1: \documentstyle[12pt]{article}
2: \textwidth=150mm
3: \textheight=210mm
4: \topskip 0cm
5: \topmargin 0cm
6: \oddsidemargin 5mm
7: \newcommand{\gsim}{ \raisebox{-.5ex}{\mbox{$\,\stackrel{>}{\sim}\,$}} }
8: \newcommand{\lsim}{ \raisebox{-.5ex}{\mbox{$\,\stackrel{<}{\sim}$\,}} }
9: \begin{document}
10: \baselineskip 8mm
11:
12: \begin{center}
13: {\large\bf BIG ENTROPY FLUCTUATIONS\newline
14: IN STATISTICAL EQUILIBRIUM:\newline
15: THE MACROSCOPIC KINETICS}\\[5mm]
16: Boris Chirikov and Oleg Zhirov\\
17: {\it Budker Institute of Nuclear Physics \\
18: 630090 Novosibirsk, Russia}\\[1mm]
19: chirikov @ inp.nsk.su\\
20: zhirov @ inp.nsk.su\\[5mm]
21: \end{center}
22:
23: %\baselineskip 8mm
24:
25: \begin{abstract}
26: Large entropy fluctuations in an equilibrium steady state
27: of classical mechanics
28: were studied
29: in extensive numerical experiments on a simple 2--freedom strongly chaotic
30: Hamiltonian model
31: described by the modified Arnold cat map.
32: The rise and fall of a large separated fluctuation was shown to be described by
33: the (regular and stable) "macroscopic" kinetics both fast (ballistic) and
34: slow (diffusive). We abandoned a vague problem
35: of "appropriate" initial conditions
36: by observing (in a long run)
37: spontaneous birth and death of arbitrarily big fluctuations
38: for any initial state of our dynamical model.
39: Statistics of the infinite chain of fluctuations,
40: reminiscent to the Poincar\'e recurrences, was shown to be
41: Poissonian. A simple empirical relation for the mean period
42: between the fluctuations (Poincar\'e "cycle") has been found
43: and confirmed in numerical experiments.
44: A new representation of the entropy via the variance of only
45: a few trajectories ("particles") is proposed which greatly
46: facilitates the computation, being at the same time fairly
47: accurate for big fluctuations.
48: The relation of our results to a long standing debates over statistical
49: "irreversibility" and the "time arrow" is briefly discussed too.
50: \end{abstract}
51:
52: \vspace{5mm}
53:
54: PACS numbers: 05.45.+b, 05.40.+j \\
55: {\it Key words:} Chaos; Entropy; Statistical equilibrium; Fluctuations;
56: Poincar\'e recurrences
57:
58: \newpage
59:
60: %\vspace{5mm}
61:
62: \section{Introduction:\newline
63: macroscopic vs. microscopic fluctuations}
64:
65: \hspace*{\parindent} Fluctuations are inseparable part of
66: the statistical laws.
67: This is well known since Boltzmann.
68: What is apparently less known are the peculiar properties of rare
69: big fluctuations
70: (BF) as different from, and even opposite in a sense to,
71: those of small stationary
72: fluctuations.
73: In this paper we consider the simplest type of chaotic dynamical systems,
74: namely a finite--freedom Hamiltonian system which admits the (stable)
75: {\it statistical equilibrium} (SE). This class of dynamical models is
76: still popular (since Boltzmann!)
77: in debates over the dynamical foundations of statistical mechanics
78: (see, e.g., "Round Table on Irreversibility" in \cite{1}, and \cite{2}).
79:
80: A fairly simple picture of BF in such systems
81: is well understood by now, though not yet
82: well known. To Boltzmann such a picture was the basis of his fluctuation
83: hypothesis for our Universe. Again, as is well understood by now such a
84: hypothesis is completely incompatible with the present structure of the
85: Universe as it would immediately imply the notorious "heat death" (see, e.g.,
86: \cite{3}). For this reason, one may even term such systems the {\it heat
87: death models}. Nevertheless, they can be and actually are widely used
88: in the description
89: and study of local statistical processes in {\it thermodynamically} closed
90: systems. The latter term means the absence of any heat exchange with the
91: environment. Notice, however, that under conditions of the exponential
92: instability of motion, typical for chaotic systems,
93: the only {\it dynamically} closed system would be the "whole Universe".
94: Particularly, this excludes
95: the hypothetical "velocity reversal" also popular in debates
96: over "irreversibility" since Loschmidt
97: (for discussion see, e.g., \cite{4}).
98:
99: In any case, dynamical models with SE do not tell us the whole story of
100: either the Universe or even a typical macroscopic process therein.
101: The principal solution of this problem, unknown to Boltzmann, is quite
102: clear by now, namely, the "equilibriumfree" models are wanted.
103: Various classes of such models are intensively studied today.
104: Moreover, the celebrated cosmic microwave background tells us that
105: our Universe {\it was born} already in the state of a heat death which,
106: however, fortunately to us all became {\it unstable}
107: due to the well--known Jeans gravitational instability
108: \cite{5}. This resulted in developing of a rich variety of collective
109: processes, or {\it synergetics}, the term recently introduced or,
110: better to say, put in use by Haken \cite{6}.
111: The most important peculiarity of such a collective instability is in that
112: the total overall relaxation (to somewhere ?) with ever increasing total
113: entropy
114: is accompanied by an also increasing phase space {\it inhomogeneity}
115: of the system, particularly in temperature.
116: In other words, the whole system as well as its local parts become
117: more and more
118: {\it nonequilibrium} to the extent of the birth of a {\it secondary} dynamics
119: which may be, and is sometime, as perfect as, for example, the celestial
120: mechanics
121: (for general discussion see, e.g., \cite{4,7,8}).
122:
123: We stress that all these inhomogeneous nonequilibrium structures are not
124: BF like in SE but are a result of regular collective instability,
125: so that they are immediately formed under a certain condition.
126: Besides, they are typically {\it dissipative structures} in Prigogine's
127: term \cite{9} due to exchange of energy and entropy with the
128: {\it infinite} environment. The latter is the most important feature of such
129: processes, and at the same time the main difficulty in studying the
130: dynamics of those
131: models both theoretically and in numerical experiments which are so much
132: simpler for SE systems.
133:
134: In the latter case a BF consists of the two symmetric parts:
135: the rise
136: of a fluctuation followed by its return, or relaxation,
137: back to SE (see Figs.1 and 2 below).
138: Both parts are described by the same kinetic (e.g., diffusion) equation,
139: the only difference being in the sign of time. This relates the
140: time--symmetric dynamical equations to the time--antisymmetric
141: {\it kinetic} (but not statistical!) equations. The principal difference
142: between the both, some times overlooked, is in that the kinetic equations
143: are widely understood as describing the relaxation only, i.e.
144: {\it increase} of the entropy in a closed system, whereas in fact they do so
145: (at least in SE)
146: for the rise of BF as well, i.e. for the entropy {\it decrease}.
147: All this was qualitatively known already to Boltzmann \cite{10}.
148: The first simple example of symmetric BF
149: was considered by Schr\"odinger \cite{11}. Rigorous mathematical theorem
150: for the diffusive (slow) kinetics was proved by Kolmogorov in 1937
151: in the paper entitled "Zur Umkehrbarkeit der statistischen Naturgesetze"
152: ("Concerning reversibility of statistical laws in nature")
153: \cite{12} (see also
154: \cite{13}). Regrettably, the principal Kolmogorov theorem still remains
155: unknown to both the participants of hot debates over "irreversibility"
156: as well as the physicists actually studying
157: such BF (see, e.g., \cite{14}).
158:
159: By now, there exists the well developed ergodic theory of dynamical
160: systems (see, e.g., \cite{15}). Particularly, it proves that the relaxation
161: (correlation decay, or mixing) proceeds eventually in both directions of time
162: for almost any initial conditions of a chaotic dynamical system.
163: However, the relaxation does not need to be always monotonic
164: which simply means
165: a BF on the way, depending on the initial conditions.
166: To get rid of such an apparently confusing (to many) "freedom" we take
167: a different approach to the problem: instead discussing the "true" initial
168: conditions and/or a "necessary" restriction of those we start our numerical
169: experiments at {\it arbitrary} initial conditions
170: (most likely corresponding to SE), and do {\it observe} what the dynamics
171: and statistics of BF are like. Of course, such an approach is based on
172: a fundamental hypothesis that all the statistical laws are contained in,
173: and can be principally derived from, the underlying fundamental (Hamiltonian)
174: dynamics. To the best of our knowledge, there is as yet no contradiction
175: to this principal hypothesis. Notice, however, that such an approach
176: can be directly
177: applied to the fluctuations in finite systems
178: with statistical
179: equilibrium only (for discussion see \cite{4} and \cite{16}).
180: In such, and only such, systems infinitely many BF grow up spontaneously
181: {\it independent} of the initial conditions of the motion. This is similar
182: to the well--known Poincar\'e recurrences (for farther discussion see
183: Section 4 below).
184:
185: In spite of the essential restrictions the simple
186: SE models allow us to better understand the mechanism and role of BF in the
187: statistical physics. Besides removal of a vague problem of the initial
188: conditions such models help a lot in clarifying the relation between
189: macroscopic and microscopic description of chaotic systems.
190: Particularly, spontaneous rise of a BF out of SE is a {\it macroscopic} event
191: as well as its subsequent relaxation back to SE, even in a few--freedom
192: system. Like other macroscopic processes the BF
193: are not only perfectly regular by themselves but also
194: surprisingly stable against any perturbations, both regular and chaotic.
195: Moreover, the perturbations
196: do not need to be small. At first glance, it looks very strange
197: in a chaotic, highly unstable, dynamics. The resolution of
198: this apparent paradox is in that the dynamical instability of motion
199: does affect the BF instant of time only.
200: As to the BF evolution, it is determined by the kinetics whatever its
201: mechanism,
202: from purely dynamical one, like in model (2.2) we will use in this paper,
203: to a completely noisy (stochastic) one.
204: As a matter of fact, the fundamental Kolmogorov theorem \cite{12} is related
205: just to the latter case but remains valid in a much more general situation.
206: Surprising stability of BF is similar to the less known
207: conception of
208: robustness for the Anosov (strongly chaotic) systems \cite{17} whose
209: trajectories get only slightly deformed under a small perturbation
210: (for discussion see \cite{4}).
211:
212: In the present paper we consider a particular type of BF which is characterized
213: by a large concentration of "particles" in a small
214: phase space domain of a dynamical system. In other
215: words, "our" fluctuations are localized in phase space and
216: separated in time. A more accurate definition of these
217: fluctuations will be given below in Section 3 (see Eq.(3.6)).
218: The same fluctuations
219: in a stochastic
220: model (with noise) were studied in detail in \cite{14}.
221: There exist, of course, many other fluctuations
222: with their own peculiarities (see, e.g., \cite{18}).
223: The primary goal of our studies was the macroscopic kinetics
224: of big fluctuations on the background of small stationary
225: microscopic fluctuations. A brief outline of our results
226: was presented in \cite{16}.
227:
228: \section{A Hamiltonian model:\\
229: most simple but strongly chaotic}
230:
231: The systems with SE allow for
232: very simple models in both the theoretical analysis as well as numerical
233: experiments which the latter are even more important for us.
234: In the present paper we will use one of the most simple and popular model
235: specified
236: by the so--called Arnold cat map (see \cite{19,20}):
237: $$
238: \begin{array}{ll}
239: \overline{p}\,=\,p\,+\,x & mod\ 1 \\
240: \overline{x}\,=\,x\,+\,\overline{p} & mod\ 1
241: \end{array}
242: \eqno (2.1)
243: $$
244: which is a linear canonical map on a unit torus. It has no parameters, and is
245: chaotic and even ergodic. The rate of the local exponential instability,
246: the Lyapunov exponent $\lambda =\ln{(3/2+\sqrt{5}/2)}=0.96$, implies
247: a fast (ballistic) kinetics with relaxation time $t_r\sim 1/\lambda\approx 1$.
248: Throughout the paper $t$ denotes the time in map's iterations.
249:
250: A minor modification of this map:
251: $$
252: \begin{array}{ll}
253: \overline{p}\,=\,p\,+\,x\,-\,1/2 & mod\ C \\
254: \overline{x}\,=\,x\,+\,\overline{p}\,-\,C/2 & mod\ 1
255: \end{array}
256: \eqno (2.2)
257: $$
258: where $C$ is a circumference of the phase space torus allows us
259: to study both a fast (exponential) ballistic kinetics ( for $C=1$)
260: as well as the slow
261: (diffusive) relaxation in $p$ (for $C\gg 1$) with characteristic time
262: $t_p\sim C^2/4D_p\gg 1$ where $D_p=1/12$ is
263: the diffusion rate in $p$.
264: In contrast to slow diffusion
265: in $p$, the relaxation time in $x$ does not depend on $C$ ($t_r\sim 1$)
266: so that subsequent values of $x$ are practically uncorrelated.
267: The map (2.2) has the (unstable) fixed point at
268: $x=x_0=1/2\,,\ p=p_0=C/2$.
269:
270: A convenient characteristic of BF size is rms volume (area) in 2D phase space
271: ($x,p$)
272: $$
273: \sigma (t)=\sigma_{p}(t)\cdot \sigma_{x}(t) \eqno (2.3)
274: $$
275: occupied by a group of $N$ trajectories (particles).
276: In ergodic motion at equilibrium $\sigma =\sigma_0=C/12$.
277: Due to a severe restriction to small $N\lsim 10$ in the numerical experiments
278: (see below) we have to use simple (average) characteristics only like (2.3).
279: On the other hand, these are just the macroscopic variables we are
280: interested in.
281:
282: Below we also restrict ourselves to a particular case of BF with the fixed
283: prescribed position in the phase space:
284: $$
285: x_{fl}\,=\,x_0\,=\,{1\over 2}\,, \qquad p_{fl}\,=\,p_0\,=\,{C\over 2}
286: \eqno (2.4)
287: $$
288: Then, the variance of phase space size
289: $v=\sigma^2=\sigma_{p}^2\cdot \sigma_{x}^2$ is determined by the relation
290: $$
291: \sigma_{p}^2\,=\,\langle p^2\rangle\,-\,p_0^2\,, \qquad
292: \sigma_{x}^2\,=\,\langle x^2\rangle\,-\,x_0^2 \eqno (2.5)
293: $$
294: where brackets $\langle ...\rangle $ denote the averaging over
295: $N$ trajectories.
296: In ergodic motion at equilibrium $v=v_{SE}=C^2/12^2$. In what follows we will
297: use the dimensionless measure $\tilde{v}=v/v_{SE}\to v$,
298: and omit tilde. In diffusive approximation of the kinetic
299: equation the variable $v(t)$ is especially convenient as it
300: is varying in proportion to time. Moreover, $v\to v_p$ in
301: this case because of a quick relaxation $v_x\to 1$ in $x$.
302:
303: In all advantages of variable $v$ its relation to the fundamental conception
304: of entropy is highly desirable. The standard definition of the entropy,
305: which can be traced back to Boltzmann, reads:
306: $$
307: S\,=\,-\,\langle\,\ln{f(x,p)}\,\rangle\,+S_0 \eqno (2.6)
308: $$
309: where $f(x,p)$ is a coarse--grained distribution function, or the phase--space
310: density, and $S_0$ an arbitrary constant to be fixed later.
311: Notice that the distribution calculated from any {\it finite}
312: number of trajectories is always
313: a coarse--grained one.
314: However, the direct application of Eq.(2.6) requires too many trajectories,
315: especially for BF of a small size. Nevertheless, just in the latter case,
316: which is the main problem under consideration, we have found
317: a simple approximate relation
318: $$
319: S(t)\,\approx\,{1\over 2}\ln{v(t)} \eqno (2.7)
320: $$
321: which, at least, gives a rough estimate for the entropy evolution \cite{16}.
322: Moreover, if the distribution is Gaussian
323: $$
324: f(x,p)\,\to\,f(p)\,=\,\frac{\exp{(-(p-p_0)^2/2v)}}{\sqrt{2\pi v}}
325: \eqno (2.8)
326: $$
327: the estimate (2.7) becomes exact as it is directly followed
328: from the definition of the entropy (2.6).
329: Comparison of both relations for the entropy will be considered
330: by the end of next Section for a typical BF.
331:
332: A great advantage of (2.7) is
333: in that the computation of $S$ does not require very many trajectories
334: as does the distribution function. In fact, even a single trajectory is
335: sufficient as Fig.1 in \cite{16} and below demonstrate !
336:
337: A finite number of trajectories used for calculating the variance $v$
338: is a sort of the coarse--grained distribution,
339: as required in relation (2.6), but with a free bin size which can be
340: arbitrarily small.
341:
342: Now we may turn to the numerical experiments.
343:
344: \section{Macroscopic kinetics:\\
345: complete, regular, and stable}
346:
347: In this Section we consider the regular BF kinetics.
348: The data were obtained from simultaneous running of $N$ trajectories for
349: very
350: long time in order to collect sufficiently many BF for reliable separation of
351: the regular part of BF, or the kinetic subdynamics in Balescu's term
352: (see \cite{21} and references therein), from the stationary fluctuations.
353: The separation was done by the plain
354: averaging of individual $v_i$ values ($i=1,...,n$) over all $n$ BF collected
355: in a run.
356:
357: The size of BF chosen for further analysis was
358: fixed by the condition that current
359: $$
360: v(t) < v_b \eqno (3.1)
361: $$
362: at some time instant $t\approx t_i$, the moment of BF. Here a prescribed
363: $v_b$ is the main input parameter of the run.
364: This condition
365: determines, in fact, the border of the whole fluctuation
366: domain (FD): $0 < v < v_b$.
367:
368: The event of entering FD is the macroscopic "cause" of BF whose obvious
369: "effect" {\it will be subsequent} relaxation to the equilibrium.
370: However, and this is the main point of our study, the second "effect"
371: of the same "cause" {\it was preceding} rise of BF in apparent contradiction
372: with the "causality principle" (for discussion see \cite{16} and Section 4
373: below). In any event, the second effect requires the permanent memory of
374: trajectory segments within some time window $w$ which is another important
375: input parameter of the run.
376:
377: The exact procedure of data processing during the run was as follows.
378: Starting from arbitrary (random) initial conditions the selection rule (3.1)
379: is checked at each iteration. Suppose, it is satisfied at some instance
380: $t_{in}$ when the bundle of trajectories enters FD.
381: In the first approximation we could consider it as the fluctuation
382: maximum (or variance minimum):
383: $t_i=t_{in}$ where sub $i$ is the number of current fluctuation
384: in a run. However, such a simple procedure would cause asymmetry with
385: respect to $t=t_i$. A better choice were the rule: $t_i=(t_{in}+t_{out})/2$
386: where $t_{out}$ is the exit moment of time from FD.
387: Instead, we have accepted the following, a more complicated procedure which, as we hope, better restores the true BF symmetry.
388: Starting from the moment $t_{in}$ we search for
389: the minimum of $v(t)$ inside a rather large interval
390: $t_{in}< t < t_{in}+w$.
391: If a minimum is found at some $t=t_{min}$, we check that it is the minimum inside
392: the next interval $t_{min}< t < t_{min}+ w$ too. If so, we
393: identify this minimum with
394: the BF top, and set: $t_i=t_{min}$, otherwise we put $t_{min}$ to the time of
395: better minimum and repeat the last step again. Obviuosly, the parameter $w$
396: should be small compared to $\langle P\rangle$, the mean period of BF, but
397: sufficiently long for the trajectory to leave the FD (3.1).
398: Typically,
399: $w\gsim C^2$, the total diffusion time, was chosen.
400: After fixing the current $t_i$ value the computation within
401: the interval $t_i<t<t_i+w$ had been completed, and only then
402: the search for the next BF was continued.
403:
404: As was already mentioned above, there are two rather simple limiting cases
405: of generally very complicated kinetics, namely the fast (ballistic) and
406: slow (diffusive) ones. An example of both in one run
407: for $N=1$ (!) is presented in Fig.1 for two fluctuations of
408: different size.
409: In this case the general condition (3.1)
410: was checked separately for $p$ and $x$:
411: $$
412: v_p(t)\,<\,v_{pb} \quad and \quad v_x(t)\,<\,v_{xb} \eqno (3.2)
413: $$
414: with $v_{pb}=v_{xb}\sim 10^{-5}$ and $v_b=v_{pb}\cdot v_{xb}\sim 10^{-10}$.
415:
416: The fast part of kinetics is approximately described as
417: $$
418: v(\tau )\,\approx\,v(0)\cdot\exp{(4\lambda\tau )} \eqno (3.3)
419: $$
420: where $\tau =t-t_i,\ \lambda$ the Lyapunov exponent (Section 2),
421: and where $v(0)\sim 10^{-13}$ is the minimal
422: variance averaged over all $n$ fluctuations observed in the run.
423: Notice that the latter value is considerably smaller than the border
424: $v_b\sim 10^{-10}$. This is because of penetration of trajectories into FD.
425: Interestingly, the ratio
426: $v_b/v(0)=2000$ is the same for both runs in Fig.1
427:
428: Surprisingly sharp crossover to diffusive kinetics, clearly seen in Fig.1,
429: is related to
430: the dynamical scale of diffusion which corresponds to a certain
431: size $v_d$ of the increasing variance at which the exponential growth stops.
432: Roughly, it occurs at time $\tau =\tau_d$ when already
433: $|x-x_0|\sim |p-p_0|\sim 1/2$,
434: whence $v_{xd}\sim 12/4=3$ and $v_{pd}\sim 3/C^2$.
435: Hence, we can characterize the dynamical
436: scale as
437: $$
438: v(\tau_d)\,=\,v_d\,=\,F_d\cdot v_{pd}\cdot v_{xd}\,=\,{9F_d\over C^2}\,,
439: \qquad \tau_d = {\ln{(v_d/v(0))}
440: \over 4\lambda}
441: \eqno (3.4)
442: $$
443: where $F_d$ is an empirical factor, and $\tau_d$ is found
444: from Eq.(3.3).
445: From data in Fig.1 the dynamical scale
446: $v_d\approx 0.015$ independent of $v_b$ which gives the empirical factor
447: $F_d\approx 1/3$.
448:
449: In the diffusion region ($v>v_d$) the initial kinetics is described by a simple
450: relation for the free diffusion (see Section 2):
451: $$
452: v(\tau )\,\approx{\tau\pm\tau_d\over C^2}\,+\,v_d\,,
453: \qquad \tau_d\,<\,\tau\,\ll\,C^2
454: \eqno (3.5)
455: $$
456: which is also shown in Fig.1.
457: It includes two corrections, $\tau_d$ and $v_d$, due to the
458: exponential ballistic kinetics. The first one with
459: opposite signs for two symmetric parts of the fluctuation
460: takes account for the "lost" time after (or prior to)
461: anti--diffusion (diffusion) while the second correction
462: describes a finite fluctuation size at the crossover
463: from (to) the diffusion. The mean empirical $\tau_d=7$, used
464: in Fig.1, is close to the value $\tau_d=6.5$ found from
465: Eq.(3.4) with another empirical quantity $v_d=0.015$.
466:
467: The large ratio
468: $$
469: B\,=\,{\langle P\rangle\over C^2}\,\gg\,1 \eqno (3.6)
470: $$
471: of the mean fluctuation period $\langle P\rangle$ to a characteristic time of
472: diffusion relaxation (see Eq.(3.5)) is the definition of {\it big}
473: fluctuation. It guaranties the separation of successive fluctuations in time.
474:
475: Now we turn to the main subject of our studies, the purely
476: diffusive kinetics of big fluctuations. To this end we, first,
477: get rid of $x$--statistics excluding $v_x$ from the
478: selection condition (3.1) which now reads:
479: $$
480: v(t)\,=\,v_p\,<\,v_{pb}\,=\,v_b \eqno (3.7)
481: $$
482: Besides, the variance $v_b$ must now exceed the new dynamical
483: border:
484: $$
485: v_b\,>\,v_d\,=\,v_{pd}\,\approx\,f_p\cdot{12\over C^2}
486: \eqno (3.8)
487: $$
488: with some empirical factor $f_p\approx 1$ (see Eq.(3.4), and
489: discussion below).
490:
491: A typical example of diffusive BF is shown in Fig.2.
492: Both the regular macroscopic kinetics of
493: anti--diffusion/diffusion as well as irregular fluctuations around
494: are clearly seen. Notice that their
495: size is rapidly
496: decreasing toward the BF maximum. One may even get the impression that
497: the motion becomes regular over there, hence the term "optimal
498: fluctuational path" \cite{14}. In fact, the motion remains diffusive down
499: to the dynamical scale $v\sim v_d$ (3.8).
500:
501: Even though a separate BF by itself is fairly regular,
502: the time instance of its spontaneous appearance $t_i$, and
503: hence the individual period $P$, are
504: random in the chaotic system. Due to statistical independence of BF under
505: condition (3.6) the expected distribution in $P$ is
506: Poissonian (Fig.3):
507: $$
508: f(P)\,=\,{\exp{(-P/\langle P\rangle )}\over \langle P\rangle } \eqno (3.9)
509: $$
510:
511: The principal characteristic of the period statistics, $\langle P\rangle$,
512: can be estimated as follows. From the ergodicity of motion
513: in the $N$--dimensional momentum space the ratio
514: $$
515: \Phi\,=\,{T_s\over t_f}\,=\,{\langle T_s\rangle\over
516: \langle P\rangle }\,=\,{{\cal{P}}_{fl}\over {\cal{P}}_{eq}}
517: \eqno (3.10)
518: $$
519: This is exact relation (in the limit $t_{run}\to\infty$) where
520: $T_s$ is the total sojourn time of trajectories within FD
521: (under condition $v(t)<v_b$) during
522: the whole run time $t_{run}$, and $\langle T_s\rangle$ is the same per
523: fluctuation. Both ratios are equal to the ratio of
524: $N$--dimensional momentum volume
525: $\cal{P}$ of the fluctuation at $\tau =0$ to that in
526: equilibrium. The ratio $\Phi$ was also measured during the run.
527: Hence
528: $$
529: \langle P\rangle\,=\,{\langle T_s\rangle\over \Phi} \eqno (3.11)
530: $$
531:
532: The next more difficult step is evaluation of $T_s=2T_{ex}$ from the
533: diffusion equation where $T_{ex}$ is the exit (or entrance due to symmetry)
534: time from (or to) FD.
535: A simple crude estimate is: $T_{ex}\sim v_b/D_p=v_bC^2$ (Section 2).
536: However, the first numerical experiments have already revealed that the actual
537: exit time is much shorter, roughly by a factor of $1/N^2$.
538: A plausible explanation is in that the most of distribution inside FD
539: is concentrated in a relatively narrow layer at the surface
540: of $N$--dimensional sphere determined by the selection condition
541: $v(t)<v_b$ (3.7). Then, the relative width of the layer $\sim 1/N$ implies the
542: observed factor $\sim 1/N^2$.
543: Father, the ratio
544: $$
545: \Phi (v_b,N)\,=\,v_b^{N/2}\cdot \phi (N) \eqno (3.12)
546: $$
547: where the geometrical function
548: $$
549: \phi (N)\,\approx\,\left({\pi e\over 6}\right )^{N/2}
550: {(1-1/6N)\over \sqrt{\pi N}} \eqno (3.13)
551: $$
552: admits a fairly accurate approximation down to $N=1$
553: (see Fig.4).
554:
555: Collecting all the formulae above, we arrive at our final
556: empirical relation
557: $$
558: \langle P\rangle\,\approx\,{F\over \Phi}\cdot{2v_bAC^2\over N^2}\,\approx\,F\cdot{2AC^2\over N^2}\cdot
559: {v_b^{1-N/2}\over \phi(N)}
560: \eqno (3.14)
561: $$
562: with two fitting factors, $A$ for the layer width, and
563: $F$ for all other approximations above.
564: Both factors cannot be united in one because the former
565: enters a new expression for the dynamical scale which naturally
566: generalizes Eq.(3.8). Together with inequality (3.6) for big fluctuation
567: the new dynamical scale was using for selection of purely
568: diffusive BF which are described by Eq.(3.14).
569: The corresponding inequality reads (cf. Eq.(3.8)):
570: $$
571: v_b\,>\,v_d\,, \qquad v_d\cdot{A\over N^2}\,\approx\,
572: f_p\cdot{12\over C^2}
573: \eqno (3.15)
574: $$
575: which means that even a small part ($A/N^2<1$)
576: of FD must exceed the dynamical scale.
577:
578: Optimization of all empirical parameters was done as follows.
579: The values of two factors, $B$ in Eq.(3.6) and $f_p$ in (3.15),
580: are not crucial that is the larger they are the better for selection of purely diffusive BF. However, this reduces the
581: amount of empirical data available. A compromise was found
582: at $B=7$ and $f_p=1$ which leaves 36 runs of 61 done, and
583: 34429 of overall 75053 BF computed with $N=2 - 10$ for
584: comparison to Eq.(3.14). This was executed in the following
585: way. For each selected run with parameters $N,\ C,\ v_b$, and
586: computed $\langle P\rangle$ and $\Phi$ the empirical factor $F$
587: supposed to be a constant was calculated from the first
588: Eq.(3.14). The value of parameter $A$ was chosen by
589: minimizing the relative standard deviation down to
590: $\Delta F/\langle F\rangle =0.17$.
591: The result, for a given set of
592: data, was $A\approx 6$. The final dependence $F(N)$ is shown
593: in Fig.5 where the bars are the statistical errors
594: $F/\sqrt{n}$ for each run.
595:
596: Coming to analysis of our main theoretical result, the
597: second Eq.(3.14), we first remark that
598: it does not describe at all
599: a single trajectory ($N=1$). This is because we excluded
600: $v_{xb}$ from the selection condition (3.7) (cf. Eq.(3.2)),
601: and thus reduced the dimension of phase space to the minimal
602: value, the unity. In this case, a single trajectory
603: repeatedly crosses FD with a period $P\sim C^2$, the whole
604: diffusion time around the phase space torus, independent of FD size. More formally, it
605: follows also from Eq.(3.14) since condition (3.6) cannot be
606: satisfied for small $v_b$.
607:
608: In case of two trajectories ($N=2$) the period does not
609: depend on $v_b$, and for data in Fig.5 the ratio
610: $\langle P\rangle /C^2\approx 8.7$. Due to fluctuations
611: the actual values of this ratio are in the interval
612: $7.4 - 11.0$, still not too big for a Big Fluctuation.
613: Apparently, this leads to a relatively large scattering of
614: points with $N=2$ which also persists for $N=3$ too.
615:
616: The main, exponential in $N$, dependence in Eq.(3.14)
617: is readily derived from a graphic picture of $N$ statistically
618: independent particles gathering together inside a small domain
619: with probability $\sim 1/P\sim v_b^{N/2}$.
620: Such estimates are
621: known for the Poincar\'e recurrences since Boltzmann \cite{10}.
622: The estimate is especially vivid in geometrical picture of
623: $N$--dimensional sphere of radius $\sqrt{v_b}$ considered above.
624: Our empirical relation (3.14) considerably improves the simple
625: estimate including
626: a more weak power--law dependence which is evident in Fig.5.
627:
628: In our studies described above we fixed the position of BF
629: in the phase space, Eq.(2.4). If we lift this restriction,
630: the probability of BF would increase by a factor of
631: $v_b^{-1/2}$, or by decrease of $N$ by one ($N\to N-1$)
632: because now only $N-1$ trajectories remain independent.
633: With the latter change all the above relations would
634: presumably still hold true.
635:
636: Our main relation (3.14) describes the diffusive kinetics for
637: $v_b>v_d$, Eq.(3.15), when a big fluctuation is not too big.
638: In the opposite case $v_b\ll v_d$ of a very big fluctuation,
639: like in Fig.1, the dependence $\langle P(v_b)\rangle $ becomes
640: much simpler\\ (see Eqs.(3.11) - (3.13), and \cite{16}):
641: $$
642: \langle P(v_b)\rangle\,=\,{\langle T_s\rangle\over \Phi}\,
643: \approx\,{2\over v_b^{N/2}\phi (N)}\,\approx\,
644: 2v_b^{-N/2}
645: \eqno (3.16)
646: $$
647: This is explained by a fast exponential kinetics near BF top
648: (Fig.1) which implies the most short exit time
649: $T_{ex}\approx 1$ and, hence, $T_{s}\approx 2$.
650: Indeed, for both BF in Fig.1 the empirical value of the product
651: $\langle P\rangle\Phi = 1.98$.
652:
653: In conclusion of this Section we show in Fig.6 the
654: macroscopic kinetics of BF entropy, both "exact" (2.6),
655: calculated on the partition of the whole interval\\
656: ($0<p<C$) into $N_p=401$ bins, and that
657: in our approximation (2.7). Both entropies were calculated
658: for the same 5 trajectories in one run. A necessary statistics
659: for exact entropy was obtained at the expense of a large
660: number $n=4580$ of fluctuations in the run.
661: For comparison of both entropies we need, first,
662: to adjust the constant $S_0$ in Eq.(2.6).
663: As is easily verified, the Gaussian distribution (2.8) leads
664: exactly to the relation (2.7) if the constant
665: $$
666: S_0\,=\,-\,{1\over 2}\ln{(2\pi e)}\,\approx\,-1.4189\,
667: \approx\,-\sqrt{2}
668: \eqno (3.17)
669: $$
670: Approximation (2.7) holds on the most part of BF except a relatively
671: small domain near the equilibrium where the distribution in $p$
672: approaches the homogeneous one. The exact entropy
673: (with constant (3.17)) in equilibrium is
674: $$
675: S_{SE}\,=\,-\,{1\over 2}\ln{\left({\pi e\over 6}\right)}\,
676: \approx\,-0.18 \eqno (3.18)
677: $$
678: instead of zero in approximation (2.7). The difference is
679: relatively small, the smaller the bigger is the fluctuation.
680: In the main part of BF our simple relation of the entropy
681: (2.7) reproduces the exact one (2.6) to a surprisingly good
682: accuracy which confirms that the distribution in $p$ is indeed
683: very
684: close to the Gaussian one (2.8) as expected.
685:
686: \section{Conclusion: thermodynamic arrow ?}
687: In the present paper the results of extensive numerical experiments on
688: big entropy fluctuations (BF) in a statistical equilibrium (SE)
689: of classical
690: dynamical systems are presented,
691: and their peculiarities are analysed and discussed.
692:
693: All numerical experiments have been carried out on the basis of a very simple
694: model - the Arnold cat map (2.1) on a unit torus - with only two minor, but important and helpful,
695: modifications:
696:
697: (1) Expansion of the torus in $p$ direction (2.2) which allows for more
698: impressive diffusive kinetics of BF out of the equilibrium
699: (Fig.2), and
700:
701: (2) Inserting a special (unstable) fixed point for a better
702: demonstration of exponential ballistic kinetics as well (Fig.1).
703: Besides, this point was used as a fixed position of BF, thus
704: relating our studies of BF to another interesting and
705: important problem, the Poincar\'e recurrences (see Eq.(2.2)).
706:
707: The most important distinction of our approach to the problem
708: was in that we have abandoned from the beginning a vague
709: question of the initial conditions, particularly
710: a "necessary" restriction of those in statistical physics.
711: Instead, we started our numerical
712: experiments at {\it arbitrary} initial conditions
713: (most likely corresponding to SE), and did {\it observe} what the dynamics
714: and statistics of BF were like. In other words, we studied
715: the {\it spontaneous} BF only.
716:
717: What is also important, such a spontaneous rise of BF out of SE
718: as well as its subsequent relaxation back to SE
719: can be considered as a statistical {\it macroscopic} event,
720: even in a few--freedom
721: system like (2.2). The term "macroscopic" refers here to average quantities as variance, entropy, mean period,
722: distribution function, and the like.
723:
724: We consider a particular class of BF which we call the Boltzmann
725: fluctuations. They are obviously symmetric with respect
726: to time reversal (see Figs. 1, 2 and 6), so that at least in this case there is no physical reason
727: at all for the conception
728: of the notorious '{\it time} arrow'.
729: Nevertheless, a related conception, say, {\it thermodynamic} arrow, pointing
730: in the direction of the average increase of entropy, makes sense in spite
731: of the time symmetry \cite{16}.
732: The point is that the BF characteristic
733: relaxation time
734: is determined by model's parameter $C$ only, and does not depend on the BF
735: itself. On the contrary, the expectation time for a given BF,
736: or the mean period between successive fluctuations, rapidly grows
737: with BF size and with the number of trajectories (or freedoms),
738: Eq.(3.14). A large ratio of both
739: $B\,=\,\langle P\rangle / C^2\,\gg\,1$ is our {\it definition}
740: of big fluctuation, Eq.(3.6).
741: A similar result was obtained recently in \cite{22} but the
742: authors missed the principal difference between the time arrow
743: and thermodynamic arrow.
744:
745: A related notion of {\it causality} arrow, which by definition
746: points from an {\it independent} macroscopic cause to its effect,
747: also makes some
748: physical sense (for discussion see \cite{16} and Section 3 above).
749: For Boltzmann's
750: BF considered in the present paper the directions of both arrows
751: do coincide {\it independent} of the direction of time.
752: The latter statement is the most important, in our opinion,
753: philosophical "moral" the principally well--known Boltzmann
754: fluctuations do teach us.
755:
756: Even though we prefer the discussion and interpretation of our empirical results in terms of entropy ($S$), the most
757: fundamental conception in
758: statistical physics, we actually use in our studies
759: another, entropy--like, quantity, the variance $v(t)$ for a group of
760: $N$ trajectories, Eq.(2.5). One reason is technical: the
761: computation of $v$ is much simpler while that for $S(t)$ is
762: either very time--consuming in numerical experiments
763: (for exact $S$ (2.6)) or approximate (2.7). Besides, for
764: diffusive kinetics, we are mainly interested in, the variance
765: is natural variable in which the BF picture is most simple
766: and comprehensible.
767:
768: Originally, we planned to cover both sides of the BF
769: phenomenon,
770: the regular macroscopic kinetics as well as accompanying
771: microscopic fluctuations (noise) around. However, our
772: numerical experiments revealed a much more complicated structure
773: of the latter as an example in Fig.2 demonstrates.
774: The dependence $v(t)$ looks like a fractal curve on a variety
775: of time scales, from the minimal one $\sim 1$ iteration
776: up to $\sim C^2$ comparable to that of BF itself.
777: This interesting problem certainly requires and deserves
778: farther special studies.
779:
780: In the present paper the fluctuations in classical mechanics
781: are considered
782: only. Generally, the quantum fluctuations would be rather
783: different. However, according to the Correspondence Principle, the
784: dynamics and statistics of a quantum system in quasiclassics are close
785: to the classical ones on the appropriate time scales of which
786: the longest one corresponds just to the diffusive kinetics,
787: and provides the necessary
788: transition to the classical limit (for details see \cite{4,23}).
789: Curiously, the computer classical dynamics that is the simulation
790: of a classical dynamical system on digital computer is of a qualitatively
791: similar character. This is because any quantity in computer representation
792: is discrete ("overquantized"). As a result the correspondence between
793: the classical continuous dynamics and its computer representation in
794: numerical experiments is generally
795: restricted to certain finite time scales
796: like in the quantum mechanics (see two first references \cite{23}).
797:
798: Discreteness of the computer phase space leads to another peculiar phenomenon:
799: generally, the computer dynamics is irreversible due to the rounding--off
800: operation unless the special algorithm is used in numerical experiments.
801: Nevertheless, this does not affect the statistical properties of chaotic
802: computer dynamics. Particularly, the statistical laws in computer
803: representation remain time--reversible in spite of (nondissipative)
804: irreversibility of the
805: underlying dynamics.
806: This simple example demonstrates that, contrary to a common belief,
807: the statistical reversibility is a more general property than the dynamical
808: one.
809:
810:
811: \newpage
812:
813: \begin{thebibliography}{99}
814: \bibitem{1} Proc. 20th IUPAP Intern. Conference on Statistical Physics
815: (Paris, 1998), Physica A {\bf 263}, 516 - 544 (1999).
816: \bibitem{2} J. Lebowitz, Physica A {\bf 194}, 1 (1993).
817: \bibitem{3} L.D. Landau and E.M. Lifshitz,
818: {\it Statistical Physics}, Part 1,
819: Moskva, Nauka, 1995 (English translation of a previous Edition:
820: Pergamon, Oxford, 1980).
821: \bibitem{4} B.V. Chirikov, Natural Laws and Human Prediction, in:
822: {\it Law and Prediction in the Light of Chaos Research},
823: ed. by P. Weingartner and G. Schurz, Springer, Berlin, 1996, p. 10;
824: Open Systems \& Information Dynamics {\bf 4}, 241 (1997);
825: chao--dyn/9705003;
826: Wiss. Zs. Humboldt Univ. zu Berlin,
827: Ges.-Sprachw. R. {\bf 24}, 215 (1975).
828: \bibitem{5} J. Jeans, Phil. Trans. Roy. Soc. A {\bf 199}, 1 (1929).
829: \bibitem{6} H. Haken, {\it Synergetics}, Springer, Berlin, 1978
830: (Russian translation:
831: Moskva, Mir, 1980).
832: \bibitem{7} A. Turing, Phil. Trans. Roy. Soc. Lond. B {\bf 237}, 37 (1952);
833: G. Nicolis and I. Prigogine, {\it Self--Organization in
834: Nonequilibrium Systems}, Wiley, New York, 1977.
835: \bibitem{8} A. Cottrell, Emergent properties of complex systems,
836: in {\it The Encyclopedia of Ignorance}, ed. by
837: R. Duncan and M. Weston-Smith, Pergamon, Oxford, 1977, p.129.
838: \bibitem{9} P. Glansdorf and I. Prigogine, {\it Thermodynamic theory of
839: structure, stability, and fluctuations}, Wiley, New York, 1971
840: (Russian translation: Moskva, Mir, 1972).
841: \bibitem{10} L. Boltzmann, {\it Vorlesungen \"uber Gastheorie}, Leipzig, Barth,
842: 1896/98 (English translation:
843: {\it Lectures on gas theory}, Cambridge Univ. Press, 1964;
844: Russian translation: Moskva, Gostexizdat, 1956).
845: \bibitem{11} E. Schr\"odinger, \"Uber die Umkehrung der Naturgesetze,
846: Sitzungsber. Preuss. Akad. Wiss. (1931), 144.
847: \bibitem{12} A.N. Kolmogoroff, Math. Ann. {\bf 113}, 766 (1937);
848: see also ibid. {\bf 112}, 155 (1936) (Russian translation:
849: {\it Selected papers on probability theory and
850: mathematical statistics}, ed. by Yu.V. Prokhorov, Moskva, Nauka,
851: 1986, pp. 197 and 173).
852: \bibitem{13} A.M. Yaglom, Dokl. Akad. Nauk SSSR {\bf 56}, 347 (1947);
853: Mat. Sbornik {\bf 24}, 457 (1949).
854: \bibitem{14} D.G. Luchinsky, P. McKlintock and M.I. Dykman,
855: Rep. Prog. Phys. {\bf 61}, 889 (1998).
856: \bibitem{15} I.P. Kornfeld, S.V. Fomin and Ya.G. Sinai, {\it Ergodic Theory},
857: Moskva, Nauka, 1980 (English translation:
858: Springer, New York, 1982).
859: \bibitem{16} B.V. Chirikov, Big Entropy Fluctuations in Nonequilibrium Steady
860: State: A Simple Model with Gauss Heat Bath,
861: preprint Budker INP 00--54; nlin.CD/0006033;
862: JETP (in press).
863: \bibitem{17} D.V. Anosov, Dokl. Akad. Nauk SSSR {\bf 145}, 707 (1962).
864: \bibitem{18} L. Schulman, Phys. Rev. Lett. {\bf 83}, 5419 (1999);
865: G. Casati, B.V. Chirikov and O.V. Zhirov, ibid. {\bf 85}, 896
866: (2000);
867: D. Evans, D. Searles, E. Mittag, cond--mat/0008421;
868: B.V. Chirikov and O.V. Zhirov, cond--mat/0009125.
869:
870: \bibitem{19} V.I. Arnold and A. Avez, {\it Ergodic Problems of Classical
871: Mechanics}, Benjamin, New York, 1968 (Russian translation: RCD,
872: Izhevsk, 1999).
873: \bibitem{20} A. Lichtenberg and M. Lieberman, {\it Regular and Chaotic
874: Dynamics},
875: Springer, New York, 1992 (Russian translation of the 1st Edition:
876: {\it Regular and Stochastic Motion}, Moskva, Mir, 1984).
877: \bibitem{21} R. Balescu, {\it Equilibrium and Nonequilibrium Statistical
878: Mechanics}, Wiley, New York, 1975 (Russian translation:
879: Moskva, Mir, 1978).
880: \bibitem{22} R. Metzler, W. Kinzel and I. Kanter,
881: On time's arrow in Ehrenfest models with reversible
882: deterministic dynamics, cond--mat/0007382.
883: \bibitem{23} B.V. Chirikov, F.M. Izrailev and D.L. Shepelyansky,
884: Sov. Sci. Rev. {\bf C2}, 209 (1981);
885: B.V. Chirikov, Time-dependent quantum systems, Lectures in
886: Les Houches Summer School on Chaos and Quantum Physics (1989),
887: Elsevier, Amsterdam, 1991, p. 443;
888: G. Casati and B.V. Chirikov, in: {\it Quantum Chaos:
889: Between Order and Didorder}, ed. by G. Casati and B.V. Chirikov,
890: Cambridge Univ. Press, Cambridge, 1995, p. 3; Physica D {\bf 86}, 220 (1995);
891: B.V. Chirikov, Pseudochaos in statistical physics,
892: Proc. Intern. Conference on Nonlinear Dynamics, Chaotic and
893: Complex Systems (Zakopane, 1995), ed. by E. Infeld, R. Zelazny
894: and A. Galkowski, Cambridge Univ. Press, Cambridge, 1997, p.149;
895: B.V. Chirikov and F. Vivaldi, Physica D {\bf 129}, 223 (1999).
896:
897: \end{thebibliography}
898:
899: %\end{document}
900:
901: \newpage
902: \begin{center} Figure captions \end{center}
903:
904: \begin{itemize}
905:
906: \item[Fig.1]
907: Mixed kinetics for two big fluctuations of different size. Full/open
908: circles show the time dependence
909: of mean variance
910: $\langle v(t-t_i)\rangle$ around BF maximum at $t=t_i$; upper horizontal straight
911: line is equilibrium while the lower line indicates the
912: empirical value of
913: dynamical scale $v_d=0.015$, Eq.(3.4) with parameter
914: $F_d\approx 1/3$.
915: Two oblique straight lines represent
916: the expected fast kinetics, Eq.(3.3), and two solid curves do so for the
917: initial diffusive kinetics, Eq.(3.5).
918: Run parameters and results are respectively: $C=15,\ N=1,\
919: v_b=3.9\times 10^{-11}
920: \ /\ 6.25\times 10^{-10}\ (v_{xb}=v_{pb}),\ v(0)=1.96\times 10^{-14}\ /\
921: 3.1\times 10^{-13},\ n=1971\ /\ 4459,\ w=500$.
922: Average period between successive fluctuations
923: $\langle P\rangle\approx1.4\times 10^7\ /\ 3.5\times 10^6$ iterations.
924:
925: \item[Fig.2]
926: Same as in Fig.1 for
927: a typical diffusive kinetics (anti--diffusion/diffusion):
928: solid curve shows the average over all
929: $n=20259$ fluctuations in a run, and wiggle line is the same for $28$
930: first fluctuations.
931: Two oblique straight lines represent
932: the expected
933: initial diffusive kinetics, Eq.(3.5) with $\tau_d=0$ and
934: empirical $v_d^{(emp)}=0.045$ while the theory (3.15) gives
935: $v_d=0.02$.
936: Other run parameters/results: $C=50,\ N=5,\ v_b=0.0256,\ w=10^4,\ \langle P\rangle\approx 7.7\times 10^5\ /\
937: 8.7\times 10^5$, and
938: $B=306\ /\ 348;\ \langle P\rangle/w\approx 77\ /\ 87$.
939:
940: \item[Fig.3]
941: Histogram of integrated distribution (3.9) for data in Fig.2.
942: Each circle shows the number of periods $P_m >m\cdot
943: \Delta P,\ m=0,1,...$ integer;\ $P_0=n,\
944: \Delta P=1.5\times 10^5;\ P_{min}/w=1.0027;\
945: P_{max}/\langle P\rangle =12.63;\ \langle P\rangle = 765084$.
946: Straight line is expected distribution
947: $n\cdot\exp{(-P/\langle P\rangle )}$.
948:
949: \item[Fig.4]
950: Comparison of directly measured ratio $\Phi_{emp}$, Eq.(3.10), with
951: theoretical approximation $\Phi_{th}$, Eq.(3.12) for
952: $N=1\ -\ 10$:
953: $\Phi_1=\Phi_{emp}/\Phi_{th}$; average over 71 runs
954: $\langle\Phi_1\rangle = 1.015\pm 0.11$ (the standard deviation);
955: the bars show statistical errors $1/\sqrt{n}$ for each run;
956: the total number of fluctuations in all runs is 127346.
957:
958: \item[Fig.5]
959: Comparison of empirical data for 36 runs,
960: selected from 61 runs computed for $N=2\ -\ 10$ by the two rules,
961: Eq.(3.6) with $B>7$ and Eq.(3.15) with $A=6$,
962: to theoretical relation (3.14) with the main fitting factor
963: $F_m,\ m=1,...,36$ (see text). Average
964: $\langle F\rangle =1.51\cdot (1\pm 0.17)$ (the standard deviation);
965: the bars show statistical errors $F_m/\sqrt{n}$ for each run;
966: the total number of fluctuations in 36 runs is 34429.
967:
968: \item[Fig.6]
969: Macroscopic kinetics of BF entropy: lower line is "exact"
970: entropy, Eq.(2.6), to be compared with approximation (2.7),
971: middle line; the upper line is the same approximation
972: for diffusion theory, Eq.(3.5) with $\tau_d=0$ and
973: empirical $v_d^{(emp)}=0.02$.
974: Run parameters/results: $C=50,\ N=5,\ v_b=0.01,\ w=10^4,\ n=4580,\
975: \langle P\rangle\approx 3.3\times 10^6,\
976: B=1314;\ \langle P\rangle/w\approx 329$.
977: The number of partition bins for calculating (2.6) $N_p=401$.
978:
979:
980: \end{itemize}
981:
982: %\end{document}
983:
984: \end{document}
985: