1: \tolerance = 10000
2: %\documentstyle[aps,prl,twocolumn]{revtex}
3: \documentstyle[aps,prl,twocolumn,epsf]{revtex}
4: %\textwidth18cm
5:
6: \def\be{\begin{eqnarray}}
7: \def\ee{\end{eqnarray}}
8:
9: \begin{document}
10: \draft
11:
12:
13: \title{
14: Thermodynamic origin of universal fluctuations and two-power laws
15: }
16:
17: \author{Jan Naudts$^1$ and Marek Czachor$^{1,2,3}$}
18: \address{
19: $^1$ Departement Natuurkunde, Universiteit Antwerpen UIA,
20: Universiteitsplein 1, B2610 Antwerpen, Belgium\\
21: $^2$ Katedra Fizyki Teoretycznej i Metod Matematycznych,
22: Politechnika Gda\'{n}ska, 80-952 Gda\'{n}sk, Poland\\
23: $^3$ Department of Physics, Technische Universit\"at Clausthal,
24: 38678 Clausthal-Zellerfeld, Germany\\
25: E-mail: mczachor@pg.gda.pl and Jan.Naudts@ua.ac.be
26: }
27:
28: \maketitle
29:
30: \begin{abstract}
31:
32: We discuss universality of response functions in systems with
33: excited degrees of freedom. We propose a unification of two
34: existing phenomenologies, two-power law decay and deviation from
35: power law due to non-extensivity. A universal curve is derived
36: by maximizing entropy with a non-linear constraint. The same
37: formalism can explain the universal
38: fluctuation curves which have been discovered recently by
39: Bramwell, Holdsworth, and Pinton.
40:
41: \end{abstract}
42:
43: \pacs{PACS numbers: 31.70.Hq, 05.40.-a, 05.70.Ce}
44:
45:
46: %+++++++++++++++++++++++++++++++++++++++++++++++++++++++++
47: %DECAY
48:
49: Many experiments measure the probability $p(t){\rm d}t$
50: that an event takes place within the time interval
51: $[t,t+{\rm d}t]$ after excitation of the system at $t=0$.
52: Classical arguments predict exponential decay.
53: We are interested here in a power law decay at large times
54: \be
55: p(t)\sim t^{-1/(1-\alpha)}
56: \label{pldecay}
57: \ee
58: with $0<\alpha<1$.
59:
60: There is an extensive literature explaining non-exponential decay
61: by means of fractal theory. In this domain of research it is accepted
62: that a scaling law like (\ref{pldecay}) holds only
63: for a limited range $t_1<t<t_2$ of the argument $t$.
64: For $t<t_1$ collective effects presumably restore non-fractal
65: behavior. For large $t$ individual behavior
66: on microscopic scale dominates and non-universal decay
67: is expected. A nearly ideal example of this behavior has been observed
68: recently in quantum dots \cite {KFH00}, with a power law extending
69: over a range of $10^5$ of time scales.
70: Many theoretical models explain power law decay as the
71: correct asymptotics in the large time limit.
72:
73: There is more and more evidence that also non-power law decay
74: can be described by a universal probability density function
75: (PDF).
76: An early explanation \cite {ABE74} of non-exponential decay
77: involves the notion
78: of an activation energy required to trigger the decay process.
79: These authors proposed the following modification
80: of (\ref{pldecay}) (using our notations)
81: \be
82: p(t)&\sim& (1+a(1-\alpha)\omega t)^{-1/(1-\alpha)}
83: \label{ffdecay}
84: \ee
85: with $a$ some positive constant, and $\omega$ a constant with
86: dimension of a frequency, introduced to
87: obtain the dimensionless combination $\omega t$.
88: Later on, (\ref{ffdecay}) lost importance in favor of stretched
89: exponential decay \cite {KR47}
90: \be
91: p(t)\sim e^{-\kappa (\omega t)^\gamma},
92: \ee
93: which is expected in case of processes with a scaling distribution
94: of relaxation rates. Here, we restrict ourselves to the case of
95: asymptotic decay with a single power law.
96:
97: In dielectric relaxation one observes usually
98: two regions of power law response (in our notations)
99: \be
100: p(t)&\sim& (\omega t)^{-1/(1-\alpha)} \qquad \omega t \ll 1\cr
101: &\sim& (\omega t)^{-1/(1-\rho)} \qquad \omega t \gg 1.
102: \label{twodecay}
103: \ee
104: (see e.g.~Weron and Jurlewicz \cite {WJ93,JW99}).
105: The point of view of the present paper is that both behaviors,
106: as described by (\ref{ffdecay}) and (\ref{twodecay}),
107: should be combined. More precisely, the generic behavior
108: in the region $\omega t \ll 1$ should be described by (\ref{ffdecay})
109: instead of (\ref{pldecay}). This substitution makes sense
110: because (\ref{ffdecay}) describes a curve which starts off
111: linearly and bends over towards a power law decay with exponent
112: $-1/(1-\alpha)$.
113:
114: \begin{figure}
115: \epsfxsize=8.25cm
116: \epsffile{fig1.eps}
117: \caption{Log-log-plot of $p(t)$ with $a=500$ for different
118: combinations of exponents $(\alpha,\rho)$:
119: (0.1,0.1) dotted, (0.1,1) solid, (0.1,0.8) short-dashed,
120: (0.1,1.08) long-dashed.}
121: \end{figure}
122:
123:
124: The actual formula, which we will derive below,
125: and which covers the whole fractal region, is (see Fig. 1)
126: \be
127: p(t)\sim
128: \left(1-a+a\left(
129: 1+(1-\rho)\omega t
130: \right)^\frac{1-\alpha}{1-\rho}
131: \right)^{-1/(1-\alpha)}.
132: \label{pdf}
133: \ee
134: Three regions can be distinguished.
135: In regions 1 and 2 the decay is as described by (\ref{ffdecay}).
136: Regions 2 and 3 show power law decay according to (\ref{twodecay}).
137: Note that $\rho>1$ corresponds to a super-exponential decay.
138:
139: %SPECIAL CASES
140: For $\rho=\alpha$ expression (\ref{pdf}) reduces to (\ref{ffdecay}).
141: If $a=1$ then it reduces to (\ref{ffdecay}) with $\alpha$
142: replaced by $\rho$.
143: In the limit $\rho=1$ one obtains
144: \be
145: p(t)\sim
146: \left(1-a+a\exp((1-\alpha)\omega t)
147: \right)^{-1/(1-\alpha)}.
148: \label {rendecay}
149: \ee
150: In the limit $\alpha=1$ one obtains
151: \be
152: p(t)\sim
153: (1+(1-\rho)\omega t)^{-a/(1-\rho)}.
154: \ee
155: In any case, in the limit $\alpha=\rho=1$
156: the decay is exponential $p(t)\sim \exp(-a\omega t)$.
157:
158: %EVIDENCE
159: We found first evidence for (\ref{pdf}) in the work
160: of Tsallis, Bemski, and Mendes \cite{TBM99}. They reanalyze
161: the experimental data of \cite {ABE74} and
162: observe that the description using
163: (\ref{ffdecay}) improves significantly by using
164: (\ref{rendecay}), formula which was introduced on an {\sl ad hoc}
165: basis. The full data set of \cite {ABE74} could be described
166: with it. The universality of (\ref{rendecay}) is further supported
167: by recent work of Montemurro \cite {MM01}
168: using a very convincing analysis of linguistic data.
169: In both papers \cite{TBM99} and \cite {MM01} there is evidence
170: that the asymptotic behavior for large $t$ still deviates from
171: (\ref{rendecay}), and follows a power law instead of decaying exponentially.
172: In the logic of the present paper, this is a cross-over from
173: one power law to another, as described by (\ref{twodecay}).
174: Hence (\ref{pdf}) should be used with $\rho\not=1$
175: instead of (\ref{rendecay}). In \cite {TBM99} an {\sl ad hoc} modification of
176: (\ref{rendecay}) was proposed, slightly different from (\ref{pdf}).
177: The same expression was used in \cite {MM01} and fits
178: the experimental data over the whole range of $t$-values.
179: It is clear that (\ref{pdf}) will fit as well, but has a number of
180: advantages:
181: 1) there is a general formalism, which is used to derive it;
182: 2) the PDF has a workable algebraic
183: expression, which is not the case for the formula
184: proposed in \cite {TBM99};
185: 3) within the context of the formalism an average time
186: $\langle\langle \omega t\rangle\rangle$ is defined
187: using a nonlinear average -- see (\ref{knav}) below.
188: The average time characterizes the PDF $p(t)$ together with the
189: two exponents $\alpha$ and $\rho$, and the cross-over frequency
190: $\omega$. It replaces the fitting parameter $a$ which has a less
191: clear physical meaning (see Fig.~2).
192: In particular, in the case $\rho=1$, we
193: show \cite {NC01} that $\langle\langle\cdot\rangle\rangle$ satisfies
194: a property of additivity, which implies that it is meaningful
195: to add average times of different experiments all having the
196: same values of $\alpha$ and $\omega$.
197:
198: \begin{figure}
199: \epsfxsize=8.25cm
200: \epsffile{fig2.eps}
201: \caption{Kolmogorov-Nagumo average time
202: $\langle\langle \omega t\rangle\rangle$ as a function of $1/a$
203: for several values of $(\alpha,\rho)$:
204: $(0.9999,0.9999)$ straight line, short-dashed;
205: $(0.999,0.9999)$ long-dashed;
206: $(0.999,1)$ solid line;
207: $(0.8,0.9)$ thick-dotted;
208: $(0.9,0.9999)$ thin-dotted.
209: }
210: \end{figure}
211:
212: Up to now we considered decay as a function of time. However the
213: PDF (\ref{pdf}) can also be considered as a generalization of
214: the distribution of Boltzmann-Gibbs. Time $t$ is then
215: replaced by energy $E$, frequency $\omega$ by a fixed inverse
216: temperature $\beta$. With this interpretation (\ref{ffdecay}) is
217: the equilibrium distribution of non-extensive thermostatistics
218: \cite {TC88}. The entropic parameter $q$ of this formalism,
219: coincides with the exponent $\alpha$ (the relation becomes
220: $q=2-\alpha$ in the recently
221: modified formalism \cite {TMP98}).
222: In the context of non-extensive thermostatistics
223: (\ref{ffdecay}) has numerous applications
224: --- for a review see \cite {TC99} or \cite {AO00}.
225: In this context
226: (\ref{rendecay}) becomes the equilibrium distribution
227: for thermostatistics based on R\'enyi's choice of Komogorov-Nagumo
228: averages and $\alpha$-entropies -- see \cite {NC01}.
229:
230: %+++++++++++++++++++++++++++++++++++++++++++++++++++++++++
231: %FLUCTUATIONS
232:
233: Let us now discuss universal distributions of fluctuating
234: quantities. Their universality has been discovered
235: recently by Bramwell, Holdsworth, and Pinton (BHP) \cite {BHP98,PH99},
236: based on earlier work \cite{LPF96}.
237: The result received ample support in the literature
238: \cite{HKG99,B00,AG01,BFH01,CRW01,DJ01,DM01,PHSB01,ADG01}.
239: The BHP density function is of the form
240: \be
241: g_{_{\rm BHP}}(\epsilon)\sim \exp\left(b(y-e^y\right)
242: \label {univspec}
243: \ee
244: with $y=c(\epsilon-u)$.
245: Setting $b=1$
246: one finds the famous Fisher-Tippett density
247: appearing in statistics of extremes (see Gumbel \cite {GEJ58}).
248: However, the value which empirically best describes
249: fluctuation spectra is $b=\pi/2$.
250:
251: Using the standard central limit theorem (CLT) we show at the end of this
252: Letter that our formalism implies (for $\rho>2/3$)
253: \be
254: {}&{}&
255: \tilde g(\epsilon)=\Big(\exp_\rho\big[v
256: +\epsilon\big]\Big)^{\rho-\alpha}\nonumber\\
257: &{}&\phantom{===}\times
258: g\Big(\frac{\sqrt{N}}{\sigma}\big[\phi_{\alpha\rho}(v+\epsilon)-
259: \phi_{\alpha\rho}(v)
260: \big]\Big).\label{flucspec}
261: \ee
262: with $\sigma^2$ the variance of $p(t)$,
263: $g(x)$ the normal density, and the functions
264: $\phi_{\alpha\rho}$ and $\exp_\rho$ defined later in the text
265: (for $\rho<2/3$
266: the normal density function $g(\epsilon)$ should be replaced by a
267: L\'evy-stable PDF).
268:
269: Of particular interest is the limit $\rho=1$ (corresponding to R\'enyi
270: $\alpha$-entropies -- see further on).
271: Then (\ref{flucspec}) can be written as
272: \be
273: \tilde g(\epsilon)\sim\exp\left(y-\frac{1}{2}d^2(e^y-1)^2\right)
274: \label{fluctext}
275: \ee
276: with $y=(1-\alpha)\epsilon$ and
277: $d=\exp((1-\alpha)v)/N^{1/2}\sigma(1-\alpha)$.
278: This expression differs in an essential way from (\ref{univspec}).
279: However, both PDFs have a similar shape (see Fig.~3).
280: Note that (\ref{univspec}) has an additional
281: fitting parameter, the role of which is taken over here
282: by the exponent $\rho$. Very accurate
283: data will be needed to distinguish the two PDFs
284: on an experimental basis.
285: An advantage of (\ref{fluctext}) is that in the limit
286: $\alpha=1$ it reduces to the normal density with width
287: $\sigma$, as expected.
288: Note also that, if $g(\epsilon)$ were replaced by the exponential density,
289: and $\rho=1$, then (\ref {flucspec}) reduces to (\ref {univspec})
290: with $b=1$ and $c=1-\alpha$.
291: However, the exponential density is {\sl not}
292: L\'evy-stable. Therefore we do not expect it on thermodynamic grounds.
293:
294:
295: \begin{figure}
296: \epsfxsize=8.25cm
297: \epsffile{fig3.eps}
298: \caption{
299: Comparison of $\tilde g(\epsilon)$
300: and $g_{_{\rm BHP}}(\epsilon)$:
301: $\tilde g(\epsilon)$ with $\rho=0.8$ dotted,
302: with $\rho=0.9$ short-dashed,
303: with $\rho=1$ long-dashed;
304: $g_{_{\rm BHP}}(\epsilon)$ solid;
305: the curves are normalized to 1 in the origin;
306: parameters of $\tilde g(\epsilon)$ are $v=1$, $\alpha=0.5$, and
307: $\sigma^2=4N$;
308: the parameters of $g_{_{\rm BHP}}(\epsilon)$ have been adapted to fit roughly
309: $\tilde g(\epsilon)$ with $\rho=1$.
310: }
311: \end{figure}
312:
313: %+++++++++++++++++++++++++++++++++++++++++++++++++++++++++
314: %THEORY
315:
316: In the final part of this Letter we derive formulas (\ref{pdf})
317: and (\ref{flucspec}) using thermodynamic arguments.
318: Weron and Jur\-le\-wicz \cite {WJ93,JW99} express the response function $p(t)$
319: in terms of PDFs $p_i(t)$, describing relaxation
320: of individual dipoles. To do this, they use statistics of extremes \cite {GEJ58}.
321: The underlying assumption is that the whole system relaxes as soon
322: as one of the individual dipoles relaxes. Our assumptions are not
323: so drastic. We start with a thermodynamic argument,
324: and replace statistics of extremes by the usual CLT.
325:
326: About 20 years ago Montroll and Shlesinger \cite {MS83}
327: proposed a thermodynamic derivation of Einstein's
328: diffusion law. It follows by optimizing entropy of the PDF $p(t)$
329: under the constraint of zero first moment and fixed second moment.
330: They observed that anomalous diffusion can be explained
331: along similar lines using an {\sl ad hoc} constraint
332: replacing the constraint on the second moment.
333: Alemany and Zanette \cite {AZ94,ZA95},
334: followed by Tsallis et al \cite {TLS95,PT99},
335: repeat these arguments,
336: replacing Shannon's entropy by the generalized entropy
337: used in nonextensive thermodynamics \cite {TC88}
338: \be
339: S_\alpha(p)=\int\hbox{ d}t\,\frac{p(t)^\alpha-p(t)}{1-\alpha},
340: \qquad \alpha>0,\alpha\not=1.
341: \label {tsallis}
342: \ee
343: These authors introduce a suitably adapted constraint on the second moment.
344: Recently \cite {NC01}, the present authors generalized
345: nonextensive thermostatistics by founding it on nonlinear
346: Kolmogorov-Nagumo averages \cite{KA30,NM30}
347: \be
348: \langle\langle f\rangle\rangle
349: =\phi^{-1}\left(\int\hbox {d}t\,p(t)\phi(f(t))\right).
350: \label{knav}
351: \ee
352: This average depends on a monotonically increasing function $\phi(x)$.
353: In the present paper we choose $\phi=\phi_{\alpha\rho}$ with
354: \be
355: \phi_{\alpha\rho}(x)&=&\ln_\alpha(\exp_\rho(x))\cr
356: &=&\frac{1}{1-\alpha}\left[
357: (1+(1-\rho)x)^{(1-\alpha)/(1-\rho)}-1
358: \right].
359: \ee
360: The $\alpha$-deformed exponential and logarithmic functions are
361: defined by \cite {TC94,BEP98}
362: \be
363: \exp_\alpha (x)&=&\left[1+(1-\alpha)x\right]^{1/(1-\alpha)}\cr
364: \ln_\alpha (x)&=&\frac{x^{1-\alpha}-1}{1-\alpha}.
365: \ee
366: The inverse function is $\phi_{\rho\alpha}(x)$.
367: Following \cite {NC01}, the corresponding definition of entropy is
368: \be
369: S_{\alpha\rho}(p)=\phi_{\rho\alpha}\left(\int\hbox{ d}t\,p(t)\ln_\alpha(1/p(t))
370: \right).
371: \ee
372: In the limit $\rho=1$ this is the $\alpha$-entropy of R\'enyi \cite {RA60}.
373: \be
374: S_{\alpha 1}(p)=\frac{1}{1-\alpha}\ln\left(
375: \int\hbox{ d}t\,p(t)^\alpha
376: \right).
377: \ee
378: On the other hand, one has $\phi_{\alpha\alpha}(x)=x$ so that
379: $S_{\alpha\alpha}(p)$ coincides with (\ref{tsallis}).
380: Let us now optimize entropy $S_{\alpha\rho}(p)$ under the constraint
381: that the average decay time $\langle\langle \omega t\rangle\rangle$
382: has a given value. A straightforward calculation using
383: Lagrange parameters produces an implicit expression
384: for the PDF $p(t)$. By introducing a free parameter $a$
385: the latter can be made explicit, resulting in (\ref{pdf})
386: (see \cite{NC01}).
387:
388: In order to calculate the fluctuation spectrum assume that $\rho>2/3$
389: (then the second moment of $p(t)$ exists and CLT
390: holds). If $A(t)$ is an arbitrary random variable we will denote its {\it
391: linear\/} (experimental and theoretical) averages by
392: \be
393: \langle A\rangle_N
394: =
395: \sum_{k=1}^N A(t_k)/N,\quad
396: \langle A\rangle
397: =
398: \int \hbox{ d}t\, p(t)A(t)\nonumber
399: \ee
400: Denote by $P(x)$ the probability of an event $x$,
401: given the PDF $p(t)$. In this notation CLT can
402: be formulated as
403: \be
404: P\Big(\langle A\rangle_N\leq z\Big)
405: &\approx&
406: F\Big(\frac{\sqrt{N}}{\sigma}(z-\langle A\rangle\Big)
407: \ee
408: with $\sigma^2$ the variance of $p(t)$,
409: $F$ the normal distribution function,
410: and $N$ sufficiently large \cite{Feller}.
411:
412: The nonlinear
413: averages
414: \be
415: \langle\langle A\rangle\rangle_N
416: =
417: \phi^{-1}
418: \Big(\langle \phi(A)\rangle_N\Big),\quad
419: \langle\langle A\rangle\rangle
420: =
421: \phi^{-1}
422: \Big(\langle \phi(A)\rangle\Big)\nonumber
423: \ee
424: are related via $\phi^{-1}$
425: to the linear ones of the random variable
426: $B(t)=\phi\big(A(t)\big)$. Since $\phi$ is monotonically increasing we can
427: write
428: \be
429: P\Big(\langle\langle A\rangle\rangle_N\leq \langle\langle A\rangle\rangle
430: +\epsilon\Big)
431: &=&
432: P\Big(\langle \phi(A)\rangle_N\leq \phi\big(\langle\langle A\rangle\rangle
433: +\epsilon\big)\Big)\nonumber
434: \ee
435: and apply CLT to its right side. One finds
436: \be
437: {}&{}&
438: P\Big(\langle\langle A\rangle\rangle_N\leq \langle\langle
439: A\rangle\rangle
440: +\epsilon\Big)\nonumber\\
441: &{}&\phantom{==}
442: \approx
443: F\Big(\frac{\sqrt{N}}{\sigma}\big[\phi\big(\langle\langle A\rangle\rangle
444: +\epsilon\big)-
445: \phi\big(\langle\langle A\rangle\rangle\big)\Big).
446: \ee
447: Introduce a limiting PDF $\tilde g(\epsilon)$ by
448: \be
449: \frac{d}{d\epsilon}
450: P\Big(\langle\langle A\rangle\rangle_N\leq
451: \langle\langle A\rangle\rangle +\epsilon\Big)
452: &\approx&
453: \frac{\sqrt{N}}{\sigma}\tilde g(\epsilon).
454: \ee
455: Using $g(x)=dF(x)/dx$ and choosing $\phi=\phi_{\alpha\rho}$ we finally
456: obtain
457: \be
458: {}&{}&
459: \tilde
460: g(\epsilon)=\Big(\exp_\rho\big[\langle\langle
461: A\rangle\rangle
462: +\epsilon\big]\Big)^{\rho-\alpha}
463: \nonumber\\
464: &{}&\phantom{===}\times
465: g\Big(\frac{\sqrt{N}}{\sigma}\big[\phi\big(\langle\langle A\rangle\rangle
466: +\epsilon\big)-
467: \phi\big(\langle\langle A\rangle\rangle\big)
468: \big]\Big).
469: \ee
470: It is clear that for $\alpha=\rho$ one gets
471: $\tilde g(\epsilon)=g(\sqrt{N}\epsilon/\sigma)$.
472: Setting
473: $\langle\langle A\rangle\rangle=v$ we obtain
474: (\ref{flucspec}).
475:
476: %+++++++++++++++++++++++++++++++++++++++++++++++++++++++++
477: %SUMMARY
478:
479: Let us summarize the results. Applying the standard formalism
480: of maximum entropy
481: principles but formulated in terms of nonlinear averages we can explain a
482: large class of
483: response functions with asymptotic power law decay, including the
484: ubiquitous two-power law. Moreover, using the same formalism
485: and standard central limit theorem we obtain a
486: universal behavior of fluctuation spectra. In both cases the universal
487: laws we propose differ slightly from the known ones. To
488: distinguish experimentally between them one needs more
489: precise experiments.
490:
491:
492: %+++++++++++++++++++++++++++++++++++++++++++++++++++++++++
493: %ACKNOWLEDGEMENTS
494:
495:
496: The stay of MC at Universiteit Antwerpen, UIA, was made possible
497: by NATO through a research fellowship. Part of the results were
498: obtained during our visit at Technical University of Clausthal.
499: We are indebted to Prof. H. D. Doebner for his comments and his
500: interest in this work. MC thanks the Alexander von Humboldt
501: Foundation for support.
502:
503: %+++++++++++++++++++++++++++++++++++++++++++++++++++++++++
504: %BIBLIOGRAPHY
505:
506: \begin{references}
507:
508: \bibitem {KFH00} M. Kuno, D.P. Fromm, H.F. Hamann, A. Gallagher,
509: and D.J. Nesbitt, J. Chem. Phys. {\bf 112}(7), 3117 (2000).
510: \bibitem{ABE74} R.H. Austin, K. Beeson, L. Eisenstein, H. Frauenfelder,
511: I.C. Gunsalus, and V.P. Marshall, Phys. Rev. Lett. {\bf 32}, 403 (1974).
512: \bibitem{KR47} R. Kohlrausch, Ann. Phys. 12, 392 (1847).
513: \bibitem{WJ93} K Weron and A Jurlewicz,
514: J. Phys. A{\bf 26}, 395-410 (1993).
515: \bibitem {JW99} A. Jurlewicz and K. Weron,
516: Cell. Mol. Biol. Lett. {\bf 4}(1), 55 (1999).
517: \bibitem{TBM99} C. Tsallis, G. Bemski, R.S. Mendes,
518: Phys. Lett. A{\bf 257}, 93 (1999).
519: \bibitem{MM01} M. Montemurro, cond-mat/0104066.
520: \bibitem{NC01} J. Naudts and M. Czachor, cond-mat/0106324.
521: \bibitem{TC88} C. Tsallis,
522: J. Stat. Phys. {\bf 52}, 479 (1988).
523: \bibitem{TMP98} C. Tsallis, R.S. Mendes, A.R. Plastino,
524: Physica A{\bf 261}, 543 (1998).
525: \bibitem {TC99} C. Tsallis, Braz. J. Phys. {\bf 29}, 1 (1999).
526: \bibitem {AO00} {\sl Nonextensive statistical mechanics
527: and its applications,} eds. S. Abe and Y. Okamoto,
528: Lecture Notes in Physics 560
529: (Springer Verlag, 2001)
530: \bibitem {BHP98} S.T. Bramwell, P.C.W. Holdsworth, and J.-F. Pinton,
531: Nature {\bf 396}, 552 (1998).
532: \bibitem {PH99} J.-F. Pinton, P.C.W. Holdsworth,
533: Phys. Rev. E, 60, R2452, (1999).
534: \bibitem {LPF96} R. Labbe, J.-F. Pinton, S. Fauve,
535: J. Phys. II, France, 6, 1099, (1996).
536: \bibitem {HKG99} J. Harte, A. Kinzig, and J. Green,
537: Science {\bf 284}(9), 334 (1999).
538: \bibitem{B00} S.T. Bramwell, K. Christensen, J.Y. Fortin, P.C.W. Holdsworth,
539: H.J. Jensen, S. Lise, J.M. Lopez, M. Nicodemi, J.F. Pinton, and M. Sellitto,
540: Phys. Rev. Lett. {\bf 84}, 3744 (2000).
541: \bibitem {AG01} V. Aji and N. Goldenfeld,
542: Phys. Rev. Lett. {\bf 86}, 1007 (2001).
543: \bibitem {BFH01} S.T. Bramwell, J.Y. Fortin, P.C.W. Holdsworth,
544: S. Peysson, J.F. Pinton, B. Portelli, and M. Sellitto,
545: Phys. Rev. E{\bf 63}, 041106 (2001)
546: \bibitem {CRW01} S.C. Chapman, G. Rowlands, and N.W. Watkins,
547: cond-mat/0007275.
548: \bibitem {DJ01} K. Dahlstedt and H.J. Jensen, cond-mat/0108007.
549: \bibitem {DM01} D.S. Dean and S.N. Majumdar, cond-mat/0104028.
550: \bibitem {PHSB01} B. Portelli, P.C.W. Holdsworth, M. Sellitto
551: and S.T. Bramwell, cond-mat/0102283, to appear in Phys. Rev. E..
552: \bibitem{ADG01} T. Antal, M. Droz, G. Gy\"orgyi, and Z. R\'acz,
553: cond-mat/0105599.
554: \bibitem{GEJ58} E.J. Gumbel,
555: {\sl Statistics of extremes} (Columbia Press, 1958)
556: \bibitem {MS83} E.W. Montroll and M.F. Shlesinger,
557: J. Stat. Phys. {\bf 32}, 209 (1983).
558: \bibitem {AZ94} P.A. Alemany and D.H. Zanette,
559: Phys. Rev. E{\bf 49}, R956 (1994).
560: \bibitem {ZA95} D.H. Zanette and P.A. Alemany,
561: Phys. Rev. Lett. {\bf 75}, 366 (1995).
562: \bibitem {TLS95} C. Tsallis, S.V.F. Levy, A.M.C. Souza,
563: and R. Maynard, Phys. Rev. Lett. {\bf 75}, 3589 (1995).
564: \bibitem {PT99} D. Prato and C. Tsallis,
565: Phys. Rev. E{\bf 60}, 2398 (1999).
566: \bibitem{KA30}A. Kolmogorov, Atti della R. Accademia Nazionale dei
567: Lincei {\bf 12}, 388 (1930).
568: \bibitem{NM30}M. Nagumo, Japan. Journ. Math. {\bf 7}, 71 (1930).
569: \bibitem{TC94} C. Tsallis, Quimica Nova {\bf 17}, 468 (1994).
570: \bibitem{BEP98} E.P. Borges, J. Phys. A {\bf 31}, 5281 (1998).
571: \bibitem{RA60} A. R\'enyi, MTA III. Oszt. K\"ozl. {\bf 10}, 251 (1960);
572: reprinted in {\it Selected Papers of Alfr\'ed R\'enyi,\/} vol 2.
573: (Akad\'emiai Kiad\'o, Budapest, 1976), pp. 526-552.
574: \bibitem{Feller}W. Feller, {\it An Introduction to Probability Theory and
575: its Applications\/}, vol. 1, second edition, (Wiley, New York, 1957).
576: \end{references}
577:
578: \end{document}
579: