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

2: 

3: \usepackage{graphicx}% Include figure files

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

5: \usepackage{bm}% bold math

6: 

7: \begin{document}

8: 

9: \title{Epidemic spread in weighted scale-free networks}

10: \author{YAN Gang$^{1}$}

11: \author{ZHOU Tao$^{1,2}$}

12: \author{WANG Jie$^{1}$}

13: \author{FU Zhong-Qian$^{1}$}

14: \author{WANG Bing-Hong$^{2}$}

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

16: \affiliation{%

17: $^{1}$Electronic Science and Technology,\\

18: University of Science and Technology of China,\\

19: Hefei Anhui, 230026, PR China \\

20: $^{2}$Nonlinear Science Center and Department of Modern Physics,\\

21: University of Science and Technology of China,\\

22: Hefei Anhui, 230026, PR China }%

23: 

24: \date{\today}

25: 

26: \begin{abstract}

27: In this letter, we investigate the detailed epidemic spreading

28: process in scale-free networks with links' weights that denote

29: familiarity between two individuals and find that spreading

30: velocity reaches a peak quickly then decays in a power-law form.

31: Numerical study exhibits that the nodes with larger strength is

32: preferential to be infected, but the hierarchical dynamics are not

33: clearly found, which is different from the well-known result in

34: unweighed network case. In addition, also by numerical study, we

35: demonstrate that larger dispersion of weight of networks results

36: in slower spreading, which indicates that epidemic spreads more

37: quickly on unweighted scale-free networks than on weighted

38: scale-free networks with the same condition.

39: \end{abstract}

40: 

41: \pacs{89.75.-k, 89.75.Hc, 87.23.Ge, 05.70.Ln}

42: 

43: \maketitle

44: 

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

46: described as complex networks with vertices representing

47: individuals or organizations and links mimicking the interactions

48: among them. Recently, the ubiquity of a power-law degree

49: distribution in real-life networks has attracted a lot of

50: attention\cite{Reviews}. Examples of such networks (scale-free

51: networks or SF networks for short) are numerous: these include the

52: Internet, the World Wide Web, social networks of acquaintance or

53: other relations between individuals, metabolic networks, integer

54: networks, food webs, etc.\cite{Networks}. The ultimate goal of the

55: study on topological structure of networks is to understand and

56: explain the workings of systems built upon those networks, for

57: instance, to understand how the topology of the World Wide Web

58: affects Web surfing and search engines, how the structure of

59: social networks affects the spread of diseases, information,

60: rumors or other things, how the structure of a food web affects

61: population dynamics, and so on.

62: 

63: Recent studies on epidemic spreading in SF networks indicate a

64: particular relevance in the case of networks characterized by

65: complex topologies and very heterogeneous

66: structures\cite{Reviews,Stanley 01} that in many cases present us

67: with new epidemic propagation scenarios\cite{Ep,B.B.P.V 04}, such

68: as absence of any epidemic threshold\cite{Ep}, hierarchical spread

69: of epidemic outbreaks\cite{B.B.P.V 04}, and so on. The new

70: scenarios are of practical interest in computer virus diffusion

71: and the spreading of diseases in heterogeneous populations.

72: Further more, they also raise new questions on how to protect the

73: networks and find optimal strategies for the deployment of

74: immunization resources\cite{Imm}. However, so far, studies of

75: epidemic spread just focus on unweighted SF networks, and a

76: detailed inspection of epidemic spreading process in weighted SF

77: networks is still missing while real networks, such as population

78: and Internet, are obviously scale-free and with links' weights

79: that denote familiarity between two individuals(like people or

80: computers), respectively. One can easily take cognizance of how

81: the links' weights affect the epidemic spreading process. For

82: instance, if your little son gets flu, then you will be infected

83: in all probability, since you two contact each other very

84: frequently(i. e. of large familiarity). By contraries, it is

85: unlikely that you will be infected by your unfamiliar colleague

86: just because of saying hello to him this morning.

87: 

88: \begin{figure}

89: \scalebox{0.8}[0.8]{\includegraphics{1}}

90: \caption{\label{fig:epsart}Density of infected individuals versus

91: time in a BBV network with $N=10^4,\delta=3.0,\omega_{{\rm

92: 0}}=1.0$ and $m=3$, the four numerical curves $i(t)$ correspond

93: with parameter $\alpha =$0.5, 0.4, 0.33 and 0.25 respectively,

94: form bottom to top.}

95: \end{figure}

96: 

97: In this letter, we intend to provide a first analysis of the time

98: evolution of epidemic spreading in weighted SF networks. The

99: weighted SF network model used in this letter is one of the most

100: well-known model introduced by Barrat, Barthelemy, and Vespignani

101: (BBV networks)\cite{B.B.V 04}, whose degree, strength and weight

102: distributions are power-law distributions with heavy tails. The

103: BBV model suggests that two main ingredients of self-organization

104: of a network in a weighted scale-free structure are strength

105: preferential attachment and weights' dynamics. These point to the

106: facts that most networks continuously grow by the addition of new

107: vertices, new vertices are preferentially attached to existing

108: vertices with larger strength, and the creation of new links will

109: introduce variations of the existing weight distribution. More

110: precisely, the weight of each new edge is fixed to value

111: $\omega_{0}$; if a new vertex linked to an existing vertex $i$,

112: then the local rearrangement of weights between $i$ and its

113: neighbors $j$ according to the simple rule

114: \begin{equation}

115: \omega_{ij}\rightarrow \omega_{ij}+\Delta\omega_{ij}

116: \end{equation}

117: where

118: \begin{equation}

119: \Delta\omega_{ij}=\delta\frac{\omega_{ij}}{s_{i}}

120: \end{equation}

121: $s_{i}$ is the strength of node $i$, expressed by

122: $s_{i}=\sum_{j}\omega_{ij}$. This rule considers that the

123: establishment of a new edge of weight $\omega_{0}$ with the vertex

124: $i$ induces a total increase of traffic $\delta$ that is

125: proportionally distributed among the edges departing from the

126: vertex according to their weights. Since BBV networks are of the

127: same properties (e.g. power-law distribution of degree, strength

128: and weight) as many social networks (e.g. friendship networks and

129: scientists collaboration networks) and technical networks (e.g.

130: Internet and WWW), it is reasonable to investigate epidemic

131: spreading on BBV networks.

132: 

133: 

134: \begin{figure}

135: \scalebox{0.75}[0.8]{\includegraphics{2}}

136: \caption{\label{fig:epsart} Density of infected individuals versus

137: time in a BBV network with $N=10^4,\alpha=2.0,\omega_{{\rm

138: 0}}=1.0$ and $m=3$, the four numerical curves $i(t)$ correspond

139: with parameter $\delta=$0.4, 0.3, 0.2 and 0.1 respectively, form

140: bottom to top. The inset shows the relationship between the

141: dispersion of weight ($\mu$) and the value of $\delta$.}

142: \end{figure}

143: 

144: \begin{figure}

145: \scalebox{0.9}[0.8]{\includegraphics{3}}

146: \caption{\label{fig:epsart} Spreading velocity at each time $t$ in

147: a BBV network with $N=10^4,\delta=3.0,\omega_{{\rm 0}}=1.0$ and

148: $m=3$,when $\alpha=0.33(square)$ and $\alpha=0.25(circle)$. The

149: inset shows the curves in log-log plot. The data are averaged over

150: 200 experiments.}

151: \end{figure}

152: 

153: 

154: In order to study the dynamical evolution of epidemic spreading we

155: shall focus on the susceptible-infected (SI) model in which

156: individuals can be in two discrete states, either susceptible or

157: infected\cite{Aderson 92}. Each individual is represented by a

158: vertex of the network and the links are the connections between

159: individuals along which the infection may spread. The total

160: population(the size of the network) $N$ is assumed to be constant

161: thus if $S(t)$ and $I(t)$ are the number of susceptible and

162: infected individuals at time $t$, respectively, then $N=S(t)+I(t)$

163: . In weighted networks, we define the infection transmission by

164: the spreading rate,

165: \begin{equation}

166: \lambda_{ij}=(\frac{\omega_{ij}}{\omega_M})^\alpha, \alpha>0

167: \end{equation}

168: at which susceptible individual $i$ acquire the infection from the

169: infected neighbor $j$, where $\alpha$ is a positive constant and

170: $\omega_{M}$ is the largest value of $w_{ij}$ in the network.

171: Obviously, more familiar two individuals(i.e. with larger weight)

172: may infect each other with greater probability. According to

173: Eq.(3), one can quickly obtain the probability that an susceptible

174: individual $i$ will be infected at the present time step is:

175: \begin{equation}

176: \lambda_i(t)=1-\prod_{j\in N_i(t)}(1-\lambda_{ij})

177: \end{equation}

178: where $N_i(t)$ is the set of all $i$'s infected neighbors at time

179: $t$.

180: 

181: \begin{figure}

182: \scalebox{0.95}[0.8]{\includegraphics{4a}}

183: \scalebox{0.87}[0.8]{\includegraphics{4b}}

184: \caption{\label{fig:epsart} Consider the cases of $\delta<1.0$ and

185: $\delta=1.0$. (a) and (b) show the spreading velocity at each time

186: $t$ in a BBV network with $N=10^4,\omega_{0}=1.0$ and $m=3$, when

187: $\delta=0.5, \alpha=2.0$ and $\delta=1.0, \alpha=1.0$,

188: respectively. The inset shows the curves in log-log plot. The data

189: are averaged over 200 experiments.}

190: \end{figure}

191: 

192: We start by selecting one vertex randomly and assume it is

193: infected. The diseases or computer virus will spread in the

194: networks in according with the rule of Eq.(3). In \emph{Fig.1}, we

195: plot the density of infected individuals versus time in a BBV

196: network with $N=10^4,\delta=3.0,\omega_{{\rm 0}}=1.0$ and $m=3$.

197: Since $\frac{\omega_{ij}}{\omega_M}\leq1$, the smaller $\alpha$

198: is, the more quickly infection spreads. It is natural that larger

199: value of $\delta$ induces larger dispersion of weight of networks.

200: Then, a direct question is that how the value of $\delta$ impacts

201: epidemic spreading behavior. In \emph{Fig.2}, we show that

202: epidemic spreads more quickly while $\delta$ is smaller. In other

203: words, larger dispersion of weight of networks results in slower

204: spreading. That means epidemic spreads more quickly on unweighted

205: scale-free networks than on weighted scale-free networks with the

206: same condition.

207: 

208: Obviously, all the individuals will be infected in the limit of

209: long time as $\lim_{t\rightarrow \infty}i(t)=1$. For the sake of

210: finding optimal strategies to protect individuals from being

211: infected, we will study the details of spreading velocity at the

212: outbreak moment. The spreading velocity is defined as:

213: \begin{equation}

214: V_{{\rm inf}}(t)=\frac{di(t)}{dt}\approx \frac{I(t)-I(t-1)}{N}

215: \end{equation}

216: where $i(t)=\frac{I(t)}{N}$. We account the number of newly

217: infected vertices at each time step and report the spreading

218: velocity in {\it Fig.3}. Apparently, the spreading velocity goes

219: up to a peak quickly that similar to the unweighted network

220: cases\cite{B.B.P.V 04}, leaving us very short response time to

221: develop control measures. Moreover, what's new and interesting,

222: velocity decays following power-law form after the ``peak time".

223: At the moment of infection outbreaks, the number of infected

224: individuals is very small, as well as after a very long time from

225: the outbreak, the number of susceptible individuals is very small.

226: Thus when $t$ is very small(close to zero) or very large, the

227: spreading velocity is close to zero, one can see the corresponding

228: simulation result in {\it Fig.3}. One may think that the velocity

229: follows power-law behavior just because of the extreme case of

230: $\delta>1.0$. Now we consider the case of $\delta<1.0$ and

231: $\delta=1.0$. {\it Fig.4} shows spreading velocity at each time

232: $t$ in a BBV network with $N=10^4,\omega_{0}=1.0$ and $m=3$, when

233: $\delta=0.5, \alpha=2.0$ and $\delta=1.0, \alpha=1.0$,

234: respectively. It is obvious that epidemic spreading behavior dose

235: not show sensitive dependence on the parameter $\delta$, the

236: reason of that fact will be explored deeply in our future

237: publications.

238: 

239: In order to give a more precise characterization of the epidemic

240: diffusion through the weighted networks, we measure the average

241: strength of newly infected vertices at time $t$, define as:

242: \begin{equation}

243: {\bar{S}_{{\rm inf}}(t)}=\frac{\sum_{s}s[I_{{\rm s}}(t)-I_{{\rm

244: s}}(t-1)]}{I(t)-I(t-1)}

245: \end{equation}

246: where $I_{{\rm s}}(t)$ is the number of infected vertices with

247: strength $s$. {\it Fig. 4} shows the average strength of newly

248: infected vertices $\bar{S}_{{\rm inf}}(t)$ as a function of time

249: $t$, and the curves exhibit that $\bar{S}_{{\rm inf}}(t)$ displays

250: a power-law behavior for large $t$, $\bar{S}_{{\rm inf}}(t)\propto

251: t^{-\gamma}$, which is remarkably different from the clear

252: hierarchical feature on unweighted networks\cite{B.B.P.V 04}.

253: 

254: It is explicit that the individuals with larger strength are much

255: more dangerous when they are infected, rather than the ones with

256: smaller strength, thus if one want to protect most individuals

257: being infected, the susceptible individuals with larger strength

258: must be protect foremost. In {\it Fig. 5}, one can find that the

259: individuals with larger strength is preferential to be infected,

260: which means there is little time leaving us to find the ``{\it

261: Large Individuals}" and isolate them. Therefore, at the outbreak

262: moment of disease or computer virus, the dense crowd or pivotal

263: servers must be protected primarily. Of course, the outcome is not

264: a good news for practical operators, but it may be relevant for

265: the development of containment strategies.

266: 

267: \begin{figure}

268: \scalebox{0.75}[0.8]{\includegraphics{5}}

269: \caption{\label{fig:epsart} Behavior of average strength of the

270: newly infected vertices at time $t$ for SI model spreading in a

271: BBV network with $N=10^4,\delta=3.0,\omega_{{\rm 0}}=1.0$ and

272: $m=3$, the inset shows that $\bar{S}_{{\rm inf}}(t)$ represents

273: power-law behavior, $\bar{S}_{{\rm inf}}(t)\propto t^{-\gamma}$.}

274: \end{figure}

275: 

276: In summary, we have studied epidemic spreading process in BBV

277: networks, and the present results provide a clear picture of the

278: infection propagation in weighted SF networks. The numerical

279: studies show that spreading velocity $V_{{\rm inf}}(t)$ and

280: average strength of newly infected vertices $\bar{S}_{{\rm

281: inf}}(t)$ present power-law time behavior for large $t$, which is

282: remarkably different from infection propagation in unweighted

283: networks. Also by numerical study, we demonstrate that larger

284: dispersion of weight of networks results in slower spreading,

285: which indicates that epidemic spreads more quickly on unweighted

286: scale-free networks than on weighted scale-free networks with the

287: same condition. These results indicate that not only the

288: topological structures of networks but also the links' weights

289: affect the epidemic spreading process. Further more, the detailed

290: study of behavior of average strength of the newly infected

291: vertices may be relevant for the development of containment

292: strategies.

293: 

294: However, up to now, there are so many important and fundamental

295: problems that puzzle us and haven't been referred to in the

296: present letter. Some of them have been partially solved and will

297: be publicized in further publication, and others are still

298: unanswered. At the end of this letter, we will list part of them.

299: How to analyze the average density of infected individuals versus

300: time at the outbreak moment in weighted SF networks, and how about

301: the dynamic behavior after ``peak time"? Is the mean-field theory

302: appropriate to solve this problem? How to design a optimal

303: containment strategy, and how about the effective for various

304: strategies, such as to protect vertices at random, to protect

305: vertices purposefully, to cut off links at random, to cut off

306: links purposefully, and so on?

307: 

308: This work has been partially supported by the State Key

309: Development Programme of Basic Research (973 Project) of China,

310: the National Natural Science Foundation of China under Grant

311: No.70271070, 70471033 and 10472116, the Specialized Research Fund

312: for the Doctoral Program of Higher Education (SRFDP

313: No.20020358009), and the foundation for graduate students of

314: University of Science and Technology of China under Grant No.

315: USTC-SS-0501.

316: 

317: \begin{thebibliography}{Reviews}

318: 

319: \bibitem{Reviews} Albert R and Barab\'{a}si A L 2002 \emph{Rev. Mod. Phys.} {\bf 74}

320: 47 \\ Dorogovtsev S N and Mendes J F F 2002 \emph{Adv. Phys.} {\bf

321: 51} 1079 \\ Newman M E J 2003 \emph{SIAM Review} {\bf 45} 167

322: 

323: \bibitem{Networks} Vazquez A, Pastor-Satorras R and Vespignani A 2003 \emph{Preprint cond-mat/0303516}\\

324:  Albert R, Jeong H and Barab\'{a}si A L 1999 \emph{Nature} {\bf 401} 130 \\

325:  Jeong H, Tombor B, Albert R, Oltvai Z N and Barab\'{a}si A L 2000 \emph{Nature} {\bf 407} 651\\

326:   Zhou T, Wang B H, Jiang P Q, Xie Y B and Bu S L 2004 \emph{Preprint

327:   cond-mat/0405258}\\

328:   Chi L P, Wang R, Su H, Xu X P, Zhao J S, Li

329:   W and Cai X 2003 \emph{Chin. Phys. Lett.} {\bf 20} 1393\\

330:   He Y, Zhu X and He D R 2004 {\it Int. J. Mod. Phys. B} {\bf 18}

331:   2595\\

332:   Xu T, Chen J, He Y and He D R 2004 {\it Int. J. Mod. Phys. B} {\bf 18}

333:   2599\\

334:   Fan Y, Li M, Chen J, Gao L, Di Z and Wu J 2004 {\it Int. J. Mod.

335:   Phys. B} {\bf 18} 2505

336: 

337: \bibitem{Stanley 01} Liljeros F, Edling C R, Amaral L A N,

338: Stanley H E, and Aberg Y 2001 \emph{Nature} {\bf 411} 907\\

339: Lloyd A L and May R M 2001 Science {\bf 292} 1316

340: 

341: \bibitem{Ep} Moore C and Newman M E J 2000 \emph{Phys. Rev. E}

342: {\bf 61} 5678\\

343: Abramson G and Kuperman M 2001 \emph{Phys. Rev. Lett.} {\bf 86}

344: 2909\\

345: Pastor-Satorras R and Vespignani A 2001 \emph{Phys. Rev. Lett.}

346: {\bf 86} 3200\\

347: Pastor-Satorras R and Vespignani A 2001 \emph{Phys. Rev. E} {\bf

348: 63} 066117\\

349: May R M and Lloyd A L 2001 \emph{Phys. Rev. E } {\bf 64}

350: 066112\\

351: Moreno Y, Pastor-Satorras R and Vespignani A 2002

352: \emph{Eur.Phys.J.B.} {\bf 26} 521\\

353: Newman M E J 2002 \emph{Phys. Rev. E} {\bf 64} 016128

354: 

355: \bibitem{B.B.P.V 04} Barthelemy M, Barrat A, Pastor-Satorras R,

356: Vespignani A 2004 \emph{Phys. Rev. Lett.} {\bf 92} 178701

357: 

358: \bibitem{Imm} Pastor-Satorras R and Vespignani A 2002 \emph{phys. Rev. E}

359: {\bf 63} 036104\\

360: Cohen R, Havlin S and Ben-Avraham D 2003 \emph{Phys. Rev. Lett.}

361: {\bf 91} 247901

362: 

363: \bibitem{B.B.V 04} Barrat A, Barthelemy M and Vespignani A 2004

364: \emph{Phys. Rev. Lett} {\bf 92} 228701

365: 

366: \bibitem{Aderson 92} Anderson R M and May R M 1992 \emph{Infectious disease in

367: humans} (Oxford: Oxford University Press)\\

368: Murray J D 1993 \emph{Mathematical Biology}(New York: Springer)

369: 

370: \end{thebibliography}

371: 

372: 

373: 

374: \end{document}

375: