1: \documentclass[twoside,12pt]{article}
2: \usepackage{graphicx}
3: \renewcommand\floatpagefraction{.9}
4: \renewcommand\textfraction{.1}
5: \begin{document}
6:
7: \centerline{{\it Brazilian Journal of Physics} {\bf 33}, 616 (2003)}
8: \bigskip
9:
10: \centerline{\Large \bf Corrections to Finite Size Scaling in
11: Percolation}
12: \bigskip
13: P.M.C. de Oliveira$^1$, R.A. N\'obrega$^2$ and D. Stauffer$^3$.
14:
15: \bigskip
16: Instituto de F\'\i sica, Universidade Federal Fluminense, av. Litor\^anea
17: s/n, Boa Viagem, Niter\'oi, Brasil 24210-340
18:
19: \medskip
20:
21: \noindent $^1$ pmco@if.uff.br \space $^2$ rafaella@if.uff.br
22:
23: \noindent $^3$ stauffer@thp.uni-koeln.de; Visiting from Institute for
24: Theoretical Physics, Cologne University, D-50923 K\"oln, Euroland
25:
26: \bigskip
27:
28: Abstract: A $1/L$-expansion for percolation problems is proposed, where $L$
29: is the lattice finite length. The square lattice with 27 different sizes $L
30: = 18, 22, \dots 1594$ is considered. Certain spanning probabilities were
31: determined by Monte Carlo simulations, as continuous functions of the site
32: occupation probability $p$. We estimate the critical threshold $p_c$ by
33: applying the quoted expansion to these data. Also, the universal spanning
34: probability at $p_c$ for an annulus with aspect ratio $r = 1/2$ is estimated
35: as $C = 0.876657(45)$.
36:
37:
38: \section{Introduction}
39:
40: In reference \cite{nz} a square lattice is viewed as a torus, thus
41: without frontiers, where sites are randomly occupied with probability $p$.
42: {\sl Wrapping} percolation probabilities can be defined within this
43: geometry, counting configurations which wrap along the horizontal and/or/xor
44: vertical directions. The probability $R_h$, for instance, counts all
45: configurations wrapping along the horizontal direction, no matter which
46: occurs vertically. As a function of $p$, $R_h$ corresponds to a plot like
47: figure 1 (to be quoted later) for a finite lattice. In the thermodynamic
48: limit $L \to \infty$, this plot approaches a step function, $R_h = 0$ below,
49: and $R_h = 1$ above the critical threshold $p_c$. The point $p_L(\tau)$
50: exemplified on the plot serves as an estimator for $p_c$: by measuring a
51: sequence of such values, for larger and larger sizes $L_1, L_2 \dots L_N$,
52: one can extrapolate this sequence for $L \to \infty$. Reference \cite{nz}
53: presents the figure $p_c = 0.59274621(13)$, the most accurate available
54: today, obtained from more than $7 \times 10^9$ Monte Carlo samples. Starting
55: from an empty lattice, filled site by site at random, each sample provides a
56: single number to the statistics, namely the precise number $n$ of occupied
57: sites for which the proper wrap (horizontal for $R_h$) appears for the first
58: time. This computational strategy of filling up the lattice site by site and
59: storing data onto bell-shaped $n$-histograms is equivalent to that of
60: \cite{elias}, and similar to early works \cite{domb}. However, the
61: multi-step strategy introduced in \cite{nz} involves many other components
62: (see \cite{nz2}). The authors of \cite{nz} assume a $p_L(\tau) - p_c \sim
63: L^{-(2+1/\nu)}$ dependence, where $\nu = 4/3$ is the correlation length
64: critical exponent. As $2+1/\nu = 2.75$ is a sufficiently high value, data up
65: to only $L = 128$ were needed. Also, the authors adopt the value $\tau^* =
66: 0.521058290$, corresponding to that particular wrapping geometry and exactly
67: known \cite{pinson} as the limiting value for $R_h$ at $p_c$, called
68: Pinson's number in \cite{zlk}. Outside this value or within other
69: geometries, the not-so-high exponent $1+1/\nu$ or even the smaller $1/\nu$
70: were observed \cite{ziff}.
71:
72:
73: \begin{figure}[hbt]
74: \begin{center}
75: \includegraphics[angle=-90,scale=0.45]{fig1.eps}
76: \end{center}
77: \caption{Spanning probability function for $L = 18$ (solid line), $38$ and
78: $74$ (dotted lines). By fixing some value $\tau$, one can find a sequence of
79: values $p_L(\tau)$ for increasing lattice sizes, approaching the critical
80: threshold $p_c$.}
81: \end{figure}
82:
83: Here, we propose the mathematical form
84:
85: \begin{equation}
86: p_L(\tau) = p_c + {1\over L^{1/\nu}} \Big[ A_0(\tau) + {A_1(\tau)\over L} +
87: {A_2(\tau)\over L^2} + \dots {A_M(\tau)\over L^M} \Big]\,\,\,\, ,
88: \end{equation}
89:
90: \noindent for estimators $p_L$ obtained from quantities like $R_h$, where
91: the cutoff $M$ is conveniently chosen according to the numerical accuracy
92: available.
93:
94:
95:
96: \begin{table}[hbt]
97: \begin{center}
98: \begin{tabular}{|l|c|c|}
99: \hline
100: $L$ & $p_L(\tau=0.9)$ & samples\\
101: \hline
102:
103: 18 & 0.55982808(075) & $10^9$\\
104: 22 & 0.56196704(122) & $10^9$\\
105: 26 & 0.56403393(062) & $10^9$\\
106: 30 & 0.56590384(082) & $10^9$\\
107: 38 & 0.56902009(083) & $10^9$\\
108: 46 & 0.57146271(087) & $10^9$\\
109: 52 & 0.57296244(060) & $10^9$\\
110: 62 & 0.57500000(066) & $10^9$\\
111: 74 & 0.57689631(175) & $10^8$\\
112: 86 & 0.57838778(129) & $10^8$\\
113: 102 & 0.57993924(157) & $10^8$\\
114: 118 & 0.58115110(109) & $10^8$\\
115: 142 & 0.58254835(133) & $5\times 10^7$\\
116: 166 & 0.58360542(109) & $5\times 10^7$\\
117: 202 & 0.58479326(182) & $3\times 10^7$\\
118: 234 & 0.58558821(076) & $3\times 10^7$\\
119: 282 & 0.58649210(124) & $10^7$\\
120: 334 & 0.58721227(157) & $10^7$\\
121: 402 & 0.58791296(095) & $10^7$\\
122: 478 & 0.58848866(104) & $10^7$\\
123: 566 & 0.58898678(159) & $6\times 10^6$\\
124: 674 & 0.58944374(126) & $6\times 10^6$\\
125: 802 & 0.58984063(146) & $4\times 10^6$\\
126: 958 & 0.59019986(149) & $4\times 10^6$\\
127: 1126 & 0.59048760(168) & $4\times 10^6$\\
128: 1354 & 0.59077660(121) & $4\times 10^6$\\
129: 1594 & 0.59100202(082) & $4\times 10^6$\\
130:
131: \hline
132: \end{tabular}
133: \caption{Values for $p_L$ obtained for fixed $\tau=0.9$, as an example. A
134: Chi-square fitting of equation (1) gives $p_c = 0.59274675(88)$, $A_0 =
135: -0.44204(11)$, $A_1 = 3.275(14)$, etc.}
136: \end{center}
137: \end{table}
138:
139:
140: We apply this formula to the $L \times L$ square lattice, with a set
141: of different increasing lengths such that the numbers of sites grow by a
142: factor of $\sqrt{2}$. Table I shows an example for $\tau = 0.9$, for which
143: the traditional Chi-square fitting \cite{recipes} gives $p_c =
144: 0.59274675(88)$, in agreement with \cite{nz} although within a larger error
145: bar. We adopted $M = 4$ in equation (1), compatible with our smallest
146: lattice size $L = 18$, since $18^{-4.75} = 1 \times 10^{-6}$ still falls
147: inside our numerical accuracy, whereas the next term $18^{-5.75} = 6 \times
148: 10^{-8}$ would be outside. The quality of this fit can be appreciated by the
149: so-called goodness-of-fit $Q$ \cite{recipes}, a quantity between 0 and 1.
150: The fit is considered believable \cite{recipes} for values of $Q > 0.1$. In
151: our example, we get $Q = 0.857$ for table I. All our data to be discussed
152: hereafter, for many other values of $\tau$ between $0.5$ and $0.99$, present
153: the same degree of accuracy, giving credit to our proposal, equation (1). In
154: spite of these accurate results, one cannot rule out some possible higher
155: terms deviating from (1), for example that proposed in \cite{hovi}.
156:
157: \section{Measured quantity}
158:
159: First, let's explain the {\sl spanning} probability we adopted
160: within the torus, instead of the {\sl wrapping} probability \cite{nz}. We
161: consider two parallel lines distant $L/2$ from each other, on the $L \times
162: L$ square lattice. For each sample --- again obtained by filling up the
163: initially empty lattice, site by site at random --- we count the precise
164: number $n$ of occupied sites for which these lines become connected for the
165: first time, no matter which occurs around the other direction. This approach
166: has a big advantage over the {\sl wrapping} probability around the whole
167: torus: From the same sample we can count $n$ just $L$ times instead of only
168: once! The parallel lines can be numbered ($i,i+L/2$) for $i = 1, 2, \dots
169: L/2$ along the horizontal direction, with the same procedure repeated
170: vertically. Thus, the statistics is multiplied by a factor of $L$. In table
171: I, for instance, the sampling counting $10^9$ for $L = 18$ corresponds to an
172: $n$-histogram with $18 \times 10^{9}$ accumulated units (the total area
173: below the bell-shaped curve). In the same table, for $L = 1594$, the much
174: smaller sampling counting $4 \times 10^6$ corresponds indeed to almost the
175: same statistics, i.e. $6 \times 10^{9}$ accumulated units below the curve.
176: This trick allows us to test a wide range of lattice sizes, and verify the
177: validity of our proposal (1). The further computational time one needs in
178: order to implement this trick is negligible: we simply keep in memory the
179: top and bottom (left and right) extreme lines for each already formed
180: cluster of neighbouring occupied sites. Thus, for each new included site,
181: only the last updated cluster must be verified.
182:
183: The {\sl spanning} probability function is obtained by superimposing
184: a lot of step functions, one for each counted $n$, and dividing the result
185: by the total number ($L \times$sampling counts). An average is then
186: performed, weighted by $C(L^2,n) p^n (1-p)^{L^2-n}$ where $C$ is a
187: combinatorial factor, yielding the $p$-continuous curves shown in figure 1.
188:
189: The 3-digits error bars shown in table I were obtained by dividing
190: the whole set of data into, say, $S = 10$ sub-sets, independently
191: calculating $p_L(\tau)$ for each sub-set. The error bars are the standard
192: deviation of this distribution divided by $\sqrt{S}$. This last division is
193: based on the supposition that the whole data is normally distributed. In
194: order to verify the validity of this approach, we repeated the same
195: procedure with $S = 20$. Indeed, the error bars are approximately the same,
196: independent of $S$. For safety, we adopted always the largest between both
197: error bars so obtained. For intermediate lattice sizes (from $L = 74$ up to
198: 402), we used $S = 10$ and 5, instead of 20 and 10.
199:
200: We also simulated larger $L \times L$ lattices for $\tau = 0.5$, up
201: to $L = 24000$, with free instead of periodic boundary conditions, within a
202: poorer statistics. The results are also compatible with equation (1).
203:
204:
205: \section{Data Analysis}
206:
207: By fitting data with equation (1), we have a set of $M+2$ parameters
208: to be determined, namely $p_c, A_0(\tau), A_1(\tau), \dots A_M(\tau)$. The
209: error bars for these quantities come from the Chi-square fitting procedure
210: \cite{recipes}, as a consequence of the primary error bars directly measured
211: for the crude data (that of table I, for instance). Thus, the accuracy
212: obtained for each parameter, in particular the interesting one $p_c$,
213: involves a series of accumulated errors. Instead of taking care of $p_c$,
214: let's turn our attention to $A_0(\tau)$ for a while. Figure 2 shows a plot
215: of this value as a function of $\tau$.
216:
217: \begin{figure}[hbt] \begin{center}
218: \includegraphics[angle=-90,scale=0.5]{fig2.eps} \end{center}
219: \caption{Expansion parameter $A_0$, equation (1), as a function of $\tau$.
220: The error bars obtained from the Chi-square fitting appear only within the
221: smaller scale of the inset, near the special point $\tau^* = 0.984786(11)$
222: where $A_0$ vanishes. }
223: \end{figure}
224:
225: One can see that $A_0$ vanishes for the particular value $\tau^* =
226: 0.984786(11)$. For the geometry we adopted, this is the equivalent of the
227: above quoted Pinson's number \cite{pinson,zlk}, i.e. the universal value of
228: our spanning probability at $p_c$ in the thermodynamic limit. Recently,
229: Cardy \cite{cardy} tried to calculate this value for an annulus, i.e. an
230: $L_1 \times L_2$ square lattice where periodic boundary condition is adopted
231: only along the direction of $L_1$. The quoted universal value corresponds to
232: the critical spanning probability between the two free-boundary lines, and
233: depends only on the ratio $r = L_2/L_1$. Unfortunately, within his
234: theoretical approach, Cardy was forced to leave out spanning configurations
235: which also wrap along the direction of $L_1$, the periodic boundary.
236:
237: Our geometry with the complete $L \times L$ torus divided in two
238: halves by the lines $i$ and $i + L/2$ corresponds to the $r = 1/2$ Cardy's
239: geometry counted twice, in parallel. Thus, our $\tau^*$ can be related to
240: Cardy's number $C$ by $1 - \tau^* = (1 - C)^2$, i.e. $C = 0.876657(45)$.
241: Compared with Cardy's exact value $C - C_{\rm wrap} = 0.7569977963$
242: \cite{cardy} we found the contribution $C_{\rm wrap} = 0.119659(45)$ for the
243: spanning configurations which also wrap around the periodic direction.
244:
245: The critical occupation probability $p_c$ can be obtained from our
246: expansion (1) by fixing any value for $\tau$. The best choice is $\tau =
247: \tau^*$, for which the leading finite-size term in (1) becomes
248: $L^{-1-1/\nu}$ instead of $L^{-1/\nu}$. In this case, we get $p_c =
249: 0.59274621(33)$, where the error bar is estimated by the same above-quoted
250: procedure of dividing the whole data set into $S = 10$ sub-sets. For some
251: unknown particular reason, the {\sl wrapping} probabilities adopted in
252: \cite{nz} works better yet, the leading finite-size term in (1) being
253: $L^{-2-1/\nu}$. In \cite{nz} a numerical evidence for that behaviour is
254: given, by plotting $\tau_L(p_c) - \tau^*$ against $L$ (assuming some
255: previously determined value for $p_c$) and verifying a power-law dependence
256: with an exponent very close to $-2$. Indeed, for our {\sl spanning}
257: probability, the same plot gives an exponent very close to $-1$.
258:
259: \section{Conclusion}
260:
261: We propose the finite-size expansion (1) for spanning probabilities
262: in percolation, where $p_L(\tau)$ is defined in figure 1.
263:
264: This approach can be used to calculate various critical quantities.
265: We applied it to a particular geometry within the site percolation problem
266: on a $L \times L$ square lattice, considered as a torus. Taking two parallel
267: lines distant $L/2$ from each other, our spanning probability counts
268: configurations for which these two lines are connected. The universal value
269: $\tau^* = 0.984786(11)$ and the critical occupation $p_c = 0.59274621(33)$
270: are obtained.
271:
272: As a by-product, we propose the universal value $C = 0.876657(45)$
273: for the critical spanning probability within a $L \times L/2$ annulus, i.e.
274: a square lattice with periodic (free) boundaries along the direction of $L$
275: ($L/2$). This probability corresponds to all configurations for which the
276: frontiers separated by $L/2$ are connected. Cardy \cite{cardy} determined
277: the exact figure $C - C_{\rm wrap} = 0.7569977963$, where $C_{\rm wrap}$
278: corresponds to spanning configurations which also wrap along the direction
279: of $L$, discounted in his approach.
280:
281: The whole computer time for obtaining these data was 20 thousand
282: hours, on a dozen computers, typically powered by an Athlon 1GHz processor.
283:
284:
285:
286: \section{Acknowledgements}
287:
288: Six months ago, the first author presented a preliminary version of
289: this work in Campos do Jord\~ao, S\~ao Paulo, Brazil, in honour of the
290: 60$^{\rm th}$ birthday of professor Silvio Salinas. Now, some other articles
291: are presented in honour of the 60$^{\rm th}$ birthday of the last author. In
292: contrast, the younger authors dedicated this work to the senior professor
293: Silvio Salinas.
294:
295: We are indebted to Carlos Tomei, Antonio Brito Serbeto, John Cardy
296: and Robert Ziff for helpful discussions, and Brazilian agencies CNPq and
297: FAPERJ for support.
298:
299: \newpage
300:
301: \begin{thebibliography} {99}
302:
303: \bibitem{nz} M.E.J. Newman and R.M. Ziff, {\it Phys. Rev.} {\bf E64}, 016706
304: (2001); {\it Phys. Rev. Lett.} {\bf 85}, 4104 (2000).
305:
306: \bibitem{elias} J.E. de Freitas, L.S. Lucena and S. Roux, {\it Physica} {\bf
307: A266}, 81 (1999); J.E. de Freitas and L.S. Lucena, {\it Int. J. Mod. Phys.}
308: {\bf C11}, 1581 (2000).
309:
310: \bibitem{domb} C. Domb, E. Stoll and T. Schneider, {\it Contemp. Phys.} {\bf
311: 577} (1980); N. Jan, T. Lookman and D. Stauffer, {\it J. Phys.} {\bf A16},
312: L117 (1983); P.M.C. de Oliveira, S. Moss de Oliveira and S.L.A. de Queiroz,
313: {\it Physica} {\bf A175}, 345 (1991).
314:
315: \bibitem{nz2} R.M. Ziff and M.E.J. Newman, {\it Phys. Rev.} {\bf E66},
316: 016129 (2002).
317:
318: \bibitem{pinson} H.T. Pinson, {\it J. Stat. Phys.} {\bf 75}, 1167 (1994).
319:
320: \bibitem{zlk} R.M. Ziff, C.D. Lorenz and P. Kleban, {\it Physica} {\bf
321: A266}, 17 (1999).
322:
323: \bibitem{ziff} R.M. Ziff, {\it Phys. Rev. Lett.} {\bf 69}, 2670 (1992).
324:
325: \bibitem{recipes} W.H. Press, B.P. Flannery, S.A. Teukolsky and W.T.
326: Vetterling, {\sl Numerical Recipes in C}, Cambridge University Press (1988),
327: section 14.3.
328:
329: \bibitem{hovi} J.-P. Hovi and A. Aharony, {\it Phys. Rev.} {\bf E53}, 235
330: (1996).
331:
332: \bibitem{cardy} J. Cardy, {\it J. Phys.} {\bf A35}, L565 (2002).
333:
334:
335:
336: \end{thebibliography}
337: \end{document}
338: