1: %\documentclass{jpsj2}
2: %%\documentclass[letter]{jpsj2} %% for letters
3: %%\documentclass[shortnote]{jpsj2} %% for short notes
4: %%\documentclass[comment]{jpsj2} %% for comments
5: %%\documentclass[addenda]{jpsj2} %% for addenda
6: %%\documentclass[errata]{jpsj2} %% for errata
7: \documentclass[twocolumn]{jpsj2} %% two-column layout
8: %%\documentclass[seceq]{jpsj2} %% It makes equation numbers included within the section number (for regular paper only).
9: %%% The following is the list of packages loaded automatically into this class file.
10: %% amsmath.sty
11: %% amssymb.sty
12: %% graphicx.sty
13: %% overcite.sty
14: %
15:
16: \title{Replica Analysis of a Preferential Urn Model}
17:
18: \author{Jun \textsc{Ohkubo}$^{1,2}$\thanks{E-mail address: jun@smapip.is.tohoku.ac.jp},
19: Muneki \textsc{Yasuda}$^{1}$
20: and Kazuyuki \textsc{Tanaka}$^{3}$
21: }
22:
23: \inst{$^{1}$Department of System Information Sciences, Graduate School of Information Sciences,\\
24: Tohoku University, 6-3-09, Aramaki-Aza-Aoba, Aoba-ku, Sendai 980-8579, Japan\\
25: $^{2}$JSPS Research Fellow\\
26: $^{3}$Department of Applied Information Sciences, Graduate School of Information Sciences,\\
27: Tohoku University, 6-3-09, Aramaki-Aza-Aoba, Aoba-ku, Sendai 980-8579, Japan\\
28: }
29:
30: \abst{
31: We analyze a preferential urn model with randomness using the replica method.
32: The preferential urn model is a stochastic model based on the concept ``the rich get richer.''
33: The replica analysis clarifies that
34: the preferential urn model with randomness shows a fat-tailed occupation distribution.
35: We discuss the intuitive picture of the urn model, a relationship between quenched and annealing systems,
36: and a trial for the application of the preferential urn model to econophysics.
37: The analytical treatments and results would be useful for various research fields
38: such as complex networks, stochastic models, and econophysics.
39: }
40:
41: \kword{urn model, replica analysis, stochastic process, statistical mechanics, Ehrenfest class,
42: Pareto's distribution, econophysics}
43:
44: \begin{document}
45: \maketitle
46:
47: \section{Introduction}
48:
49: Urn models have been used in order to explain various physical phenomena such as quantum gravity and glassy dynamics
50: \cite{Ritort1995,Bialas1997,Drouffe1998,Bialas1998,Bialas2000,Godreche2001,Leuzzi2002,Burda2002}.
51: Because of their tractability for analytical calculations,
52: the urn models have been studied extensively in statistical physics and probability theory.
53: The examples of the urn models are the backgammon model \cite{Ritort1995}
54: and the zeta urn model \cite{Bialas1997};
55: these models can be treated analytically.
56: Furthermore, it has been shown that these urn models relate to the zero-range process
57: which is also famous in the statistical physics; see a review by Evans and Hanney \cite{Evans2005}.
58:
59: The preferential urn model\cite{Ohkubo2005c} is also one of the examples of the urn models.
60: It has been revealed that the preferential urn model shows a fat-tailed occupation distribution
61: only in the case with randomness; the meanings of the ``randomness'' will be explained later in the present paper.
62: The fat-tailed behavior is similar to the Parato's distribution \cite{Newman2005}.
63: The preferential urn model stems from research on complex networks \cite{Ohkubo2005c},
64: and because of the simplicity,
65: the model would have a wide range of applications in research on not only complex networks,
66: but also many other fields such as econophysics and social sciences.
67: In econophysics or social sciences,
68: it could be reasonable to consider heterogeneous cases, e.g.,
69: in which each agent in econophysical systems has a different ability.
70:
71: In the present paper, an analysis of an urn model with randomness is presented.
72: While equilibrium properties of urn models without randomness are easily obtained \cite{Godreche2001},
73: the analysis for the urn models with randomness are not straightforward.
74: We analyze the preferential urn model
75: with randomness by using the replica method \cite{Nishimori2001}.
76: Furthermore, we will give some discussions for the analytical treatment
77: and an example of the applications of the preferential urn model for econophysics,
78: though the example is just a trial one.
79:
80: The outline of the present paper is as follows.
81: Section~2 gives a brief introduction to urn models.
82: In \S~3, we introduce the preferential urn model.
83: The general discussion for calculating occupation distributions is given in \S~4,
84: and \S~5 gives the replica analysis for the preferential urn model.
85: Some discussions for the replica analysis and an example of the applications of the model are given in \S~6.
86: Finally, we draw the main conclusions in \S~7.
87:
88: \section{Backgrounds of Urn Models}
89:
90: One of the most important properties of the urn models is that the partition function is given by a factorized form
91: \cite{Godreche2001}.
92: For example, we consider a system with $N$ urns and $M$ balls which are distributed among $N$ urns.
93: When we define an energy of each urn as $E(n_i)$
94: by using the number of balls in the urn, $n_i$,
95: the total energy of the whole system is given by
96: \begin{align}
97: E(n_1, \dots, n_N) = \sum_{i=1}^N E(n_i).
98: \end{align}
99: The partition function of the system is calculated from
100: \begin{align}
101: Z = \sum_{n_1=1}^\infty \cdots \sum_{n_N=1}^\infty p_{n_1} \cdots p_{n_N} \delta\left( \sum_{i=1}^N n_i , M \right),
102: \label{intro_partition_function_monkey}
103: \end{align}
104: where
105: \begin{align}
106: p_{n_i} = e^{-\beta E(n_i)}
107: \label{intro_boltzmann}
108: \end{align}
109: is the unnormalized Boltzmann weight attached to urn $i$, and $\beta$ is an inverse temperature.
110: The Kronecker delta of eq.~\eqref{intro_partition_function_monkey}
111: stems from the restriction of the total number of balls.
112: We can obtain various physical quantities
113: by using the partition function of eq.~\eqref{intro_partition_function_monkey};
114: for example, we can calculate an equilibrium occupation probability
115: $P(k)$ with which an urn has $k$ balls in the equilibrium state.
116:
117:
118: The above example, i.e., eq.~\eqref{intro_partition_function_monkey},
119: corresponds to \textit{monkey class} of urn models \cite{Godreche2001}.
120: In the Monkey class, an urn is selected at random, from which any ball is drawn,
121: and transfer the drawn ball to the other urn;
122: this dynamics corresponds to the image of a monkey playing at exchanging balls among urns.
123: In other words, each ball is indistinguishable in the monkey class,
124: so that the equilibrium statistics of the monkey class is considered as Bose-Einstein statistics (quantum).
125: On the other hand, there is the other type of urn models, so-called \textit{Ehrenfest class},
126: which corresponds to the Maxwell-Boltzmann statistics (classical).
127: In the Ehrenfest class, balls within an urn are distinguishable,
128: so that the partition function of the Ehrenfest class is written as
129: \begin{align}
130: Z = \sum_{n_1=1}^\infty \cdots \sum_{n_N=1}^\infty
131: \frac{p_{n_1}}{n_1!} \cdots \frac{p_{n_N}}{n_N!} \delta\left( \sum_{i=1}^N n_i , M \right),
132: \label{intro_partition_function_Ehrenfest}
133: \end{align}
134: which involves inverse factorials $1/(n_i!)$;
135: it is different from the partition function of the
136: monkey class of eq.~\eqref{intro_partition_function_monkey}.
137:
138: The choice of the energy of each urn is arbitrary,
139: and hence one can consider a lot of models
140: so that a selected model is suitable for his/her own problem.
141:
142:
143:
144: \section{Preferential Urn Model}
145:
146: \begin{figure}
147: \begin{center}
148: \includegraphics[width=7.5cm,keepaspectratio,clip]{fig1.eps}
149: \caption{
150: Illustration of the preferential urn model.
151: A ball is drawn at random,
152: and then the drawn ball is moved to an urn chosen preferentially.
153: }
154: \label{fig_model}
155: \end{center}
156: \end{figure}
157:
158:
159: We consider a system of $M$ balls distributed among $N$ urns,
160: and denote the density of the system as $\rho = M/N$.
161: The number of balls contained in urn $i$ is denoted by $n_i$, and hence $\sum_{i=1}^N n_i = M$.
162: For the preferential urn model, we define the following energy of each urn:
163: \begin{align}
164: E(n_i) = - \ln (n_i!).
165: \label{eq_energy}
166: \end{align}
167: From the definition of the energy,
168: it is clear that each urn tends to get more and more balls
169: because an urn is stable when the urn obtains a large number of balls.
170: Note that there is a restriction for the total number of balls,
171: so that an urn cannot get an infinitely large number of balls.
172: It would be clarified that the transition rate in the urn model with the above definition of the energy obeys
173: a preferential concept, as described later,
174: and hence the urn model is called the \textit{preferential urn model}.
175: Furthermore, we here assume that
176: the preferential urn model belongs to the Ehrenfest class.
177:
178: The unnormalized Boltzmann weight of eq.~\eqref{intro_boltzmann} is written by
179: \begin{align}
180: p_{n_i} = \left( n_i! \right)^{\beta_i},
181: \end{align}
182: where $\beta_i$ is an inverse temperature of urn $i$.
183: Note that each urn has own inverse temperature and
184: each inverse temperature could be different from each other;
185: we denote the distribution of the inverse temperatures as $\phi(\beta)$.
186: Though it might seem strange in terms of physics that each urn has a different temperature,
187: there should be no restriction for a homogeneity of temperature
188: in information physics, social sciences, and econophysics
189: (Or one could say that the system is in nonequilibrium.)
190: We call the inverse temperature ``the inverse \textit{local} temperature'' because of its locality.
191: It will be shown in \S~4 and \S~5 that
192: the locality is important for generating a fat-tailed occupation distribution,
193: and, furthermore,
194: these inverse local temperatures allow us to understand results of an equilibrium occupation distribution intuitively.
195:
196: Using the heat-bath rule,
197: we derive the transition rate $W_{n_i \to n_i + 1}$ from the state $n_i$ to $n_i + 1$ as follows \cite{Drouffe1998}:
198: \begin{align}
199: W_{n_i \to n_i + 1} & \propto \frac{p_{n_i+1}}{p_{n_i}} \nonumber \\
200: &= (n_i+1)^{\beta_i}.
201: \end{align}
202: The transition rate indicates that an urn with more balls tends to get more balls,
203: and this means the preferential concept.
204: The dynamics of the preferential urn model is illustrated in Fig.~1.
205:
206: In summary, the construction of the preferential urn model is as follows:
207: \begin{enumerate}
208: \item[(i)]
209: Set $N$ urns and $M$ balls.
210: The $M$ balls are distributed among the $N$ urns,
211: and the initial distribution is arbitrary (usually distributed at random).
212: Each inverse local temperature $\beta_i$ is chosen by using a distribution $\phi(\beta)$.
213: \item[(ii)]
214: Choose a ball at random (the Ehrenfest class).
215: \item[(iii)]
216: Move the drawn ball to an urn selected by using the transition rate $W_{n_i \to n_i + 1}$.
217: \item[(iv)]
218: Repeat the procedures (ii) and (iii) until the system reaches an equilibrium state.
219: \end{enumerate}
220:
221: If all urns have the same temperatures, i.e., $\beta_i = \beta$ for all $i$,
222: equilibrium occupation distributions are easily calculated \cite{Bialas1997,Godreche2001,Ohkubo2005c}.
223: For instances, the equilibrium occupation distribution in the high-temperature limit ($\beta_i \to 0$ for all $i$)
224: is a Poisson distribution \cite{Godreche2001,Ohkubo2005c}.
225: However, it is generally difficult to calculate equilibrium occupation distributions of urn models with randomness.
226: In the following sections,
227: we give an analytical treatment to calculate the equilibrium occupation distribution of the preferential urn model
228: with randomness.
229:
230:
231: \section{Occupation Distribution}
232:
233: In this section, we give a general discussion for calculating occupation distributions,
234: and it will be revealed that the case with randomness is difficult to be analyzed.
235: In the next section, the replica analysis is given,
236: and an explicit form of the occupation distribution will be obtained in the case with randomness.
237:
238: By using the integral representation of the Kronecker delta, $2 \pi i \delta(a,b) = \oint \textrm{d}z \, z^{a-b-1}$,
239: the partition function of eq.~\eqref{intro_partition_function_Ehrenfest} is calculated as follows:
240: \begin{align}
241: Z_1 &= \sum_{n_1 = 0}^\infty \dots \sum_{n_N = 0}^\infty (n_1!)^{\beta_1 - 1}
242: \dots (n_N!)^{\beta_N - 1} \nonumber \\
243: & \qquad \times \frac{1}{2\pi i}\oint\!\textrm{d}z \, z^{\sum_{i=1}^N n_i - M - 1}.
244: \label{eq_Z_N_M}
245: \end{align}
246: Note that all values of the inverse local temperatures
247: $\{\beta_1, \dots, \beta_N\}$ are clamped at certain values in eq.~\eqref{eq_Z_N_M}.
248: Here, we define $f_{k}^{(\beta_1)}$ as an equilibrium occupation probability of urn $1$
249: with the inverse local temperature $\beta_1$ in a certain configuration:
250: \begin{align}
251: f_{k}^{(\beta_1)} &= \frac{1}{Z_1}
252: \sum_{n_1 = 0}^\infty \dots \sum_{n_N=0}^\infty \delta(n_1,k) \frac{p_{n_1}}{n_1!} \dots \frac{p_{n_N}}{n_N!}
253: \nonumber \\
254: & \qquad \times \delta\left( \sum_{i=1}^{N} n_i , M\right).
255: \label{eq_fk_not_simple}
256: \end{align}
257: Then, $f_{k}^{(\beta_1)}$ indicates the probability with which urn $1$ has $k$ balls in a certain configuration.
258: Next, we define $Z_2$ as
259: \begin{align}
260: Z_2 &=
261: \sum_{n_2 = 0}^\infty \dots \sum_{n_N = 0}^\infty
262: (n_2!)^{\beta_2 - 1} \dots (n_N!)^{\beta_N - 1} \nonumber \\
263: & \qquad \times \oint \frac{\textrm{d}z}{2\pi i}
264: z^{k + \sum_{i=2}^N n_i - M - 1} ,
265: \end{align}
266: and simplify $f_{k}^{(\beta_1)}$ of eq.~\eqref{eq_fk_not_simple} as follows:
267: \begin{align}
268: f_{k}^{(\beta_1)} = (k!)^{\beta_1 - 1} \frac{Z_2}{Z_1}.
269: \label{eq_urn1_clamped}
270: \end{align}
271:
272: In order to obtain the equilibrium occupation distribution of urn $1$ with $\beta_1$
273: in the configurational average, $P(k, \beta_1)$,
274: we average $f_{k}^{(\beta_1)}$ over the configuration $\{\beta_2, \cdots, \beta_N \}$:
275: \begin{align}
276: P(k, \beta_1)
277: &= \left\langle f_{k}^{(\beta_1)}
278: \right\rangle_{\{\beta_2, \dots, \beta_N\}} \nonumber \\
279: &= (k!)^{\beta_1 - 1}
280: \left\langle \frac{Z_2}{Z_1} \right\rangle_{\{ \beta_2, \dots, \beta_N \}},
281: \label{eq_p_k_0}
282: \end{align}
283: where $\langle \cdots \rangle_{\{ \beta_2, \dots, \beta_N \}}$
284: means that the configurational average is taken only over $\{\beta_2, \dots, \beta_N\}$;
285: the configurational average of a quantity $A(\beta_1, \beta_2, \cdots, \beta_N)$ is defined as
286: \begin{align}
287: &\left\langle A(\beta_1, \beta_2, \dots, \beta_N) \right\rangle_{\{\beta_2, \dots, \beta_N\}}
288: \nonumber \\
289: &\quad = \int \phi(\beta_2) \textrm{d}\beta_2 \cdots \int \phi(\beta_N) \textrm{d}\beta_N \,
290: A(\beta_1, \beta_2, \dots, \beta_N).
291: \end{align}
292: Finally, we take the configurational average of $P(k,\beta)$ for $\beta$.
293: Then, the equilibrium occupation distribution is calculated from
294: \begin{align}
295: P(k) &= \int \textrm{d}\beta \,\, \phi(\beta) P(k,\beta).
296: \label{eq_distribution_1}
297: \end{align}
298:
299: In eq.~\eqref{eq_p_k_0},
300: it is difficult to take the configurational average of the numerator and denominator at the same time.
301: Therefore, we use replica analysis for evaluate the equilibrium occupation distribution.
302:
303:
304: \section{Replica Analysis}
305: The replica method \cite{Nishimori2001} is one of the powerful methods for analysing random systems.
306: In the replica analysis, we prepare $m$ replicas of the original system,
307: and evaluate the configurational average of the partition functions, assuming the thermodynamic limit.
308: After that, the limit $m \to 0$ is taken,
309: which has successfully clarified typical properties of many random systems.
310: While mathematical validation of the replica analysis still remains unresolved,
311: the replica analysis has been empirically recognized as a reliable scheme in physics;
312: for instance, we can use the replica analysis in order to calculate a free energy or order parameters.
313: In this section,
314: we use the replica analysis to obtain the equilibrium occupation distribution of the preferential urn model
315: with randomness.
316: The aim of the present paper is to propose the analytical treatment for the preferential urn model with randomness,
317: and hence we here restrict ourselves only to the case
318: in which the distribution of the inverse local temperatures, $\phi(\beta)$,
319: follows an uniform distribution, i.e., $\phi(\beta) = 1$ for $\beta \in [0,1]$.
320:
321: First, we assume the following expression in order to calculate eq.~\eqref{eq_p_k_0}:
322: \begin{align}
323: \left\langle f_{k}^{(\beta_1)}
324: \right\rangle_{\{\beta_2, \dots, \beta_N\}}
325: = \exp\left(
326: \left\langle \ln f_{k}^{(\beta_1)}
327: \right\rangle_{\{\beta_2, \dots, \beta_N\}}
328: \right).
329: \label{eq_p_k}
330: \end{align}
331: The expression is obtained from the \textit{self-averaging property}\cite{Nishimori2001}
332: because of the equilibrium occupation probability $f_{k}^{(\beta_1)}$ is intensive.
333: Therefore, it is necessary only to calculate the configurational average of
334: $\ln f_{k}^{(\beta_1)}$; this term is calculated from
335: \begin{align}
336: \left\langle \ln f_{k}^{(\beta_1)} \right\rangle_{\{ \beta_2, \dots, \beta_N\}}
337: = &(\beta_1 - 1) \ln k!
338: - \left\langle \ln Z_1 \right\rangle_{\{ \beta_2, \dots, \beta_N\}} \nonumber \\
339: &+ \left\langle \ln Z_2 \right\rangle_{\{ \beta_2, \dots, \beta_N\}}.
340: \label{eq_ln_f}
341: \end{align}
342: In the replica analysis,
343: we use the following manipulations to evaluate the configurational average:
344: \begin{align}
345: \left\langle \ln Z_i \right\rangle_{\{ \beta_2, \dots, \beta_N\}}
346: =& \lim_{m\to 0} \left( \frac{\left\langle Z_i^m \right\rangle_{\{ \beta_2, \dots, \beta_N\}} - 1}{m}
347: \right) \nonumber \\
348: &\qquad \qquad \qquad \qquad (i = 1,2).
349: \label{eq_replica}
350: \end{align}
351: We should calculate the following quantities to evaluate eq.~\eqref{eq_ln_f}
352: using eq.~\eqref{eq_replica}.
353: Therefore, we first calculate the following quantities:
354: \begin{align}
355: &\left\langle Z_1^m \right\rangle_{\{ \beta_2, \dots, \beta_N\}} \nonumber \\
356: &\quad = \oint \prod_{\alpha = 1}^m \left\{ \frac{\textrm{d}z_\alpha}{2\pi i} z_\alpha^{-(M+1)}
357: G(z_\alpha)^{N-1} H(z_\alpha, \beta_1) \right\},
358: \label{eq_replica_1} \\
359: &\left\langle Z_2^m \right\rangle_{\{ \beta_2, \dots, \beta_N\}} \nonumber \\
360: &\quad = \oint \prod_{\alpha = 1}^m \left\{ \frac{\textrm{d}z_\alpha}{2\pi i} z_\alpha^{-(M+1-k)}
361: G(z_\alpha)^{N-1} \right\},
362: \label{eq_replica_2}
363: \end{align}
364: where
365: \begin{align}
366: G(z) &= \int_0^1 \textrm{d}\beta \,\, \phi(\beta) \sum_{n=0}^\infty (n!)^{\beta-1} z^n \nonumber \\
367: &= \int_0^1 \textrm{d}\beta \,\, \sum_{n=0}^\infty (n!)^{\beta-1} z^n
368: \label{eq_g0}
369: \end{align}
370: and
371: \begin{align}
372: H(z,\beta) = \sum_{n=0}^\infty (n!)^{\beta-1} z^{n}.
373: \end{align}
374: As mentioned above, we here treat the case with uniform randomness, i.e., $\phi(\beta) = 1, \,\, \beta \in [0,1]$,
375: so that the second line of eq.~\eqref{eq_g0} can be derived.
376: Next, we assume \textit{replica symmetry},
377: i.e., $z_\alpha = z$,
378: and derive the replica symmetric solution.
379: Then, by using the saddle-point method, eqs~\eqref{eq_replica_1} and \eqref{eq_replica_2} become as follows:
380: \begin{align}
381: &\left\langle Z_1^m \right\rangle_{\{ \beta_2, \dots, \beta_N\}} \nonumber \\
382: &\quad \simeq \exp\left\{
383: -m \ln z_\textrm{s} - m\ln G(z_\textrm{s}) + m \ln H(z_\textrm{s},\beta_1) \right. \nonumber \\
384: & \quad \qquad \left. - m\rho N \ln z_\textrm{s} + mN \ln G(z_\textrm{s}) + m\ln C
385: \right\} \nonumber \\
386: &\quad \simeq 1 - m\ln z_\textrm{s} - m\ln G(z_\textrm{s}) + m \ln H(z_\textrm{s},\beta_1) \nonumber \\
387: & \quad \qquad - m\rho N \ln z_\textrm{s} + mN \ln G(z_\textrm{s}) + m\ln C,
388: \label{Z_1_approximate}
389: \end{align}
390: where $C$ is a constant;
391: \begin{align}
392: &\left\langle Z_2^m \right\rangle_{\{ \beta_2, \dots, \beta_N\}} \nonumber \\
393: &\quad \simeq 1 - m(1-k)\ln z_\textrm{s} - m\ln G(z_\textrm{s}) \nonumber \\
394: & \qquad - m \rho N \ln z_\textrm{s} + mN \ln G(z_\textrm{s}) + m\ln C.
395: \label{Z_2_approximate}
396: \end{align}
397: In deriving the last expression of eqs.~\eqref{Z_1_approximate} and \eqref{Z_2_approximate},
398: we consider a second-order Taylor expansion because the limit $m \to 0$ will be taken.
399: The saddle-points $z_\textrm{s}$ in $\left\langle Z_1^m \right\rangle_{\{ \beta_2, \dots, \beta_N\}}$
400: and $\left\langle Z_2^m \right\rangle_{\{ \beta_2, \dots, \beta_N\}}$
401: are the same and calculated by using the following saddle-point equation:
402: \begin{align}
403: \rho &= \frac{z_\textrm{s} G'(z_\textrm{s})}{G(z_\textrm{s})},
404: \label{eq_saddle}
405: \end{align}
406: where
407: \begin{align}
408: G(z_\textrm{s})
409: &= 1 + z_\textrm{s} + \sum_{n=2}^\infty \left[ \frac{1}{\ln n!} \left( 1 - \frac{1}{n!} \right)\right] z_\textrm{s}^n
410: \label{eq_g1}
411: \end{align}
412: and
413: \begin{align}
414: G'(z_\textrm{s})
415: &= 1 + \sum_{n=2}^\infty \left[ \frac{1}{\ln n!} \left( 1 - \frac{1}{n!} \right)\right] n z_\textrm{s}^{n-1}.
416: \label{eq_g2}
417: \end{align}
418: Equations~\eqref{eq_g1} and \eqref{eq_g2} are easily obtained from eq.~\eqref{eq_g0}.
419: From the above calculations, we obtain the following expression by rearranging eq.~\eqref{eq_ln_f}:
420: \begin{align}
421: &\left\langle \ln f_{k}^{(\beta_1)} \right\rangle_{\{ \beta_2, \dots, \beta_N\}} \nonumber \\
422: &\quad = (\beta_1 - 1) \ln k! - \ln H(z_\textrm{s},\beta_1) + k \ln z_\textrm{s}.
423: \label{eq_av_ln_fk}
424: \end{align}
425: Therefore, the equilibrium occupation distribution of urn $1$ with $\beta_1$
426: in the configurational average is obtained
427: from eqs.~\eqref{eq_p_k_0}, \eqref{eq_p_k}, and \eqref{eq_av_ln_fk} as follows:
428: \begin{align}
429: P(k, \beta_1)
430: &= (k!)^{\beta_1-1} \frac{z_\textrm{s}^k}{H(z_\textrm{s},\beta_1)} \nonumber \\
431: &= \frac{(k!)^{\beta_1-1} z_\textrm{s}^k}{\sum_{n=0}^\infty (n!)^{\beta_1-1} z_\textrm{s}^n}.
432: \end{align}
433: Finally, we obtain the equilibrium occupation distribution of the whole system as
434: \begin{align}
435: P(k)
436: &= \int_0^1 \textrm{d}\beta \,\, \frac{(k!)^{\beta-1} z_\textrm{s}^k}{\sum_{n=0}^\infty (n!)^{\beta-1} z_\textrm{s}^n}.
437: \label{eq_distribution}
438: \end{align}
439: Note that the equilibrium occupation distribution is suitably normalized.
440:
441:
442: A large density $\rho$ leads to $z_\textrm{s} \simeq 1$,
443: and hence we consider that the saddle-point is approximately $z_\textrm{s} = 1$ for $\rho \gg 1$.
444: Substituting $z_\textrm{s} = 1$ into eq.~\eqref{eq_distribution},
445: we obtain the asymptotic form of the equilibrium occupation distribution as follows (see Appendix~A):
446: \begin{align}
447: P(k) &\sim k^{-2} \frac{1}{(\ln k)^2}.
448: \label{eq_asymptotic}
449: \end{align}
450: Hence, the equilibrium occupation distribution follows a generalized power law
451: with a squared inverse logarithmic correction.
452:
453:
454:
455: \begin{figure}
456: \begin{center}
457: \includegraphics[width=8cm,keepaspectratio,clip]{fig2.eps}
458: \caption{
459: Equilibrium occupation distributions of the preferential urn model.
460: The numerical results for the density $\rho = 1.0$ are shown by squares,
461: and those for $\rho = 4.0$ by triangles.
462: The solid lines correspond to eq.~\eqref{eq_distribution} for respective cases.
463: In both cases, the number of urns $N$ is $1000$, and that of the transfer process of balls is $2000N$.
464: The data are averaged over $20$ different realizations.
465: }
466: \label{fig_results}
467: \end{center}
468: \end{figure}
469:
470: To confirm the above analytical treatments, we performed numerical experiments.
471: Because initial distributions are arbitrary, we put each ball to urns at uniformly random as initial configurations.
472: In the numerical experiments,
473: the number of urns $N$ is $1000$,
474: and that of the transfer process of balls is $2000N$.
475: We checked that the $2000N$ transfer processes are enough to give the equilibrium states.
476: Figure~\ref{fig_results} shows the comparison between the equilibrium occupation distributions
477: of numerical experiments and those of the replica analysis.
478: For the density $\rho = 1.0$, we obtain the saddle-point $z_s = 0.660$ from eq.~\eqref{eq_saddle};
479: $z_s = 0.967$ for $\rho = 4.0$.
480: The numerical experiments indicate that the replica analysis is adequate.
481:
482: \section{Discussions}
483: \subsection{Intuitive understanding of the local temperatures}
484: Here, we discuss an intuitive understanding of the preferential urn model.
485: The heterogeneity of the inverse local temperatures indicates
486: that each urn has different ability to obtain balls.
487: The inverse local temperature allows us to obtain the following intuitive picture;
488: urns with high-temperature tend to kick out balls,
489: and those with low-temperature are likely to stock balls.
490: Therefore, when there is no randomness,
491: all urns compete for balls, so that a fat-tailed behavior does not occur.
492: On the contrary, when an urn has a lower-temperature,
493: the urn tends to get a lot of balls.
494: Hence, it is easy to imagine that urns with a lot of balls can emerge due to randomness.
495:
496:
497: \subsection{Restriction for the inverse local temperature $\beta$}
498:
499: We note that the value of the inverse local temperature $\beta_i$ has a restriction:
500: the value of $\beta_i$ can not be over $1$.
501: For example, eq.~\eqref{eq_g0} does not converge if $\beta > 1$.
502: The restriction of $\beta < 1$ means that the urn with $\beta > 1$ is too cold to kick out balls in the urn,
503: so that the model does not work well.
504: Actually, we have confirmed by numerical experiments that almost all balls concentrate in an urn with $\beta > 1$.
505: Therefore, the restriction of $0 \leq \beta \leq 1$ is necessary.
506:
507: \subsection{Annealing systems}
508:
509: As mentioned in \S~3,
510: it is difficult to calculate the configurational average of the numerator and denominator
511: at the same time, i.e., the configurational average of eq.~\eqref{eq_p_k_0}.
512: Hence, we use the replica method to calculate the configurational average.
513: The difficulty stems from the fact that the preferential urn model has \textit{quenched disorder};
514: evaluation of a physical quantity is taken for a given fixed (quenched) set of $\beta_i$,
515: so that the inverse local temperatures are time-independent.
516: However, if we consider an annealing system, i.e., the inverse local temperatures are not quenched
517: and changed with time enough fastly,
518: we can easily take the average of eq.~\eqref{eq_p_k_0},
519: and there is no need to use the replica method.
520:
521: For the annealing system,
522: \begin{align}
523: \left\langle \frac{Z_2}{Z_1} \right\rangle_{\{ \beta_2, \dots, \beta_N \}}
524: = \frac{\left\langle Z_2 \right\rangle_{\{ \beta_2, \dots, \beta_N \}}}
525: {\left\langle Z_1 \right\rangle_{\{ \beta_2, \dots, \beta_N \}}},
526: \end{align}
527: and these quantities of $\left\langle Z_1 \right\rangle_{\{ \beta_2, \dots, \beta_N \}}$
528: and $\left\langle Z_2 \right\rangle_{\{ \beta_2, \dots, \beta_N \}}$
529: are easily obtained.
530: Finally, we can get the equilibrium occupation distribution for the annealing system
531: as the same form of eq.~\eqref{eq_distribution}.
532: It is worth pointing out that
533: the case of the annealing system gives the same results as for the quenched system.
534: For the annealing system,
535: we can straightforwardly get the equilibrium occupation distribution without using the replica method,
536: then one might consider that there is no need to use replica method to analyze urn models with randomness.
537: However, when we have performed numerical experiments for the annealing system
538: in which all inverse local temperature vary at each time step,
539: the fat-tailed behavior does not occur.
540: Hence, it is an open question why the result for the annealing system agree in that for the quenched system,
541: and how situation corresponds to the annealing system.
542: At least, we consider that the replica method is adequate for analyzing urn models with quenched randomness.
543:
544:
545: \subsection{Econophysical application}
546:
547: The preferential urn model stems from the analysis of the nongrowing models for complex networks \cite{Ohkubo2005c},
548: but it would be a useful model for various phenomena.
549: For example, we consider an econophysical model in which many agents exchange their wealth.
550: Each agent corresponds to each urn,
551: and the transfer procedure of their wealth is represented by the transfer process of balls.
552: We here assume the transition rate $W_{k_i \to k_i+1} \propto (k_i+1)^{\beta_i}$,
553: where $k_i$ is the amount of wealth of agent $i$,
554: and $\beta_i$ represents an ability of agent $i$ for obtaining wealth.
555: The above dynamics indicates the following statements:
556: \begin{itemize}
557: \item A wealth to be transferred is selected at random.
558: It means that agents with large wealth are likely to loose their wealth
559: (Agents with large wealth have a lot of opportunity to consume their wealth.)
560: \item Agents with larger wealth have the chance to get wealth in higher probability.
561: \item Each agent has a different ability for obtaining wealth.
562: \end{itemize}
563: Hence, the above model seems reasonable for an econophysical one.
564: The econophysical model can be treated analytically, as shown in the present paper;
565: supposing that the distribution of their abilities $\beta_i$ is an uniform one,
566: we conclude that the wealth distribution has a fat-tailed form of eq.~\eqref{eq_asymptotic}
567: when total wealth in the whole system is enough large.
568: Though the fat-tailed form has the squared inverse logarithmic correction,
569: the cumulative distribution of eq.~\eqref{eq_asymptotic} is similar to the Parato's distribution\cite{Newman2005}.
570: The preferential urn model has not been proposed in order to the Parato's distribution,
571: so that it is unclear that the above discussion is suitable; the model is just a trial one.
572: However, it seems reasonable to assume that a system includes some randomness;
573: each agent is in various environments and has a different ability.
574: Furthermore, the preferential concept also seems to be adequate for real econophysical systems.
575:
576: \section{Concluding Remarks}
577:
578: In summary, we analyzed the preferential urn model with randomness using the replica method.
579: While it is in general difficult to analyze urn models with randomness,
580: the preferential urn models with uniform randomness for inverse local temperatures
581: can be treated analytically by using the replica method.
582: From the analysis,
583: it was revealed that the equilibrium occupation distribution has a fat-tailed form when the density is enough large.
584: The fat-tailed distribution does not follow a pure power-law distribution;
585: the distribution has a generalized power law with a squared inverse logarithmic correction.
586:
587: The preferential urn model would produce various further studies,
588: such as nonequilibrium properties, other variation of the urn models with randomness,
589: and applications of the urn models for various other research fields.
590: It will also be important to discuss relationships with interacting particle systems,
591: such as the zero-range process \cite{Evans2005}.
592:
593: \section*{Acknowledgment}
594: The authors thank Naoki Masuda for his helpful comments on the manuscript.
595: This work was supported in part by grant-in-aid for scientific research (No. 14084203 and No. 17500134)
596: from the Ministry of Education, Culture, Sports, Science and Technology, Japan.
597: J. O. was partially supported by Research Fellowship from Japan Society
598: for the Promotion of Science (JSPS) for Young Scientists.
599:
600:
601: \appendix
602: \section{Derivation of eq.~\eqref{eq_asymptotic}}
603: When we approximate the saddle-point $z_\textrm{s} = 1$ for $\rho \gg 1$,
604: the equilibrium occupation distribution of eq.~\eqref{eq_distribution} is given by
605: \begin{align}
606: P(k) &= \int_0^1 \textrm{d}\beta\frac{(k!)^{\beta-1}}{\sum_{n=0}^\infty (n!)^{\beta-1}}.
607: \end{align}
608: By using the partial integration twice,
609: the approximate equilibrium occupation distribution for large $k$ region is obtained as follows:
610: \begin{align}
611: P(k)
612: &= \frac{1}{\ln (k!)} \left[ \frac{1}{\sum_{n=0}^\infty 1} -
613: e^{-\ln k!}\frac{1}{\sum_{n=0}^\infty (n!)^{-1}} \right] \nonumber \\
614: & \quad + \frac{1}{(\ln (k!))^2} \left[ \frac{H'(1,1)}{H(1,1)^2}
615: - e^{- \ln (k!) } \frac{H'(1,0)}{H(1,0)^2} \right] \nonumber \\
616: & \quad + \frac{1}{O( (\ln (k!))^3 )} \nonumber \\
617: &\sim \frac{1}{(\ln (k!))^2} \frac{\sum_{n=0}^\infty \ln (n!)}{\left( \sum_{n=0}^\infty 1 \right)^2} \nonumber \\
618: & = \frac{1}{(\ln (k!))^2}
619: \lim_{S\to \infty }\frac{\sum_{n=0}^S \ln (n!)}{\left( \sum_{n=0}^S 1 \right)^2} \nonumber \\
620: & \sim \frac{1}{(\ln (k!))^2} \lim_{S\to \infty }\frac{O(S^2)}{O(S^2)} \nonumber \\
621: & \sim k^{-2} \frac{1}{(\ln k)^2},
622: \end{align}
623: where we use the following relation \cite{Prudnikov1986},
624: \begin{align}
625: \lim_{S \to \infty} \sum_{n=0}^S \ln (n!) \sim \lim_{S \to \infty } O(S^2),
626: \end{align}
627: and the Stirling formula, $\ln (k!) \sim k \ln k - k$.
628:
629: %\bibliographystyle{unsrt.bst}
630: %\bibliography{urn_replica}% Produces the bibliography via BibTeX.
631:
632: \begin{thebibliography}{99}
633:
634: \bibitem{Ritort1995} F. Ritort: Phys. Rev. Lett. {\bf 75} (1995) 1190.
635: \bibitem{Bialas1997} P. Bialas, Z. Burda and D. Johnston: Nucl. Phys. B {\bf 493} (1997) 505.
636: \bibitem{Drouffe1998} J. M. Drouffe, C. Godr{\` e}che and F. Camia: J. Phys. A {\bf 31} (1998) L19.
637: \bibitem{Bialas1998} P. Bialas, Z. Burda, and D. Johnston: Nucl. Phys. B {\bf 542} (1999) 413.
638: \bibitem{Bialas2000} P. Bialas, L. Bogacz, Z. Burda, and D. Johnston: Nucl. Phys. B {\bf 575} (2000) 599.
639: \bibitem{Godreche2001} C. Godr{\` e}che and J. M. Luck: Eur. Phys. J. B {\bf 23} (2001) 473.
640: \bibitem{Leuzzi2002} L. Leuzzi and F. Ritort: Phys. Rev. E {\bf 65} (2002) 056125.
641: \bibitem{Burda2002} Z\ Burda, D. Johnston, J. Jurkiewicz, M. Kami{\'n}ski, M. A. Nowak, G. Papp, and I. Zahed:
642: Phys. Rev. E {\bf 65} (2002) 026102.
643: \bibitem{Evans2005} M. R. Evans and T. Hanney: J. Phys. A: Math. Gen. {\bf 38} (2005) R195.
644: \bibitem{Ohkubo2005c} J. Ohkubo, M. Yasuda and K. Tanaka: Phys. Rev. E {\bf 72} (2005) 065104(R).
645: \bibitem{Newman2005} M. E. J. Newman: Contemporary Physics {\bf 46} (2005) 323.
646: \bibitem{Nishimori2001} H. Nishimori:
647: {\it Statistical Physics of Spin Glasses and Information Processing: An Introduction}
648: (Oxford University Press, Oxford, 2001).
649: \bibitem{Prudnikov1986}
650: A. P. Prudnikov, Y. A. Brychkov and O. I. Marichev:
651: {\it Integrals and series; translated from the Russian by N.M. Queen}
652: (Gordon and Breach Science Publishers, New York, 1986) Vol.1.
653:
654:
655: \end{thebibliography}
656:
657: \end{document}
658: