1: \documentstyle[aps,epsf,pre,multicol]{revtex}
2: \tightenlines
3: \begin{document}
4: \title{ Persistence in higher dimensions : a finite size scaling
5: study}
6:
7: \author{ G. Manoj$^{1}$ and P. Ray $^{1,*}$}
8:
9: \address{ $^1$ The Institute of Mathematical Sciences, C. I. T. Campus,
10: Taramani, Chennai 600 113, India \\
11: $^*$ The Abdus Salam International Centre for Theoretical Physics, Trieste, Italy}
12:
13:
14: \maketitle
15:
16: \begin{abstract}
17: We show that the persistence probability $P(t,L)$, in a coarsening system of linear size $L$
18: at a time $t$, has the finite size scaling form $P(t,L)\sim
19: L^{-z\theta}f(\frac{t}{L^{z}})$ where $\theta$ is the persistence
20: exponent and $z$ is the coarsening exponent.
21: The scaling function $f(x)\sim x^{-\theta}$ for $x \ll 1$ and is constant for large $x$.
22: The scaling form implies a fractal distribution of persistent sites with
23: power-law spatial correlations.
24: We study the scaling numerically for Glauber-Ising model at dimension $d = 1$ to
25: 4 and extend the study to the diffusion problem. Our finite size scaling
26: ansatz is satisfied in all these cases providing a good estimate of the
27: exponent $\theta$.
28: \end{abstract}
29: \pacs{05.40.+j, 05.50.+q, 05.70.Ln}
30: %\newpage
31: %\bigskip
32:
33: %\bigskip
34: \begin{multicols}{2}
35: Persistence decay has been the subject of considerable research activity in recent years.
36: The basic quantity under investigation is the persistence probability $P(t)$, which is the
37: probability that a given stochastic variable with zero mean retains its sign throughout
38: the time interval [0 : $t$]. For a large number of systems, it was found that
39: at asymptotic times $t$, $P(t)\sim t^{-\theta}$, where $\theta$ is in general, a
40: dimension dependent, non-trivial
41: exponent, believed to be unrelated to the other known exponents\cite{REVIEW}.
42: The non-triviality of $\theta$ is particularly true for spatially extended systems
43: where the time evolution of the stochastic field at one lattice site is coupled to that
44: of its neighbours, making the effective single-site evolution non-Markovian.
45:
46: In recent times, the spatial aspects of the persistence problem has also come under
47: study. In particular, the spatio-temporal evolution of the set of persistent sites has
48: been studied by several authors. These include the diffusion problem in $d=1$\cite{ZAN},
49: Ising models in spatial dimension $d=1$\cite{MR0,MR1,MR2} and $d=2$\cite{JAIN} and
50: the generalised $q$-state Potts model in $d=1$\cite{BRST}.
51: It was found that the interplay between persistence decay and the underlying coarsening
52: process leads to dynamical scaling and
53: fractal formation in the spatial distribution of the persistent sites.
54: Such fractal structure has also been reported in an experimental study
55: of breath figures\cite{BREATH}.
56:
57: In the present paper we propose a scaling form for the persistence probability
58: $P(t,L)$ as a function of the lattice size $L$ and time $t$.
59: We use physical arguments to motivate the scaling form in the context of the Ising
60: model and show that the scaling reflects the fractal nature and power-law
61: correlations in the spatial distribution of persistent sites.
62: We provide numerical evidence for its validity through simulations in
63: spatial dimension $d=1$ to 4. The analysis is further
64: extended to the diffusion problem where approximate analytic theories have
65: been used to predict $\theta$ in all dimensions.
66: We argue that fractal formation in diffusion should take place
67: in all dimensions and provide supportive results from simulations.
68:
69: Let us consider the Ising model
70: in a $d$-dimensional geometry of linear size $L$. We start from an initial random
71: configuration and quench the system, say, to the temperature $T=0$.
72: As a result, the spins
73: evolve in time following the Glauber dynamics, lowering the total
74: energy of the configuration in
75: the process. In course of time, domains of positive and negative spins form, with
76: characteristic length scale $\xi(t)$ growing as a power-law in time ie., $\xi(t)\sim t^{1/z}$,
77: where $z$ is the dynamical exponent for the coarsening process\cite{PHOD}.
78: The fraction of persistent
79: spins decays as power of time :
80: $P(t,L)\sim t^{-\theta}$ as long as $t\ll t^{*} \sim L^{z}$.
81: For $t \gg t^{*}$, the domain cannot grow any further because of the finite system size
82: and persistence probability stops decaying, attaining a limiting value
83: $P(\infty,L)\sim L^{-z\theta}$. This happens as long as
84:
85: \begin{equation}
86: \frac{z\theta}{d}< 1
87: \label{eq:FRAC}
88: \end{equation}
89:
90: For $z\theta > d$, persistence probability will decay to zero for
91: any lattice size $L$. Also we assume that there is no `blocking', whereby a finite
92: fraction of spins never flip, leading to a limiting value $P_{\infty}$ independent
93: of finite size effects. Such a situation is believed to occur in Ising model for dimensions
94: $d > 4$\cite{STAUFFER} and in disordered systems \cite{CMNEWMAN}.
95:
96: The above behaviour of the persistent
97: fraction $P(t,L)$ for finite lattice sizes can be summarised in the following dynamical
98: scaling form.
99:
100: \begin{equation}
101: P(t,L)=L^{-z\theta}f(t/L^{z})
102: \label{eq:SCALFORM}
103: \end{equation}
104:
105: where the scaling function $f(x)\sim x^{-\theta}$ for $x \ll 1$ and $f(x)\to$ constant
106: at large $x$. Similar finite size scaling ideas have been used in a previous
107: work in the context of global persistence exponent for nonequilibrium critical
108: dynamics\cite{SATYA1}.
109:
110: The finite-size scaling form given by Eq.\ref{eq:SCALFORM}
111: implies the presence of scale-invariant spatial correlations in the system, characteristic
112: of fractals. To show this, we consider the two-point
113: correlation function $C(r,t)$, which we define as the probability of finding a
114: persistent spin at a distance $r$ from another persistent spin.
115: For a $d$-dimensional system, $C(r,t)$ satisfies the normalisation condition
116: $\int_{0}^{L}C(r,t)d^{d}r=L^{d}P(t,L)$. After substituting
117: Eq.\ref{eq:SCALFORM}, this becomes
118:
119: \begin{equation}
120: \int_{0}^{L}C(r,t)r^{d-1}dr\sim L^{d-z\theta}f(t/L^{z})
121: \end{equation}
122:
123: Let us rewrite this equation in terms of
124: a new function $F(a,b)=a^{z\theta}C(a,b)$ and
125: dimensionless variables $x=r/L$ and $\tau=t/L^{z}$.
126:
127: \begin{equation}
128: \int_{0}^{1}F(Lx,L^{z}\tau)x^{d-1-z\theta}dx\sim f(\tau)
129: \end{equation}
130:
131: Since the RHS of the equation has no explicit $L$-dependence, LHS should also
132: be likewise. This is possible only if $F(a,b)=g(ba^{-z})$, where $g(\eta)$ is
133: given by the integral relation
134:
135: \begin{equation}
136: \tau^{\frac{d}{z}-\theta}\int_{\tau}^{\infty}\eta^{\theta-(1+\frac{d}{z})}g(\eta)d\eta
137: \sim zf(\tau)
138: \label{eq:INTEGRAL}
139: \end{equation}
140:
141: Using the above equation, the limiting behaviour of the function $g(\eta)$ for small
142: and large values of the argument could be deduced from the known behaviour
143: of the function $f(\tau)$. Consider $\tau\gg 1$, where $f(\tau)$ is constant.
144: From Eq. \ref{eq:INTEGRAL}, this implies that $g(\eta)$ is constant for large $\eta$.
145: In the other extreme of $\tau\ll 1$, $f(\tau)\sim \tau^{-\theta}$. We split
146: the integral in Eq. \ref{eq:INTEGRAL} as $\int_{\tau}^{\infty}=
147: \int_{\tau}^{\alpha} + \int_{\alpha}^{\infty}$ and note that $g(\eta)$ is
148: constant in the second integral for sufficiently large $\alpha$.
149: The second integral
150: vanishes as $\tau^{\frac{d}{z}-\theta}$ as $\tau\to 0$, whereas the RHS diverges as
151: $\tau^{-\theta}$. This can be consistent only if the first integral diverges as
152: $\tau^{-\theta}$, which would imply that $g(\eta)\sim \eta^{-\theta}$ as $\eta\to 0$.
153: This leads to the following dynamical scaling form for $C(r,t)$.
154:
155: \begin{equation}
156: C(r,t)=r^{-z\theta}g(\frac{t}{r^{z}})
157: \label{eq:CORR}
158: \end{equation}
159:
160: For small separations $r\ll t^{1/z}$, this scaling form implies
161: scale-free correlations, ie., $C(r,t)\sim r^{-z\theta}$, characteristic of a fractal
162: with fractal dimension $d_{f}=d-z\theta$. On the other hand, over larger length scales,
163: $C(r,t)\sim t^{-\theta}$, which is indicative of the absence of any spatial
164: correlations. This scaling description was introduced by us\cite{MR1,MR2}
165: in the context of $A+A\to\emptyset$ model, and later verified
166: numerically in 2-dimensional Ising model\cite{JAIN} also.
167:
168: To check the finite-size scaling form given by Eq. \ref{eq:SCALFORM},
169: we simulate Ising spin
170: systems of various sizes in spatial dimension $d=1$ to 4. Starting from a
171: random initial configuration, the spins are quenched to zero temperature and are
172: updated sequentially using the Glauber updating rule by which
173: a spin is always flipped if the resulting energy change $\Delta E < 0$, never flipped if
174: $\Delta E >0$, and flipped with probability $\frac{1}{2}$ if $\Delta E=0$.
175: One MC time step was counted
176: after every spin in the lattice was updated once.
177: The persistence probability at any time $t$ was determined as the fraction
178: of spins that did not flip even once till time $t$ since the time evolution started.
179: The data is averaged typically over 1000 starting random configurations for small
180: $L$ and low $d$ and over $50$ starting configurations for large $L$ and
181: high $d$.
182:
183: \vspace{-0.5cm}
184: \begin{figure}[a]
185: \narrowtext
186: \epsfxsize=1.7in
187: \hspace{-1.5cm}
188: \epsfbox{fig1.ps}
189: \vspace{-0.5cm}
190: \caption{The persistence probability $P(t,L)$ is plotted against time $t$
191: (measured in MC steps) for three different lattice sizes $L$ in $d=2$ Glauber Ising model.}
192: \end{figure}
193:
194: \vspace{-0.5cm}
195: \begin{figure}[b]
196: \narrowtext
197: \epsfxsize=1.7in
198: \hspace{-1.5cm}
199: \epsfbox{fig2.ps}
200: \vspace{-0.5cm}
201: \caption{
202: Same as Fig. 1, except that the scaling function $f(x)=L^{z\theta}P(t,L)$
203: is plotted against the dimensionless scaling variable $x=t/L^{z}$. The data for different
204: $L$ values were found to collapse well to a single curve for $\theta=0.21$
205: and $z=2.12\pm 0.05$.
206: }
207: \end{figure}
208:
209:
210: For $T=0$ Glauber dynamics of Ising model, the persistence exponent $\theta$ is
211: exactly
212: known to be $3/8$ in $d=1$\cite{DERRIDA}. In higher dimensions, simulations predict
213: $\theta\simeq 0.22(d=2)$\cite{STAUFFER,DERRIDA1,SATYA} and $\theta\simeq 0.16 (d=3)$
214: \cite{STAUFFER}.
215: In our finite size scaling analysis of the simulation data, we adopt the following
216: procedure. For $d=1,2$ and 3, we fix $\theta$ at its known value and adjust
217: $z$ to find the value which gives the best data collapse. In all cases, we find
218: $z\simeq 2$, which is the accepted value of the coarsening exponent for
219: non-conserved scalar models\cite{PHOD}.
220: (In $d=3$ Galuber dynamics, a slower $t^{1/3}$ coarsening has been
221: observed before\cite{SETHNA}. This is presumably due to
222: lattice effects, but we have not seen any signature of this effect
223: in our simulations).
224: In $d=4$, on the other hand,
225: we fix $z$ at 2, and adjust $\theta$ to collapse the data to a single curve.
226: The results are displayed in Figs. 1 to 4.
227:
228:
229: %In $d=3$, we find good scaling with $\theta=0.166$\cite{STAUFFER} and $z=2.05\pm
230: %0.03$.
231: %We do not see evidence for an effective value $z=3$, observed in earlier $d=3$
232: %simulations, and believed to be caused by lattice effects \cite{SATYA}.
233:
234: In $d=4$, we find that for $z=2$, $\theta=0.12\pm 0.02$ gives reasonably good data
235: collapse over the time scales and system sizes studied. Fig.4 shows the
236: scaled data in $d=4$. It may be mentioned that in $d=4,$ earlier simulations
237: had suggested that the persistence decay might be slower than a power-law,
238: and perhaps logarithmic\cite{STAUFFER}. However, the agreement of our data
239: with the scaling form Eq.\ref{eq:SCALFORM} suggests that persistence follows
240: a power-law decay in $d=4$ also. For $d > 4 $, blocking of spins has been
241: shown to lead to a limiting value of $P(t,L)$ as $t\to\infty$, which is
242: independent of $L$\cite{STAUFFER}. We could simulate only small lattice sizes for
243: $d=5$ from which we cannot make any conclusive remark at this stage.
244:
245: \vspace{-0.5cm}
246: \begin{figure}[c]
247: \narrowtext
248: \epsfxsize=1.7in
249: \hspace{-1.5cm}
250: \epsfbox{fig3.ps}
251: %\vspace{-10.0cm}
252: \vspace{-0.5cm}
253: \caption{The scaling function $f(x)=L^{z\theta}P(t,L)$ is plotted
254: against the dimensionless scaled time $x=t/L^{z}$ for three $L$-values in $d=3$ Glauber
255: Ising model.
256: The observed data collapse has been obtained for $z=2.05$
257: and $\theta=0.166$.}
258: \end{figure}
259:
260: \vspace{-0.5cm}
261: \begin{figure}[d]
262: \narrowtext
263: \epsfxsize=1.7in
264: \hspace{-1.5cm}
265: \epsfbox{fig4.ps}
266: \vspace{-0.5cm}
267: \caption{ The figure shows the scaled probability plotted against the
268: dimensionless scaled time in $d=4$ Glauber Ising model. We have fixed $z=2$, and find that
269: $\theta=0.12\pm 0.02$ gives the best data collapse.}
270: \end{figure}
271:
272:
273:
274: In the diffusion problem, we have a scalar field $\phi({\bf x},t)$ evolving according to
275: the diffusion equation. The initial values $\phi({\bf x},0)$ are taken to be independent
276: random variables with zero mean.
277:
278: \begin{equation}
279: \frac{\partial \phi({\bf x},t)}{\partial t}={\nabla}^{2}\phi({\bf x},t)
280: \hspace{0.3cm};\hspace{0.3cm}\langle\phi({\bf x},0)\phi({\bf x}^{\prime},0)\rangle=
281: \delta^{d}({\bf x}-{\bf x}^{\prime})
282: \label{eq:DIFF}
283: \end{equation}
284:
285: For this problem, it has been shown using approximate analytic
286: theories\cite{DIFF1,DIFF2,DIFF3}that $P(t)\sim t^{-\theta}$ in
287: all dimensions. The predicted exponent values in low dimensions were
288: in good agreement with simulation results.
289: The exponent was found to increase with dimension, and has been suggested
290: to have the asymptotic value
291: $\theta(d)\simeq \alpha\sqrt{d}$ as $d\to\infty$. The constant $\alpha$
292: has been estimated
293: to be $\simeq 0.14$\cite{DIFF1,DIFF2} and $\simeq 0.18$\cite{DIFF3} by different authors.
294: For $d=1,2$ and 3, the exponent values are found to be $\theta \simeq 0.12, 0.18$ and $0.23$
295: respectively.
296:
297: To simulate Eq. \ref{eq:DIFF} numerically, we use the finite difference Euler discretization
298: scheme on cubic lattices of $L^{d}$ sites \cite{DIFF1,DIFF2}.
299:
300: \begin{equation}
301: \phi({\bf x},t+\Delta t)=\phi({\bf x},t)+a\left[\sum_{{\bf x}^{\prime}}\phi({\bf
302: x}^{\prime},t)-2d\phi({\bf x},t)\right]
303: \label{eq:DISCRETE}
304: \end{equation}
305:
306: where ${\bf x}^{\prime}$ runs over all the $2d$ nearest neighbour lattice sites
307: of ${\bf x}$ in the cubic lattice and $a=\frac{\Delta t}{(\Delta x)^{2}} <
308: \frac{1}{2d}$
309: for stability of the discretization scheme. We have taken $a=\frac{1}{4d}$ in our
310: simulations as this value has been observed to provide the fastest approach to
311: the asymptotic regime\cite{DIFF1}.
312:
313: For the diffusion problem, simple scaling arguments suggest that the dynamical
314: exponent $z=2$ in all dimensions. In all dimensions studied,
315: we found excellent scaling collapse with $z\simeq 2$ and the $\theta$ values quoted
316: above. Upon substitution of the
317: exponent values into Eq. \ref{eq:FRAC}, it can be easily seen that the condition for fractal
318: formation is satisfied for $d=1,2$ and 3. For $d=1$, this has already been confirmed
319: by an earlier numerical study\cite{ZAN}. Our results for the persistence probability
320: and the scaling function for three different lattice sizes
321: in $d=2$ is displayed in Fig. 5 and 6.
322:
323: It is also possible to extrapolate these results to the $d\to\infty$ limit using the
324: asymptotic form suggested for $\theta$. We see that in this limit, the LHS of Eq. \ref{eq:FRAC}
325: vanishes as $\frac{1}{\sqrt{d}}$, leading us to conjecture that fractal formation persists
326: in all dimensions for the diffusion problem.
327:
328:
329: \vspace{0.5cm}
330: \begin{figure}[e]
331: \narrowtext
332: \epsfxsize=1.7in
333: \hspace{-1.5cm}
334: \epsfbox{fig5.ps}
335: \vspace{-0.5cm}
336: \caption{The persistence probability $P(t,L)$ is plotted against time $t$
337: (measured as the number of MC steps) for three different lattice sizes $L$ in $d=2$ diffusion problem.}
338: \end{figure}
339:
340: \vspace{-0.5cm}
341: \begin{figure}[f]
342: \narrowtext
343: \epsfxsize=1.7in
344: \hspace{-1.5cm}
345: \epsfbox{fig6.ps}
346: \vspace{-0.5cm}
347: \caption{
348: Same as Fig. 5, except that the scaling function $f(x)=L^{z\theta}P(t,L)$
349: is plotted against the dimensionless scaling variable $x=t/L^{z}$. The data for different
350: $L$ values were found to collapse well to a single curve for $\theta=0.186$
351: and $z=2.05\pm 0.04$.
352: }
353: \end{figure}
354:
355:
356:
357: To conclude, we have proposed a finite size scaling ansatz for the persistence
358: probability in a coarsening system. The scaling form corresponds to the fractal
359: structure and dynamic scaling characterising the spatio-temporal evolution
360: of the persistent set. We check the scaling form numerically for Glauber-Ising model
361: and for the diffusion problem. Finite size scaling enables us to study persistence
362: reliably in higher dimensions. Our results agree with the known values of $\theta$ in
363: the case of Ising model(from $d=1$ to 3) and in the diffusion problem (we have checked
364: upto $d=3$). For $d=4$ Ising model, we find the signature of algebraic decay of persistence
365: with $\theta\simeq 0.12$, in contrast with what had been reported earlier\cite{STAUFFER}.
366:
367: We thank G. I. Menon and D. Dhar for a critical reading of the manuscript
368: and valuable suggestions.
369: G.M gratefully acknowledges the hospitality at The Abdus Salam ICTP, Trieste,
370: Italy where this work was done.
371: G. M also thanks C. Sire, S. N. Majumdar and A. J. Bray for helpful discussions
372: and illuminating remarks.
373:
374: %\Acknowledgement
375:
376:
377: \begin{references}
378: \bibitem{REVIEW} For a recent review, S. N. Majumdar, Curr. Sci. {\bf 77}, 370 (1999)
379: (preprint cond-mat/9907407).
380: \bibitem{ZAN} D. H. Zanette, Phys. Rev. E {\bf 55} 2462 (1997).
381: \bibitem{MR0} G. Manoj and P. Ray, preprint cond-mat/9901130.
382: \bibitem{MR1} G. Manoj and P. Ray, J. Phys. A {\bf 33} L109 (2000).
383: \bibitem{MR2} G. Manoj and P. Ray, J. Phys. A {\bf 33} 5489 (2000).
384: \bibitem{JAIN} S. Jain and H. Flynn, preprint cond-mat/0004148.
385: \bibitem{BRST} A. J. Bray and S. J. O'Donoghue, preprint cond-mat/0005180.
386: \bibitem{BREATH} M. Marcos-Martin, D. Beysens, J-P. Bouchad, C. Godr\`{e}che and
387: I. Yekutieli, Physica D {\bf 214}, 396 (1995).
388: \bibitem{PHOD} A. J. Bray, Adv. Phys. {\bf 43}, 357 (1994).
389: \bibitem{STAUFFER} D. Stauffer, J. Phys. A {\bf 27}, 5029 (1994).
390: \bibitem{CMNEWMAN} C. M. Newman and D. L. Stein, Phys. Rev. Lett. {\bf 82}, 3944 (1999) ; S. Jain, Phys. Rev. E {\bf 60}, R2445 (1999).
391: \bibitem{SATYA1} S. N. Majumdar, A. J. Bray, S. J. Cornell and C. Sire, Phys. Rev. Lett. {\bf 77}, 3704 (1996).
392: \bibitem{DERRIDA} B. Derrida, V. Hakim and V. Pasquier, Phys. Rev. Lett. {\bf 75}, 751 (1995).
393: \bibitem{DERRIDA1} B. Derrida, A. J. Bray and C. Godr\`{e}che, J. Phys. A
394: {\bf 27}, L357 (1994).
395: \bibitem{SATYA} S. N. Majumdar and C. Sire, Phys. Rev. Lett. {\bf 77}, 1420 (1996).
396: \bibitem{SETHNA} J. D. Shore, M. Holzer and J. P. Sethna, Phys. Rev. B
397: {\bf 46}, 11376 (1992).
398: \bibitem{DIFF1} S. N. Majumdar, C. Sire, A. J. Bray and S. J. Cornell,
399: Phys. Rev. Lett. {\bf 77} 2867 (1996).
400: \bibitem{DIFF2} B. Derrida, V. Hakim and R. Zeitak, Phys. Rev. Lett. {\bf 77} 2871 (1996).
401: \bibitem{DIFF3} T. J. Newman and Z. Toroczkai, Phys. Rev. E {\bf 58}, 2685 (1998).
402: \end{references}
403:
404:
405:
406:
407:
408:
409:
410: \end{multicols}
411:
412:
413:
414:
415:
416:
417:
418:
419: \end{document}
420: