0810.1436/z2.tex
1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2: \documentclass[aps,prl,twocolumn,showpacs]{revtex4}
3: \usepackage{epsfig}
4: 
5: \usepackage{amsmath,amssymb}
6: \newcommand{\ignore}[1]{}
7: 
8: 
9: \begin{document}
10: 
11: 
12: \title{Systems with two symmetric absorbing states: relating the 
13: microscopic dynamics with the macroscopic behavior}
14: 
15: \author{Federico Vazquez}
16: \email[E-mail: ]{federico@ifisc.uib-csic.es}
17: \author{Crist\'obal L\'opez}
18: %\email[E-mail: ]{clopez@ifisc.uib.es}
19: 
20: \affiliation{IFISC, Instituto de F\'{\i}sica Interdisciplinar y Sistemas
21: Complejos (CSIC-UIB), E-07122 Palma de Mallorca, Spain}
22: \homepage{http://www.ifisc.uib.es}
23: 
24: \date{\today}
25: 
26: 
27: \begin{abstract} We propose a general approach to study spin models with two
28: symmetric absorbing states.  Starting from the microscopic dynamics on a
29: square lattice, we derive a Langevin equation for  the time evolution of the
30: magnetization field, that successfully explains coarsening properties of a
31: wide range of  nonlinear voter models and systems with intermediate states.
32: We find that the  macroscopic behavior only depends on the first derivatives
33: of the spin-flip probabilities.  Moreover, an  analysis of the  mean-field
34: term reveals the three types of transitions commonly observed in these systems
35: -generalized voter, Ising and directed percolation-.  Monte Carlo simulations
36: of the spin dynamics qualitatively agree with theoretical predictions.
37: \end{abstract}
38: 
39: \pacs{02.50.-r, 05.70.Fh, 64.60.Ht, 87.23.-n}
40: 
41: \maketitle
42: 
43: \emph{Introduction.}  The non-equilibrium dynamics of interacting particle
44: systems with absorbing states is a central issue in modern Statistical
45: Mechanics \cite{Hinrichsen00,Odor03}.  In the last years much interest has
46: been given to the special case of models with two symmetric ($Z_2$-symmetry)
47: absorbing states~\cite{Hammal05,Droz03,Dornic01}, due to the relevance in many
48: different contexts of a subclass of them, the so-called voter models.  They
49: have been widely used in diverse disciplines to study different dynamics, such
50: as species competition~\cite{Clifford73}, allele frequency in
51: genetics~\cite{Baxter07}, kinetics of heterogeneous
52: catalysis~\cite{Krapivsky92-Frachebourg96}, and more recently, opinion
53: formation~\cite{Castellano07} and language  spreading~\cite{Abrams03}.  In
54: these models, the state of a particle in a lattice site evolves according  to
55: the density of states in its near neighborhood. Interesting dynamical
56: behaviors arise depending on the specific updating rules, the number of states
57: and the functional form of the transition probabilities between
58: configurations.  For instance, it has been found that the addition of memory
59: or inertia in the spin  dynamics \cite{Dall'Asta07,Stark08}, the introduction
60: of intermediate states  \cite{Castello06,Baronchelli06,Dall'Asta08}, or the
61: use of non-linear transitions \cite{Schweizer}, result in a drastic  change of
62: the coarsening properties and final outcome of the system.
63: 
64: Despite that the dynamical rules of the models are  very
65: different in nature, many of them seem to share the same macroscopic
66: behavior, such as coarsening and criticality.  However, the minimal conditions 
67: that a microscopic dynamics must hold in order to observe a particular 
68: behavior have not been clearly identified yet.  In other words, given a spin 
69: model defined by its flipping transition probability and interaction range, 
70: can we anticipate how the system will evolve over time ? 
71: 
72: In this article, we try to answer this question by developing an approach that
73: connects the microscopic dynamics with the macroscopic space-time evolution of
74: the system in square lattices.  We derive a Langevin equation for the
75: magnetization field, and find that, at the macroscopic level, the properties
76: are only determined by the first three derivatives of the transition 
77: probabilities.
78: 
79: Our Langevin equation coincides with that postulated by Hammal et. al.
80: \cite{Hammal05} by symmetry arguments, but now the coefficients of the
81: different terms have a clear explanation in terms of the  transition
82: probabilities.  The analysis of this equation helps to understand some of the
83: open questions about phase ordering in these systems, that is, whether the
84: coarsening is driven by curvature like in the Ising model \cite{Bray02}, or it
85: is without surface tension like in the original voter model (VM) 
86: \cite{Dornic01}.
87: This approach also explains, from a different perspective than
88: in~\cite{Dall'Asta08}, why adding intermediate states to the VM leads
89: to an effective surface tension.  Moreover, numerical simulations of the
90: spin dynamics with different interaction ranges confirm the three
91: possible classes of phase transitions unveiled in~\cite{Hammal05}, and
92: clarifies an apparent controversy found between previous works
93: \cite{Dornic01,Droz03}.
94: 
95: \emph{The model and the Langevin equation.}  Each 
96: site ${\bf r}= (r_1,..,r_d)$
97: of a $d$-dimensional square lattice is occupied by one particle with a spin
98: that can assume either value  $1$ (up) or $-1$ (down).  The dynamics consists
99: of choosing, at each time step, a site ${\bf r}$ at random and flipping the
100: spin $S_{\bf r}$ at this site with a probability  that is a function, $f(-
101: S_{\bf r} \psi_{\bf r})$, of the product between $S_{\bf r}$ and the
102: particle's local magnetization  $\psi_{{\bf r}} \equiv \frac{1}{z} \sum_{{\bf
103: r'/r}}S_{{\bf r'}}$, where the sum is over the $z$ nearest-neighboring sites
104: ${\bf r'}$ of ${\bf r}$.  $z$ is an arbitrary integer number that defines the
105: interaction range (e.g. $z=4$ for first-nearest neighbor interactions). In
106: order to ensure that the fully ordered configurations $S_{\bf r} = 1$ or
107: $S_{\bf r}=-1$ for all {\bf r} are absorbing, the \emph{flipping probability}
108: $f$ must vanish when the spin is aligned with all its neighboring  spins, i.e,
109: $f(-1)=0$.
110: 
111: We want to derive a Langevin equation for the field  $\phi_{\bf r}(t)$, that 
112: is a continuous representation of the spin at site  ${\bf r}$, at
113: time $t$.   For this we follow a standard approach (see 
114: \cite{Dall'Asta08}), and consider an  ensemble of  $\Omega$ copies of the
115: system, each copy representing a particular spin configuration.  This is
116: equivalent to assume $\Omega$ spin particles at each site of the lattice (our
117: microscopic model corresponds exactly to $\Omega=1$, but this substitution can
118: be made at the end of the calculation).  In this approach
119: $\phi_{\bf r}(t)$ is replaced by the average spin value  $\phi_{\bf r}(t) \to
120: \frac{1}{\Omega} \sum_{j=1}^\Omega S_{\bf  r}^j$, and $\psi_{\bf r}$ by the
121: average local field  $\psi_{\bf r} \to \frac{1}{z} \sum_{{\bf r'/r}}\phi_{\bf
122: r'}(t)$, at site  ${\bf r}$ at time $t$.  In a time step, a site ${\bf x}$ and
123: one particle  from that site with spin $S_{\bf x}$ are randomly chosen.   Then
124: the spin attempts to flip with probability $f(-S_{\bf x} \psi_{\bf x})$.  If
125: the flipping occurs, then the field on the  entire lattice, represented as
126: $\{\phi\}$, changes only at site  ${\bf x}$ by $-2 S_{\bf x}/\Omega$.  Thus
127: the rising and lowering  transition  rates are $W\left(\{\phi\} \to~\{\phi\}
128: \pm \frac{2}{\Omega} \delta_{{\bf x},{\bf r}}\right)= W^{\pm}(\phi,{\bf x},t)
129: =  \frac{1}{2} \left( 1\mp  \phi_{\bf x} \right) f(\pm \psi_{\bf x})$.   We
130: can write a master equation for the time evolution of the probability
131: distribution $\mathcal P(\{\phi\},t)$, which after an expansion to second
132: order in $1/\Omega$ leads to the following  Fokker-Planck equation
133: \begin{eqnarray}
134: &&\frac{\partial}{\partial t} \mathcal P(\{\phi\},t)= \\ 
135: && \sum_{\bf r} -\frac{1}{\Omega} \frac{\partial}{\partial \phi } 
136: \Bigl\{ 2 \left[ W^+(\phi,{\bf r},t)- W^-(\phi,{\bf r},t) \right] 
137: \mathcal P(\{\phi\},t) \Bigr\}
138: \nonumber \\
139: && +\frac{1}{\Omega^2} \frac{\partial^2}{\partial \phi^2} \Bigl\{2 
140: \left[W^+(\phi,{\bf r},t)+W^-(\phi,{\bf r},t) \right] \mathcal P(\{\phi\},t) 
141: \Bigr\}. \nonumber
142: \end{eqnarray}
143: From the above equation, using the expressions for  $W^{\pm}(\phi,{\bf r},t)$
144: and rescaling time with $\Omega$, we arrive to the Langevin equation
145: \begin{equation}
146: \frac{\partial \phi_{\bf r}(t)}{\partial t} = \left[1-\phi_{\bf r}(t)\right] 
147: f(\psi_{\bf r})-
148: \left[1+\phi_{\bf r}(t)\right] f(-\psi_{\bf r}) + \eta_{\bf r}(t),
149: \label{dphi-dt}
150: \end{equation}
151: where $\eta_{\bf r}(t)$ is a Gaussian white noise with correlations
152: \begin{eqnarray}
153: && \langle \eta_{\bf r}(t) \eta_{\bf r'}(t') \rangle = 
154:  \Bigl\{ \left[1-\phi_{\bf r}(t)\right]  
155: f(\psi_{\bf r}) \nonumber \\ 
156: &&+ \left[1+\phi_{\bf r}(t)\right] f(-\psi_{\bf r}) \Bigr\} 
157: \delta_{\bf r,r'} \delta(t-t')/{\Omega^{1/2}}. 
158: \end{eqnarray}
159: Note that up to now our derivation is completely general, and the fact that
160: the system has two absorbing states is only present in the  condition for the
161: flipping probability.  Now we look for an approximation to  Eq.(\ref{dphi-dt})
162: which, however, captures the behavior of a wide range of absorbing $Z_2$
163: models.  As  explained in \cite{Hammal05}, this is obtained with a $\phi^6$
164: model, i.e. when the right hand side of Eq.~(\ref{dphi-dt}) is proportional to
165: $\phi^5$. Thus, we expand $f$ around $\psi_{\bf r}=0$ up to forth order in
166: $\psi_{\bf r}$, but also making $f$ vanish at $\psi_{\bf r} = -1$ (the
167: condition for existence of absorbing states at $\pm 1$):
168: \begin{equation}
169: \label{f}
170: f(\psi_{\bf r}) = \frac{1}{2} (1+\psi_{\bf r}) \left(c+a \psi_{\bf r}+
171: d \psi_{\bf r}^2- b \psi_{\bf r}^3 \right),
172: \end{equation}
173: where the real coefficients $a, b, c$ and $d$ are, for convenience, defined as
174: (primes denoting derivatives)
175: \begin{eqnarray}
176: \label{coef}
177: c \equiv 2 f(0), ~ a \equiv 2 f'(0)- c,~ \nonumber \\ 
178: d \equiv f''(0)-a,~ b \equiv -\frac{f'''(0)}{3}+d .  
179: \end{eqnarray}
180: To obtain a closed equation for $\phi_{\bf r}(t)$, we replace  
181: expression~(\ref{f}) for $f(\psi_{\bf k})$ into
182: Eq.~(\ref{dphi-dt}) and make the substitution $\psi_{\bf r} = \phi_{\bf r} +
183: \Delta \phi_{\bf r}$, where we define the Laplacian operator 
184: $\Delta \phi_{\bf r} \equiv  \frac{1}{z} \sum_{\bf r'/r} \left( \phi_{\bf
185:   r'}-\phi_{\bf r} \right) =\psi_{\bf r}-\phi_{\bf r}$.  We then expand 
186: to first order in $\Delta \phi_{\bf r}$
187: (assuming that $\phi_{\bf r}$ is a smooth  function of ${\bf r}$ in the long  
188: time
189: limit, so that $\Delta \phi_{\bf r} \ll \phi_{\bf r} < 1$), and obtain the 
190: following Langevin equation for $\phi_{\bf r}$ 
191: \begin{eqnarray}
192: \label{dphi-dt1}
193: \frac{\partial \phi}{\partial t}&=&(1-\phi^2) (a \phi - b \phi^3) \nonumber \\ 
194: &+&\left[a+c+(d-2a-3b)\phi^2 \right] \Delta \phi + \eta, 
195: \end{eqnarray}
196: with correlations for the noise 
197: \begin{eqnarray}
198: \label{corr}
199: &&\langle \eta_{\bf r}(t) \eta_{\bf r'}(t') \rangle = \\  
200: &&\Bigl\{ (1-\phi^2) (c+d\phi^2)+(a-c+2d)\phi \Delta \phi \Bigr\} 
201: \delta_{\bf r,r'} \delta(t-t'), \nonumber 
202: \end{eqnarray}
203: where $\phi$ denotes $\phi_{\bf r}(t)$.
204: Equations~(\ref{dphi-dt1}) and (\ref{corr}) agree  with the Langevin
205: equation proposed in \cite{Hammal05}, based on symmetry arguments, to
206: describe order-disorder phase transitions in general models  with two
207: symmetric absorbing states (the noise correlation in their equation
208: is a simplified version of Eq.(\ref{corr})).  We have derived this expression
209: from  the microscopic dynamics, and therefore the different parameters of the
210: theory have a clear interpretation as a function of the transition rates.
211: The first two terms of Eq.~(\ref{dphi-dt1}) can also be obtained by
212: identifying the field  $\phi_{\bf r}$ with the average value $\langle S_{\bf
213: r} \rangle$ of the spin  at site ${\bf r}$ over all spin configurations, and
214: following the moments  approach technique used by Krapivsky et al. for the
215: original VM \cite{Krapivsky92-Frachebourg96}.  This corresponds to
216: the  $\Omega \to \infty$ limit, for which fluctuations are neglected and the
217: equation for $\phi$ becomes deterministic.
218: 
219: We now use Eq.~(\ref{dphi-dt1}) to gain insight into the macroscopic
220: ordering dynamics.  At the mean-field (MF) level, where the noise term is
221: neglected,  Eq.~(\ref{dphi-dt1}) takes the form of a time-dependent
222: Ginzburg-Landau equation \cite{Bray02} $\frac{\partial \phi}{\partial t} =  D
223: \Delta \phi -\frac{\partial V}{\partial \phi}$, with  the potential
224: $V(\phi)=-\frac{a}{2} \phi^2 + \frac{a+b}{4} \phi^4 - \frac{b}{6} \phi^6$ and
225: $D$ an effective diffusion constant.  In Fig.~\ref{f-phi} we sketch the shape
226: of $V$ and its associated $f(\psi)$ for different values of $a$ and $b$.   
227: When $a>0$ ($f'(0)>f(0)$), $V$ has two symmetric  minima, thus the
228: system coarsens driven by surface tension \cite{Bray02}.   On the contrary,
229: when $a<0$, ($f'(0)<f(0)$), the minimum is at $\phi=0$, then the system 
230: remains in an active disordered state with  particles continuously flipping 
231: their spins, and a global magnetization that  fluctuates around zero.
232: 
233: To illustrate our previous results, we now analyze a general class of 3-state
234: models \cite{Castello06,Baronchelli06}, known to exhibit curvature 
235: driven  by
236: surface tension, as recently shown in \cite{Dall'Asta08}.  They are
237: composed  by two external absorbing states $S=\pm 1$, and an intermediate
238: state $S=0$.  The transition of a particle from  $S=-1$ to $S=1$ happens in
239: two stages.  If we denote by  $\rho_-, \rho_0$ and $\rho_+$, the densities of
240: nearest-neighboring particles  in states $-1,0$ and $1$, respectively, the
241: particle first switches from $S=\mp 1$ to $S=0$ with probability $\rho_{\pm} +
242: \rho_0/2$, and then with the same probability $\rho_{\pm} + \rho_0/2$ from 
243: $S=0$
244: to $S=\pm 1$.  Hence, disregarding the intermediate state, this model can be
245: thought as one with two absorbing states $S=\pm 1$, and an effective
246: transition  probability from $S=-1$ to $S=1$ equals to $(\rho_+ +
247: \rho_0/2)^2$.  In the  same  way, we believe that models with many
248: intermediate states behave as  equivalent 2-state models with effective
249: transition probabilities that are  \emph{ non-linear} in the local densities,
250: and our theory can be applied.  We test this by considering a 2-state model
251: proposed by Abrams and Strogatz to study the competition between two
252: languages~\cite{Abrams03}. The transition probabilities are given by  $P(\mp 1
253: \to \pm 1) = \rho_{\pm}^q$, where $q$ is a positive real number, so that the
254: $q=1$ case reduces to the original VM, while $q \ge 2$ corresponds to models
255: with one or more intermediate states.  In terms of the local field  $\psi = 2
256: \rho_+-1$,  the flipping probability  is  $f(\psi) =
257: \left(\frac{1+\psi}{2}\right)^q$.  Calculating the coefficients  defined in
258: Eq.~(\ref{coef}), and replacing them  into Eq.~(\ref{dphi-dt1}) we obtain
259: \begin{eqnarray}
260: \frac{\partial \phi}{\partial t} &=& \frac{(q-1)}{3 \times 2^{q}}(1-\phi^2) 
261: \left[ 6 \phi + (q-2)(q-3) \phi^3 \right] \nonumber \\ 
262: &+& \frac{q}{2^{q}}
263: \left[ 2+ (q-1)(q-4) \phi^2 \right] \Delta \phi + \eta.
264: \end{eqnarray}
265: When $q<1$ ($f'(0)<f(0)$), the stable solution is $\phi=0$, corresponding to a
266: disordered active system. When $q>1$ ($f'(0)>f(0)$), the stable solutions are
267: $\phi=\pm 1$, thus the system orders driven by surface tension until it
268: reaches one of the absorbing states ($\phi=1,-1$ for all ${\bf r}$).   In
269: particular, this type of ordering is observed when $q=2$, for which the
270: Langevin equation has a similar form as the one derived by Dall'Asta et al.
271: for 3-state models
272: \begin{equation}
273: \frac{\partial \phi}{\partial t} = \frac{1}{2} \left(\phi - \phi^3 \right) +
274: (1-\phi^2) \Delta \phi + \sqrt{1-\phi^2}~\eta,
275: \end{equation} 
276: confirming the above mentioned equivalence with a 2-state model with quadratic
277: transition probability.  The special case $q=1$ corresponds to the VM, with 
278: Langevin equation
279: \begin{equation}
280: \frac{\partial \phi}{\partial t} = \Delta \phi +  \sqrt{1-\phi^2-\Delta \phi}
281: ~\eta.
282: \end{equation} 
283: Neglecting the Laplacian in the noise term, this equation is the same as the
284: one suggested by Dickman et al. \cite{Dickman95}.  Even though the potential
285: is zero, the system still orders due to the presence of the Laplacian, but
286: without surface tension.
287: 
288: \emph{Numerical simulations of the microscopic dynamics.}
289: Equation~(\ref{dphi-dt1}) shows that, at the MF level, the macroscopic
290: behavior of a particular model defined by the probability $f$ is only
291: determined by the coefficients $a$ and $b$, that are expressed in terms of the
292: derivatives of $f$ around $\psi_{\bf k}=0$.  As qualitatively predicted by the
293: MF theory, and confirmed in \cite{Hammal05}, for a fix value $b<b^*$, there is
294: a unique generalized voter (GV) order-disorder transition at a critical value  
295: $a_{GV}$, that
296: separates an active stationary state with absolute  magnetization $m=0$ for
297: $a<a_{GV}$, from a frozen ordered state with $m=1$ for $a>a_{GV}$.  For
298: $b>b^*$, as $a$ is increased, a symmetry breaking Ising transition is observed
299: at a value $a_I$, followed by a directed percolation (DP) transition at a
300: value $a_{DP}>a_I$.  For $a<a_I$ the system is disordered ($m=0$), whereas for
301: $a_I<a<a_{DP}$ it gets partially magnetized ($0<m<1$).  Above $a_{DP}$, the
302: system relaxes to the fully ordered state ($m=1$).  Since we  know the
303: connection between the macroscopic and the microscopic dynamics  expressed in
304: Eq.~(\ref{dphi-dt1}) and Eq.~(\ref{f}) respectively, we now study  these
305: transitions by a Monte  Carlo simulation of the model. This approach is
306: complementary to the one followed by Hammal \emph{et al.} in which they
307: integrate the Langevin equation; and it also allows to test the field theory.
308: 
309: \begin{figure}[t] 
310: \begin{center}
311: \includegraphics[width=3.4in, clip=true]{Fig1.eps}
312: \end{center}
313: \caption{Flipping probability $f$ (left boxes) and its associated potential $V$
314:   (right boxes) vs local field $\psi$. Curves correspond to coefficient values 
315:   $a=-0.3$ (dotted), $a=0$ (solid), $a=0.3$ (dashed), for $b=-0.25$ (top) and
316:    $b=1.0$ (bottom).  For both values of $b$, a single-well and double-well
317:   potentials are obtained for $a<0$ and $a>0$ respectively.}
318: \label{f-phi}
319: \end{figure}
320: 
321: We performed numerical simulations on a $2$-dimensional square lattice with
322: first-nearest neighbors (1st-NNs) interactions ($z=4$).  We used flipping
323: probabilities that are polynomial functions of the form of Eq.~(\ref{f}), for
324: $b=-0.25$, $0.5$ and $3.0$, and various values of $a$ (see Fig.~\ref{f-phi}).
325: Coefficients $c$  and $d$ were set in order to make $f$ an increasing function
326: of $\psi$, and to  arbitrarily fix the point  $f(\psi=1)=1$.   Starting from
327: an ordered configuration of down spins (initial quenching), we flipped the
328: spins of four neighboring sites at the center of the lattice and let the
329: system evolve.  We found that the average density of up spins $N$ and the
330: survival probability $P$, for the three values of $b$, decay at the
331: critical transition point $a_{GV}$ as  $N \sim t^{\eta}$ and 
332: $P \sim t^{-\delta}$ (not shown), 
333: where   $\eta \simeq 0$ and $\delta \simeq 0.95$ agree with the
334: exponents $0$  and $1.0$ respectively, of the GV universality
335: class~\cite{Odor03,Dickman95}.  Also, we found that the average density of
336: interfaces has, at $a_{GV}$, a logarithmic decay with time ($\rho \simeq
337: \pi/(2 \ln t)$), as in the VM \cite{Krapivsky92-Frachebourg96}.
338: Surprisely, only a  GV transition was observed for all three values of $b$,
339: in contradiction with \cite{Hammal05}.
340: 
341: \begin{figure}[t]
342: \begin{center}
343: \includegraphics[width=3.2in, clip=true]{Fig2.eps}
344: \end{center}
345: \caption{GV, DP and Ising transitions on a $2D$ square lattice of side $L=400$
346: with up to 3rd-NNs ($z=12$).  (a) $1/\rho$ vs time on a log-linear scale, for
347: $b=-0.25$  and values of $a$ around the GV critical point $a_{GV} \simeq
348: -0.1105$. (b)   $\rho$ vs time on a log-log scale for $b=0.5$ and values of
349: $a$ around the DP critical point $a_{DP} \simeq 0.2127$.   (c) Binder cumulant
350: $U$ vs $a$ for $b=0.5$. Curves  cross at $a_I \simeq 0.205$, where $U \simeq
351: 0.56$, close to  the universal value $0.6107$ of the $d=2$ Ising model
352: (horizontal dashed line). Insets: (a) survival probability $P$ and (b) density
353: of up spins $N$, on a $800^2$ lattice,  starting from a quenched
354: configuration.  Curves are  averages over $10^5$ realizations. Dashed straight
355: lines have slopes $2/\pi$ and $-1$ in (a) and its inset respectively, whereas
356: the slopes are $-0.45$ and $0.2295$ in (b).}
357: \label{12NN}
358: \end{figure}  
359: 
360: However, our results are in agreement with that of Dornic et. al. 
361: \cite{Dornic01}
362: where, by studying the dynamics of coarsening without surface tension (also
363: with 1st-NNs interactions), they conjectured that all models with
364: $Z_2$-symmetry and without bulk noise exhibit GV transitions.  This apparent
365: disagreement between theory (three transitions) and simulations (only GV
366: transition) comes from the fact that Ising
367: dynamics is observed when there is bulk noise, and this happens in our model
368: when the interaction distance is equal or larger than $2$.  Thus, a spin
369: surrounded by $8$ parallel spins can  still flip if at least one of the four
370: 3-rd NNs is antiparallel.  Indeed,  simulations taking up to 2-nd NNs ($z=8$)
371: also revealed GV only, but  increasing the interactions up to 3-rd NNs
372: ($z=12$), all GV, Ising and DP transitions were observed.  This agrees with
373: the work by Droz et al. \cite{Droz03}, in which they studied an absorbing
374: Ising model in two dimensions, and found that extending the interaction range
375: up to $z=12$ neighbors, the transition from disorder to order splits into a
376: first Ising transition  that breaks the symmetry, and then a DP transition to
377: the unique absorbing  state selected by the spontaneous symmetry breaking.
378: 
379: In Fig.~\ref{12NN} we plot the numerical results for $z=12$ neighbors.  We see 
380: that for $b=-0.25$ (Fig.\ref{12NN}(a)), the decay of $\rho$ and $P$
381: at $a_{GV} \simeq -0.1105$ correspond to that of a GV transition, whereas for
382: $b=0.5$ (Fig.~\ref{12NN}(b)), the transition to complete order happens at a 
383: value $a_{DP} \simeq
384: 0.2127$ at which  $\rho \sim t^{-\delta}$, $N \sim t^{\eta}$ and $P \sim
385: t^{-\delta}$ (not  shown), with $\delta \simeq 0.45$ and $\eta \simeq 0.2295$,
386: i.e. DP critical exponents.  In order to find the Ising transition, we
387: calculated the Binder cumulants $U=1-m_4/3 m_2^2$ (Fig.~\ref{12NN}(c)), where
388: $m_4$ and $m_2$ are the fourth and second moments of the magnetization, as a
389: function of $a$.  As we can see, at the critical point
390: $a_I \simeq 0.205$, the curves of the Binder cumulants for different system
391: sizes cross each other at the value $U \simeq 0.56$, similar to the universal
392: value $0.6107$ of the $2D$ Ising model.
393: 
394: \emph{Conclusions.}  Summing up, we have derived from the microscopic
395: dynamics, the Langevin equation for the magnetization field of general
396: non-equilibrium spin systems with two symmetric absorbing states.  This
397: equation agrees with the one introduced in previous work, but now the
398: dependence of the different terms on the flipping probability is explicitly
399: stated.   This methodology allows to predict the macroscopic behavior, such as
400: critical properties and ordering dynamics, by simply knowing the derivatives
401: of the transition probabilities.   A large class of models in many different
402: disciplines can be studied in this way.  The generalization of this approach
403: to models with an arbitrary number of symmetric absorbing states seems to be
404: challenging.
405: 
406: We are very grateful to Maxi San Miguel and Miguel A. Mu\~noz for fruitful
407: discussions.  We acknowledge  support from project FISICOS (FIS2007-60327) of
408: MEC and  FEDER, and NEST-Complexity project PATRES (043268).
409: 
410: 
411: 
412: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
413: 
414: \begin{thebibliography}{}
415: 
416: \bibitem{Hinrichsen00} H. Hinrichsen, Adv. Phys. {\bf 49}, 815 (2000).
417: 
418: \bibitem{Odor03} G. \'Odor, Rev. Mod. Phys. {\bf 76}, 663 (2004).   
419: 
420: \bibitem{Hammal05} O. Al Hammal, H. Chat\'e, I. Dornic, and M. A. Mu\~noz, 
421:   Phys. Rev. Lett. {\bf 94}, 230601 (2005).
422: 
423: \bibitem{Dornic01} I. Dornic, H. Chat\'e, J. Chave, and H. Hinrichsen,
424:   Phys. Rev. Lett. {\bf 87}, 045701 (2001).
425: 
426: \bibitem{Droz03} M. Droz, A. L. Ferreira, and A. Lipowski, Phys. Rev. E 
427:   {\bf 67}, 056108 (2003).
428: 
429: \bibitem{Clifford73} P. Clifford and A. Sudbury, \emph{Biometrika}, 
430:   {\bf 60}, 581 (1973). 
431: 
432: \bibitem{Baxter07} G. J. Baxter, R. A. Blythe and A. J. MacKane, Math. Biosci. 
433:   {\bf 209}, 124 (2007).
434: 
435: \bibitem{Krapivsky92-Frachebourg96} P. L. Krapivsky, Phys. Rev. A {\bf 45}, 
436:   1067 (1992); \\L. Frachebourg and P. L. Krapivsky, Phys. Rev. E {\bf 53}, 
437:   R3009 (1996).
438: 
439: \bibitem{Castellano07} C. Castellano, S. Fortunato and V. Loreto, 
440: {\tt arXiv:0710.3256} (2007). 
441: 
442: \bibitem{Abrams03} D. M. Abrams and S. H. Strogatz, Nature {\bf 424}, 900 
443: (2003).
444: 
445: \bibitem{Dall'Asta07} L. Dall'Asta and C. Castellano, Europhys. Lett. 
446: {\bf  77}, 60005 (2007).
447: 
448: \bibitem{Stark08} H.-U. Stark, C. J. Tessone, and F. Schweitzer,
449:   Phys. Rev. Lett. {\bf 101}, 018701 (2008).  
450: 
451: \bibitem{Castello06} X. Castell\'o, V. M. Egu\'iluz and M. San Miguel, 
452:   New Journal of Physics {\bf 8}, 308 (2006).
453: 
454: \bibitem{Baronchelli06} A. Baronchelli, L. Dall'Asta, A. Barrat, and
455:   V. Loreto, Phys. Rev. E {\bf 73}, R015102 (2006).
456: 
457: \bibitem{Dall'Asta08} L. Dall'Asta and T. Galla, {\tt arXiv:0806.0817} (2008).
458: 
459: \bibitem{Schweizer} F. Schweitzer and L. Behera, {\tt arXiv:cond-mat/0307742}
460:   (2008).
461: 
462: \bibitem{Bray02} A. J. Bray, Adv. Phys. {\bf 51}, 481 (2002).
463: 
464: \bibitem{Dickman95} R. Dickman, and A. Yu. Tretyakov, Phys. Rev. E {\bf 52},
465:   3218 (1995).
466: 
467: \end{thebibliography}{}
468: 
469: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
470: 
471: 
472: \end{document}
473: 
474: