1: \documentclass[a4paper,aps,pre,amsmath,amsfonts,showpacs,preprint]{revtex4}
2: \usepackage{graphicx}
3: \usepackage{amssymb}
4:
5: \begin{document}
6:
7: \title{Inhomogeneous sandpile model: Crossover from multifractal scaling
8: to finite size scaling }
9:
10:
11: \author{Jozef \v{C}ern\'{a}k }
12:
13:
14: \email{jcernak@kosice.upjs.sk}
15:
16:
17: \affiliation{University of P. J. \v{S}af\'{a}rik in Ko\v{s}ice, Department
18: of Biophysics, Jesenn\'{a} 5, SK-04000 Ko\v{s}ice, Slovak Republic}
19:
20:
21: \date{\today{}}
22:
23: \begin{abstract}
24: We study an inhomogeneous sandpile model in which two different toppling
25: rules are defined. For any site only one rule is applied corresponding
26: to either the Bak, Tang and Wiesenfeld model {[}P.Bak, C. Tang, and K. Wiesenfeld,
27: Phys. Rev. Lett. \textbf{59}, 381 (1987){]} or the Manna two-state sandpile model
28: {[}S. S. Manna, J. Phys. A \textbf{24}, L363 (1991){]}. A parameter $c$ is introduced
29: which describes a density of sites which are randomly deployed and where
30: the stochastic Manna rules are applied. The results show that the
31: avalanche area exponent $\tau_{a}$, avalanche size exponent $\tau_{s}$,
32: and capacity fractal dimension $D_{s}$ depend on the density $c$.
33: A crossover from multifractal scaling of the Bak, Tang, and Wiesenfeld
34: model ($c=0$) to finite size scaling was found. The critical density
35: $c$ is found to be in the interval $0<c<0.01$. These results demonstrate
36: that local dynamical rules are important and can change the global
37: properties of the model.
38: \end{abstract}
39:
40: \pacs{05.65.+b, 05.40.-a, 64.60.Ak}
41:
42: \maketitle
43:
44: \section{Introduction}
45:
46: Bak, Tang, and Wiesenfeld (BTW) \cite{BTW_1987} introduced a concept
47: of self-organized criticality (SOC) as a common feature of different
48: dynamical systems where the power-law temporal or spatial correlations
49: are extended over several decades. Dynamical systems with many interacting
50: degrees of freedom and with short range couplings naturally evolve
51: into a critical state through a self-organized process. They proposed
52: a simple cellular automaton with deterministic rules, which is known
53: as a sandpile model, to demonstrate this new phenomenon. In this model
54: the relaxation rules are conservative, no dissipation takes place
55: during relaxation, and correspond to a nonlinear diffusion equation
56: \cite{BTW_1987}. Generally, the sandpile model is represented by
57: a $d$-dimensional hypercube of the finite linear size $L$. Its boundaries
58: are open and allow an energy dissipation, which takes place only at
59: the boundaries.
60:
61: Manna proposed a two-state version of the sandpile model \cite{Manna_1991}
62: where no more than one particle is allowed to be at a site in the
63: stationary state. If one particle is added to a randomly chosen site,
64: then relaxation starts depending on the occupancy of the site. If
65: the site is empty, a particle is launched. In the case when the site
66: is not empty, a hard core interaction throws the particles out from
67: the site and the particles are redistributed in a random manner among
68: its neighbours. All sites affected by this redistribution create an
69: avalanche. An avalanche is stopped if any site reached the stationary
70: state, i.e. no more than one particle occupies a site.
71:
72: The first systematic study of scaling properties, universality and
73: classification of deterministic sandpile models was carried out by
74: Kadanoff \emph{et al.} \cite{Kadanoff}. Using numerical simulations
75: and by varying the underlying microscopic rules which describe how
76: an avalanche is generated they investigated whether different models
77: have the same universal properties. Applying finite-size scaling (FSS)
78: and multifractal scaling techniques they studied how a finite-size
79: of the system affects scaling properties.
80:
81: The real-space renormalization group calculations \cite{Pietronero}
82: suggested that deterministic \cite{BTW_1987} and stochastic \cite{Manna_1991}
83: sandpile models belong to the same universality class. On the other
84: hand, many numerical results \cite{Ben-Hur,Biham,Lubeck,Menech,Milsh}
85: show clearly two different universality classes. They do not confirm
86: the hypothesis that small modifications in the dynamical rules of
87: the models do not change the universality class, presented by Chessa
88: \emph{at al}. \cite{Chessa}.
89:
90: This study was motivated by the results published by Tebaldi \emph{et
91: al.} \cite{Tebaldi_1999}, and Stella and Menech \cite{Stella_2001},
92: where a multifractal scaling of an avalanche size distribution of
93: the BTW model was demonstrated. They assume that a multifractal character
94: for SOC models like the BTW model is a crucial step towards the solution
95: of universality issues. By applying the moment analysis they found
96: FSS for the two-state Manna model \cite{Stella_2001}. Based on these
97: results they conclude that the 2D BTW model and the Manna model belong
98: to qualitatively different universality classes. This assumption was
99: confirmed recently \cite{Karma,Karma_E}, where a precise toppling
100: balance has been investigated in more detail.
101:
102: In this paper we report the results of disturbing the dynamics of
103: the BTW model using stochastic Manna sites which are randomly deployed.
104: They can introduce stochastic events during an avalanche propagation.
105: Our model was derived from the inhomogeneous sandpile model \cite{Cer}
106: in witch two different deterministic toppling rules were defined.
107: In the proposed model the first toppling rule corresponds to the BTW
108: model \cite{BTW_1987} and the second rule is now stochastic and corresponds
109: to the two-state Manna model \cite{Manna_1991}. The model is similar
110: to that in Ref. \cite{Karma_E}, however we applied the original toppling
111: rules of the listed sandpile models.
112:
113: The paper is organized as follows. The inhomogeneous sandpile model
114: is introduced in Sec. \ref{sec:Mathematical-model}. The avalanche
115: scaling exponents, capacity fractal dimensions and crossover from
116: multifractal to FSS are investigated with numerical simulations and
117: the results are presented in Sec. \ref{sec:Results}. The Sec. \ref{sec:Discussion}
118: is devoted to a discussion which is followed by conclusions in Sec.
119: \ref{sec:Conclusion}.
120:
121:
122: \section{\label{sec:Mathematical-model}Mathematical model}
123:
124: We consider a $d$-dimensional hyper-cubic lattice of linear size
125: $L$, and a notation presented by Ben-Hur \emph{et al.} \cite{Ben-Hur}
126: is followed to define a sandpile model. Each site $\mathbf{\mathbf{i}}$
127: has assigned a dynamical variable $E(\mathbf{\mathbf{i}})$ that generally
128: represents a physical quantity such as energy, grain density, stress,
129: etc. A configuration $\{ E(\mathbf{i})\}$ is classified as stable
130: if for all sites $E(\mathbf{\mathbf{i}})<E_{c}$, where $E_{c}$ is
131: a threshold value. We note that the two-state Manna model \cite{Manna_1991}
132: has no threshold $E_{c}$. The Manna model has defined a hard core
133: repulsion interaction among different particles at the same position.
134: This hard core repulsion interaction can be described by a threshold
135: where the threshold value $E_{c}=2$ is assigned to any site. In our
136: inhomogeneous sandpile model, the threshold values $E_{c}$ depend
137: on the site position $\mathbf{i}$, $E_{c}(\mathbf{i})$ \cite{Cer}.
138: The conditions for a stationary state, a stable configuration $\{ E(\mathbf{i})\}$
139: (no avalanche), are now $E(\mathbf{i})<E_{c}(\mathbf{i})$, where
140: the threshold $E_{c}(\mathbf{i})$ at the site $\mathbf{i}$ was randomly
141: chosen from two allowed values
142:
143: \begin{equation}\label{eq:thresholds} E_{c}(\mathbf{i})=\begin{cases} E_{c}^{I}\ =4\\
144:
145: E_{c}^{II}=2. \end{cases}\end{equation}For any site $\mathbf{i}$ the threshold $E_{c}(\mathbf{i})$ [Eq.
146: (\ref{eq:thresholds})] is defined in such a manner that $n$ randomly
147: chosen sites have the value $E_{c}^{II}$ and the remaining $L^{d}-n$
148: sites have the value $E_{c}^{I}$. The density of sites with the threshold
149: value $E_{c}^{II}$ is denoted $c$, and $c=n/L^{d}$.
150:
151: Let us assume that a stable configuration $\{ E(\mathbf{j})\}$ is
152: given, and then we select a site $\mathbf{i}$ at random and increase
153: $E(\mathbf{i})$ by some amount $\delta E$. We now consider $\delta E=1$
154: for any site. When an unstable configuration is reached, $E(\mathbf{i})\geq E_{c}(\mathbf{i})$,
155: a relaxation takes place. An unstable site $\mathbf{i}$ lowers its
156: energy, that is distributed among the neighbor sites. The directions
157: to the neighbor sites are defined by the vectors $\mathbf{e}_{1}=(0,1)$,
158: $\mathbf{e}_{2}=(0,-1)$, $\mathbf{e}_{3}=(1,0)$, and $\mathbf{e}_{4}=(-1,0)$.
159: The relaxation is defined by the following rules
160:
161: \begin{equation}
162: E(\mathbf{i})\rightarrow E(\mathbf{i})-\sum_{e}\Delta E(\mathbf{i}),\label{eq:relax1}\end{equation}
163:
164:
165: \begin{equation}
166: E(\mathbf{i}+\mathbf{e})\rightarrow E(\mathbf{i}+\mathbf{e})+\Delta E(\mathbf{e}),\label{eq:relax2}\end{equation}
167:
168:
169: \begin{equation}
170: \sum_{e}\Delta E(\mathbf{e})=E_{c}(\mathbf{i}),\label{eq:relax3}\end{equation}
171:
172:
173: \begin{equation}\label{eq:relax4} \mathbf{e}=\begin{cases} \{\mathbf{e}_{1},\ \mathbf{e}_{2},\ \mathbf{e}_{3},\ \mathbf{e}_{4}\} & if\ E_{c}(\mathbf{i})=E_{c}^{I}\\
174:
175: \{\mathbf{e}_{\zeta},\ \mathbf{e}_{\eta}\} & if\ E_{c}(\mathbf{i})=E_{c}^{II}\end{cases} \end{equation}where $\mathbf{e}$ is a set of vectors from the site $\mathbf{i}$
176: to its neighbors. The indexes $\zeta$ and $\eta$ are integers $1,2,3$,
177: and $4$ randomly chosen at any relaxation. The neighbors that receive
178: the energy can became unstable and topple, thus generating an avalanche.
179: The distribution of energy is described by Eqs. (\ref{eq:relax1})
180: and (\ref{eq:relax2}), we added additional rules Eqs. (\ref{eq:relax3})
181: and (\ref{eq:relax4}) which specify the manner how the energy is
182: distributed depending on the position $\mathbf{i}$, threshold $E_{c}(\mathbf{i})$
183: [Eq. (\ref{eq:relax3})], and corresponding sandpile model [Eq. (\ref{eq:relax4})].
184: The relaxation rules Eqs. (\ref{eq:relax1})-(\ref{eq:relax4}) are
185: applied until that moment when a new stable configuration is reached
186: again, for all sites $E(\mathbf{i})<E_{c}(\mathbf{i})$. Obviously,
187: during one avalanche an arbitrary unstable site $\mathbf{i}$ can
188: transfer the energy $E_{c}(\mathbf{i})$ a few times to became stable,
189: $E(\mathbf{i})<E_{c}(\mathbf{i})$. A $d$-dimensional lattice has
190: open boundaries so added energy can flow outside the system, and an
191: energy dissipation takes place only at the boundaries.
192:
193: This model has been designed to enable a well defined change between
194: two well known nondirected sandpile models: deterministic \cite{BTW_1987}
195: and stochastic \cite{Manna_1991} (nondirected only on average) similarly
196: as in Ref. \cite{Karma}. The model belongs to the critical height
197: models with conservative relaxation rules and with undirected energy
198: transfer where the two thresholds are randomly frozen. It can be characterized
199: as a sandpile with a possibility to modify its scaling behaviors.
200:
201:
202: \section{\label{sec:Results}Results}
203:
204: We shall report the results obtained using numerical simulation of
205: the conservative, undirected, critical height sandpile model defined
206: by Eqs. (\ref{eq:relax1})-(\ref{eq:relax4}). The simulations were
207: carried out for the following parameters: $d=2$, two-dimensional
208: lattice of linear sizes $L=256,\ 512$ and $1024$, randomly added
209: energy $\delta E=1$, two thresholds either $E_{c}^{I}=4$ or $E_{c}^{II}=2$,
210: and with density of sites with threshold $E_{c}^{II}$ in the interval
211: $0\leq c\leq1$. In our simulations we have used the density $c$
212: as a model parameter. For densities of stochastic sites $c=0$ and
213: $1$ the model behaves as the BTW model \cite{BTW_1987} and Manna
214: model \cite{Manna_1991}, respectively, which are both considered
215: to be \emph{Abelian} \cite{Dhar}.
216:
217: Avalanches can be characterized by such properties as their size,
218: area, lifetime, linear size, and perimeter. We concentrate only on
219: a minimal number of parameters which are necessary to demonstrate
220: the investigated phenomena: the avalanche area $a$ and avalanche
221: size $s$. Here the avalanche area $a$ is the number of lattice sites
222: that have relaxed at least once during the avalanche. The avalanche
223: size $s$ is the total number of relaxations that occurred during
224: the avalanche. The probability distributions of these variables are
225: usually described as power-laws with cutoff
226:
227: \begin{equation}
228: P(x)=x^{-\tau_{x}}F(x/x_{c}),\label{eq:FSS}\end{equation}
229: where $x=a,s.$ When the system size $L$ goes to infinity, the cutoff
230: $x_{c}$ diverges as $x_{c}\sim L^{D_{x}}$. If we assume FSS, then
231: the set of exponents ($\tau_{x},D_{x}$) from Eq. (\ref{eq:FSS})
232: defines the universality class of the model \cite{Chessa}.
233:
234: The avalanche area probability distribution $P(a)$ and avalanche
235: size probability distributions $P(s)$ have been analyzed at finite
236: lattice sizes $L=256,512$, and $1024$. It is expected that these
237: distributions follow a power-law $P(x)\sim x^{\tau_{x}}$ [Eq. (\ref{eq:FSS})].
238: For any lattice size $L$ and density $c$ the corresponding scaling
239: exponents $\tau_{x,L}(c)$ were determined. The scaling exponents
240: found in the numerical simulations for the largest lattice size $L=1024$
241: and for selected densities $c$ are presented in Table \ref{cap:The-scaling-exponents}.
242: It is evident that the exponents are increasing with $c$ in the interval
243: $0<c<0.1$ and then for densities $c>0.1$ they are almost constant.
244:
245: %
246: \begin{table}[b]
247:
248: \caption{\label{cap:The-scaling-exponents}The scaling exponents $\tau_{x,L=1024}(c)$
249: for the finite lattice size $L=1024$ and selected densities in the
250: interval $0\leq c\leq1$. The statistical errors are $\pm0.001$.}
251:
252: \begin{tabular}{lccccc}
253: \hline
254: density $c$&
255: $0$&
256: $0.01$&
257: $0.10$&
258: $0.50$&
259: $1$\tabularnewline
260: \hline
261: $\tau_{a,L=1024}$&
262: $1.131$&
263: $-$&
264: $1.291$&
265: $1.315$&
266: $1.338$\tabularnewline
267: $\tau_{s,L=1024}$&
268: $1.137$&
269: $1.240$&
270: $1.263$&
271: $1.266$&
272: $1.283$\tabularnewline
273: \hline
274: \end{tabular}
275: \end{table}
276:
277:
278: The scaling exponents $\tau_{x,L}$ show a finite size-effect when
279: the lattice size $L$ is changed. Their dependences on lattice sizes
280: $L$ are approximated by a formula proposed by Manna \cite{Manna_PA}
281:
282: \begin{equation}
283: x=x_{L\rightarrow\infty}-\frac{const.}{\ln(L)}.\label{eq:fit}\end{equation}
284: This approximation was used to extrapolate the scaling exponents
285: $\tau_{x,L\rightarrow\infty}$ for the infinite lattice $L\rightarrow\infty$.
286:
287: %
288: \begin{figure}[t]
289: \includegraphics{fig1.eps}
290:
291:
292: \caption{\label{cap:exponents}(Color online) The avalanche area and size
293: scaling exponents $\tau_{a,L\rightarrow\infty}$ and $\tau_{s,L\rightarrow\infty}$
294: are approximated for the infinite lattice size $L\rightarrow\infty$.
295: The exponents depend on the density $c$ of Manna sites. }
296: \end{figure}
297:
298:
299: The avalanche size probability distributions $P(s)$ obey the power-law
300: dependence for any density $c$. The corresponding scaling exponents
301: $\tau_{s,L\rightarrow\infty}(c)$ are shown in the Fig. \ref{cap:exponents}.
302: In the range of densities $0.01\leq c\leq0.1$ these scaling exponents
303: decrease from $\tau_{s,L\rightarrow\infty}(0.01)=1.37$$\pm0.025$
304: to $\tau_{s,L\rightarrow\infty}(0.1)=1.29\pm0.025$ and then, for
305: higher densities $c>0.1$, are almost constant.
306:
307: The avalanche area scaling exponents $\tau_{a,L\rightarrow\infty}$
308: show a more complex dependence on the density $c$. For densities
309: $0.09\leq c\leq0.5$ they decrease from $\tau_{a,L\rightarrow\infty}(0.09)=1.49$$\pm0.025$
310: to $\tau_{a,L\rightarrow\infty}(0.5)=1.38\pm0.025$, then for higher
311: densities $c>0.5$ the exponents $\tau_{a,L\rightarrow\infty}(c)$
312: are almost constant. It was observed that for densities $0.01\leq c\leq0.09$
313: the avalanche area distributions $P(a)$ do not follow exactly a power-law
314: dependence as it is expected from Eq. (\ref{eq:FSS}). Therefore the
315: exponents $\tau_{a,L\rightarrow\infty}(c)$ from this density interval
316: are not included in Fig. \ref{cap:exponents}. One typical example
317: is shown in Fig. \ref{cap:avalanche-area} where the density of random
318: toppling sites is $c=0.01$ and the lattice size is $L=1024$. The
319: double-log plot of area distribution function $P(a)$ clearly shows
320: that a possible approximation function is not a straight line which
321: must correspond to the simple power-law dependence.
322:
323: %
324: \begin{figure}[t]
325: \includegraphics{fig2.eps}
326:
327:
328: \caption{\label{cap:avalanche-area}(Color online) The avalanche area distribution
329: $P(a)$ does not follow exactly a power-law function. The parameters
330: used in the numerical simulation were: density $c=0.01$ and linear
331: lattice size $L=1024$. }
332: \end{figure}
333:
334:
335: For the two well known sandpile models, BTW ($c=0$) and Manna ($c=1$)
336: the scaling exponents $\tau_{a,L\rightarrow\infty}(0)=1.26$, $\tau_{s,L\rightarrow\infty}(0)=1.23$,
337: $\tau_{a,L\rightarrow\infty}(1)=1.36$, and $\tau_{s,L\rightarrow\infty}(1)=1.27$
338: were found. In addition, for all densities $c$ (see Fig. \ref{cap:exponents})
339: the relation $\tau_{a,L\rightarrow\infty}(c)>\tau_{s,L\rightarrow\infty}(c)$
340: is valid.
341:
342: %
343: \begin{figure}[t]
344: \includegraphics{fig3.eps}
345:
346:
347: \caption{\label{cap:multi}(Color online) The extrapolated spectra of the
348: avalanche area $a$ and avalanche size $s$ for various densities
349: of Manna toppling sites: (a) $c=0$ BTW model, which shows multifractal
350: scaling, (b) $c=0.01$ at which the multifractal scaling of BTW model
351: is destroyed, and (c) $c=0.95$ where the model shows the FSS near
352: the two-state Manna model ($c=1$). The maximal error bars of $f(\alpha,c)$
353: are for $q\approx0$, and are approximately $\pm0.05$, but for a
354: higher $q$ they are smaller. The $\alpha$ values are determined
355: within errors $\pm0.025$. }
356: \end{figure}
357:
358:
359: The scaling exponents $\tau_{x,L}$ as functions of the lattice size
360: $L$ show a finite-size scaling effect [Eq. (\ref{eq:fit})]. An exact
361: determination of scaling exponents $\tau_{x,L\rightarrow\infty}$
362: from numerical experiments is therefore a difficult task. A new method
363: was introduced \cite{Lubeck} to increase the numerical accuracy of
364: the exponents based on their direct determination. We found that the
365: method gives slightly larger exponents than a simple extrapolation
366: of Eq. (\ref{eq:fit}). However, the exponents $\tau_{s}$ do not
367: fluctuate around their mean values as it was observed in the paper
368: \cite{Lubeck}. Our error bars were larger, therefore we have to repeat
369: this analysis again in more details.
370:
371: Tebaldi \emph{et al.} \cite{Tebaldi_1999} found that in the BTW model
372: the avalanche area distributions $P(a)$ show FSS and avalanche size
373: distribution $P(s)$ scale as a multifractal. To describe these scaling
374: properties rather a multifractal spectrum $f(\alpha)$ versus $\alpha$
375: than the single scaling exponent $\tau_{s}$ [Eq. (\ref{eq:FSS})]
376: is necessary. Thus, the scaling exponent $\tau_{s}$ loses the importance
377: and is replaced by a spectrum of exponents. Despite this fact, the
378: avalanche size scaling exponents $\tau_{s,L\rightarrow\infty}(0)$
379: are determined. They enable a comparison with the previous results,
380: since the whole point is that the exponent $\tau_{s,L\rightarrow\infty}(0)$
381: does not exist. The recent studies \cite{Tebaldi_1999,Stella_2001}
382: led us to analyze the multifractal properties of the model given by
383: Eqs. (\ref{eq:relax1})-(\ref{eq:relax4}) for various densities $c$.
384: To determine the multifractal spectra a method presented in the paper
385: \cite{Stella_2001} was useful. There, for any finite-size lattices
386: $L$, the quantities $\alpha_{x}(q,L)=\left\langle log(x)x^{q}\right\rangle /\left[log(L)\left\langle x^{q}\right\rangle \right]$
387: and $\sigma_{x}(q,L)\sim log\left(\left\langle x^{q}\right\rangle \right)/log(L)$
388: were computed. It was observed that $\alpha_{x}(q,L)$ and $\sigma_{x}(q,L)$
389: show a finite-size dependence on the system size $L$, which is well
390: approximated by Eq. (\ref{eq:fit}) and this relation was used to
391: extrapolate $L\rightarrow\infty$ quantities. Based on the Legendre
392: structure relating $f_{x}$ to $\sigma_{x}$, a parametric representation
393: of $f_{x}(\alpha_{x})$ by plotting $f_{x}(q)=\sigma_{x}(q)-\alpha_{x}(q)q$
394: versus $\alpha_{x}(q)$ can be obtained \cite{Stella_2001}.
395:
396: Some significant spectra of $f_{x}(\alpha_{x},c)$ extrapolated for
397: an infinite lattice size $L\rightarrow\infty$ are shown for illustration
398: in Fig. \ref{cap:multi}. The $f_{x}(\alpha_{x},c)$ values were determined
399: for the parameter $q$ in the range $-3.5<q<3.5$ and they are limited
400: by errors about $\pm0.08$, similarly as in Ref. \cite{Stella_2001}.
401: We have observed that if $f_{x}(\alpha_{x},c)$ spectra are computed
402: for all avalanches where $a>50$ then the errors of $f_{x}(\alpha_{x},c)$
403: are $\pm0.05$. The multifractal scaling of the avalanche size probability
404: distribution $P(s)$ and FSS of avalanche area probability distribution
405: $P(a)$ were found at density $c=0$ (see Fig. \ref{cap:multi} (a)).
406: The avalanche probability distributions $P(x)$ show FSS for densities
407: $c=0.01$ Fig. \ref{cap:multi} (b), and for $c=0.95$ Fig. \ref{cap:multi}
408: (c) which is close to the Manna model ($c=1$). The spectra for $c=0$
409: and $1$ agree well with the previous results \cite{Stella_2001}.
410: It was found that the multifractal scaling of $P(s)$ was destroyed
411: (Fig. \ref{cap:multi}(b)) at a relatively small density of Manna
412: sites $0<c<0.01$. %
413: \begin{figure}[t]
414: \includegraphics{fig4.eps}
415:
416:
417: \caption{\label{cap:-crossover}(Color online) A crossover from multifractal
418: scaling to finite size scaling takes place at $0<c<0.01$. The hatched
419: area border an interval of $\Delta f(c)$ in which $\Delta f(c)\doteq0$
420: and probability distributions $P(x)$ show finite size scaling. We
421: note that $\Delta f(c)=f_{a}^{min}(\alpha_{a},c)-f_{s}^{min}(\alpha_{s},c)$. }
422: \end{figure}
423:
424:
425: Stella \emph{et al.} \cite{Stella_2001} claim that if probability
426: distributions $P(x)$ satisfied FSS the large $q$ data accumulate
427: in the same value $f_{x}(\alpha_{x})$ where $\alpha_{x}^{max}=D_{x}$
428: and $f_{x}=-(\tau_{x}-1)D_{x}$. However, for probability distribution
429: showing the multifractal scaling there is no accumulation point and
430: $f_{x}(\alpha_{x})$ points shift progressively down as the parameter
431: $q$ is increasing and the parameter $q$ approaches $D_{x}$. This
432: fact is utilized as a simple criterion to recognize which probability
433: distributions show either multifractal scaling or FSS \cite{Stella_2001}.
434: The equality $f_{a}^{min}(\alpha_{a},c)\doteq f_{s}^{min}(\alpha_{s},c)$
435: is considered to be an attribute that probability distributions $P(x)$
436: show FSS. To test this equality the differences $\Delta f(c)$ defined
437: as $\Delta f(c)=f_{a}^{min}(\alpha_{a},c)-f_{s}^{min}(\alpha_{s},c)$
438: were determined. The equality $\Delta f(c)\doteq0$ is considered
439: for true if $\left|\Delta f(c)\right|\leq0.10$ which reflects numerical
440: errors. The differences $\Delta f(c)$ are shown in Fig. \ref{cap:-crossover}
441: where the hatched area limits the region where the equality is true
442: and thus the avalanche probability distributions $P(x)$ show FSS
443: behavior. It is clearly evident that only one value of $\Delta f(c)$
444: at the density $c=0$, is outside the region $\left|\Delta f(0)\right|>0.10$,
445: and it corresponds to multifractal scaling of the BTW model \cite{Tebaldi_1999,Stella_2001}.
446: We have no data from the interval of densities $0<c<0.01$ and thus
447: we may only expect that a crossover from multifractal to FSS takes
448: place in this interval.
449:
450: %
451: \begin{figure}[b]
452: \includegraphics{fig5.eps}
453:
454:
455: \caption{\label{cap:capacity}(Color online) The capacity fractal dimensions
456: $D_{x=a,s}(c)$ as functions of the density $c$. The error bars are
457: $\pm0.025$. }
458: \end{figure}
459:
460:
461: The $f_{x}(\alpha_{x},c)$ spectra enable us to determine the capacity
462: fractal dimensions $D_{x}(c)$ as $D_{x}(c)=\alpha_{x}^{max}(c)$.
463: The results $D_{x}(c)$ for densities $0\leq c\leq1$ are shown in
464: the Fig. \ref{cap:capacity}. For the BTW model $D_{s}(0)=2.88\pm0.025$
465: and $D_{a}(0)=2.02\pm0.025$, and for the Manna model $D_{s}(1)=2.77\pm0.025$
466: and $D_{a}(1)=2.03\pm0.025$ were found. The avalanche area capacity
467: fractal dimensions $D_{a}(c)$ are almost constant $D_{a}(c)\doteq2$,
468: for any density $c$, and $D_{a}(0)\doteq D_{a}(1)$. In the interval
469: of densities $0.01<c<0.15$ the avalanche size dimension $D_{s}(c)$
470: is decreasing from $D_{s}(0.01)=2.90$ to the value $D_{s}(0.15)=2.78$
471: and is then almost constant for $c>0.15$, finally $D_{s}(0)>D_{s}(1)$.
472:
473: The moment analysis method \cite{Stella_2001} was used to clarify
474: interesting properties of the scaling exponents $\tau_{x,L\rightarrow\infty}(c)$
475: which are shown in Fig. \ref{cap:exponents}. The values of the functions
476: $f_{x}^{min}(c)$ and $D_{x}(c)$ ( Fig. \ref{cap:capacity}) are
477: determined from the $f_{x}(\alpha_{x},c)$ plots. For specific densities
478: $c=0$ (the BTW model) and $c=1$ (the Manna model) $f_{a}^{min}(0)=-0.43\pm0.05$
479: and $f_{s}^{min}(1)=-0.784\pm0.05$ were found. Then the scaling exponents
480: are given $\tau_{x}(c)=1-f_{x}^{min}(c)/D_{x}(c)$ and are shown in
481: the Fig. \ref{cap:f_exponets}. For the density $c=0$, it was found
482: $\tau_{a}(0)=1.213\pm0.0125$. For the densities $0.01\leq c\leq0.15$,
483: the exponents decrease from $\tau_{a}(0.01)=1.441\pm0.0125$ and $\tau_{s}(0.01)=1.329\pm0.0125$
484: to the values $\tau_{a}(0.15)=1.394\pm0.0125$ and $\tau_{s}(0.15)=1.299\pm0.0125$,
485: which are subsequently constant for $c>0.15$. For the density $c=1$,
486: they are $\tau_{a}(1)=1.386\pm0.0125$ and $\tau_{s}(1)=1.297\pm0.0125$.
487: These results are similar to those determined directly from the distribution
488: functions $P(x)\sim x^{-\tau_{x}}$(Fig. \ref{cap:exponents}).
489:
490: %
491: \begin{figure}
492: \includegraphics{fig6.eps}
493:
494:
495: \caption{\label{cap:f_exponets}(Color online) The scaling exponents were
496: determined using $\tau_{x}(c)=1-f_{x}^{min}(c)/D_{x}(c)$, for the
497: moment analysis all avalanches where $a>50$ were taken into account. }
498: \end{figure}
499:
500:
501:
502: \section{\label{sec:Discussion}Discussion}
503:
504: The plots of $\tau_{x,L}$ vs. $1/\ln L$ and an approximation given
505: by Eq. (\ref{eq:fit}) were used to extrapolate scaling exponents
506: $\tau_{x,L\rightarrow\infty}$ \cite{Manna_PA,Lubeck}. L\"{u}beck
507: and Usadel \cite{Lubeck} have analyzed an influence of an uncertainty
508: in the determination of the exponents $\tau_{x,L}$ on the precision
509: of the extrapolated exponents $\tau_{x,L\rightarrow\infty}$. Their
510: results show that this method is not very accurate. However, this
511: approximation enables us to make a comparison of our results with
512: previous ones. The scaling exponents of the BTW model $\tau_{a,L\rightarrow\infty}(0)=1.26$
513: and $\tau_{s,L\rightarrow\infty}(0)=1.23$ (Fig. \ref{cap:exponents})
514: are approximately the same as those found in Ref. \cite{Lubeck} ($\tau_{a}=1.258$
515: and $\tau_{s}=1.247$) using the same method. The exponents of the
516: Manna model $\tau_{a,L\rightarrow\infty}(1)=1.36$, and $\tau_{s,L\rightarrow\infty}(1)=1.27$
517: are comparable with the previous results, $\tau_{s,L=1024}=1.28\pm0.02$
518: \cite{Manna_1991} and with $\tau_{a}=1.373$ and $\tau_{s}=1.275$,
519: which were found by direct determination of exponents \cite{Lubeck}
520: or calculated from the moment analysis $\tau_{a}\doteq1.36$ and $\tau_{s}\doteq1.28$
521: \cite{Lubeck_2}. The results obtained by the moment analysis \cite{Stella_2001},
522: $f_{a}^{min}(0)=-0.43\pm0.05$ and $f_{s}^{min}(1)=-0.784\pm0.05$,
523: agree well with the previous results, ${\textstyle \sigma_{a}=-0.391\pm0.011}$
524: and $\sigma_{s}=-0.7900\pm0.002$ \cite{Lubeck_2}. We may conclude
525: that the experimental data for two known densities, $c=0$ and $1$,
526: and data analysis methods give approximately the same exponents as
527: were found in previous numerical experiments \cite{Manna_1991,Lubeck,Lubeck_2}.
528:
529: The scaling exponents defined by Eq. (\ref{eq:FSS}) \cite{Kadanoff}
530: and the conditional exponents $\gamma_{xy}$\cite{Christensen,Ben-Hur}
531: can characterize the sandpile models. The theory predicts $\tau_{s}=1.253$
532: \cite{Pietronero} and a few numerical experiments show $D_{s}\simeq2.7$
533: and $D_{a}\simeq2$ \cite{Pietronero,Chessa}. The conditional exponents
534: $\gamma_{sa}$ determined directly from the numerical experiments
535: are $\gamma_{sa}(0)=1.06$ and $\gamma_{sa}(1)=1.23$ \cite{Ben-Hur}.
536:
537: Let us assume that the BTW and Manna models belong to the same universality
538: class. Then the scaling exponents $\tau_{x}(c),D_{x}(c)$ [Eq. (\ref{eq:FSS}]
539: of the model (Eqs. (\ref{eq:relax1})-(\ref{eq:relax4})) must be
540: independent on the density $c$, i.e. $\tau_{x}(c)=const.$ and $D_{x}(c)=const.$
541: This means that knowing only the scaling exponents ($\tau_{x}(c),D_{x}(c)$),
542: we could not distinguish how many sites are toppling by deterministic
543: or stochastic manner [Eq. (\ref{eq:relax4})].
544:
545: We observed that the capacity fractal dimensions $D_{a}(c)$ is constant
546: for any density $c$, $D_{a}(c)\doteq2$. The capacity fractal dimension
547: $D_{s}(0)=2.88$ is the same as was found in the Ref. \cite{Stella_2001},
548: $D_{s}\doteq2.86$ (determined from the Fig. 1(a) in \cite{Stella_2001}).
549: Our capacity fractal dimension $D_{s}(1)=2.77$ is higher than the
550: value $D\simeq2.7$ \cite{Manna_1991,Chessa}, however it is closer
551: to the $D\simeq2.75$ \cite{Karma}. In addition, for densities $0.01\leq c\leq0.1$,
552: the scaling exponents $\tau_{x,L\rightarrow\infty}(c)$, $\tau_{x}(c)$
553: (Figs. \ref{cap:exponents} and \ref{cap:f_exponets}) and $D_{s}(c)$
554: (Fig. \ref{cap:capacity}) depend on the density $c$. These scaling
555: exponents and capacity fractal dimension are not constant. They demonstrate
556: that the assumption about a single universality class is wrong and
557: thus confirm the existence of different universality classes.
558:
559: The conditional scaling exponents $\gamma_{xy}$ \cite{Christensen}
560: can be determined as $\gamma_{xy}(c)=(\tau_{y}(c)-1)/(\tau_{x}(c)-1)$
561: \cite{Lubeck_2}. Substituting the known scaling exponents $\tau_{x}(c)$
562: (Fig. \ref{cap:f_exponets}), we determined $\gamma_{sa}(0.01)\doteq1.34$
563: and for the Manna model, $\gamma_{sa}(1)\doteq1.29$. We note that
564: the scaling exponent $\tau_{s}(0)$ does not really exist.
565:
566: To determine the exact scaling exponents of the probability distribution
567: functions $P(x)$, the experimental data must show a power-law dependence
568: given by Eq. (\ref{eq:FSS}). However, the avalanche area size distributions
569: $P(a)$ do not follow exactly power-law distributions for densities
570: $0<c\leq0.1$ in the whole range of avalanche area sizes, a typical
571: example is shown in the Fig. \ref{cap:avalanche-area}. Chessa \emph{et
572: al.} \cite{Chessa} found that the area size distribution $P(a)$
573: of the BTW model ($c=0$) is not compatible with the FSS hypothesis
574: in the whole range of avalanches. However, for large size of avalanches
575: the FSS form must be approached. They assume that the scaling in the
576: BTW model needs sub-dominant corrections of the form $P(x)=(C_{1}x^{-\tau_{1}}+C_{2}x^{-\tau_{2}}+\ldots)F(x/x_{c})$
577: where $C_{i}$ are nonuniversal constants and that these corrections
578: do not determine universality class. The asymptotic scaling behavior
579: is determined by the leading power law. We assume that the deviation
580: from a simple power-law for densities $0<c\leq0.1$ (Sec. \ref{sec:Results})
581: could be explained by this correction. We observed that the exponents
582: for large avalanches $a$ are larger than the approximate exponents
583: ($\tau_{a,L=1024}=1.23$ in the Fig. \ref{cap:avalanche-area}) which
584: cover the whole range. As a consequence, the leading exponents $\tau_{a,L}(c)$
585: for densities $0<c<0.1$ are higher than the approximate exponents
586: which we found (they are not shown in the Fig.\ref{cap:avalanche-area}
587: for $0<c\leq0.09$). It is evident that the leading scaling exponents
588: $\tau_{a,L}$ are different and are not constant (Fig. \ref{cap:exponents})
589: as in the case of the BTW model or the Manna model and thus the model
590: for these densities belongs to a different class than the BTW model
591: or the Manna model.
592:
593: Divergences from the expected power-law behaviour of the BTW model
594: and a need of sub-dominant correction were observed in another inhomogeneous
595: sandpile model \cite{Cer}. Here the avalanche dynamic was disturbed
596: by sites which had the second higher threshold. The effect was significant
597: for thresholds $E_{C}\geq32$ and low concentration of such sites
598: \cite{Cer}.
599:
600: The multifractal properties (Fig. \ref{cap:multi}) of the model given
601: by Eqs. (\ref{eq:thresholds})-(\ref{eq:relax4}) for the density
602: $c=0$ (the BTW model), and FSS for the density $c=1$ (the Manna
603: mode) agree well with the recent results \cite{Stella_2001}. In addition,
604: the crossover from multifractal to FSS was observed in the Fig. \ref{cap:-crossover}.
605: Our results can only predict that a critical density is expected to
606: be found in the interval of densities $0<c<0.01$ (Figs. \ref{cap:multi}
607: and \ref{cap:-crossover}). This interval is five times smaller than
608: what was found in Ref. \cite{Karma} where the results are based on
609: the autocorrelation function of the avalanche wave time series \cite{Menech_2}.
610:
611: We assume that divergences from power-law dependences in inhomogeneous
612: conservative models, \cite{Cer} and Eqs. (\ref{eq:thresholds})-(\ref{eq:relax4}),
613: have a common reason which is connected to the crossover from multifractal
614: scaling to FSS \cite{Karma}. In both models a disorder is induced
615: by deployment of disturbing sites. These disturbing sites either increase
616: the short range coupling during relaxations in deterministic model
617: \cite{Cer} or introduce the random toppling [Eq. (\ref{eq:relax4})].
618: In these models toppling imbalance \cite{Karma_E,Karma} only for
619: a few such sites can change character of waves in the models from
620: coherent to more fragmented waves \cite{Ben-Hur,Biham,Milsh,Stella_2001}.
621:
622: In this study, the multifractal properties of the BTW model which
623: is initially homogeneous, are destroyed at very low concentrations
624: of such disturbing sites. In the opposite case, the Manna model shows
625: the FSS and resistance to disturbance caused by presence of BTW sites
626: because all significant exponents from Eq. (\ref{eq:FSS}) are approximately
627: constant in a broad range of densities $0.15\leq c\leq1$. One possible
628: explanation for this is that the nature of the small perturbation
629: of the model is not the same when we perform changes around the densities
630: at $c=0$ and $c=1$. A small perturbation of the dynamical rules
631: of the BTW model ($c=0$) breaks the toppling symmetry \cite{Karma}
632: and this may explain why the changes in the scaling exponents $\tau_{x}(c)$
633: and capacity fractal dimension $D_{s}(c)$ are so unexpected. On the
634: other hand, for the Manna model ($c=1$), decreasing of the density
635: $c$ cannot influence the unbalanced toppling symmetry of the Manna
636: model \cite{Karma}. For sandpile models which show FSS this is an
637: expected result and agrees well with the theory \cite{Pietronero,Chessa},
638: where a small modification of toppling rules cannot change the scaling
639: exponents.
640:
641: We can clearly identify two universality classes which correspond
642: to the classes proposed in papers \cite{Ben-Hur} or \cite{Karma}:
643: (a) nondirected models, for density $c=0$ (BTW model, the multifractal
644: scaling \cite{Tebaldi_1999,Stella_2001,Menech}), and they show a
645: precise toppling balance \cite{Karma} and they are sensitive on disturbance
646: of avalanche dynamics, (b) random relaxation models, for densities
647: $0.1<c<1$ where FSS of $P(x)$ is verified, they are nondirected
648: only on average (Manna two-state model $c=1$ \cite{Ben-Hur}). In
649: these models breaking of the precise toppling balance \cite{Karma}
650: is observed, the scaling exponents are resistant to disturbance of
651: avalanches. The classification for densities $0<c<0.1$ is not so
652: clear. If we follow the proposed classifications then the model is
653: a random relaxation model \cite{Ben-Hur} with broken precise toppling
654: balance \cite{Karma} and it belongs in the same class as the Manna
655: model. On the other hand, the scaling exponents differ from the Manna
656: model and they are not universal ($\tau_{x}(c)\neq const$., $D_{s}(c)\neq const.$),
657: and the reasons of the sub-dominant approximation of area probability
658: distribution functions \cite{Chessa} can play an important role.
659: We assume that a new universality class between the BTW ($c=0$, multifractal
660: scaling) and the Manna ($c>0.5$, FSS) classes \cite{Karma,Karma_E}
661: could be identified for densities $0<c<0.1$. However, a more detailed
662: study is necessary to verify this classification.
663:
664: Our additional arguments to the previous results \cite{Ben-Hur,Biham,Cer,Karma,Milsh,Menech,Lubeck}
665: show that small modifications of the dynamical rules of the model
666: can lead to different universality classes what is considered to be
667: unusual from a theoretical standpoint \cite{Chessa}.
668:
669:
670: \section{\label{sec:Conclusion}Conclusion}
671:
672: In these computer simulations multifractal scaling of the BTW model
673: \cite{Tebaldi_1999} and FSS of the Manna model \cite{Stella_2001}
674: were confirmed. In addition, a crossover from multifractal scaling
675: to FSS \cite{Karma} was observed when avalanche dynamics of the BTW
676: model was disturbed by Manna sites which were randomly deployed in
677: the lattice, as their density was increased. This crossover takes
678: place for a certain density $c$ in the interval $0<c<0.01$. This
679: interval is five times smaller than what was found recently \cite{Karma}.
680: The scaling exponents $\tau_{x}(c)$ and the capacity fractal dimension
681: $D_{s}(c)$ are not constant for all densities $c$ which is necessary
682: if the models \cite{BTW_1987,Manna_1991} belong to the same universality
683: class. These result agree well with the previous conclusions that
684: multifractal properties of the BTW model \cite{Stella_2001,Tebaldi_1999,Menech},
685: toppling wave character \cite{Ben-Hur,Biham,Milsh} and precise toppling
686: balance \cite{Karma,Karma_E} are important properties for solving
687: the universality issues.
688:
689: An open question remains about how to characterize the universality
690: class for densities $0.01<c<0.1$, where the scaling exponents are
691: not universal ($\tau_{x}(c)\neq const$. and $D_{s}(c)$$\neq const.$)
692: and in addition, the avalanche probability distributions $P(a)$ do
693: not show exact power-law behavior since the sub-dominant corrections
694: of $P(a)$ \cite{Chessa} are important. In this interval of densities
695: $c$, our model belongs to the random relaxation models \cite{Ben-Hur}
696: and to the models with unbalanced toppling sites \cite{Karma,Karma_E},
697: however, its scaling exponents are not equal to the exponents of the
698: Manna model.
699:
700: Based on the previous findings \cite{Karma,Karma_E} and our results
701: we assume that the avalanche dynamics of undirected conservative models,
702: in which some of the probability distribution functions show a multifractal
703: scaling (the BTW model), is disturbed by suitable toppling rules which
704: are different from the two-state Manna model (for example a stochastic
705: four-state Manna model \cite{Ben-Hur,Milsh}), then a local manner
706: for the energy distribution during the relaxation can be important
707: and can change the scaling exponents. However, the models which show
708: the FSS for all probability distribution functions (the Manna model)
709: are not sensitive to the details of the toppling rules and are consistent
710: with theoretical predictions \cite{Pietronero,Chessa}.
711:
712:
713: \section*{Acknowledgments}
714:
715: The author thanks G. Helgesen for his comments to the manuscript.
716: The numerical simulations were carried out using the ARC middleware
717: and NorduGrid infrastructure \cite{Nordugrid}. We acknowledge the
718: financial support from the Slovak Ministry of Education: Grant NOR/SLOV2002.
719:
720: \begin{thebibliography}{10}
721: \bibitem{BTW_1987}P. Bak, C. Tang, and K. Wiesenfeld, Phys. Rev. Lett. \textbf{59},
722: 381 (1987); Phys. Rev. A \textbf{38}, 364 (1988).
723: \bibitem{Manna_1991}S. S. Manna, J. Phys. A \textbf{24}, L363 (1991).
724: \bibitem{Kadanoff}L. P. Kadanoff, S. R. Nagel, L. Wu and S. Zhou, Phys. Rev. A \textbf{39},
725: 6524 (1989).
726: \bibitem{Pietronero}L. Pietronero, A. Vespignani, and S. Zapperi, Phys. Rev. Lett. \textbf{72},
727: 1690 (1994); A. Vespignani, S. Zapperi, and L. Pietronero, Phys. Rev.
728: \textbf{}E, \textbf{51}, 1711 (1995).
729: \bibitem{Menech}M. De Menech, Phys. Rev. E \textbf{70}, \textbf{}028101 (2004).
730: \bibitem{Lubeck}S. L\"{u}beck and K. D. Usadel, Phys. Rev. E \textbf{55}, 4095 (1997).
731: \bibitem{Ben-Hur}A. Ben-Hur and O. Biham, Phys. Rev. E \textbf{53}, R1317 (1996).
732: \bibitem{Biham}E. Milshtein, O. Biham, and S. Solomon, Phys. Rev. E \textbf{58},
733: 303 (1998).
734: \bibitem{Milsh}O. Biham, E. Milshtein, and O. Malcai, Phys. Rev. E \textbf{63}, 061309
735: (2001).
736: \bibitem{Chessa}A. Chessa, H. E. Stanley, A. Vespignani, and S. Zapperi, Phys. Rev.
737: E \textbf{59}, \textbf{}R12 (1999).
738: \bibitem{Tebaldi_1999}C. Tebaldi, M. De Menech and A. L. Stella, Phys. Rev. Lett. \textbf{83},
739: 3952 (1999).
740: \bibitem{Stella_2001}A. L. Stella and M. De Menech, Physica A \textbf{295}, 101 (2001).
741: \bibitem{Karma}R. Karmakar, S. S. Manna, and A. L. Stella, Phys. Rev. Lett. \textbf{94},
742: 088002 (2005).
743: \bibitem{Karma_E}R. Karmakar and S. S. Manna, Phys. Rev. E \textbf{71}, 015101 (2005).
744: \bibitem{Cer}J. \v{C}ern\'{a}k, Phys. Rev. E \textbf{65}, 046141 (2002).
745: \bibitem{Dhar}D. Dhar, Physica A \textbf{263}, 4 (1999).
746: \bibitem{Manna_PA}S. S. Manna, Physica A \textbf{179}, 249 \textbf{}(1991).
747: \bibitem{Lubeck_2}S. L\"{u}beck, Phys. Rev. E \textbf{61}, 204 (2000).
748: \bibitem{Christensen}K. Christensen, H. C. Fogedby, and H. J. Jensen, J. Stat. Phys. \textbf{63},
749: 653 (1991).
750: \bibitem{Menech_2}M. De Menech and A. L. Stella, Phys. Rev. E \textbf{62}, R4528 (2000).
751: \bibitem{Nordugrid}For more information see the web site\\ http://www.nordugrid.org
752: \end{thebibliography}
753:
754: \end{document}
755: