1: \documentclass[pre,showpacs,preprint]{revtex4}
2: \usepackage{graphicx}
3:
4:
5: \begin{document}
6:
7: \title{Non-Abelian self-organized criticality model with one stochastic
8: site in each avalanche shows multifractal scaling }
9:
10:
11: \author{Jozef \v{C}ern\'{a}k}
12:
13:
14: \address{Institute of Physics, P. J. \v{S}af\'{a}rik University in Ko\v{s}ice,
15: Jesenná 5, Ko\v{s}ice, Slovak Republic}
16:
17: \begin{abstract}
18: I have proposed a\textit{ non-Abelian }and stochastic self-organized
19: criticality model in which each avalanche contains one stochastic
20: site and all remaining sites in the avalanche are deterministic with
21: a constant threshold $E_{c}^{I}$. Studies of avalanche structures,
22: waves and autocorrelations, size moments and probability distribution
23: functions of avalanche size, for the thresholds $4\leq E_{c}^{I}\leq256$,
24: were performed. The shell-like avalanche structures, correlated waves
25: within avalanches, complex size moments and probability distribution
26: functions show multifractal scaling like the \textit{Abelian} and
27: deterministic BTW model despite the fact that the model is \textit{non-Abelian
28: }and stochastic with unbalanced relaxation rules at each stochastic
29: site.
30: \end{abstract}
31:
32: \pacs{45.70.Ht, 05.65.+b, 05.70.Jk, 64.60.Ak}
33:
34: \maketitle
35:
36: \section{Introduction}
37:
38: Bak, Tang, and Wiesenfeld (BTW) \cite{BTW} introduced a concept
39: of self-organized criticality (SOC) to study dynamical systems with
40: temporal and spatial degrees of freedom. They designed a simple cellular
41: automaton with conservative and deterministic relaxation rules to
42: demonstrate the SOC phenomenon. Manna (M) \cite{Manna} proposed
43: another conservative SOC model in which stochastic relaxation rules
44: instead of deterministic rules were defined.
45:
46: A stable configuration (see below) in the BTW model does not depend
47: on the order of relaxations so the model is Abelian \cite{Dhar}.
48: On the other hand, a stable configuration in the M model depends on
49: the order of relaxations and the model is thus non-Abelian. Dhar \cite{Dhar_abel}
50: theoretically proved the Abelian property of the M model for the case
51: when probabilities of many final stable configurations are considered.
52:
53: Based on the real-space renormalization group calculations, Pietronero
54: \emph{at al.} \cite{Pietro} claimed that both deterministic \cite{BTW}
55: and stochastic \cite{Manna} models belong to the same universality
56: class, i.e. a small modification in the relaxation rules of the models
57: cannot change the universality class. It was assumed that both models
58: show a finite size scaling (FSS) \cite{Chessa}. With FSS the avalanche
59: size, area, lifetime, and perimeter follow power laws with cutoffs
60: \cite{Chessa}:
61:
62: \begin{equation}
63: P(x)=x^{-\tau_{x}}F(x/L^{D_{x}}),\label{eq:FSS}\end{equation}
64: where $P(x)$ is the probability density function of $x$, $F$ is
65: the cutoff function, and $\tau_{x}$ and $D_{x}$ are the scaling
66: exponents. The set of scaling exponents $(\tau_{x},\: D_{x})$ defines
67: the universality class \cite{Chessa}.
68:
69: Avalanche structure studies \cite{Ben} and numerical simulations
70: \cite{Lub} showed that the BTW and M models do not belong to the
71: same universality class. Later, Tebaldi \textit{at al.} \cite{Teb}
72: found that avalanche size distributions in the BTW model follow multifractal
73: scaling. They concluded that the avalanche size exponent $\tau_{s}$
74: (Eq. \ref{eq:FSS}) does not apply to the BTW model. An avalanche
75: wave decomposition approach \cite{Ivash} was applied \cite{Menech}
76: to demonstrate the different wave features of the BTW and M models.
77: Karmakar \emph{at al.} \cite{Karmar} found that models with balanced
78: relaxation rules are similar to the deterministic BTW model and models
79: with unbalanced relaxation rules are similar to stochastic M-like
80: models. These results \cite{Karmar,Teb,Ben,Lub,Menech} support the
81: idea that the BTW and Manna models are prototypes of different universality
82: classes.
83:
84: The BTW and M models are well understood \cite{Dhar,Dhar_abel,Lub,Teb,Ben}.
85: On the hand, we know very little about the transition from multifractal
86: to fractal scaling \cite{Cer,Karmar}. I modified the BTW model \cite{BTW}
87: to study this transition. I allowed stochastic relaxations in one
88: site in each avalanche. The stochastic site is located in the place
89: where the avalanche is initiated. Avalanches are randomly initiated
90: in various lattice sites so the position of stochastic site is changed
91: at every new avalanche and the stochastic M-like site can visit all
92: lattice sites for sufficiently large set of avalanches. An avalanche
93: dynamics study \cite{Menech} is useful to understand the transition
94: from multifractal to fractal scaling and to test the hypothesis of
95: precise relaxation balance \cite{Karmar}. The results suggest that
96: the model cannot belong to either BTW or M universality classes.
97:
98: In Sec. \ref{sec:model} I introduce a \textit{non-Abelian} and stochastic
99: model. In Sec. \ref{sec:Results} I apply numerical simulations and
100: statistical methods to find avalanche structures, autocorrelation
101: functions, Hurst exponents, avalanche size moments and probability
102: density functions of avalanche sizes. Section \ref{sec:Discussion}
103: is devoted to a discussion which is followed by conclusions in Sec.
104: \ref{sec:Conclusion}.
105:
106:
107: \section{Stochastic Self-organized criticality model\label{sec:model}}
108:
109: The stochastic SOC model is defined on a two dimensional (2D) lattice
110: $L\times L$ where each site $\mathbf{i}$ has assigned two dynamical
111: variables $E(\mathbf{i})$ and $E_{c}(\mathbf{i})$ \cite{Cer}.
112: The variable $E(\mathbf{i})$ represents for example, energy and variable
113: $E_{c}(\mathbf{i})$ represents a threshold at site $\mathbf{i}$.
114: All thresholds $E_{c}(\mathbf{i})$ are equal to the same value $E_{c}^{I}$
115: in the interval $4\leq E_{c}^{I}\leq256$. Relaxation rules are undirected,
116: conservative and deterministic for all sites $\mathbf{i}$ with the
117: thresholds $E_{c}(\mathbf{i})=E_{c}^{I}$. At each site $\mathbf{i}$
118: the relaxation rules are precisely balanced \cite{Karmar}. Thus
119: the sites behave as the BTW sites \cite{BTW} for the threshold $E_{c}^{I}=4$.
120: In a stable configuration (a stationary state), all sites $\mathbf{i}$
121: follow $E(\mathbf{i})<E_{c}(\mathbf{i})$. Let us assume that from
122: a stable configuration we iteratively select $\mathbf{i}$ at random
123: and increase $E(\mathbf{i})\rightarrow E(\mathbf{i})+1$. If an unstable
124: configuration is reached, i.e. $E(\mathbf{i})\geq E_{c}^{I}$, then
125: the site $\mathbf{i}$ is labeled as $\mathbf{i_{\mathrm{M}}}$. The
126: initial threshold at site $\mathbf{i_{\mathrm{M}}}$, $E_{c}(\mathbf{i_{\mathrm{M}}})=E_{c}^{I}$
127: is changed to the new value $E_{c}(\mathbf{i_{\mathrm{M}}})=E_{c}^{II}=2$
128: and stochastic relaxation rules \cite{Manna} are assigned to this
129: site. In each avalanche only one site $\mathbf{i_{\mathrm{M}}}$ undergoes
130: stochastic relaxations while all remaining sites relax as undirected,
131: deterministic and conservative sites. All unstable sites $\mathbf{i}$
132: (including $\mathbf{i_{\mathrm{M}}}$) where $E(\mathbf{i})\geq E_{c}(\mathbf{i})$
133: undergo relaxations to reach the stable configuration $E(\mathbf{i})<E_{c}(\mathbf{i})$.
134: If a stable configuration is reached, then the threshold $E_{c}(\mathbf{i_{\mathrm{M}}})$
135: at the site $\mathbf{i_{\mathrm{M}}}$ is set to $E_{c}(\mathbf{i_{\mathrm{M}}})=E_{c}^{I}$
136: and deterministic relaxation rules \cite{BTW} are assigned to the
137: site $\mathbf{i_{\mathrm{M}}}$. The site $\mathbf{i_{\mathrm{M}}}$
138: disappears and all sites $\mathbf{i}$ of the lattice are BTW-like
139: sites. Stable and unstable states are repeated many times. Adding
140: of energy ($E(\mathbf{i})\rightarrow E(\mathbf{i})+1$) takes place
141: randomly thus stochastic sites $\mathbf{i_{\mathrm{M}}}$ could visit
142: all lattice sites. Thus stochastic sites introduce an annealed disorder
143: in the initial deterministic model \cite{BTW}.
144:
145:
146: \section{Results\label{sec:Results}}
147:
148: One stochastic site in each avalanche makes the model stochastic \cite{Biham}.
149:
150: All sites, except one, relax energy as\textit{ Abelian} sites \cite{Dhar_1990,Dhar},
151: however I classified the model as non-Abelian. To demonstrate the\textit{
152: non-Abelian} property consider two critical sites \cite{Dhar_1990}
153: (Fig. \ref{fig:non-abelian}) with thresholds $E_{c}^{II}=2$ (the
154: stochastic site) and $E_{c}^{I}=4$ (the deterministic site). One
155: can verify (Fig. \ref{fig:non-abelian}) that the order of relaxations
156: of unstable sites leads to different final stable configurations i.e.
157: the OSS model is \textit{non-Abelian}.
158:
159: Relaxations in all deterministic sites are precisely balanced \cite{Karmar}.
160: However, one site in each avalanche shows unbalanced relaxations.
161: Thus we consider the OSS model for the model with unbalanced relaxation
162: rules.
163:
164: %
165: \begin{figure}
166: \includegraphics[width=7cm]{fig1}
167:
168: \caption{\label{fig:non-abelian}(color online) Order of relaxations of the
169: critical sites shows the\textit{ non-Abelian} property of the OSS
170: model. The circle denotes a stochastic site and the star ({*}) denotes
171: a site which relaxes as first. In the initial state there are two
172: critical sites $E_{c}^{II}=2$ (stochastic) and $E_{c}^{I}=4$. (a)
173: The stable configuration in the case when the stochastic site ($E_{c}^{II}=2$)
174: relaxed as first. (b) The stable configuration in the case when the
175: deterministic site ($E_{c}^{I}=4$) relaxed as first. Arrows show
176: directions of energy diffusion in the stochastic site. }
177:
178: \end{figure}
179:
180:
181: Analyzing the avalanche structures can provide an important initial
182: information \cite{Ben,Karmar,Santra} about the nature of the SOC
183: model. Several avalanches of the OSS model, at the threshold $E_{c}^{I}=4$,
184: have been decomposed into clusters with equal numbers of relaxations
185: (Fig. \ref{fig:aval}). Avalanche structures of two types: (i) without
186: holes Fig. \ref{fig:aval}(a) and (ii) with rare holes in clusters
187: Fig. \ref{fig:aval}(b) were observed. These structures are similar
188: to shell-like structures one finds in the BTW model.
189:
190: %
191: \begin{figure}
192: \includegraphics[width=7cm]{fig2a}
193:
194: \includegraphics[width=7cm]{fig2b}
195:
196: \caption{\label{fig:aval}(color online) Two types of avalanche structures
197: on a 2D lattice of size $128\times128$: (a) without holes in the
198: clusters and (b) with holes in clusters. Lattice sites with the same
199: numbers of relaxations are shown by the same color (rainbow pseudo-color
200: coding). Only one site in any avalanche relaxes in a stochastic manner
201: as a M site (the red area), however all remaining sites relax in a
202: deterministic manner as BTW sites ($E_{c}^{I}=4$). }
203:
204: \end{figure}
205:
206:
207: The study of mathematical SOC models allows avalanches to be decomposed
208: avalanches into waves \cite{Ivash}. This approach enables the investigation
209: of correlation of waves within the avalanche \cite{Menech}. I demonstrated
210: that the OSS model is \textit{non-Abelian,} thus the avalanche waves
211: could be defined as in the case of the M model \cite{Menech}. On
212: the other hand, waves initiated by relaxations of the stochastic site
213: ($\mathbf{i_{\mathrm{M}}}$) propagate though the lattice of BTW sites.
214: Thus waves would be defined as well as waves of the BTW model. An
215: avalanche of size $s$ is decomposed into $m$ waves with sizes $s_{k}$,
216: where $s={\textstyle \sum_{k=1}^{m}}s_{k}$. A time-sequence of avalanche
217: waves $s_{k}$ is used to determine the autocorrelation function \cite{Menech,Stella}
218:
219: \begin{equation}
220: C(t,L)=\frac{\langle s_{k+t}s_{k}\rangle_{L}-\langle s_{k}\rangle_{L}^{2}}{\langle s_{k}^{2}\rangle_{L}-\langle s_{k}\rangle_{L}^{2}},\label{eq:autoco}\end{equation}
221: where time is $t=1$, $2$, $\ldots$, and the time averages are
222: taken over $5\times10^{6}$ waves for lattice sizes $L=128$,$\:$$256$,$\:$$512$,$\:$$1024,\:2048$
223: and $4096$. The autocorrelations $C^{BTW}(t,\: L)$ of the BTW model
224: (Fig. \ref{fig:Autocorrelation}(a)) and $C^{OSS}(t,\: L)$ of the
225: OSS model at threshold $E_{c}^{I}=4$ (Fig. \ref{fig:Autocorrelation}(b))
226: approach zero only for times $t_{max}^{BTW}$ and $t_{max}^{OSS}$
227: exceeding the maximum number of waves in avalanches. This result is
228: a consequence of correlated waves in avalanches \cite{Menech}. I
229: have observed (Fig. \ref{fig:Autocorrelation}) that at a given lattice
230: size $L$, the time $t_{max}^{OSS}$ is approximately as large as
231: time $t_{max}^{BTW}$ ($t_{max}^{OSS}\doteq t_{max}^{BTW}$). I note
232: that avalanche waves in the M model are uncorrelated due to autocorrelation
233: functions $C^{M}(t,\: L)\doteq0$ for $t\geq1$ \cite{Menech,Karmar}.
234:
235: %
236: \begin{figure}
237: \includegraphics[width=7cm]{fig3}
238:
239: \caption{\label{fig:Autocorrelation}(color online) Log-lin plots of autocorrelation
240: functions $C^{BTW}(t,\: L)$ and $C^{OSS}(t,\: L)$ (threshold $E_{c}^{I}=4$)
241: for lattices: $L=128,\:256,\:512,\:1024$,$\:2048$ and $4096$ (from
242: left to right). The autocorrelation functions $C^{X}(t,\: L)$ for
243: given lattice sizes $L$ and for time $t>5$ are approximated by an
244: exponential function $C^{X}(t,\: L)\sim\exp(-\alpha_{L}t)$ (thick
245: lines) where decay rates are: (a) $\alpha_{4096}^{BTW}=0.0244$ (BTW
246: model) and (b) $\alpha_{4096}^{OSS}=0.027$ (OSS model). }
247:
248: \end{figure}
249:
250:
251: The autocorrelations $C^{BTW}(t,\: L)$ (Eq. \ref{eq:autoco}) of
252: the BTW model were approximated by a power law $f(t)\sim t^{-\tau_{c}}$
253: and cutoff function $g(t/L^{D_{c}})$ \cite{Menech}:
254:
255: \begin{equation}
256: C^{BTW}(t,\: L)=f(t)g(t/L^{D_{c}}).\label{eq:autoco_approx}\end{equation}
257: The autocorrelations $C^{BTW}(t,\: L)$ (Fig. \ref{fig:Autocorrelation}(a))
258: agree well with the previous results \cite{Menech,Karmar}. However,
259: I have found that the power law approximation $f(t)\sim t^{-\tau_{c}}$
260: does not approximate the autocorrelation $C^{BTW}(t,\: L)$. I have
261: verified (Fig. \ref{fig:Autocorrelation}) that the exponential function
262: $f(t,\: L)\sim\exp(-\alpha_{L}t)$ better approximates not only the
263: autocorrelation $C^{BTW}(t,\: L)$ but also $C^{OSS}(t,\: L)$.
264:
265: %
266: \begin{figure}
267: \includegraphics[width=6.5cm]{fig4}
268:
269: \caption{\label{fig:timemoments} The plots of $\sigma_{c}(q)$ are approximated
270: by the linear function $\sigma_{c}(q)=p_{1}q+p_{0}$. For the BTW
271: model (thin line) the parameters are $p_{1}=1.06\pm0.05,$$\:$$p_{0}=1.06\pm0.05$
272: and for the OSS model at threshold $E_{c}^{I}=4$ (thick line) the
273: parameters are $p_{1}=1.09\pm0.05$, and $p_{0}=1.06\pm0.05$ (this
274: approximation is shown with dashed line). The inset shows log-log
275: plots of $<t^{q}>$ vs. $L$ for the OSS model and for the exponents
276: $q=-1,\:0,\:1,\:2,\:3$ and $4$ (from bottom to top). }
277:
278: \end{figure}
279:
280:
281: The autocorrelations $C(t,\: L)$ from Eq. \ref{eq:autoco_approx}
282: are used to compute the time moments \cite{Menech}:
283:
284: \begin{equation}
285: \langle t^{q}\rangle_{L}={\displaystyle \sum_{t}C(t,L)t^{q}\sim L^{\sigma_{c}(q)}}.\label{eq:c_moments}\end{equation}
286: For lattice sizes $L=128-2048$, threshold $E_{c}^{I}=4$ and for
287: several values of $q$, the plots of $\log<t^{q}>$ vs. $\log L$
288: (Fig. \ref{fig:timemoments}, inset) were used to determine the functions
289: $\sigma_{c}^{BTW}(q)$ and $\sigma_{c}^{OSS}(q)$. The plots $\sigma_{c}^{BTW}(q)$
290: and $\sigma_{c}^{OSS}(q)$ exhibit linear dependence for $q$ in the
291: range $1.0\leq q\leq4.0$. From these plots (Fig. \ref{fig:timemoments})
292: the parameters $D_{c}^{BTW}=1.06\pm0.05$, $D_{c}^{OSS}=1.09\pm0.05$,
293: $\tau_{c}^{BTW}\doteq0$ and $\tau_{c}^{OSS}\doteq0$ were determined
294: \cite{Menech}.
295:
296: Stochastic process are often characterized by Hurst exponents \cite{Benoit}.
297: Fluctuations $F(t,\: L)$ \cite{Menech}:
298:
299: \begin{equation}
300: F(t,L)=[\langle\Delta y(t)^{2}\rangle_{L}-\langle\Delta y(t)\rangle_{L}^{2}]^{1/2},\end{equation}
301: are used to determine Hurst exponents where $y(t)=\sum_{k=1}^{t}s_{k}$
302: and $\Delta y(t)=y(k+t)-y(k)$. If fluctuations $F(t,\: L)$ should
303: scale as $F(t,\:\infty)\sim t^{H}$ then $H$ is the Hurst exponent
304: \cite{Menech}. Two exponents $H^{BTW}=0.89\pm0.02$ and $H^{BTW}\doteq1/2$
305: (Fig. \ref{fig:Hurst}) were determined for the BTW model and for
306: times $t<t_{max}^{BTW}$ and $t>t_{max}^{BTW}$. Similarly, the Hurst
307: exponents $H^{OSS}=0.88\pm0.02$ and $H^{OSS}=0.610\pm0.001$ were
308: determined for the OSS model (threshold $E_{c}^{I}=4$) and for times
309: $t<t_{max}^{OSS}$ and $t>t_{max}^{OSS}$. Fluctuations $F(t,\: L)$
310: were also determined for the other thresholds $8\leq E_{c}^{I}\leq256$
311: (Fig. \ref{fig:Hurst}). For all thresholds $4\leq E_{c}^{I}\leq256$
312: of the OSS model two scaling regions of $F(t,\: L)$ were identified
313: in contrast to the fluctuation of the M model which exhibits the single
314: scaling with the Hurst exponent $H^{M}=0.53\pm0.05$ (it is not shown
315: in Fig. \ref{fig:Hurst}).
316:
317: %
318: \begin{figure}
319: \includegraphics[width=6.5cm]{fig5}
320:
321: \caption{\label{fig:Hurst}(color online) The fluctuations $F(t,\: L=4096)$
322: of the BTW and OSS model with thresholds $4\leq E_{c}^{I}\leq256$.
323: The Hurst exponents were determined for the threshold $E_{c}^{I}=4$:
324: $H^{BTW}=0.89\pm0.01$ and $H^{OSS}=0.88\pm0.01$ , the time $1.0<t<50.0$,
325: $H^{BTW}=0.503\pm0.001$ and $H^{OSS}=0.610\pm0.001$ for the time
326: $1000.0<t<10000.0$. }
327:
328: \end{figure}
329:
330:
331: Moment analysis \cite{Karmar,Teb,Menech} was used to study scaling
332: properties of both BTW and OSS models. A property $x$ in the FSS
333: system obeys the scaling given by Eq. \ref{eq:FSS}. The $q$ moments
334: of $x$ are defined as
335:
336: \begin{equation}
337: \langle x^{q}\rangle=\intop_{0}^{x_{max}}x^{q}P(x,L)dx\sim L^{\sigma_{x}(q)},\label{eq:moments}\end{equation}
338: where $\sigma_{x}(q)=(q+1-\tau_{x})D_{x}$ and $x_{max}\sim L^{D_{x}}$.
339: I calculated the moments only for avalanche size $s$. The plots $\log\langle s^{q}\rangle$
340: versus $\log L$ for approximately five hundred values of the exponent
341: $q$ were used to determine the functions $\sigma_{s}^{BTW}(q)$ and
342: $\sigma_{s}^{OSS}(q)$ (Eq. \ref{eq:moments}). These plots scale
343: precisely for the BTW model (Fig. \ref{fig:Avalanche-size-moments}(a))
344: and for the OSS model (threshold $E_{c}^{I}=4,$ Fig. \ref{fig:Avalanche-size-moments}(b))
345: for all exponents $0.0\leq q\leq4.0$ and lattice sizes $128\leq L\leq4096$.
346:
347: %
348: \begin{figure}
349: \includegraphics[width=7cm]{fig6a}
350:
351: \includegraphics[width=7cm]{fig6b}
352:
353: \caption{\label{fig:Avalanche-size-moments}Avalanche size moments $\langle s^{q}\rangle_{L}$
354: at lattice size $L$ versus lattice size $L$ display scaling behaviour
355: for wide range of exponents $q$. The results are shown for (a) the
356: BTW model ($128\leq L\leq2048$) and (b) the OSS model with threshold
357: $E_{c}^{I}=4$ ($128\leq L\leq4096$) whereas selected exponents $q$
358: are $q=1.0,\:2.0,\:3.0,\:3.1$ and $4.0$ (from bottom to top). }
359:
360: \end{figure}
361:
362:
363: The results show that $\sigma_{s}^{BTW}(1.0)=2.03$ and $\sigma_{s}^{OSS}(1.0)=2.04$
364: for threshold $E_{c}^{I}=4$ (Fig. \ref{fig:dsigma}, inset), close
365: to the expected value $\sigma_{s}^{BTW}(1)\doteq2.0$ \cite{Santra,Karmar}.
366: The function $\sigma_{s}^{OSS}(q)$ grows faster than functions $\sigma_{s}^{BTW}(q)$
367: and $\sigma_{s}^{M}(q)$ (Fig. \ref{fig:dsigma}, inset) when the
368: exponent $q>1.0$ increases. The function $\partial\sigma_{s}^{OSS}(q)/\partial q$
369: increases with increasing $q>1.0$. At $q=2.07$ it reaches the maximum
370: $D_{s}^{OSS}(2.07)=3.17\pm0.01$ and for $q>2.07$ is almost constant
371: or slowly decreases. For $q=4.0$, the capacity dimensions are $D_{s}^{M}(4)=2.76$,
372: $D_{s}^{BTW}(4)=2.88$ \cite{Cer} and $D_{s}^{OSS}(4)=3.11\pm0.02$.
373: The functions $\partial\sigma_{s}^{BTW}(q)/\partial q$ and $\partial\sigma_{s}^{OSS}(q)/\partial q$
374: continuously increase when the exponent $q$ increases in the range
375: $1.0<q<2.07$. This fact is considered to be a signature of multifractal
376: scaling \cite{Karmar,Teb}. The capacity dimension of the Manna model
377: $D_{s}^{M}(q)$ is constant for exponents $q>1.0$: $D_{s}^{M}(q)=2.76\pm0.01$
378: \cite{Karmar,Cer,Santra}, which is a typical property of the FSS
379: models \cite{Karmar,Santra}.
380:
381: %
382: \begin{figure}
383: \includegraphics[width=7cm]{fig7}
384:
385: \caption{\label{fig:dsigma}Plots of $\partial\sigma_{s}(q)/\partial q$ versus
386: $q$ for the M (dashed curves, $128\leq L\leq2048$), BTW (thin line,
387: $128\leq L\leq2048$) and OSS models for the thresholds $E_{c}^{I}=4$
388: (thick solid, $128\leq L\leq4096$). The inset shows plots $\sigma_{s}(q)$
389: versus $q$ for the M (dashed line, $128\leq L\leq2048$), BTW (thin
390: line, $128\leq L\leq2048$) and OSS (thick solid, $128\leq L\leq4096$)
391: models. }
392:
393: \end{figure}
394:
395:
396: Karmakar \textit{et al.} \cite{Karmar} claimed that if local avalanche
397: dynamics meets criterion of a precise relaxation balance, then the
398: model must show the same behaviors as the BTW model. If we increase
399: the threshold $E_{c}^{I}>4$ and we modify the relaxation rules to
400: meet the criterion of the precise relaxation balance, then the model
401: with all deterministic sites \cite{Cer2009} has the same properties
402: as the BTW model. In this model, for thresholds $8\leq E_{c}^{I}\leq256$,
403: I introduced one stochastic site in each avalanche to compare the
404: behaviors of the modified SOC model (see Sec. \ref{sec:model}) with
405: the behaviors of the BTW model.
406:
407: %
408: \begin{figure}
409: \includegraphics[width=6.5cm]{fig8}
410:
411: \caption{\label{fig:Autocorr2}(color online) Autocorrelations $C(t,\: L=4096)$
412: of the OSS model for thresholds $4\leq E_{c}^{I}\leq256$.}
413:
414: \end{figure}
415:
416:
417: Autocorrelations $C(t,\: L)$ Eq. \ref{eq:autoco} were determined
418: for the OSS model with the thresholds $4\leq E_{c}\leq256$. Avalanches
419: waves within avalanches are correlated (Fig. \ref{fig:Autocorr2})
420: for all thresholds because for the time $t<t_{max}$ autocorrelations
421: $C(t,\: L)$ are greater than $0$ ($C(t,\: L)>0$) and for the time
422: $t>t_{max}$ the autocorrelations approach the value $C(t,\: L)\doteq0.0$.
423:
424: Functions $\partial\sigma_{s}^{OSS}(q)/\partial q$ of the OSS model
425: for thresholds $4\leq E_{c}\leq256$ and function $\partial\sigma_{s}^{BTW}(q)/\partial q$
426: of the BTW model are shown in Fig. \ref{fig:dsigma2}. The results
427: show that $\partial\sigma_{s}^{OSS}(q)/\partial q>\partial\sigma_{s}^{BTW}(q)/\partial q$
428: for thresholds $4\leq E_{c}\leq256$ and for exponents $1.0\leq q\leq4.0$.
429: A difference between functions $\partial\sigma_{s}^{OSS}(q)/\partial q-\partial\sigma_{s}^{BTW}(q)/\partial q$
430: at given exponent $q$ is observed to be higher than an expected experimental
431: error \cite{Karmar}.
432:
433: %
434: \begin{figure}
435: \includegraphics[width=6.5cm]{fig9}
436:
437: \caption{\label{fig:dsigma2}(color online) The plots of $\partial\sigma_{s}(q)/\partial q$
438: of the BTW and OSS ($4\leq E_{c}\leq256$) models ($128\leq L\leq4096$). }
439:
440: \end{figure}
441:
442:
443: Probability density functions $P(s)$ of avalanche size $s$ were
444: determined for different lattice sizes $L=128,\:512$ and $4096$
445: (Fig. \ref{fig:pdf} (a)) and thresholds $E_{c}^{I}=4,\:8$ and $256$
446: (Fig. \ref{fig:pdf} (b)) to know the impact of the lattice size $L$
447: or thresholds $E_{c}^{I}$ on the probability density functions $P(s)$.
448: The probability density functions $P(s)$ of the OSS model show small
449: increases of their slopes for avalanches of size $s<10$.
450:
451: %
452: \begin{figure}
453: \includegraphics[width=7cm]{fig10a}
454:
455: \includegraphics[width=7cm]{fig10b}
456:
457: \caption{\label{fig:pdf}(color online) Probability distribution functions
458: $P(s)$ of avalanche size $s$ for (a) the constant threshold $E_{c}=4$
459: and increasing lattice size $L=128,\:512$ and $4096$ and for (b)
460: the constant lattice size $L=4096$ and increasing thresholds $E_{c}=4,\:8$
461: and $256$. }
462:
463: \end{figure}
464:
465:
466:
467:
468:
469: \section{Discussion\label{sec:Discussion}}
470:
471: The avalanche structures of the OSS model (Fig. \ref{fig:aval}) are
472: similar to the shell-like structures of the BTW model\cite{Ben,Teb}.
473: However, detailed analysis of these structures shows that the structures
474: do not have the same properties as the structures of the BTW model.
475: For example, we can see holes inside OSS structures (Fig. \ref{fig:aval}
476: (b)) which are not possible in the BTW model \cite{Karmar}. These
477: holes support our classification of the OSS model as a model with
478: unbalanced relaxation rules \cite{Karmar}.
479:
480: Autocorrelations $C(t,\: L)$ of the OSS model for thresholds $4\leq E_{c}^{I}\leq256$
481: (Figs. \ref{fig:Autocorrelation}(b) and \ref{fig:Autocorr2}) are
482: $C(t,\: L)>0$ for times $t_{max}$. This is a consequence of correlated
483: avalanche size waves within avalanche \cite{Manna,Stella}. Autocorrelations
484: of the BTW model are the same as in the paper\cite{Menech}. However,
485: I cannot confirm that these autocorrelations are approximated by the
486: power law approximation in Eq. \ref{eq:autoco_approx}. I have observed
487: that an exponential function $C(t,\: L)\sim\exp(-\alpha t)$ where
488: $\alpha$ is a decay rate, is a better approximation of the autocorrelations
489: $C(t,\: L)$ then the power law approximation Eq. \ref{eq:autoco_approx}.
490: This finding is supported by the fact that time-moment analysis (Fig.
491: \ref{fig:timemoments}) for the BTW and OSS models leads to the time
492: exponents $\tau^{BTW}\doteq0$ and $\tau^{OSS}\doteq0$. The reason
493: for this discrepancy with the previous results \cite{Menech} is
494: not clear and additional study is necessary. On the other hand, the
495: maximum number of waves scales as $t_{max}^{BTW}\sim L$ and this
496: result confirms the previous observation \cite{Menech}. Similarly,
497: for the OSS model $t_{max}^{OSS}\sim L$ (Fig. \ref{fig:timemoments}).
498:
499: The fluctuations $F(t,\: L)$ and corresponding Hurst exponents $H_{1}^{BTW}$
500: and $H_{2}^{BTW}$of the BTW model (Fig. \ref{fig:Hurst}) agree well
501: with the previous results \cite{Menech,Karmar}. Fluctuations $F(t,\: L)$
502: of the OSS model for thresholds $4\leq E_{c}^{I}\leq256$ have two
503: scaling regions (Fig. \ref{fig:Hurst}) and corresponding Hurst exponents
504: $H_{1}^{OSS}$ and $H_{2}^{OSS}$. An existence of two scaling regions
505: confirms the correlated avalanche size waves.
506:
507: The moments of avalanche size Fig. \ref{fig:dsigma} for the BTW and
508: M models agree well with previous results \cite{Menech,Karmar}.
509: The plots $\partial\sigma_{s}(q)/\partial q$ of the OSS model in
510: Figs. \ref{fig:dsigma} and \ref{fig:dsigma2}, for thresholds $4\leq E_{c}^{I}\leq256$
511: and exponents $q>1.0$, are not constant as in the case of the M model.
512: The increase of $\partial\sigma_{s}(q)/\partial q$ for $q>1.0$ when
513: the exponent $q$ increases is considered for a signature of multifractal
514: scaling \cite{Menech,Teb}. Based on the moment analysis (Figs. \ref{fig:dsigma}
515: and \ref{fig:dsigma2}) for the thresholds $4\leq E_{c}^{I}\leq256$
516: and previous conclusions \cite{Menech,Teb,Stella}, I claim that
517: avalanche size distributions show multifractal scaling.
518:
519: Probability density functions of avalanche size $P(s)$ of the OSS
520: model show a moderate increase of small avalanches $s<10$ for thresholds
521: $4\leq E_{c}^{I}\leq256$ (Fig. \ref{fig:pdf}). I assume that these
522: changes of probability density functions $P(s)$ do not influence
523: scaling of avalanche size moments (Fig. \ref{fig:Avalanche-size-moments}).
524:
525: Holes in some avalanches (Fig. \ref{fig:aval}(b)) are characteristic
526: for the models with unbalanced relaxation rules which exhibit uncorrelated
527: avalanche waves \cite{Karmar,Ben}. On the other hand, the existence
528: of holes in the OSS model is not sign of uncorrelated waves beacuse
529: avalanche size waves are correlated (see correlations $C^{OSS}(t,L)$
530: (Fig. \ref{fig:Autocorrelation} (b) and \ref{fig:Autocorr2}) and
531: fluctuations $F^{OSS}(t,L)$ (Fig. \ref{fig:Hurst})) exhibit two
532: scaling regions.
533:
534: Shell-like avalanche structures \cite{Ben,Karmar}, avalanche wave
535: correlations $C^{OSS}(t,L)$ \cite{Menech,Karmar}, avalanche wave
536: fluctuations $F^{OSS}(t,L)$ \cite{Menech} and avalanche size moments
537: $\sigma_{s}^{OSS}(q)$ \cite{Menech,Karmar} of the OSS model support
538: the conclusion that the OSS model shows multifractal scaling for thresholds
539: $4\leq E_{c}^{I}\leq256$. I can reproduce the plots $\partial\sigma_{s}^{BTW}(q)/\partial q$
540: and $\partial\sigma_{s}^{OSS}(q)/\partial q$ (Fig. \ref{fig:dsigma})
541: for thresholds $4\leq E_{c}^{I}\leq256$. Comparison of the functions
542: $\partial\sigma_{s}^{BTW}(q)/\partial q$ with the previous results
543: of the BTW model and undirected model \cite{Karmar} shows that these
544: functions collapse to a single function. I demonstrated that the OSS
545: model shows multifractal scaling but the functions $\partial\sigma_{s}^{OSS}(q)/\partial q$,
546: for thresholds $4\leq E_{c}^{I}\leq256$, are not identical with the
547: function $\partial\sigma_{s}^{BTW}(q)/\partial q$ of the BTW model
548: (Figs. \ref{fig:dsigma}, \ref{fig:dsigma2}). The differences between
549: functions $\partial\sigma_{s}^{BTW}(q)/\partial q$ and $\partial\sigma_{s}^{OSS}(q)/\partial q$
550: at given exponent $q$ are larger than the method error \cite{Karmar}.
551: Based on these evidences, I conclude that OSS and BTW models belong
552: to the multifractal universality class. However, the models do not
553: have identical properties despite the fact that in the OSS model only
554: one site in each avalanche undergoes stochastic relaxations as the
555: M site.
556:
557:
558: \section{Conclusion\label{sec:Conclusion}}
559:
560: The OSS model has been developed to study properties of the inhomogeneous
561: sand pile model \cite{Cer} at very low densities of M sites \cite{Cer_next}.
562: Based on traditional classification schemes \cite{Karmar}, one can
563: expect that the model will belong to the M universality class. However,
564: I have demonstrated that one stochastic M site in each avalanche is
565: not enough to change multifractal scaling of the model to the FSS
566: (Fig. \ref{fig:dsigma}). The OSS model is stochastic, \textit{non-Abelian}
567: with unbalanced relaxation rules \cite{Karmar}, the classification
568: schema implies that the model belongs to the M universality class.
569: However, the OSS models exhibits correlated avalanche waves and multifractal
570: scaling which is not allowed for the models in the M universality
571: class. The OSS model exhibits multifractal scaling and the classification
572: scheme implies that the model belongs to the BTW universality class,
573: however the model is stochastic, non-Abelian, and has an unbalanced
574: relaxation rule \cite{Karmar}, thus it cannot belong to the BTW
575: class. I think that it could be more convenient to consider multifractal
576: or FSS scaling as a main criterion for model classification to solve
577: this paradox. Then the models which show multifractal scaling (Fig.
578: \ref{fig:dsigma}, BTW and OSS models) despite the fact that they
579: are not identical (see Sec. \ref{sec:Results}) could belong to the
580: multifractal universality class. Models that show FSS scaling Eq.
581: \ref{eq:FSS} (M model) could belong to the FSS universality class.
582: I have analyzed another SOC model where the results support this classification
583: scheme \cite{Cer2009}.
584:
585: \
586:
587: Thanks to Alex Read for reading the manuscript and fort discussion.
588: Computer simulations were carried out with NorduGrid community resources
589: and in the KnowARC project. This work was supported by the Slovak
590: Research and Development Agency under contract No. RP EU-0006-06.
591:
592: \begin{thebibliography}{10}
593: \bibitem{BTW}P. Bak, C. Tang, and K. Wiesenfeld, Phys. Rev. Lett.
594: \textbf{59}, 381 (1997); Phys. Rev. A \textbf{38}, 364 (1988).
595:
596: \bibitem{Manna}S. S. Manna, J. Phys. A \textbf{24}, L363 (1991).
597:
598: \bibitem{Dhar} D. Dhar, Physica A \textbf{263}, 4 (1999); Physica
599: A \textbf{369}, 29 (2006).
600:
601: \bibitem{Dhar_abel}D. Dhar, Physica A \textbf{270}, 69 (1999).
602:
603: \bibitem{Pietro} L. Pietronero, A. Vespignani, and S. Zapperi, Phys.
604: Rev. Lett. \textbf{72}, 1690 (1994); A. Vespignani, S. Zapperi and
605: L. Pietronero, Phys. Rev. E \textbf{51}, 1711 (1995).
606:
607: \bibitem{Chessa}H. E. Stanley, A. Vespignani, and S. Zapperi, Phys.
608: Rev. E \textbf{59}, R12 (1999).
609:
610: \bibitem{Ben}A. Ben-Hur and O. Bihami, Phys. Rev. E \textbf{53},
611: R1317 (1996).
612:
613: \bibitem{Lub}S. Lübeck and K. Usadel, Phys. Rev. E \textbf{55}, 4095
614: (1997).
615:
616: \bibitem{Teb} C. Tebaldi, M. de Menech, and A. L. Stella, Phys. Rev.
617: Lett. \textbf{83}, 3952 (1999).
618:
619: \bibitem{Ivash}E. V. Ivashkevich, D. V. Ktitarev, and V. B. Priezzhev,
620: Physica A \textbf{209}, 347 (1994).
621:
622: \bibitem{Menech}M. De Menech and A. L. Stella, Phys. Rev. E \textbf{62},
623: R4528 (2000).
624:
625: \bibitem{Karmar}R. Karmakar, S. S. Manna, and A. L. Stella, Phys,
626: Rev. Lett. \textbf{94}, 088002 (2006).
627:
628: \bibitem{Cer}J. \v{C}ernák, Phys. Rev. E \textbf{73}, 066125 (2006).
629:
630: \bibitem{Biham}O. Biham, E. Milshtein, and O. Malcai, Phys. Rev.
631: E \textbf{63}, 061309 (2001).
632:
633: \bibitem{Dhar_1990}D. Dhar, Phys. Rev. Lett. \textbf{64}, 1613 (1990).
634:
635: \bibitem{Santra} S.B. Santra, S. R. Chanu, and D. Deb, Phys. Rev.
636: E \textbf{75}, 041122 (2007).
637:
638: \bibitem{Stella} A. L. Stella and M. De Menech, Physica A \textbf{295},
639: 101 (2001).
640:
641: \bibitem{Benoit}Benoit B. Mandelbrot, \textit{The fractal geometry
642: of nature} (W. E. Freeman and Company, New York, 1983).
643:
644: \bibitem{Cer2009} J. \v{C}ernák, arXiv:0908.0318v1 {[}cond-mat.stat-mech].
645:
646: \bibitem{Cer_next}J. \v{C}ern\'{a}k and J. Bobot, manuscript is
647: being prepared.
648: \end{thebibliography}
649:
650: \end{document}
651: