1: \documentclass[twocolumn,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}

2: 

3: 

4: \usepackage{graphicx}% Include figure files

5: \usepackage{dcolumn}% Align table columns on decimal point

6: \usepackage{bm}% bold math

7: 

8: \begin{document}

9: 

10: \title{Scale-free networks without growth}

11: \author{Yan-Bo Xie}

12: \author{Tao Zhou}

13: \author{Bing-Hong Wang}

14: \email{bhwang@ustc.edu.cn, Fax:+86-551-3603574.}

15: \affiliation{%

16: Department of Modern Physics and Nonlinear Science Center and ,

17: University of Science and Technology of China, Hefei, 230026, PR

18: China

19: }%

20: 

21: \date{\today}

22: 

23: \begin{abstract}

24: In this letter, we proposed an ungrowing scale-free network model,

25: wherein the total number of nodes is fixed and the evolution of

26: network structure is driven by a rewiring process only. In spite

27: of the idiographic form of $G$, by using a two-order master

28: equation, we obtain the analytic solution of degree distribution

29: in stable state of the network evolution under the condition that

30: the selection probability $G$ in rewiring process only depends on

31: nodes' degrees. A particular kind of the present networks with $G$

32: linearly correlated with degree is studied in detail. The analysis

33: and simulations show that the degree distributions of these

34: networks can varying from the Possion form to the power-law form

35: with the decrease of a free parameter $\alpha$, indicating the

36: growth may not be a necessary condition of the self-organizaton of

37: a network in a scale-free structure.

38: \end{abstract}

39: 

40: \pacs{89.75.Hc, 64.60.Ak, 84.35.+i, 05.40.-a}

41: 

42: \maketitle

43: 

44: 

45: Many social, biological, and communication systems can be properly

46: described as complex networks with nodes representing individuals

47: and edges mimicking the interactions among them

48: \cite{Review1,Review2,Review3}. Examples are numerous: these

49: include the Internet, the World Wide Web, social networks,

50: metabolic networks, brain networks, food webs, and many others.

51: Recent empirical studies indicate that the networks in various

52: fields have some common characteristics, the most important of

53: which are called small-world effect \cite{WS} and scale-free

54: property \cite{BA}. The ubiquity of complex networks and the

55: discoveries of the common network properties inspire scientists to

56: construct general models. In the simplest way, these models can be

57: categorized into two classes, the growing models and the ungrowing

58: ones. In the growing models, the number of nodes in network grows

59: continuously, while in ungrowing cases, the number of nodes is

60: fixed. The Watts-Strogatz (WS) model is the pioneer and

61: representative one of ungrowing models, which can be constructed

62: by starting with a regular network and randomly moving one

63: endpoint of each edge with probability $p$ \cite{WS}. WS networks

64: display small-world effect, that is, they are highly clustered and

65: of small average distance. The most successful one of growing

66: models is the Barab\'{a}si-Albert (BA) model \cite{BA}, which

67: suggests that two main ingredients of the self-organization of a

68: network in a scale-free structure are growth and preferential

69: attachment. These points to the facts that most networks grow

70: continuously by adding new nodes, which are preferentially

71: attached to existing nodes with a large number of neighbors.

72: 

73: By using the mean-field theory on a toy model, Barab\'{a}si

74: \emph{et al} assert that the growth is one of the necessary

75: conditions for the emergence of scale-free property \cite{BAJ},

76: and the networks without growth may display an exponential degree

77: distribution. This hypothesis is now widely accepted. However,

78: some real-life networks having fixed size (or almost fixed size)

79: are of scale-free property. For instance, consider the friendship

80: networks of school children wherein the students are represented

81: by nodes and two nodes are connected by an edge if the

82: corresponding students have good friendship

83: \cite{friend1,friend2}. These networks do not grow with time since

84: the number of students in a class is almost fixed. Some students

85: have many good friends, while many others have only a few friends.

86: The students' heterogeneous societal ability leads to a skewed

87: degree distribution (or a skewed strength distribution for

88: weighted case \cite{friend2}), approximated to a power-law form. A

89: typical example in biological systems is the functional networks

90: of human brain \cite{Brain}. Although its structure varies with

91: time, the brain functional network's sizes is unchanged \cite{ex1}

92: while it displays scale-free property. Another significant

93: examples are the relationship networks of countries like world

94: trade web \cite{WTW1,WTW2} and world exchange arrangements web

95: \cite{WEAW1,WEAW2}. These networks are scale-free, while their

96: sizes almost do not change in the recent a few years.

97: 

98: Although these networks mentioned above have (almost) fixed size,

99: they are not static since their structures vary with time. These

100: motions can be considered as rewiring processes, that is, some

101: existing edges are removed while some new ones are added. For

102: example, in friendship networks, good friends may quarrel about

103: beliefs, money, or some other things, and become impassive to each

104: other; while some ones will become good friends for their common

105: interests and difficulties. The economic globalization promotes

106: different countries' economies to be integrated together in terms

107: of trade and capital flow, making countries more interdependent

108: than ever. In order to avoid financial crises, more flexible

109: economic regime should be implemented, and the government may

110: change the exchange-rate/trade regime for certain economic and

111: political reasons, leading redistribution of degree/weight in

112: world exchange arrangement networks and world trade networks.

113: These redistribution/rewiring processes make the structures

114: varying ceaselessly in recent years, but the scale-free property

115: is always observed. Notice that, in the ungrowing model proposed

116: by Barab\'{a}si \emph{et al} \cite{BAJ}, no edges will be removed

117: thus no rewiring processes will occur. In this letter, for the

118: first time to our knowledge, a scale-free network model without

119: growth is proposed, and we argue that the ungrowing networks can

120: have power-law degree distributions, which attributes to the

121: rewiring processes.

122: 

123: \begin{figure}

124: \scalebox{0.8}[0.9]{\includegraphics{graph1}} \caption{(Color

125: online) The degree distribution of the present model for the case

126: $\alpha \gg 1$. The black hollow squares and red solid circles

127: represent the analytic and simulation results, respectively. The

128: analytic result agrees with the simulation accurately, and both of

129: them obey Possion forms. }

130: \end{figure}

131: 

132: For simplicity, we consider a network model with both fixed number

133: of nodes $N$ and fixed number of edges $E$. The initial state of

134: the network can be a random graph or any other types of graph.

135: Then the network evolves based on the following rewiring

136: processes: At each time step, an edge is randomly selected and

137: removed from the network.  At the same time, a node is selected

138: with the preferential probability $G(k_i)$, where $k_i$ is the

139: degree of the $i$th node. Another node is selected also with the

140: above preferential probability. Then a new edge connecting these

141: two nodes is created. Notice that in this rewiring process, the

142: total number of edges is unchanged. The above process is repeated

143: at each time step. Finally, we expect that the network reaches an

144: equilibrium state which is independent of the initial state.

145: 

146: Let $P(k,t)$ represents the number of nodes with degree $k$ at the

147: time step $t$.  $P(k,t)$ satisfies the following normalization

148: conditions

149: \begin{equation}

150: \sum_k P(k,t)=N,

151: \end{equation}

152: \begin{equation}

153: \sum_k kP(k,t)=2E,

154: \end{equation}

155: \begin{equation}

156: \sum_k G(k)P(k,t)=1.

157: \end{equation}

158: It is straightforward to write

159: down the master equation for $P(k,t)$,

160: \begin{eqnarray}

161: &&P(k,t+1)-P(k,t)={(k+1)P(k+1,t)-k P(k,t)\over E}\nonumber\\

162: &&+2G(k-1)P(k-1,t)-2G(k)P(k,t).

163: \nonumber\\

164: \end{eqnarray}

165: It may be helpful to explain the physical meaning of various terms

166: in the above equation.  The first two terms on the right hand side

167: represent the net gain of $P(k,t)$ due to the edge removing

168: process.  In particular, the first term represents the net gain of

169: $P(k,t)$ when the removing edge connects a node with the degree

170: $k+1$.  Notice that when an edge is randomly selected, the

171: connecting nodes have more chance to be of large degree.

172: Explicitly, the node with degree $k$ is selected with the

173: probability $k/\sum_k kP(k,t)=k/2E$.  Since each edge connects to

174: two nodes, the first term is thus $(k+1)P(k+1,t)/E$.  Similarly,

175: the second term represents the net loss of $P(k,t)$ when the

176: removing edge connects a node with the degree $k$. The third and

177: fourth terms represents the net gain of $P(k,t)$ due to the edge

178: adding process.  The third term represents the net gain of

179: $P(k,t)$ when the adding edge connects a node with the degree

180: $k-1$.  Notice that a node with degree $k$ is selected with the

181: probability $G(k)$. The fourth term represents the net loss of

182: $P(k,t)$ when the adding edge connects a node with the degree $k$.

183: 

184: When $t$ is sufficiently large, we expect that $P(k,t)$ approaches a

185: stationary state denoted by $P_s(k)$ that satisfies

186: \begin{widetext}

187: \begin{equation}

188: {(k+1)P_s(k+1)-kP_s(k)\over 2E}+G(k-1)P_s(k-1)-G(k)P_s(k)=0,

189: \end{equation}

190: \end{widetext}

191: where $P_s(k)$ also satisfy the normalization conditions

192: Eqs.(1-3).  Notice that this master equation is a second order

193: equation in which $P_s(k+1)$ is determined by $P_s(k)$ and

194: $P_s(k-1)$. Eqs.(1-3,5) can be solved by the following method.

195: Define

196: \begin{equation} H(k)=G(k)P_s(k)

197: \end{equation} and

198: \begin{equation} W(k)=kP_s(k)/2E.

199: \end{equation}

200: Then Eqs.(2-3,5) can be rewritten as

201: \begin{equation}

202: \sum_k W(k)=1, \end{equation} \begin{equation} \sum_k

203: H(k)=1,\end{equation} \begin{equation} H(k)=H(k-1)+W(k+1)-W(k).

204: \end{equation}

205: By iteration, one can obtain \begin{equation}

206: H(k)=H(0)+W(k+1)-W(1).

207: \end{equation}

208: Since $W(0)=0$ and $$\sum_k H(k)=\sum_k W(k)=1,$$ one immediately

209: have \begin{equation} W(k+1)=H(k) \end{equation} for $k\geq 0$.

210: Substituting Eqs.(6-7) into Eq.(12), one obtains the following

211: solution

212: \begin{equation}

213: P_s(k)={(2E)^k\over k!}G(k-1)G(k-2)...G(0)P_s(0)

214: \end{equation}

215: for $k\geq 1$ with $P_s(0)$ determined by the normalization

216: condition Eq.(1).  It should be stressed that given the dependence

217: of $G(k)$ on $k$, there is still a proportional coefficient in

218: $G(k)$ determined by Eq.(3).

219: 

220: Hereinafter, we focus on a special form, also one of the simplest

221: cases, of $G(k)$, where $G(k)$ is linearly correlated with $k$,

222: \begin{equation}

223: G(k)=\frac{k+\alpha}{\sum^N_{i=1}(k_i+\alpha)}={k+\alpha\over

224: 2E+N\alpha},

225: \end{equation}

226: where $k_i$ denotes the degree of node $i$, and $\alpha$ is a

227: constant (Notice that $G(k)$ satisfies the normalization condition

228: Eq. (3).). Substituting the above equation into Eq. (13), one can

229: straightforwardly obtain, in this special case, that

230: \begin{equation}

231: P_s(0)=N(\frac{\alpha}{\alpha+\gamma})^\alpha

232: \end{equation}

233: \begin{equation}

234: P_s(k)=N({\alpha\over \alpha+\gamma})^{\alpha} ({\gamma\over

235: \alpha+\gamma})^k{1\over k!}\alpha(\alpha+1)...(\alpha+k-1), k\geq

236: 1

237: \end{equation}

238: where $\gamma=2E/N$.

239: 

240: When $\alpha \gg 1$, the mechanism of preferential attachment is

241: destroyed, that is, since $\alpha$ probably plays the main part in

242: the sum $\alpha+k_i$, the probability a high degree node being

243: selected is approximately the same as that of a low degree node

244: (see Eq. 14). Therefore, one expects that the present network will

245: approach a completely random network with its degree distribution

246: obeying the Possion distribution \cite{RG}

247: \begin{equation}

248: P_s(k)=N\frac{\texttt{e}^{-\gamma}}{k!}\gamma^k.

249: \end{equation}

250: When $\alpha \ll 1$ and $\gamma \ll 1$, most nodes are isolated

251: and the nodes with large degree are very few. However, when

252: $\alpha \ll 1$ and $\gamma \succeq 1$, interesting result emerges.

253: The corresponding degree distribution obeys a power-law form with

254: exponent approximated to 1 when $k\ll \gamma / \alpha$

255: \begin{equation}

256: P_s(k)=N(\frac{\alpha}{\gamma})^\alpha\frac{\alpha}{k}.

257: \end{equation}

258: 

259: \begin{figure}

260: \scalebox{0.8}[0.9]{\includegraphics{graph2}} \caption{The degree

261: distribution of the present model with $\alpha=0.01$ and

262: $\gamma=5$. The hollow squares and solid curve represent the

263: simulation and analytical results, respectively. The analytical

264: result agrees with the simulation well, obeying an approximately

265: power-law form. The inset exhibits the case of $\alpha=0.01$ and

266: $\gamma=10$ for comparison.}

267: \end{figure}

268: 

269: 

270: 

271: 

272: In succession, we will show some simulations. $10^6$ time steps

273: are performed in all simulations while only the final $2\times

274: 10^5$ time steps are used for statistical average. The initial

275: graph is simply the completely random graph with given $N$ and

276: $E$. The network size if fixed as $N=1000$ unless a special

277: statement is addressed. In figure 1, we report the case for very

278: large $\alpha$, where $p_s(k):=P_s(k)/N$ denotes the normalized

279: probability function for degree distribution. The simulation

280: result strongly supports the analytic one (see Eq. (16)), and both

281: of them obey Possion forms. In this figure, one will find that

282: there are about 5\% nodes with degree zero. However, this

283: probability only reflects the average number of isolated nodes

284: over all configurations. Because of the rewiring mechanism, no

285: nodes are always isolated. Although real-life networks are not

286: always connected, most previous studies have focused on connected

287: graphs. Therefore, hereinafter, we neglect the isolated nodes.

288: 

289: \begin{figure}

290: \scalebox{0.8}[0.9]{\includegraphics{graph3}} \caption{(Color

291: online) The degree distributions of the present model. The

292: parameter $\gamma=10$ is fixed, and the black squares and red

293: circles represent the cases with $\alpha=1.0$ and $\alpha=0.1$,

294: respectively. The degree distribution with $\alpha=1.0$ obeys

295: approximately an exponential form rather than a power-law form.}

296: \end{figure}

297: 

298: 

299: In figure 2, we report the analytical and simulation results about

300: degree distributions generated by very small $\alpha=0.01$ and

301: $\gamma=5$. The degree distribution displays obviously scale-free

302: property with a cut-off at $k \approx 120$. The solid curve

303: denotes the analytical solution, which agrees with the simulation

304: well. We also have checked that the departure from Eq.(18) will

305: get greater if the parameter $\alpha$ becomes larger. In the

306: inset, the case of $\gamma=10$ is shown for comparison. As

307: mentioned above, the two extreme cases with $\alpha\approx 0$ and

308: $\alpha\gg 1$ will lead to the power-law and Possion

309: distributions, respectively. In figure 3, we report the simulation

310: results for $\alpha=0.1$ and $\alpha=1.0$. The departure from the

311: power law becomes larger as the increase of $\alpha$, and the

312: degree distribution with $\alpha=1.0$ obeys approximately an

313: exponential form rather than a power-law form.

314: 

315: In conclusion, we proposed an ungrowing model and obtained the

316: analytic results for arbitrary forms of selection probability

317: $G(k)$. One special kind of the present networks is investigated

318: in detail, which can generate networks with different classes of

319: degree distributions by tuning two parameters. Especially, when

320: $\alpha$ is close to zero and $\gamma \succeq 1$, the present

321: networks display scale-free property with its exponent

322: approximately equal to that of world exchange arrangement web

323: \cite{WEAW2}.

324: 

325: The previous empirical studies have demonstrated the extensive

326: existence of scale-free networks. Most of them are growing at all

327: times, while some others are of (almost) fixed size. Most previous

328: models \cite{Review1,Review2,Review3,BA,BAJ} suggested the growth

329: a key ingredients of the emergence of scale-free structure, thus

330: fail to provide an underlying mechanism by which the ungrowing

331: networks exhibiting scale-free property. Here we argue that the

332: rewiring mechanism widely existed in real-life networks like

333: friendship networks \cite{friend1,friend2}, brain functional

334: networks \cite{Brain}, world trade web \cite{WTW1,WTW2}, and world

335: exchange arrangements web \cite{WEAW1,WEAW2}, may be the origin of

336: the emergence of scale-free structure in fixed-size networks.

337: 

338: 

339: Finally, we would like to point out that many different classes of

340: degree distributions can be obtained from different forms of

341: $G(k)$ \cite{unpub}. In particular, one can obtain the power law

342: degree distribution $P_s(k)\sim k^{-\beta-1}$ with tunable

343: exponent $\beta$ by setting $G(k)\sim

344: \frac{k^{\beta+1}}{(k+1)^\beta}$ when the number of edges $E$ is

345: properly chosen \cite{unpub}.

346: 

347: This work is support by the National Natural Science Foundation of

348: China under Nos. 70271070, 10472116, 70471033, and 70571074, and

349: the Specialized Research Fund for the Doctoral Program of Higher

350: Education under No. 20020358009.

351: 

352: \begin{thebibliography}{Reviews}

353: 

354: \bibitem{Review1} R. Albert and A. -L. Barab\'{a}si, Rev. Mod. Phys. {\bf 74}, 47(2002).

355: \bibitem{Review2} S. N. Dorogovtsev and J. F. F. Mendes, Adv. Phys. {\bf 51}, 1079(2002).

356: \bibitem{Review3} M. E. J. Newman, SIAM Review {\bf 45}, 167(2003).

357: \bibitem{WS} D. J. Watts and S. H. Strogatz, Nature {\bf 393}, 440(1998).

358: \bibitem{BA} A. -L. Barab\'{a}si and R. Albert, Science {\bf 286}, 509(1999).

359: \bibitem{BAJ} A. -L. Barab\'{a}si, R. Albert, and H. Jeong, Physica A {\bf 272}, 173(1999).

360: \bibitem{friend1} A. Rapoport, and W. J. Horvath, Behav. Sci. {\bf 6}, 279(1961).

361: \bibitem{friend2} B. Hu, X. -Y. Jiang, J. -F. Ding, Y. -B. Xie, and B. -H. Wang, Physica A {\bf 353}, 576(2005).

362: \bibitem{Brain} V. M. Egu\'{i}luz, D. R. Chialvo, G. A. Cecchi, M. Baliki, and A. V. Apkarian, Phys. Rev. Lett. {\bf 94}, 018102(2005).

363: \bibitem{ex1} In human brain functional networks, each node

364: represents a voxels of dimension $3\times 3.475\times 3.475

365: \texttt{mm}^3$ in the picture taken by functional magnetic

366: resonance imaging (FMRI) in humans. Therefore, the network size is

367: fixed as $N=36\times 64\times 64$.

368: \bibitem{WTW1} M. A. Serrano, and M. Bogu\~{n}\'{a}, Phys. Rev. E

369: {\bf 68}, 015101(2003).

370: \bibitem{WTW2} X. Li, Y. -Y. Jin, and G. Chen, Physica A {\bf 328}

371: 287(2003).

372: \bibitem{WEAW1} M. Mussa, P. Masson, A. Swoboda, E. Jadresic, P. Mauro, and A. Berg, \emph{Exchange Rate Regimes in an

373: Increasing Integrated World Economy}, International Monetary Fund,

374: Washington DC, 2002.

375: \bibitem{WEAW2} X. Li, Y. -Y. Jin, and G. Chen, Physica A {\bf 343}

376: 573(2004).

377: \bibitem{RG} B. Bollob\'{a}s, {\it Random Graphs} (Academic Press, New York, 2001).

378: %\bibitem{MGT} B. Bollob\'{a}s, {\it Morden Graph Theory} (Springer-Verlag, New York, 2002).

379: \bibitem{unpub} Y. -B. Xie, T. Zhou, and B. -H. Wang, to be

380: published.

381: 

382: 

383: \end{thebibliography}

384: 

385: \end{document}

386: