1: %\documentclass[aps,preprint]{revtex4}
2: \documentclass[aps,twocolumn]{revtex4}
3: \usepackage{epsfig}
4:
5: \begin{document}
6:
7: \title{Entropy production in the majority-vote model}
8:
9: \author{Leonardo Crochik and T\^ania Tom\'e}
10:
11: \affiliation{Instituto de F\'{\i}sica,
12: Universidade de S\~{a}o Paulo,
13: Caixa Postal 66318 \\
14: 05315-970 S\~{a}o Paulo, S\~{a}o Paulo, Brazil}
15:
16: \date{\today}
17:
18: \begin{abstract}
19:
20: We analyzed the entropy production in the majority-vote model by
21: using a mean-field approximation and Monte Carlo simulations.
22: The dynamical rules of the model do not obey detailed balance
23: so that entropy is continuously being produced.
24: This nonequilibrium stochastic model is known to have a critical
25: behavior belonging to the universality class of the equilibrium Ising model.
26: We show that the entropy production also
27: exhibits a singularity at the critical point
28: similar to the one occurring in the entropy, or the energy, of the
29: equilibrium Ising model.
30:
31: PACS numbers: 05.70.Ln, 05.50.+q, 02.50.Ga
32:
33: \end{abstract}
34:
35: \maketitle
36:
37: \section{Introduction}
38:
39: Irreversible systems in the stationary state
40: are in a process of continuous production
41: of entropy. In systems, like the one studied here,
42: governed by a master equation, that is, defined by a
43: continuous time Markov process, the irreversibility
44: is characterized by the lack of detailed balance.
45: When a system governed by a Markov process is reversible,
46: that is, when the dynamic rules are such that detailed
47: balance is obeyed, the production of entropy vanishes at
48: the equilibrium state.
49: This is indeed the case of the Glauber model
50: and any other dynamics used to simulate the equilibrium Ising model.
51: The production of entropy is then a signature
52: of irreversibility.
53:
54: The rate of change of the entropy $S$ of a system
55: can be properly decomposed into two contributions \cite{nico77}
56: \begin{equation}
57: \frac{dS}{dt} = \Pi - \Phi,
58: \label{1}
59: \end{equation}
60: where $\Pi$ is the entropy production due to irreversible
61: processes occurring inside the system and $\Phi$ is the entropy flux
62: from the system to the environment.
63: The quantity $\Pi$ is positive definite whereas $\Phi$
64: can have either sign. In the stationary state
65: the entropy $S$ of the system remains constant so
66: that $\Phi=\Pi$. Notice that the quantity $\Phi$
67: is defined here as the flux from inside to outside the system.
68: So that, it will be positive in the stationary state.
69:
70: In this work we study the steady state production of
71: entropy in a nonequilibrium lattice model, namely,
72: the majority-vote model \cite{ligg85,gray85,oliv92}.
73: This is a polling
74: model in which individuals in a community take the opinion
75: of their neighbors with a certain probability $p$
76: and the opposite opinion with probability $q=1-p$.
77: In two or more dimensions, this model displays a continuous
78: phase transition described by the same critical exponents
79: as the equilibrium Ising model \cite{oliv92}.
80: This is in agreement with the conjecture by
81: Grinstein et al. \cite{grin85}
82: according to which nonequilibrium stochastic systems with
83: up-down symmetry fall in the universality class of the
84: equilibrium Ising model.
85:
86: The flux of entropy is determined by means of an expression
87: which is the average of a function of the rates of
88: transition from one state to another and its reverse
89: \cite{maes99,lebo99,maes00,maes03,leco05}.
90: As the entropy production equals the entropy flux in
91: the stationary state, the former can be determined
92: from the expression for the latter.
93: We use mean-field and Monte Carlo simulations to calculate the
94: entropy flux.
95: In the stationary regime, and at the critical point, the
96: entropy flux, or equivalently,the entropy production,
97: displays a singularity which we assume
98: to be the same singularity occurring in the entropy of the
99: equilibrium Ising model.
100:
101: \section{Entropy production}
102:
103: Let us consider a system described by a continuous Markov process
104: with stochastic variables defined over the sites of a regular lattice.
105: A configuration of the system is denoted by
106: $\sigma=(\sigma_1,\sigma_2,\ldots,\sigma_N)$ where $N$ is the number
107: of sites of the lattice and $\sigma_i=\pm1$ is the spin variable
108: associated to site $i$.
109: We will be concerned only with one spin flip dynamics,
110: defined by a transition rate $w_i(\sigma)$ in which
111: the spin variable $\sigma_i$ changes its sign.
112: The time evolution of the probability $P(\sigma,t)$
113: is governed by the master equation,
114: \begin{equation}
115: \frac{d}{dt} P(\sigma,t)=\sum_i
116: \{ w_i(\sigma^i) P(\sigma^i,t) - w_i(\sigma) P(\sigma,t) \},
117: \label{2}
118: \end{equation}
119: where $\sigma^i=(\sigma_1,\sigma_2,\ldots,-\sigma_i,\ldots,\sigma_N)$.
120:
121: The Gibbs entropy $S(t)$ of the system at time $t$
122: is defined by
123: \begin{equation}
124: S(t) = - \sum_\sigma P(\sigma,t) \ln P(\sigma,t).
125: \label{3}
126: \end{equation}
127: Using the master equation (\ref{2}), its
128: time derivative can be written as
129: \[
130: \frac{d}{dt}S(t) =\frac12 \sum_{\sigma} \sum_i
131: \ln\frac{P(\sigma^i,t)}{P(\sigma,t)} \times
132: \]
133: \begin{equation}
134: \times \{ w_i(\sigma^i) P(\sigma^i,t) - w_i(\sigma) P(\sigma,t) \}.
135: \end{equation}
136: In agreement with Eq. (\ref{1}),
137: the righ-hand side of this expression should be decomposed into two terms,
138: the entropy production $\Pi$ and the entropy flux $\Phi$.
139: These two quantities have the following expressions
140: \cite{maes99,lebo99,maes00,maes03}
141: \[
142: \Pi = \frac12\sum_{\sigma} \sum_i
143: \ln\frac{w_i(\sigma^i)P(\sigma^i,t)}{w_i(\sigma)P(\sigma,t)} \times
144: \]
145: \begin{equation}
146: \times \{ w_i(\sigma^i) P(\sigma^i,t) - w_i(\sigma) P(\sigma,t) \},
147: \label{5}
148: \end{equation}
149: and
150: \begin{equation}
151: \Phi = \sum_{\sigma} \sum_i w_i(\sigma) P(\sigma,t)
152: \ln\frac{w_i(\sigma)}{w_i(\sigma^i)}.
153: \label{6}
154: \end{equation}
155: The right-hand side of Eq. (\ref{5}) is always positive as can
156: be easily proved,
157: and the right-hand side of Eq. (\ref{6}) can be written as the average
158: over the stationary probability distribution, that is,
159: \begin{equation}
160: \Phi = \sum_i \langle w_i(\sigma)
161: \ln\frac{w_i(\sigma)}{w_i(\sigma^i)} \rangle.
162: \label{7}
163: \end{equation}
164:
165: This is a particularly useful equation
166: because it can be employed to estimate $\Phi$
167: from a Monte Carlo simulation. As it
168: is well known only quantities that can be
169: written as averages can be determined numerically
170: in a Monte Carlo simulation. In this sense
171: it is not possible to determine $S$ given by Eq. (\ref{3})
172: nor $\Pi$, given by Eq. (\ref{5}),
173: but it is actually possible to determine $\Phi$
174: from Eq. (\ref{6}).
175: We remark finally that in the steady state
176: $\Pi=\Phi$ so that it is possible to determine
177: the entropy production in this regime
178: by a Monte Carlo simulation.
179:
180: \section{Majority-vote model}
181:
182: The majority-vote model is a one-spin flip stochastic
183: dynamics defined by the following transition rate
184: \begin{equation}
185: w_i(\sigma)
186: = \frac{1}{2} \{ 1-\gamma\sigma_i {\cal F}(\sum_\delta \sigma_{i+\delta}) \},
187: \label{8}
188: \end{equation}
189: where ${\cal F}(x)$ is a function that equals $-1$, $0$ or $+1$
190: according to whether $x<0$, $x=0$ or $x>0$, and the summation
191: is over the nearest neighbor sites of site $i$.
192: Notice that the transition rate $w_i(\sigma)$ has the
193: up-down symmetry, that is, it is invariant
194: under the sign change of all spin variables $\sigma_i$.
195: At each time interval, a site $i$ is chosen at random.
196: If the majority of the neighbors are in state $+1(-1)$
197: then the site takes the value $+1(-1)$ with probability
198: $p$ and the opposite sign with probability $q=1-p$
199: where $p=(1+\gamma)/2$.
200: We will restrict ourselves to the case $0\leq q \leq 1/2$ so that
201: $1\geq\gamma\geq 0$.
202: The model can also be interpreted as an Ising system in contact with
203: two heat reservoir at temperatures $0^+$ and $0^-$.
204: Putting it in a different way, one reservoir always provides energy
205: and the other takes it away. Spin systems in contact with two heat
206: baths at different temperatures
207: \cite{masi85,dick87,garr87,marq89,tome89,tome91,racz94,figu00,leco05}
208: are perhaps the simplest models with nonequilibrium steady states
209: exhibiting dynamic phase transitions.
210:
211: In the stationary regime, the present
212: model displays a continuous phase transition from an ordered
213: (ferromagnetic) state
214: to a disordered (paramagnetic) state .
215: On a square lattice it is found by numerical simulation that
216: the critical point occurs at $q_c=0.075(1)$ \cite{oliv92}.
217: The ordered state occurs when $0\leq q <q_c$ and
218: the disordered state when $q_c<q\leq 1/2$.
219: For $0<q<1/2$, this model does not obey detailed balance
220: and we expect a strictly positive entropy production.
221: When $q=1/2$ the system is completely disordered and
222: corresponds to a reversible system so that the
223: entropy production vanishes in this case.
224: The critical behavior \cite{oliv92} puts this model in the same
225: universality class as the equilibrium Ising model.
226: This result comes from the conjecture by Grinstein et al.
227: \cite{grin85} which states that models with stochastic evolution
228: rules with up-down symmetry belongs to the Ising universality class.
229:
230: From the transition rate $w_i(\sigma)$ given by Eq. (\ref{8})
231: it is straightforward to show that
232: \begin{equation}
233: B_i(\sigma) =
234: \ln\frac{w_i(\sigma)}{w_i(\sigma^i)} =
235: \left(\ln\frac{q}{p} \right)
236: \sigma_i {\cal F}(\sum_\delta \sigma_{i+\delta}).
237: \label{9}
238: \end{equation}
239: Therefore, the entropy flux per site $\phi=\Phi/N$
240: for the majority-vote model can be
241: determined as the average
242: \begin{equation}
243: \phi = \langle B_i(\sigma) w_i(\sigma) \rangle.
244: \label{10}
245: \end{equation}
246: Notice that for a square lattice the function ${\cal F}(x)$
247: reads
248: \begin{equation}
249: {\cal F}(\sigma_1+\sigma_2+\sigma_3+\sigma_4)=
250: \frac18(\sigma_1+\sigma_2+\sigma_3+\sigma_4)
251: (3-\sigma_1\sigma_2\sigma_3\sigma_4).
252: \label{10a}
253: \end{equation}
254:
255: \section{Mean-field results}
256:
257: %------------------ Figure 1 --------------------
258: \begin{figure}[t]
259: \epsfig{file=fluxmf.eps,height=6cm}
260: \caption{Entropy flux $\phi$ for the majority-vote model
261: in the simple mean-field approximation as a function
262: of the external parameter $q$.
263: The dotted line indicates the position of the
264: critical point occurring at $q_c=1/6$.}
265: \end{figure}
266: %------------------------------------------------
267:
268: From the master equation we get the following equations
269: for the time evolution of the magnetization $\langle\sigma_i\rangle$
270: \begin{equation}
271: \frac{d}{dt} \langle\sigma_i\rangle =
272: -2\langle\sigma_i w_i(\sigma)\rangle.
273: \end{equation}
274:
275: In the first order dynamic mean-field approximation,
276: or simple mean-field approximation,
277: the correlations are neglected and we need only this equation.
278: We apply the approximation to the case of a regular
279: lattice of coordination four.
280: In the stationary state, the magnetization
281: $m=\langle\sigma_i\rangle$ is given by the equation
282: \begin{equation}
283: m= \frac{\gamma}{2} m (3-m^2).
284: \end{equation}
285: Using this approximation, we derive the following expression for
286: the entropy flux
287: \begin{equation}
288: \phi = \left(\ln\frac{q}{p}\right)
289: \{ \frac14(3m^2-m^4)-\frac{\gamma}{16}(5+6m^2-3m^4) \}.
290: \end{equation}
291:
292: The paramagnetic solution, $m=0$,
293: gives the following expression for the
294: entropy flux in the paramagnetic phase
295: \begin{equation}
296: \phi = \frac{5}{16}(1-2q)\ln\frac{1-q}{q}.
297: \end{equation}
298: The ferromagnetic solution is given by the expression
299: \begin{equation}
300: m = \sqrt{\frac{1-6q}{1-2q}},
301: \end{equation}
302: which is valid for $q<q_c=1/6$.
303: From this result it follows that the entropy flux
304: in the ferromagnetic state is
305: \begin{equation}
306: \phi = \frac{q(1-q)}{1-2q}\ln\frac{1-q}{q}.
307: \end{equation}
308: The stationary entropy flux, or equivalently the entropy production,
309: is a continuous function of the
310: parameter $q$ as shown in Fig. 1. At the critical
311: point it presents a singularity represented
312: in this mean-field approximation by a discontinuity in the first
313: derivative.
314:
315: \section{Numerical simulations}
316:
317: We have simulated the majority-vote model on a square
318: lattice with periodic boundary conditions for
319: different values of the size $N=L\times L$ of the system.
320: The simulation was performed as follows. At each time
321: step a site is chosen at random. It takes the
322: value of the majority sign of its neighbors with probability
323: $p=1-q$ and the opposite sign with probability $q$
324: in accordance with the prescription given by Eq. (\ref{8}).
325: After discarding the first Monte Carlo steps
326: the stationary properties are calculated.
327: We used from $10^6$ to $10^7$ Monte Carlo steps
328: to calculate the averages such as the flux $\phi$
329: given by Eq. (\ref{10}). The magnetization and other
330: quantities such as the susceptibility have
331: already been determined by Monte Carlo simulations \cite{oliv92}
332: and will not concern us here. It is found that
333: a continuous phase transition takes place at
334: $q_c=0.075(1)$ and that the critical exponents
335: are the same as those of the two-dimensional Ising model \cite{oliv92}.
336:
337: %------------------ Figure 2 --------------------
338: \begin{figure}[t]
339: \epsfig{file=fluxmc.eps,height=6cm}
340: \caption{Stationary entropy flux $\phi$ for the majority
341: vote model by Monte Carlo simulation as a function
342: of the external parameter $q$ for various values
343: of the system size $L$.
344: The dotted line indicates the position of the
345: critical point occurring at $q_c=0.075$.}
346: \end{figure}
347: %------------------------------------------------
348:
349: %------------------ Figure 3 --------------------
350: \begin{figure}[t]
351: \epsfig{file=fluc.eps,height=6cm}
352: \caption{Enlargement of Fig. 2 around the critical point.}
353: \end{figure}
354: %------------------------------------------------
355:
356: In Fig. 2 we show the numerical results for
357: the entropy flux $\phi$ for several values
358: of the size of the system $L$.
359: The entropy flux is finite and continuous.
360: It has a maximum and vanishes when $q\to 1/2$ and when
361: $q\to 0$ as expected since in these two limits
362: the system reaches an equilibrium stationary state.
363: When $q\to 1/2$ we found numerically
364: that the flux vanishes according to
365: \begin{equation}
366: \phi = b (\frac12-q)^2,
367: \end{equation}
368: with $b=0.190(3)$ and when $q\to 0$, it vanishes
369: according to
370: \begin{equation}
371: \phi = a q^2 \ln\frac{1-q}{q},
372: \end{equation}
373: with $a=1.83(5)$.
374:
375: We remark that the critical point
376: does not correspond to the maximum of $\phi$.
377: Actually, it corresponds to the point of inflexion
378: occurring just before the maximum as can be seen in Fig. 3.
379: At the critical point the flux is finite but
380: has a singularity which we assume to be
381: the same type as that of the energy or the entropy of
382: the equilibrium Ising model, namely,
383: of the form
384: \begin{equation}
385: \phi = \phi_c + A_{\pm}|q-q_c|^{1-\alpha},
386: \label{19}
387: \end{equation}
388: where $\phi_c$ is the value of the entropy flux
389: at the critical point.
390: The amplitudes $A_+$ and $A_-$ correspond
391: to the regimes below ($q<q_c$) and above ($q>q_c$)
392: the critical point.
393:
394: The energy of the equilibrium Ising model
395: is related to the short range correlations of the even type.
396: From Eqs. (\ref{9}) and (\ref{10}) and the result given by Eq. (\ref{10a})
397: we see that the entropy flux is also related
398: to the short range correlations of the even type:
399: $\langle\sigma_i\sigma_j\rangle$
400: and $\langle\sigma_i\sigma_j\sigma_k\sigma_\ell\rangle$
401: where $i,j,k$ and $\ell$ are nearest and next-nearest neighbor
402: sites. We expect, therefore, that the critical behavior of the entropy
403: flux of nonequilibrium models with up-down symmetry
404: and the critical behavior of energy of the equilibrium
405: Ising model are described by the same critical exponent.
406:
407: %------------------ Figure 4 --------------------
408: \begin{figure}[t]
409: \epsfig{file=flucder.eps,height=5.8cm}
410: \caption{Derivative $d\phi/dq$ of the entropy flux
411: $\phi$ with respect to the external parameter $q$
412: for several values of the system size $L$.
413: The dotted line indicates the position of the
414: critical point occurring at $q_c=0.075$.}
415: \end{figure}
416: %------------------------------------------------
417:
418: %------------------ Figure 5 --------------------
419: \begin{figure}[t]
420: \epsfig{file=scalmax.eps,height=5.8cm}
421: \caption{Maximum of the derivative $d\phi/dq$ of the entropy flux
422: $\phi$ with respect to the external parameter $q$
423: as a function of the logarithm of the system size $L$.
424: }
425: \end{figure}
426: %------------------------------------------------
427:
428: To determine the critical behavior it is convenient
429: to study the derivative of the entropy flux
430: with respect to the external parameter $q$
431: which, as follows from Eq. (\ref{19}), behaves as
432: \begin{equation}
433: \frac{d\phi}{dq} \sim |q-q_c|^{-\alpha}.
434: \end{equation}
435: The exponent $\alpha$ is the same exponent
436: related to the specific heat of the equilibrium
437: Ising model. Since, for the square lattice,
438: the singularity is of the logarithm ($\alpha=0$) type
439: then we assume that for the present two-dimensional
440: majority-vote model
441: \begin{equation}
442: \frac{d\phi}{dq} \sim |\ln|q-q_c||.
443: \label{21}
444: \end{equation}
445:
446: To test the assumption given by Eq. (\ref{21})
447: we have numerically determined $d\phi/dq$ for several lattice
448: sizes $L$ as shown in Fig. 4.
449: Using a finite-size scaling theory \cite{ferd69}, then
450: this quantity as a function of $L$ should diverges as
451: $\ln L$
452: %\begin{equation}
453: %\left(\frac{d\phi}{dq}\right)_c \sim \ln L
454: %\end{equation}
455: at the critical point and a similar behavior
456: at the maximum, that is,
457: \begin{equation}
458: \left(\frac{d\phi}{dq}\right)_{\rm max} \sim \ln L.
459: \end{equation}
460: From Fig. 5 we see that this behavior is indeed
461: followed when $L\ge20$.
462:
463: \section{conclusion}
464:
465: We have determined by mean-field approximation and by
466: Monte Carlo simulations the stationary entropy production
467: of a nonequilibrium
468: model with up-down symmetry, namely, the majority-vote model.
469: The calculation of this quantity by means of Monte Carlo simulations
470: was possible because it equals the entropy flux, in the
471: stationary state, and so this quantity can
472: be written as an average over a
473: stationary probability distribution.
474: The mean-field analysis as well as the Monte Carlo simulations
475: show that the stationary entropy production
476: is positive for nonequilibrium situation, it
477: vanishes when the system attains an equilibrium
478: stationary state, and
479: exhibits a singularity at the critical point.
480: The mean-field results gives a
481: singularity represented by a discontinuous first derivative
482: as usually happens in mean-field calculations of the energy
483: as a function of temperature in equilibrium spin models.
484: The Monte Carlo data, analyzed by a finite-size scaling theory,
485: have shown that the stationary entropy production has the same singular
486: behavior at the critical point as the energy of the
487: equilibrium Ising model. In the present case, namely,
488: the model defined on a square lattice, the singularity
489: of the derivative of the entropy production
490: is characterized by a logarithmic divergence.
491:
492: \section*{Acknowledgment}
493:
494: T. T. and L. C. acknowledge financial support from CNPq
495: and FAPESP, respectively.
496:
497: \begin{thebibliography}{99}
498:
499: \bibitem{nico77} G. Nicolis and I. Prigogine, {\it Self-Organization in
500: Nonequilibrium Systems} (Wiley, New York, 1977).
501:
502: \bibitem{ligg85} T. M. Ligget, {\it Interacting Particle Systems}
503: (Springer-Verlag, New York, 1985).
504:
505: \bibitem{gray85} L. Gray, in {\it Particle Systems, Random Media and
506: Large Deviations} edited by R. Durret (American Mathematical
507: Society, Providence, Rhode Island, 1985), p. 149.
508:
509: \bibitem{oliv92} M. J. de Oliveira, J. Stat. Phys. {\bf 66}, 273 (1992).
510:
511: \bibitem{grin85} G. Grinstein, C. Jayaprakash, and Yu He,
512: Phys. Rev. Lett. {\bf 55}, 2527 (1985).
513:
514: %\bibitem{gall95} G. Gallavotti and E. G. D. Cohen,
515: %Phys. Rev. Lett. {\bf 74}, 2694 (1995).
516: %G. Gallavotti and E. G. D. Cohen,
517: %J. Stat. Phys. {\bf 80}, 931 (1995).
518:
519: %\bibitem{kurc98} J. Kurchan, J. Phys. A {\bf 31}, 3719 (1998).
520:
521: \bibitem{maes99} C. Maes, J. Stat. Phys. {\bf 95}, 367 (1999).
522:
523: \bibitem{lebo99} J. L. Lebowitz and H. Spohn,
524: J. Stat. Phys. {\bf 95}, 333 (1999).
525:
526: \bibitem{maes00} C. Maes, F. Redig, and A. Van Moffaert,
527: J. Math. Phys. {\bf 41}, 1528 (2000).
528:
529: \bibitem{maes03} C. Maes and K. Neto\v{c}n\'y
530: J. Stat. Phys. {\bf 11}, 269 (2003).
531:
532: \bibitem{masi85} A. DeMasi, P. A. Ferrari, and J. L. Lebowitz,
533: Phys. Rev. Lett. {\bf 55}, 1947 (1985).
534:
535: \bibitem{dick87} R. Dickman, Phys. Lett. A {\bf 122}, 463 (1987).
536:
537: \bibitem{garr87} P. L. Garrido, A. Labarta, and J. Marro,
538: J. Stat. Phys. {\bf 49}, 551 (1987).
539:
540: \bibitem{marq89} M. C. Marques, J. Phys. A {\bf 22}, 4493 (1989).
541:
542: \bibitem{tome89} T. Tom\'e and M. J. de Oliveira,
543: Phys. Rev. A {\bf 40}, 6643 (1989).
544:
545: \bibitem{tome91} T. Tom\'e, M. J. de Oliveira, and M. A. Santos,
546: J. Phys. A {\bf 24}, 3677 (1991).
547:
548: \bibitem{racz94} Z. R\'acz and R. K. P. Zia,
549: Phys. Rev. E {\bf 49}, 139 (1994).
550:
551: \bibitem{figu00} W. Figueiredo and B. C. S. Grandi,
552: Braz. J. Phys. {\bf 30}, 58 (2000).
553:
554: \bibitem{leco05} V. Lecomte, Z. R\'acz, and F. van Wijland,
555: J. Stat. Mech. (2005) P02008.
556:
557: \bibitem{ferd69} A. E. Ferdinand and M. E. Fisher,
558: Phys. Rev. {\bf 185}, 832 (1969).
559:
560: \end{thebibliography}
561:
562:
563: \end{document}
564: