j)$---associated at each site. Grains are added at the top row only,
345: and mass flow is only down. The toppling
346: at a site $(i,j)$ occurs deterministically whenever $h(i,j) \geq h_c=2$,
347: and two grains are transferred downward, i.e.,
348: \begin{equation}
349: h(i,j)\to h(i,j)-2 \ ;\ h(i+1,j_\pm )\to h(i+1,j_\pm )+1 \ ,
350: \label{rules}
351: \end{equation}
352: where $(i+1,j_\pm )$ represents two downward neighboring sites
353: to the site $(i,j)$.
354:
355: The probability distribution of duration of avalanches in zero driving
356: rate $P_1(t) \sim t^{-\tau _t^0}{\cal{P}}(tL^{-1})$ with $\tau _t^0=3/2$
357: given in Eq.\ (\ref{p1_0}) becomes exact at large $t$ \cite{DR}.
358: In addition the dynamic exponent $z^0=1$. This implies that
359: the average duration $\langle t\rangle _0 \sim L^{1/2} \to \infty $.
360: Similarly, the area $s$ enclosed by the boundary of an avalanche is given
361: by the distribution $D(s,L) \sim
362: s^{-\tau _s^0}{\cal{D}}(sL^{-D_s^0})$, where $\tau _s^0=4/3$ and the
363: fractal dimension $D_s^0=3/2$. Note that the number of toppled grains
364: at each active site is two, then the distribution of the number of toppled
365: grains within an avalanche, $D(n)$, is described by the same exponents,
366: i.e., $\tau _n^0=4/3$ and $D_n^0=3/2$ at zero driving rate.
367:
368: A finite driving rate $r>0$ is implemented as follows. An avalanche is
369: started from the top and at each step of the avalanche progress a new
370: particle is added with probability $r$ at a random site at first
371: row. We also consider a deterministic addition of particles, i.e.,
372: we add a particle at regular intervals $\Delta t$. Both approaches
373: lead to the same results when the statistics is high enough.
374: An added particle
375: may trigger a new elementary avalanche before the previous one stops,
376: thus making a pattern of active sites distributed over lattice. A
377: snapshot of growth of a combined avalanche with marked active sites is
378: shown in Fig.\ 2 (top).
379: A combined avalanche---avalanche queue---is thus determined by a
380: non-interrupted activity on the lattice and it stops when no more active
381: sites occur. Then a new avalanche is started.
382: It should be noted that when $r>0$ the number of added grains is higher
383: than the number of combined avalanches. Another important remark is that
384: the repeated toppling at
385: a site may occur as soon as $r>0$. This leads to the inequality
386: $\langle n\rangle > \langle s\rangle $, and thus $D_n >D_s$, and $z>1$.
387: The numerical simulations confirm these conclusions (see below).
388: Typically, we consider $2\times 10^6$ combined avalanches at each
389: driving rate and lattice size. Periodic boundary conditions are
390: applied in the perpendicular direction.
391:
392: \begin{figure}
393: \epsfxsize=52mm\epsffile[188 278 421 515]{rchr_fig2a.ps}
394:
395: \epsfxsize=52mm\epsffile[156 256 455 536]{rchr_fig2b.ps}
396: \caption{\label{fig2} Top: Growth of a combined avalanche in the case
397: of conservative dynamics ($c=0$) for driving rate $r=0.05$ and $L=100$.
398: Nine fronts of active sites (dark) are visible.
399: Bottom: A complete combined avalanche in the case of dissipative dynamics with
400: $c=0.02$ and driving rate $r=0.5$. Different colors correspond to distinct
401: toppling waves.}
402: \end{figure}
403:
404:
405:
406:
407: In this Section we perform numerical simulations for $r>0$ and
408: finite lattice sizes $L$. The limit $L\to \infty$ will be discussed in
409: Sec.\ IV. In main Figs.\ 3 and 4 are shown the integrated probability
410: distributions of duration $P(t^\prime \ge t,L)$ and number of topplings
411: $D(n^\prime \ge n,L)$ for fixed driving rate $r=0.05$ and various
412: lattice sizes $L$. It should be noted that both slopes and cut-offs
413: of these distributions appear to be different compared to ones of
414: the elementary avalanches. In particular, slopes decrease with $r$
415: (see more detailed discussion in next Section). For instance at $r=0.05$
416: we find $\tau _t =0.4$ and $\tau _n=0.3$, in the steep part, and
417: $\tau _t =0.31$ and $\tau _n=0.2$ in the flat part near the cut-off.
418: A cut-off in the probability
419: distribution of durations appear (cf. main Fig.\ 3). The characteristic
420: jump at $t=L$ is related to the conditional probability: an activity
421: lasts longer than $L$ steps only if the preceding avalanches last
422: not shorter than $t=L$. The jump decreases with increasing lattice size.
423:
424: \begin{figure}[thb]
425: \epsfxsize=80mm\epsffile[43 69 508 564]{rchr_fig3.ps}
426: \caption{\label{fig3}Double logarithmic plot of the integrated probability
427: distribution of duration (busy periods) $P(t,L)$ of queues of Dhar-Ramaswamy
428: avalanches vs. duration $t$, measured in Monte Carlo steps [MCS].
429: Fixed driving rate $r=0.05$ and various
430: lattice sizes $L=$24,48,96,192, and 384 are used.
431: Inset: Multifractal spectral function $\Phi _t(\alpha _t)$ vs. $\alpha _t$
432: obtained from the data in main figure according to Eq.\ (6) using
433: $X_0=1$, $L_0=2$.}
434: \end{figure}
435: It is interesting that these distributions can not be scaled according
436: to a simple finite-size scaling (with new exponents), as
437: one may naively expect. Instead, we find that a multifractal scaling
438: applies according to the law
439: \begin{equation}
440: P(X,L,r) = \left({L\over{L_0}}\right)^{\Phi _X(\alpha _X)} \ ; \
441: \alpha _X = {{\log(X/X_0)}\over{\log(L/L_0)}} \ ,
442: \label{mulifractal}
443: \end{equation}
444: where, as before, $X$ stands for $t$ or $n$. The corresponding spectral
445: functions $\Phi _t(\alpha _t)$ and $\Phi _n(\alpha _n)$ are determined
446: numerically for $r=0.05$ and shown in the insets to Figs.\ 3 and 4,
447: respectively. The spectrum depends on the driving rate. For driving
448: rates close to the line $r\sim L^{-1/2}$ extremely large avalanches
449: may appear and the scaling fits fail.
450: The origin of multifractality in the queues of Dhar-Ramaswamy avalanches
451: can be found in the fact that an unrestricted multiple toppling at
452: each site may occur, and that a toppling at a given site releases a local
453: avalanche which propagates from that site downwards.
454:
455: \begin{figure}[thb]
456: \epsfxsize=80mm\epsffile[43 69 508 564]{rchr_fig4.ps}
457: \caption{\label{fig4}Same as Fig.\ 3 but for the integrated distribution
458: of number of topplings (mass) $D(n,L)$ vs. mass $n$ [number of grains].
459: Shown are only curves for $L=$96,192, and 384. Inset: Spectral function
460: $\Phi _n(\alpha _n)$ vs. $\alpha _n$. $X_0=0.97\pm 0.02$, $L_0=2.15
461: \pm 0.05$.}
462: \end{figure}
463:
464: \begin{figure}[thb]
465: \epsfxsize=80mm\epsffile[38 73 449 298]{rchr_fig5.ps}
466: %\vskip 8mm
467: \caption{\label{fig5} Attempted finite-size scaling fit of the data
468: from main Fig.\ 3 and 4 for the distribution of avalanche duration (right)
469: and mass (left).}
470: \end{figure}
471:
472: For comparison we show how a finite-size scaling fit of the
473: data fails. In Fig.\ 5 we present an attempt of scaling collapse of the
474: data shown in Figs.\ 3 and 4 above, according to the formula $P(X,L)=
475: L^{-\lambda }{\cal{P}}(XL^{-D_X})$.
476: Note that the best collapse of the tails of distributions are obtained
477: by fixing the fractal dimensions as $D_t=1.20$ and $D_n=1.75$, which
478: correspond to $\alpha _X$ at the shoulder of
479: the spectrum $\Phi _t(\alpha _t)$ and $\Phi _n(\alpha _n)$,
480: respectively. Fixing a smaller (larger) value for the fractal dimension
481: leads to systematic shifts of the distribution tails
482: to the right (left) with increasing $L$.
483: The best fit shown in Fig.\ 5 is obtained for $\lambda =0.35$, which
484: satisfies (within numerical accuracy)
485: the scaling relation $\lambda =z(\tau_t-1) =D_n(\tau_n-1)$
486: with $\tau _X-1$ determined at the flat part of the distribution.
487: However, as the Fig.\ 5 shows, fixing $D_X$ and $\lambda $ leads to
488: the systematic shifts of the 'horizontal' part of the distributions
489: to the right with decreasing $L$. Fixing the exponents independently
490: from the scaling relation results in crossing of the lines for
491: different $L$ values.
492:
493:
494:
495: For driving rates $r> L^{-1/2}$ an ever-running avalanche may occur,
496: representing a continuous activity on the lattice. The flow phase
497: can be characterized by an average number of topplings per site,
498: which is expected to have a nontrivial $L$-dependence. The probability
499: of occurrence of the flow phase in the limit $L\to \infty $ will be
500: discussed in Sect.\ IV.
501:
502:
503: \section{Nonuniversality in dissipative dynamics}
504:
505: In the presence of dissipative defects $c>0$ the distribution of
506: elementary avalanches, which is given in Eq.\ (\ref{p1_c}), appears to
507: have a finite characteristic length $\xi <\infty $. Thus the average
508: duration at zero driving rate is finite $\langle t\rangle _0 <\infty $.
509: Precise value of the average duration is controlled by an external
510: parameter---probability of dissipation $c$, and not by system size
511: $L$, provided that $L> \xi$. The screening length $\xi \sim
512: 1/c$ was first estimated numerically in Ref.\ \cite{Tetal}. A more
513: precise analytical expression can be derived (see below and Ref.\
514: \cite{JT}) as $\xi ^{-1} \sim -\ln (1-c)$.
515: Here we perform numerical simulations in the case $L > \xi $ at
516: driving rates $0 < r \leq 1$.
517: In this range of driving rates we expect the role of lattice size
518: in the analysis of Sec.\ II is to be replaced by the characteristic
519: length $\xi $. In addition, competition between dissipation and
520: driving rate leads to new phenomena.
521:
522: In Fig.\ 2 (bottom) an example of a combined avalanche is shown
523: for $c=0.02$ and driving rate $r=0.5$. It is remarkable that the
524: number of topplings per site decrease with distance from the top.
525: Intermittency of the dynamics as well as the occurrence of the
526: long-range correlations can be seen by direct examination of
527: the recorded activity of the system $n(t)$ at each time step. For
528: the server queue the quantity $n(t)$ is interesting as the measure
529: of the energy which is dissipated by the server at a given moment of
530: time $t$. In Fig.\ 6 we show an example of the recorded signal for
531: certain choice of parameters $r$, $c$, and $L$ corresponding to
532: the region (N) of the phase diagram (cf. Fig.\ 1).
533: A combined avalanche on this recording
534: is represented by a set of peaks between two consecutive drops of
535: the signal to the base line. The Fourier spectrum of the signal (shown
536: on top panel in Fig.\ 6) exhibits a power-law behavior. The slope
537: $\varphi \approx 0.9$ weakly increases with driving rate $r$.
538:
539:
540: \begin{figure}[thb]
541: \epsfxsize=80mm\epsffile[43 69 508 746]{rchr_fig6.ps}
542: %\vskip 8mm
543: \caption{\label{fig6} Sandpile noise $n(t)$---number of active sites at
544: time $t$, plotted against time $t$ [MCS] (lower panel) and its
545: Fourier spectrum (top panel) for fixed driving rate r=0.33.
546: Dissipation rate $c=0.01$ and $L=128$. Shaded area shown in the lower
547: panel is an example of unit signal, corresponding to a combined avalanche.}
548: \end{figure}
549:
550:
551:
552:
553:
554:
555: The scaling properties of the distributions of avalanche queues
556: depend on the mutual ratio of the driving and dissipation rates.
557: In particular, the scaling function in Eq.\ (\ref{scal-general})
558: exhibits a nontrivial dependence on both arguments $x\equiv t\xi ^{-1}$
559: and $y\equiv r\xi ^{1/2}$. (Another suitable choice of variables
560: would be $(tc,rt^{1/2})$.)
561: In Fig.\ 7 we show the distribution of the avalanche mass (dissipated energy)
562: $n$ for fixed $c=0.01$ and various driving rates $r$. In general, the
563: cut-offs of the distributions increase and slopes decrease with increased
564: driving rates $r$. More detailed analysis of the slopes shows that for
565: the range of values of driving rates
566: the scaling behavior of the distributions can be described by the
567: scaling exponents which depend on the driving rate.
568: The slopes of various distributions and the corresponding fractal
569: dimensions, which are defined in Eq.\ (\ref{fract-dim}), are
570: shown against driving rate in Fig.\ 8. The dissipation rate
571: is fixed $c=0.01$. Note that $\tau _{\ell}$ in Fig.\ 8 represents
572: slope of the distribution of the largest linear length reached in a
573: combined avalanche. For a range of values of driving rates
574: the scaling exponents decrease and the fractal
575: dimensions increase with $r$, while the
576: scaling relations between various exponents are found to be satisfied
577: within numerical error bars.
578: \begin{figure}
579: \epsfxsize=80mm\epsffile[43 69 508 564]{rchr_fig7.ps}
580: %\vskip 8mm
581: \caption{\label{fig7}
582: Distribution of avalanche mass $D(n,r)$ vs. mass $n$ [number of grains]
583: shown in double logarithmic scale for fixed dissipation rate $c=0.01$
584: and $ L=192$ and for
585: various driving rates $r$= 0.021,0.041,0.083,0.125,0.166,0.25, and 0.33
586: (left to right). Inset: Average mass (top), area (middle) and duration
587: (bottom) of combined avalanches plotted against driving rate [MCS$^{-1}$]
588: for fixed $c=0.01$ and $L=192$. At each point average is taken over
589: $2\times 10^6$ combined avalanches. Fitting curves: $\langle n\rangle
590: =167\times \exp{(11.4r^{1.28})}$, and $\langle t\rangle =23\times
591: \exp{(9.6r^{1.42})}$.}
592: \end{figure}
593: The variation of the scaling exponents
594: can be approximated by a linear dependence of $r$.
595: It is interesting to note that a qualitatively similar behavior---linear
596: variation of the scaling exponents with driving rate, has been measured
597: experimentally in the
598: case of Barkhausen noise in driven disordered ferromagnets \cite{BN-r}.
599: Our present analysis suggests that such behavior can be related
600: to an interplay of driving rate and dissipation at defects, and that
601: it applies more universally.
602: Dependence of the fractal dimensions (and of the slope exponents
603: via scaling relations) of the avalanche queues on driving rate $r$ can
604: be linked to the $r$-dependence of the average length of the queue
605: $<N> =1+r<t>_0$.
606: It has been shown recently \cite{Romu} that the fractal spectrum of the
607: series of elementary signals in the case of transit times in the ricepile
608: model varies as a power of the length of series.
609: A precise $r$-dependence of the exponents
610: in the case of avalanche queues requires additional work and will be
611: given elsewhere.
612:
613:
614:
615:
616:
617:
618:
619: \begin{figure}
620: \epsfxsize=80mm\epsffile[52 70 449 564]{rchr_fig8.ps}
621: %\vskip 8mm
622: \caption{\label{fig8} Lower panel: Various scaling exponents of the
623: avalanche queues plotted against driving rate for fixed dissipation
624: $c=0.01$ and $L=192$. $*$ indicates the products
625: $z(\tau _t-1)\approx D_n(\tau _n-1)\approx D_s(\tau_s-1) \approx
626: 1.1(\tau _{\ell}-1)$. Also shown are linear fits of the data.
627: Top panel: Amplitudes $\xi_n$ (upper curve) and $\xi_t$ defined in
628: Eq.\ (7) vs driving rate. Lines: fit curves satisfying
629: Eq.\ (8) with $B_n=6.29\pm 0.46$ and $B_t=4.5\pm 0.51$.}
630: \end{figure}
631:
632:
633: The opposite effects on the exponents are obtained by increasing concentration
634: of defects $c$ at fixed driving rate. In particular, the slopes of the
635: distributions increase and fractal dimensions decrease with {\it increasing}
636: concentration of defects in a limited range \cite{comment}.
637:
638: An exact expression for the scaling function ${\cal{P}}(x,y)$ in Eq.\
639: (\ref{scal-general}) can not be guessed.
640: It appears that the cut of the surface ${\cal{P}}(x,y)$ at
641: $r=const$ for well balanced values of $c$ and $r$ can be
642: approximated by a stretch-exponential function, so that we have
643: \begin{equation}
644: P(X,c,r) = X^{-\tau _X(r)}\exp{(-X^{\sigma _X}/\xi _X(r))} \ .
645: \label{stretch-ex}
646: \end{equation}
647: Here $X$ stands for $n$ or $t$, and we find $\sigma _n =1.14 \pm 0.04$
648: and $\sigma _t =1.28\pm 0.06$, for the distribution of energy and duration,
649: respectively. The amplitudes $\xi_X(r)$ can be fitted (see top panel
650: of Fig.\ 8) by the following function of driving rate $r$
651: \begin{equation}
652: \xi_X(r) = A_X(c)\exp(rB_X(c))\ ,
653: \label{xi-r}
654: \end{equation}
655: for a fixed dissipation rate $c$.
656:
657: The observed parameter-dependence of the probability distributions is also
658: reflected in the behavior of the average duration
659: and energy of combined avalanches. Notice that the average duration
660: $\langle t\rangle $ represents the average busy period of a server in the
661: queue theory. Beside the average duration $\langle t\rangle $, we also
662: compute the average values of number of topplings (energy) $\langle
663: n\rangle $ and area $\langle s\rangle $ (total number of distinct sites)
664: affected by the processing of a combined avalanche. These average values are
665: shown vs. $r$ in the inset to Fig.\ 7 for $c=0.01$. The values
666: $\langle t\rangle $ and $\langle n\rangle $ increase with driving rate
667: faster than an exponential function. Fitting the data in the inset to Fig.\ 7
668: we find
669: \begin{equation}
670: \langle X\rangle = a_{0X}\exp{(a_{1X}(c)r^{\sigma })} \ ,
671: \label{averages}
672: \end{equation}
673: where $X\equiv n, t$ and we estimate $\sigma = 1.2 \pm 0.1$.
674: Note that, in contrast to durations and energies, the average area
675: of an avalanche is bounded
676: by the number of cells in the system $\langle s\rangle \leq L^2$.
677:
678: As already mentioned above, for a given dissipation rate $c$ and
679: $L\gg \xi$ a driving rate $r_0(c)$ exists such that fast addition
680: of grains compensates the losses in the bulk. In fact along an extremal line
681: \begin{equation}
682: r_0(c) \sim \kappa \xi ^{-1/2} \approx \kappa \sqrt{c} \
683: \label{comp-line}
684: \end{equation}
685: the coherence length remains constant.
686: The existence of the compensation line Eq.\ (\ref{comp-line}) can be
687: demonstrated by considering sets of data for average durations and
688: energies against driving rate $r$, obtained for different characteristic
689: length $\xi \sim 1/c$. These data can be scaled according to the
690: following scaling form
691: \begin{equation}
692: \langle X\rangle \xi ^{-D_X(2-\tau _X)} =
693: {\cal{G}}\left(r\xi ^{z(2-\tau _t)} - \kappa \right) \ .
694: \label{FSS-c}
695: \end{equation}
696: The corresponding scaling fits for the two cases $X\equiv t$ and
697: $X\equiv n$ are shown in Fig.\ 9. Notice that the respective exponents
698: $D_n(2-\tau _n)=1$ and $z(2-\tau _t)=1/2$ are exact values,
699: thus leaving only one parameter, namely $\kappa$, to be determined by
700: the fitting procedure. This is an advantage of having the exact solution
701: for the elementary avalanches \cite{DR}.
702: >From the best fit we find $\kappa =1.14 \pm 0.1$ in the given range of
703: values of $c$ (see caption to Fig.\ 9).
704: It is evident from Fig.\ 9 that the scaling function defined in
705: Eq.\ (\ref{FSS-c}) increases faster than an exponential.
706:
707:
708:
709:
710:
711: In the simulations a continuous flow may occur in the case of
712: dissipative dynamics at finite lattice size $L$
713: when the driving rate is increased. However, with increased system
714: size $L$ the behavior is different
715: from the one in the case of conservative dynamics discussed above.
716: >From the numerical simulations alone we can not distinguish if the
717: affected area of a continuous avalanche diverges
718: with the system size $L\to \infty $, or it remains finite for the range of
719: driving rates considered here.
720: We will also discuss large $L$ limit in Sec.\ IV.
721:
722:
723:
724: We have restricted our analysis to the case where the degree of dissipation
725: is such that $\xi \ll L$. For $c \to 0$, however, we have that
726: $\xi \to \infty $,
727: thus the role of $ L$ and $\xi$ is interchanged at some finite $L$.
728: In the reverse limit when $L \ll \xi $ the behavior is expected to be
729: similar to the case of conservative dynamics at finite $L$, studied in
730: Sec.\ II.
731:
732:
733: \begin{figure}[thb]
734: \epsfxsize=78mm\epsffile[43 69 508 564]{rchr_fig9.ps}
735: \caption{\label{fig9} Finite-size scaling plot of the average
736: duration $\langle t\rangle $ (top) and mass
737: $\langle n\rangle $ (bottom) according to Eq.\ (\ref{FSS-c}).
738: Data are taken for values of $c$=0.01,0.0125,0.015, and 0.02
739: such that $\xi <L=192$ is satisfied.
740: Note that the exponents are exact and the fitted value
741: $\kappa = 1.14 $ within numerical error bars.}
742: \end{figure}
743:
744:
745:
746:
747:
748:
749:
750:
751:
752:
753:
754:
755:
756:
757: \section{Analytical results}
758:
759: We start the analytical description of the model with the
760: assumption that individual avalanches, which form each combined
761: avalanche, are statistically independent. By the definition of the model,
762: each toppling in a single avalanche occurs later than those in the
763: previous avalanches, so that the individual avalanches never intersect
764: in the space-time points. Nevertheless, the next avalanche is sensitive
765: to the configuration of occupation numbers left by the previous avalanches.
766: In this way the individual avalanches are dependent on preceding avalanches.
767: On the other hand, it is known from Abelian properties
768: of the directed sandpile model \cite{DR}, that the recurrent state is
769: characterized by the independent distribution of occupation numbers
770: zero and one at each site. Hence, one can expect that this property
771: of recurrent state provides independent distribution
772: of single avalanches at least for asymptotically large systems.
773: This assumption allows us to consider the process of driving of
774: the directed sandpile automaton as a sequence of independent identically
775: distributed (i.i.d.) events. Another important consequence of the
776: statistical independence is that stops of combined
777: avalanches can be considered as recurrent events, i.e., the probability of
778: two successive stops $Prob(t_1,t_2)$ at the moments $t_1$ and $t_1+t_2$ is
779: given by the product $Prob(t_1)Prob(t_2)$.
780:
781:
782: We consider the
783: probability distribution $F(t)$ that a combined avalanche
784: starting at the moment $t=0$ stops {\it for the first time}
785: at the discrete moment $t$. This means that $F(t)$ is the probability
786: that the stop of all preceding elementary avalanches occurs
787: till the moment $t$. Thus, $F(t)$ coincides with the probability
788: distribution of duration of combined avalanches, when an ensemble of
789: events is considered.
790: Along with $F(t)$, it is useful to consider the function $U(t)$ which
791: is defined as the probability that all preceding single avalanches stop
792: to the moment $t$ {\it regardless} of how many stops of
793: combined avalanches occurred before $t$.
794: Noting that stops of combined avalanches are recurrent events, we can
795: write for $F(t)$ and $U(t)$ the following identity \cite{Feller}
796: \begin{equation}
797: U(t)=F(1)U(t-1)+F(2)U(t-2)+...+F(t)U(0)
798: \label{2}
799: \end{equation}
800: where it is convenient to put F(0)=0 and U(0)=1. For the generating
801: functions defined by
802: \begin{equation}
803: \label{3}
804: u(s)=\sum_{t=0}^{\infty} s^t U(t)
805: \end{equation}
806: and
807: \begin{equation}
808: \label{4}
809: f(s)=\sum_{t=1}^{\infty} s^t F(t)
810: \end{equation}
811: one easily gets from Eq. (\ref{2}) the known equation of the theory of
812: recurrent events \cite{Feller}
813: \begin{equation}
814: \label{5}
815: f(s)=\frac{u(s)-1}{u(s)} \ .
816: \end{equation}
817: The total probability that a combined avalanche ever stops is
818: given by $f(1)$.
819: Therefore, the probability that a combined avalanche never stops,
820: i.e., the probability of a continuous flow, which is given by
821: \begin{equation}
822: \label{6}
823: F(\infty)=1-f(1),
824: \end{equation}
825: does not vanish if $u(1)$ in Eq.\ (\ref{5}) is finite $u(1)<\infty$ .
826:
827:
828: \subsection{Case $c=0$}
829:
830: In the case of conservative dynamics ($c=0$) the
831: probability distribution of durations of elementary avalanches
832: is given by Eq.\ (\ref{p1_0}).
833: Then we can estimate the probability $U(t)$ as follows.
834: Let $\Delta t \equiv 1/r$ be the average
835: time interval between addition of successive particles to the first row.
836: The probability $Prob(x \leq t)$ that a single avalanche has duration
837: less than $t$ is
838: \begin{equation}
839: \label{8}
840: Prob(x \leq t) \sim 1- \frac{b}{t^{1/2}}
841: \end{equation}
842: for large $t$, where $b$ is a constant of the order unity when $L$
843: is large. Then for times $t \gg \Delta t$ we have
844: \begin{equation}
845: \label{9}
846: U(t) \sim \left(1- \frac{b}{t^{1/2}}\right)\left(1-
847: \frac{b}{(t-\Delta t)^{1/2}}\right)\cdots \left(1-
848: \frac{b}{(\Delta t)^{1/2}}\right) \ .
849: \end{equation}
850: Introducing $k=t/\Delta t$ we can write Eq.\ (\ref{9}) as the sum
851: \begin{equation}
852: \label{10}
853: \ln U(t) = \sum_{n=1}^{k}\ln \left(1-\frac{b}{(n \Delta t)^{1/2}}\right)
854: \leq - \frac{k^{1/2}}{(\Delta t)^{1/2}} \ .
855: \end{equation}
856: Approximating the sum by an integral leads to
857: \begin{equation}
858: \label{11}
859: U(t)\sim \left({{\sqrt{rt}-b\sqrt{r}}\over{1-
860: b\sqrt{r}}}\right)^{-b^2r}\exp{[-2b(r\sqrt{t}-\sqrt{r})]} \
861: \end{equation}
862: for $t\gg 1$ and $0 < r \leq 1$.
863: For the infinite lattice there exists a constant $c_2$ such that
864: \begin{equation}
865: \label{12}
866: u(1) \leq c_2 \sum_{t=0}^{\infty} \exp (- r\sqrt{t}) < \infty \ .
867: \end{equation}
868: Therefore, we can find that for all finite driving rates $r>0$
869: there is a non-zero probability of continuous flow. In particular,
870: the sum in Eq. (\ref{12}) diverges at small $r$ as
871: \begin{equation}
872: \label{13}
873: u(1) \sim \frac{1}{r^2}
874: \end{equation}
875: leading to the probability of stop $f(1) \sim 1-2r^2b^2 $, which
876: decreases from unity as soon as a finite driving rate is applied. Then the
877: probability of continuous flow Eq.\ (\ref{5}) increases from zero by the
878: same amount, i.e.,
879: \begin{equation}
880: \label{14}
881: F(\infty) \sim r^2, r \to 0 \ .
882: \end{equation}
883: For large $r$ we expect that
884: \begin{equation}
885: \label{15}
886: 1 - F(\infty)\sim \exp (-r) \ .
887: \end{equation}
888: If the size of the system is finite ($L<\infty $) the probability of
889: stops $U(t)$ is bounded from {\it above} by a finite value
890: \begin{equation}
891: \label{16}
892: U(t) \le (rL)^{-b^2r/2}\exp (-2brL^{1/2} )\ ,
893: \end{equation}
894: which follows from Eq.\ (\ref{11}) taking only the dominant $t$-behavior
895: for $t=L \gg 1$.
896:
897:
898:
899: \subsection{Case $c>0$}
900:
901: In the case of finite dissipation rate $c>0$ the dissipation
902: leads to a finite characteristic length $\xi $ in the
903: distribution of elementary avalanches in Eq.\ (\ref{p1_c}).
904: This can be easily demonstrated using mean-field arguments
905: \cite{Ivash.Priez} in reaction-diffusion systems.
906: Let us suppose that at each site of the lattice
907: one of species $A$ or $B$ is living. These species represent two
908: possible states of the original model: $A$ corresponds to the empty
909: site, $B$ to the occupied site.
910: Due to the external driving force, new particles $\phi$ are added
911: to the first row of the lattice at rate $r$. The
912: propagation of particles can be described by the following rules
913: \begin{equation}
914: \label{18}
915: A+\phi \rightarrow B,\ B+\phi \rightarrow A+2\phi \ .
916: \end{equation}
917: The kinetic equations corresponding to this scheme of ``chemical'' reactions
918: are
919: \begin{equation}
920: \label{19}
921: \dot{n}_A(\ell )=n_{\phi}(\ell )[n_B(\ell )-n_A(\ell )]
922: \end{equation}
923: \begin{equation}
924: \label{20}
925: \dot{n}_B(\ell )=n_{\phi}(\ell )[n_A(\ell )-n_B(\ell )]
926: \end{equation}
927: \begin{equation}
928: \label{21}
929: \dot{n}_{\phi}(\ell )=-n_{\phi}(\ell )+2n_{\phi}(\ell -1)n_B(\ell )(1-c)
930: \end{equation}
931: where $n_A(\ell )$, $n_B(\ell )$, and $n_{\phi}(\ell )$ are concentrations
932: of species $A$, $B$ and $\phi$, respectively, at the $\ell $-th row.
933: In the steady state we have
934: $\dot{n}_A=\dot{n}_B=\dot{n}_{\phi}=0$ and Eqs.\ (\ref{19})-(\ref{21})
935: lead to the simple conditions for the concentrations
936: $n_A=n_B=1/2$ and \cite{comment-addition}
937: \begin{equation}
938: \label{23}
939: n_{\phi}(\ell )=r(1-c)^\ell \ .
940: \end{equation}
941: For $c>0$, the density of particles $\phi$ and, hence, the number of
942: topplings in an avalanche decay exponentially with the distance $\ell $
943: from the top as $n_{\phi}(\ell )=r\exp(-\ell/\xi )$. This implies that
944: the characteristic length of the avalanche
945: is $\xi ^{-1} \sim -\ln(1-c) \sim c$.
946: Therefore the above results, in particular Eq.\ (\ref{16}), obtained for
947: the case of finite lattices $L$ and $c=0$
948: apply also for the case $c>0$ by the substitution $L\to L_c$ with
949: \begin{equation}
950: \label{24}
951: L_c =\min\{\xi ,L \} \ .
952: \end{equation}
953:
954: \subsection{Bounds for the busy time}
955:
956:
957: If the combined avalanches are finite (i.e., there is no continuous flow),
958: we can estimate
959: their average duration using the known theorem from the theory of recurrent
960: events (Ref.\ \cite{Feller}, Ch. XIII, Theorem 3 ). According to this theorem,
961: the inverse average time of combined avalanches $\langle t \rangle ^{-1}$
962: coincides with the limit of the sequence $U(t)$ when $t \rightarrow \infty$.
963: Using the bound for $U(t)$ given by Eq.\ (\ref{16}), we get
964: \begin{equation}
965: \label{29}
966: \langle t \rangle \geq (rL)^{b^2r/2}\exp (2brL^{1/2} )
967: \end{equation}
968: The true asymptotics of $\langle t \rangle$ possibly contains an
969: additional prefactor $L^{1/2}$ like in the case for $r=0$. Notice that
970: we get numerically that the average duration as a function of $r$ (cf.
971: inset to Fig.\ 7) increases faster than
972: the exponential, which agrees with Eq.\ (\ref{29}).
973:
974: The combination $rL^{1/2}$ appears as a characteristic parameter
975: determining the duration of combined avalanches. Thus, for $\xi < L$ it
976: follows from Eqs.\ (\ref{29}) and (\ref{23}) that the coherence length
977: remains constant if $r$ varies with $c$ as $r \sim \sqrt{c}$, i.e. the
978: increasing driving rate compensates the dissipation.
979:
980:
981:
982: Another interesting feature of the probability distributions at finite
983: driving rates is the occurrence of a stretch-exponential cutoffs both
984: in dissipative and non-dissipative case (cf. Figs.\ 3,4,7).
985: Indeed, we can see from Eq.\ (\ref{2}) that $U(t)>F(t)$ for all finite $t$.
986: Therefore, for non-dissipative case we have an exponential decay of
987: combined avalanches
988: \begin{equation}
989: \label{25}
990: P(t) \sim F(t) < (rt)^{-b^2r/2}\exp(-2br\sqrt{t})
991: \end{equation}
992: in the thermodynamic limit $L \rightarrow \infty$, which follows directly
993: from Eq.\ (\ref{11}) for large $t$.
994:
995: For finite lattice sizes $L$ or finite dissipation $c>0$ the function
996: $U(t)$ is bounded from above by a constant given in Eq.\ (\ref{16})
997: with $L$ replaced by $L_c$. In this case we can find the
998: origin of an exponential cut-off in the following way.
999: Consider an enveloping process which corresponds to propagation of the
1000: front of combined avalanches. Duration of an
1001: elementary avalanche starting at $t_{i0}$ is a simple linear function
1002: of time, $t_i =t-t_{i0}$. The enveloping process consists of those
1003: topplings which occur at the maximal distance from the time axis
1004: at each moment of time $t$. The position $x$
1005: of the front is an one-dimensional random walk confined to the interval
1006: $[0,\xi ]$. Starting from $x=0$, the walk performs a step ahead with some
1007: probability, and a step back, the length of which is a random variable.
1008: The combined avalanche stops if the random walk returns to the origin
1009: where it is trapped. The probability of the return to the origin from
1010: an arbitrary position $x$ is not smaller than $U(\xi )$, where
1011: \begin{equation}
1012: \label{26}
1013: U(\xi ) = (r\xi )^{-b^2r/2}\exp(-2br\xi ^{1/2}) \ ,
1014: \end{equation}
1015: and $\xi \sim 1/c$ as above.
1016: The survival time $t$ of the random walk under consideration does not
1017: exceed the period of successful tests in the Bernoulli scheme with the
1018: probability of ``success'' $1-U(\xi )$. The period of tests in the
1019: Bernoulli process has the geometrical distribution
1020: \begin{equation}
1021: \label{27}
1022: Prob(x=t)=(1-U(\xi ))^t U(\xi )\ .
1023: \end{equation}
1024: Hence, the time distribution for combined avalanches is bounded from
1025: above by the exponential function in Eq.\ (\ref{27}). Using Eq.\
1026: (\ref{26}), we obtain
1027: \begin{equation}
1028: \label{28}
1029: P(t) \sim F(t)< A(1-U(\xi ))^t \sim A \exp(-tU(\xi ))\ ,
1030: \end{equation}
1031: where $A$ is a constant and $U(\xi )$ is given by Eq.\ (\ref{26}).
1032: Note that this expression represents an upper bound of the distribution
1033: of busy periods in Eq.\ (\ref{stretch-ex}). Therefore $1/U(\xi )$ plays
1034: the role of an effective correlation length at finite $r$, in a qualitative
1035: agreement with the numerical data and Eq.\ (\ref{xi-r}) of previous Section.
1036:
1037:
1038:
1039:
1040:
1041:
1042:
1043:
1044:
1045:
1046:
1047: \section{Conclusions and discussion}
1048:
1049:
1050: We have shown that a finite driving rate $r$ is a relevant perturbation,
1051: which alters self-organized critical states in the directed sandpile
1052: automata. In the case of conservative dynamics $r$ couples to $\langle
1053: t\rangle _0 \sim L^{1/2}$, thus leading to enhanced effects when the
1054: length scale is increased $L\to \infty$. A continuous flow eventually
1055: occurs, in which critical long-range correlations are destroyed.
1056: On a finite length scale, $L_c =\min \{\xi ,L\} <\infty $
1057: either due to finite screening length $\xi $ or finite system size $L$,
1058: the critical states occur with qualitatively new correlation properties,
1059: which is manifested in (i) a multifractal scaling of combined avalanches
1060: when $L \ll \xi $, and (ii) occurrence of compensation between driving
1061: and dissipation along a line $r_0(c)\sim \xi ^{-1/2}\sim \sqrt{c}$,
1062: when $L\gg \xi $.
1063: How precisely the effective coherence length $\xi_{eff}(r)$ of combined
1064: avalanches increases with driving rate depends on details of the
1065: relaxation process and grain addition. In the case of a finite
1066: input current at each site of the system, we find a finite toppling rate at
1067: all scales, $n_{\phi }(\ell )\sim r/c$, compatible with $\xi _{eff}(r)
1068: \to \infty$. However, if grains
1069: are added only at the top, the correlation length increases
1070: exponentially with $r$ in the range $0<r\le 1$, but remains finite
1071: presumably up to large driving rates. Here we restricted the
1072: discussion to the case $r\leq 1$, where queues of Dhar-Ramaswamy
1073: avalanches occur. Avalanche queuing for this range of driving rates
1074: is peculiar to our model, due to strictly local critical height rule
1075: and the directed transport. In the rice-pile and in the symmetric
1076: Abelian models \cite{Al-Maya,BTW-r} a perfect queuing is prevented by
1077: the collision of avalanches, which occurs at any finite $r>0$.
1078: Owing to the exact solution for behavior of elementary avalanches \cite{DR},
1079: we were able to study properties of the queue in detail. In particular,
1080: we find that average busy period is bounded from below by an exponential
1081: function $\langle t\rangle \geq (rL)^{b^2r/2}\exp{(2brL^{1/2})}$. The
1082: avalanche queue
1083: can be regarded as a multifractal set, in which the average length is
1084: regulated by the driving rate as $\langle N\rangle =1+rL^{1/2}$.
1085: There are no waiting times for elementary avalanches, therefore
1086: our model corresponds to the realization known in the queue theory as
1087: ``infinite number of servers''. Hence, the average number of jobs $q_n$ that
1088: can be served in parallel is unlimited. It should be stressed that our
1089: cellular automaton represents a new example in the queue theory in which
1090: queuing jobs are distributed according to a power-law distribution with
1091: the exponent $\nu <2$ and average duration of jobs is limited by a
1092: control parameter $L_c=\min \{\xi ,L \}$. We hope that more practical
1093: examples of this class can be found.
1094: We also believe that the study of the scaling properties of the queues,
1095: as we have done in this work, adds a new aspect which has not been
1096: considered so far in the queue theory.
1097: The observed nonuniversal scaling properties of avalanche queues can be
1098: related to variation of the average length of the queue with driving
1099: rate. The scaling exponents are found to vary approximately
1100: linearly with the driving rate $r$. A similar $r$-dependence was
1101: observed experimentally in other driven disordered systems, and seems
1102: to apply more generally.
1103:
1104: A continuous activity on the lattice, corresponding to a flow phase
1105: occurs for $r\geq L^{-1/2}$ in the case of conservative
1106: dynamics. On an infinite lattice $L\to \infty $ probability of continuous
1107: flow increases from zero as $F(\infty )\sim 2b^2r^2$, whereas probability
1108: of an intermittent avalanche-like flow decreases from unity with the same
1109: rate $f(1)\sim 1- 2b^2r^2$. For the case of dissipative dynamics $c>0$
1110: an effective coherence length increases with $r$ for $1/c < L\to \infty$.
1111: Our results suggest that the probability $f(1)$ remains finite even
1112: at high driving rates. In the limit $L\to \infty $ the compensation line
1113: extends to the point $c\to 0$, $r\to 0$.
1114: For the range of driving rates studied in this work we expect that the
1115: transport properties of grains in this model remain
1116: unchanged (looked at the time scale of avalanche propagation), compared
1117: to the transport at zero driving rate \cite{BT-unp}. Collision of
1118: avalanches, which occurs first at rates $r>1$ may accelerate the grain
1119: transport, possibly resulting in a new scaling behavior of the
1120: distribution of transit times. Notice that due to local critical height
1121: rules and deterministic topplings the depth of the active zone (defined
1122: in Ref.\ \cite{Al-Maya}) does not change in the flow phase of our model.
1123:
1124: The analytical results in Sec.\ IV are
1125: derived assuming that elementary avalanches may be considered as
1126: independent events. We checked by computing numerically correlation
1127: function between events in a queue for finite $L$ that
1128: rather weak correlations occur. The correlations increase with
1129: the ``distance'' $\tau $ between avalanches as $\tau ^\eta $, where
1130: $\eta =0.05\pm 0.01$ with statistical error bars.
1131:
1132:
1133:
1134:
1135: \acknowledgments
1136: Work of B.T. was supported by the Ministry
1137: of Science and Technology of the Republic of Slovenia.
1138: Work of V.P. was supported in part by the International
1139: Slovenian-Russian project. B.T. wishes to thank A. Corral for
1140: discussions.
1141:
1142: \begin{references}
1143:
1144: \bibitem[*]{*}Electronic address: Bosiljka.Tadic@ijs.si
1145: \bibitem[**]{**}Electronic address: priezzvb@thsun1.jinr.ru
1146:
1147: \bibitem{DD-rev}D. Dhar, Physica A {\bf 263}, 4 (1999).
1148: \bibitem{DR}D. Dhar and R. Ramaswamy, Phys. Rev. Lett. {\bf 63},
1149: 1659 (1989).
1150: \bibitem{exact}See Ref.\ \cite{DD-rev} for more exact results.
1151: \bibitem{TD}B. Tadi\'c and D. Dhar, Phys. Rev. Lett. {\bf 79}, 1519
1152: (1997).
1153: \bibitem{Tetal}B. Tadi\'c, U. Nowak, K.D. Usadel, R. Ramaswamy, and
1154: S. Padlewski, Phys. Rev. A {\bf 45}, 8536 (1992).
1155: \bibitem{BN-r}G. Bertotti, G. Durin, and A. Magni, J. Appl. Phys. {\bf
1156: 75}, 5490 (1994);G. Durin and A. Magni, in {\it Fractal Reviews of the
1157: Natural and Applied Sciences}, M. Novak Ed., London (1996).
1158:
1159: \bibitem{IFO}A. Corral, C.J. P\'erez, A. D\'{\i}az-Guilera, and
1160: A. Arenas, Phys. Rev. Lett. {\bf 74}, 118 (1995).
1161: \bibitem{drums}H.M. Jaeger, C.-heng Liu, and S.R. Nagel,
1162: Phys. Rev. Lett. {\bf 62}, 40 (1989); M. Bretz, J.B. Cunningham, P.L.
1163: Kurczynski, and F. Nori, Phys. Rev. Lett. {\bf 69}, 2431 (1992).
1164:
1165: \bibitem{queue-books}A.A. Borovkov {\it Vero\-yat\-nost\-nie pro\-cessi v
1166: teo\-rii mas\-so\-vovo obslu\-zhi\-vani\-ya}, Nauka, Moscow 1972;
1167: J. W. Cohen, {\it The Single Server Queue}, Second edition, North Holland,
1168: 1982.
1169: \bibitem{noncons-local}See, e.g., Ref.\ \cite{Tetal} and B. Tadi\'c and
1170: R. Ramaswamy, Phys. Rev. E {\bf 54}, 3157 (1996).
1171: \bibitem{noncons-global}See Ref.\ \cite{IFO} and S.S. Manna, L. Kiss,
1172: and J. K\'ertez,
1173: J. Stat. Phys. {\bf 61}, 923 (1990); K. Christensen and Z. Olami,
1174: Phys. Rev. A {46}, 1892 (1992); S.S. Manna (unpublished).
1175: \bibitem{Al-Maya}A. Corral and M. Paczuski, Phys. Rev. Lett. {\bf 83 }
1176: 572 (1999).
1177: \bibitem{BTW-r}A. Barrat, A. Vespignani, and S. Zapperi, e-print
1178: cond-mat/9903269.
1179: \bibitem{comment-sb}This implementation correspond to site defects.
1180: A bond-type of defects are expected to have qualitatively similar
1181: effects [A. Povolotsky, unpublished].
1182: \bibitem{JT}J. Theiler, Phys. Rev. E {\bf 47}, 733 (1993).
1183: \bibitem{Zwart}A. P. Zwart, preprint, Memorandum COSOR 99-12 (1999),
1184: Eindhoven University of Technology.
1185:
1186: \bibitem{comment-n}Number of topplings $n$ occurring during an
1187: avalanche (also called avalanche mass) in this model
1188: is proportional to the energy
1189: dissipated in the avalanche. Here we use all three terms equally.
1190:
1191: \bibitem{Romu}R. Pastor-Satorras, Phys. Rev. E {\bf 56}, 5284 (1997).
1192:
1193: \bibitem{comment}Somewhat similar effects of disorder have been
1194: observed disordered ferromagnets at finite driving rates
1195: [B. Tadi\'c, Phys. Rev. Lett. {\bf 77}, 3843 (1996)].
1196:
1197:
1198: \bibitem{Feller}W. Feller, {\it An introduction to probability theory and
1199: its applications}, Second edition, J.W., New-York, 1971.
1200: \bibitem{Ivash.Priez}E. V. Ivashkevich and V. B. Priezzhev, Physica A
1201: {\bf 254}, 97 (1998).
1202: \bibitem{comment-addition}Note that the behavior is different
1203: in the case of addition of particles at {\it each site} on the lattice,
1204: when we have a constant number of topplings at
1205: all distances $n_{\phi}(\ell ) = (r/c)[1+(1-c)^{\ell -1}]$ controlled by the
1206: ratio $r/c$.
1207:
1208: \bibitem{BT-unp} B. Tadi\'c (unpublished).
1209:
1210:
1211:
1212: \end{references}
1213:
1214:
1215: \end{multicols}
1216:
1217: \end{document}
1218: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1219: %\newpage%%
1220:
1221: \narrowtext
1222: \twocolumn
1223:
1224: \begin{figure}
1225: \epsfxsize=82mm\epsffile[56 75 392 298]{rchr_fig1.ps}
1226: \caption{\label{fig1}Schematic phase diagram for a finite system size
1227: $L$. Cross-hatched area represents a crossover region between
1228: $\xi \ll L$ (right) and $\xi \gg L$ (left). Regions with distinct
1229: behavior of the avalanche queues are shown: subcritical (S),
1230: nonuniversal (N), and multifractal scaling region (M), and flow phase (F).
1231: Solid line represents the compensation line. Transition to the flow
1232: phase is marked by $\diamond$ for $c=0$ and by dotted line for small
1233: $c>0$. In the origin ($\bullet$) only single Dhar-Ramaswamy avalanches occur.}
1234: \end{figure}
1235:
1236:
1237: \begin{figure}
1238: \epsfxsize=52mm\epsffile[188 278 421 515]{rchr_fig2a.ps}
1239:
1240: \epsfxsize=52mm\epsffile[156 256 455 536]{rchr_fig2b.ps}
1241: \caption{\label{fig2} Top: Growth of a combined avalanche in the case
1242: of conservative dynamics ($c=0$) for driving rate $r=0.05$ and $L=100$.
1243: Nine fronts of active sites (dark) are visible.
1244: Bottom: A complete combined avalanche in the case of dissipative dynamics with
1245: $c=0.02$ and driving rate $r=0.5$. Different colors correspond to distinct
1246: toppling waves.}
1247: \end{figure}
1248:
1249:
1250: \begin{figure}[thb]
1251: \epsfxsize=80mm\epsffile[43 69 508 564]{rchr_fig3.ps}
1252: \caption{\label{fig3}Double logarithmic plot of the integrated probability
1253: distribution of duration (busy periods) $P(t,L)$ of queues of Dhar-Ramaswamy
1254: avalanches vs. duration $t$, measured in Monte Carlo steps [MCS].
1255: Fixed driving rate $r=0.05$ and various
1256: lattice sizes $L=$24,48,96,192, and 384 are used.
1257: Inset: Multifractal spectral function $\Phi _t(\alpha _t)$ vs. $\alpha _t$
1258: obtained from the data in main figure according to Eq.\ (6) using
1259: $X_0=1$, $L_0=2$.}
1260: \end{figure}
1261:
1262:
1263:
1264: \begin{figure}[thb]
1265: \epsfxsize=80mm\epsffile[43 69 508 564]{rchr_fig4.ps}
1266: \caption{\label{fig4}Same as Fig.\ 3 but for the integrated distribution
1267: of number of topplings (mass) $D(n,L)$ vs. mass $n$ [number of grains].
1268: Shown are only curves for $L=$96,192, and 384. Inset: Spectral function
1269: $\Phi _n(\alpha _n)$ vs. $\alpha _n$. $X_0=0.97\pm 0.02$, $L_0=2.15
1270: \pm 0.05$.}
1271: \end{figure}
1272:
1273:
1274: \begin{figure}[thb]
1275: \epsfxsize=80mm\epsffile[38 73 449 298]{rchr_fig5.ps}
1276: \caption{\label{fig5} Attempted finite-size scaling fit of the data
1277: from main Fig.\ 3 and 4 for the distribution of avalanche duration (right)
1278: and mass (left).}
1279: \end{figure}
1280:
1281:
1282:
1283:
1284: \begin{figure}[thb]
1285: \epsfxsize=78mm\epsffile[43 69 508 746]{rchr_fig6.ps}
1286: %\vskip 8mm
1287: \caption{\label{fig6} Sandpile noise $n(t)$---number of active sites at
1288: time $t$, plotted against time $t$ [MCS] (lower panel) and its
1289: Fourier spectrum (top panel) for fixed driving rate r=0.33.
1290: Dissipation rate $c=0.01$ and $L=128$. Shaded area shown in the lower
1291: panel is an example of unit signal, corresponding to a combined avalanche.}
1292: \end{figure}
1293:
1294:
1295:
1296:
1297: \begin{figure}[thb]
1298: \epsfxsize=80mm\epsffile[43 69 508 564]{rchr_fig7.ps}
1299: %\vskip 8mm
1300: \caption{\label{fig7}
1301: Distribution of avalanche mass $D(n,r)$ vs. mass $n$ [number of grains]
1302: shown in double logarithmic scale for fixed dissipation rate $c=0.01$
1303: and $ L=192$ and for
1304: various driving rates $r$= 0.021,0.041,0.083,0.125,0.166,0.25, and 0.33
1305: (left to right). Inset: Average mass (top), area (middle) and duration
1306: (bottom) of combined avalanches plotted against driving rate [MCS$^{-1}$]
1307: for fixed $c=0.01$ and $L=192$. At each point average is taken over
1308: $2\times 10^6$ combined avalanches. Fitting curves: $\langle n\rangle
1309: =167\times \exp{(11.4r^{1.28})}$, and $\langle t\rangle =23\times
1310: \exp{(9.6r^{1.42})}$.}
1311: \end{figure}
1312:
1313:
1314: \begin{figure}[thb]
1315: \epsfxsize=78mm\epsffile[52 70 449 564]{rchr_fig8.ps}
1316: %\vskip 8mm
1317: \caption{\label{fig8} Lower panel: Various scaling exponents of the
1318: avalanche queues plotted against driving rate for fixed dissipation
1319: $c=0.01$ and $L=192$. $*$ indicates the products
1320: $z(\tau _t-1)\approx D_n(\tau _n-1)\approx D_s(\tau_s-1) \approx
1321: 1.1(\tau _{\ell}-1)$. Also shown are linear fits of the data.
1322: Top panel: Amplitudes $\xi_n$ (upper curve) and $\xi_t$ defined in
1323: Eq.\ (7) vs driving rate. Lines: fit curves satisfying
1324: Eq.\ (8) with $B_n=6.29\pm 0.46$ and $B_t=4.5\pm 0.51$.}
1325: \end{figure}
1326:
1327:
1328:
1329:
1330:
1331: \begin{figure}[thb]
1332: \epsfxsize=78mm\epsffile[43 69 508 564]{rchr_fig9.ps}
1333: \caption{\label{fig9} Finite-size scaling plot of the average
1334: duration $\langle t\rangle $ (top) and mass
1335: $\langle n\rangle $ (bottom) according to Eq.\ (\ref{FSS-c}).
1336: Data are taken for values of $c$=0.01,0.0125,0.015, and 0.02
1337: such that $\xi <L=192$ is satisfied.
1338: Note that the exponents are exact and the fitted value
1339: $\kappa = 1.14 $ within numerical error bars.}
1340: \end{figure}
1341:
1342:
1343:
1344:
1345: %\end{multicols}
1346: \end{document}
1347:
1348: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1349:
1350:
1351: \begin{figure}[thb]
1352: \epsfxsize=80mm\epsffile[48 71 513 561]{rchr_fig5.ps}
1353: %\vskip 8mm
1354: \caption{\label{fig5} Same as Fig.\ 4 but for fixed driving rate $r=0.5$
1355: and for varying dissipation rate $c=$0.01, 0.02, 0.03,0.04, 0.05,
1356: 0.1, 0.16, and 0.25 (right to left). $L=192$. Data for $c=0.04$ are
1357: shown by empty circles and a fitting curve Eq.\ (\ref{stretch-ex}) as a
1358: full line.
1359: Inset: Effective amplitudes $\xi _X$ defined in Eq.\ (\ref{stretch-ex})
1360: for the case of duration (lower) and energy (upper curve) of combined
1361: avalanches, plotted against driving rate $r$ for fixed dissipation
1362: $c=0.01$. Full lines: Fitting curves according to Eq.\ (8) with
1363: $B_n=6.29\pm 0.46$ and $B_t=4.5\pm 0.51$.}
1364: \end{figure}
1365:
1366:
1367:
1368: #!/bin/csh -f
1369: # this uuencoded Z-compressed .tar file created by csh script uufiles
1370: # for more information, see e.g. http://xxx.lanl.gov/faq/uufaq.html
1371: # if you are on a unix machine this file will unpack itself:
1372: # strip off any mail header and call resulting file, e.g., rchrfigs.uu
1373: # (uudecode ignores these header lines and starts at begin line below)
1374: # then say csh rchrfigs.uu
1375: # or explicitly execute the commands (generally more secure):
1376: # uudecode rchrfigs.uu ; uncompress rchrfigs.tar.Z ;
1377: # tar -xvf rchrfigs.tar
1378: # on some non-unix (e.g. VAX/VMS), first use an editor to change the
1379: # filename in "begin" line below to rchrfigs.tar_Z , then execute
1380: # uudecode rchrfigs.uu
1381: # compress -d rchrfigs.tar_Z
1382: # tar -xvf rchrfigs.tar
1383: #
1384: uudecode $0
1385: chmod 644 rchrfigs.tar.Z
1386: zcat rchrfigs.tar.Z | tar -xvf -
1387: rm $0 rchrfigs.tar.Z
1388: exit
1389:
1390: