cond-mat0509290/a.tex
1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2: %% Trim Size: 9.75in x 6.5in
3: %% Text Area: 8in (include Runningheads) x 5in
4: %% ws-mplb.tex   :   18-6-2004
5: %% TeX file to use with ws-mplb.cls written in Latex2E.
6: %% The content, structure, format and layout of this style file is the
7: %% property of World Scientific Publishing Co. Pte. Ltd.
8: %% Copyright 1995, 2002 by World Scientific Publishing Co.
9: %% All rights are reserved.
10: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
11: 
12: \documentclass{ws-ijmpb}
13: \begin{document}
14: 
15: \markboth{Jian-Guo Liu, et. al}{Optimization of scale-free network
16: for random failures}
17: 
18: %%%%%%%%%%%%%%%%%%%%% Publisher's Area please ignore %%%%%%%%%%%%%%%
19: %
20: \catchline{}{}{}{}{}
21: %
22: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
23: 
24: \title{OPTIMIZATION OF SCALE-FREE NETWORK FOR RANDOM
25: FAILURES
26: %\footnote{For the title, try
27: %not to use more than 3 lines. Typeset the title in 10 pt Times
28: %roman, uppercase and boldface.}
29: }
30: 
31: \author{JIAN-GUO LIU$^\dag$, ZHONG-TUO WANG and YAN-ZHONG
32: DANG}
33: %\footnote{Typeset names in
34: %10~pt Times roman, uppercase. Use the footnote to indicate the
35: %present or permanent address of the author.}
36: 
37: 
38: \address{Institute of System Engineering, Dalian University of
39: Technology, 2 Ling Gong Rd.,\\ Dalian 116024, P R China\\
40: %\footnote{State completely without abbreviations, the affiliation
41: %and mailing address, including country. Typeset in 8~pt
42: %Times italic.}\\
43: $^\dag$liujg004@tom.com}
44: 
45: 
46: \maketitle
47: 
48: \begin{history}
49: \received{(8 February 2005)} \revised{(31 Aug 2005)}
50: \end{history}
51: 
52: \begin{abstract}
53: It has been found that the networks with scale-free distribution are
54: very resilient to random failures. The purpose of this work is to
55: determine the network design guideline which maximize the network
56: robustness to random failures with the average number of links per
57: node of the network is constant. The optimal value of the
58: distribution exponent and the minimum connectivity to different
59: network size are given in this paper. Finally, the optimization
60: strategy how to improve the evolving network robustness is given.
61: \end{abstract}
62: 
63: \keywords{Scale-free network; optimal programme; power-law degree
64: distribution; random failures.}
65: 
66: \section{Introduction}
67: 
68: Recently, much attention has been focused on the topic of
69: scale-free networks which characterize many social, information,
70: technological and biological systems.\cite{1,2,3,4} The
71: qualitative properties of many interesting real-world examples,
72: such as the internet network, the power grid network and the
73: protein interaction network, are as following:
74: \begin{description}
75: \vspace{4pt} \item[{(1)}] the degree distribution has power-law
76: tail; \vspace{4pt}\item[{(2)}] local clustering of edges: graph is
77: not locally tree-like; \vspace{4pt}\item[{(3)}] small average
78: distance.
79: \end{description}
80: The networks can be visualized by nodes representing individuals,
81: organizations, computers and by links between them representing
82: their interactions. For the purpose of analyzing topology, we
83: ignore the variation in the type of links. Robustness of the
84: network topology comes from the presence of alternate paths, which
85: ensures the communication remains possible in spite of the damages
86: to the network.
87: 
88: Designers of the networks must assume that networks have random
89: failures or might be attacked, and some of these attacks can
90: result in damage. The robust networks will continue functioning in
91: spite of such damages. Although many literatures have discussed
92: what the optimal network topology would be, many real-world
93: networks present the power-law degree distribution.
94: %With the real network
95: %size become larger and the average connectivity change, what
96: %should we do to improve the network robustness to random failures.
97: 
98: When the scale-free networks are subjected to random breakdowns,
99: with a fraction $p$ of the nodes and their connections are removed
100: randomly, the network's integrity might be compromised: when the
101: exponent of the power-law degree distribution $\gamma>3$, there
102: exists a critical threshold $p_c$, such that for $p>p_c$, the
103: network would break into smaller and disconnected parts, but the
104: networks with $\gamma<3$ are more resilient to random breakdowns.
105: Cohen {\textit {et. al}}\cite{10} presented a criterion to calculate
106: the percolation critical threshold to random failures to scale-free
107: networks. If we attack the scale-free networks intentionally: the
108: removal of sites is not random, but rather sites with the highest
109: connectivity are targeted first. The numerical simulations suggest
110: that scale-free networks are highly sensitive to this kind of
111: attack.\cite{9} Cohen {\textit {et. al}}\cite{11} studied the exact
112: value of the critical fraction needed for disruption. Thus
113: scale-free networks are highly robust against random failures of
114: nodes and hypersensitive to intentional attacks against the system's
115: largest nodes. So a randomly chosen node has low degree with high
116: probability, but removal of a highly connected node might produce
117: large effect to the network. This situation is often compared to
118: that of the classical random graph presented by Erd$\check{o}$s and
119: R\'{e}nyi.\cite{5,6} Such graphs have a Poisson degree distribution.
120: This makes the random graphs less robust to random failures than
121: comparable networks with power-law degree distribution, but much
122: more robust against attacks on hubs.
123: 
124: %An interesting equation is
125: 
126: 
127: 
128: In this paper, we specifically focus on the robustness of the
129: network topology to random failures. We use the percolation theory
130: and the optimization method to investigate the guideline which can
131: maximize the robustness of the scale-free networks to random
132: failures of nodes with the constrained condition that the average
133: connectivity of per node in the network is constant. The percolation
134: theory provides the measures of distribution which are possible ways
135: for measuring robustness. We examine the relationship between the
136: average connectivity per node and the network robustness to random
137: failures. Then, we investigate the trend of the network robustness
138: to random failures with the network size $N$. The work may provide
139: the theoretical evidence that if the minimal connectivity and the
140: exponent of the power-law degree distribution take in more optimal
141: way, the robustness of the scale-free networks can be optimized.
142: 
143: If we construct and maintain a network with a given number of
144: nodes as being proportional to the average number of links
145: $\langle k \rangle$ per node in the network, our goal then becomes
146: how to maximize the robustness of a network with $N$ nodes to
147: random failures with the constraint that the number of links
148: remains constant but the nodes are connected in a different and
149: more optimal way.
150: 
151: 
152: \section{Optimal Strategy for Random Failures}
153: 
154: 
155: 
156: Our goal is to maximize the threshold for random removal with the
157: condition that the average degree $\langle k \rangle$ per node is
158: constant. We construct the following model.
159: \begin{equation}\label{F2.7}
160:   \left\{\begin{array}{rlc}
161:     \max & p_c^{\rm rand}, \\[5pt]
162:     {\rm s.t.} & \langle k \rangle={\rm constant.}
163:   \end{array}
164:   \right.
165: \end{equation}
166: 
167: For any degree distribution $P(k)$, the threshold for random
168: removal of nodes is$^{9}$
169: 
170: \begin{equation}\label{F2.1}
171: p_c^{\rm rand}=1-\frac{1}{\kappa_0-1},
172: \end{equation}
173: where $\kappa_0\equiv \frac{\langle k^2\rangle}{\langle k
174: \rangle}$ is calculated from the original connectivity
175: distribution. A wide range of networks have the power-law degree
176: distribution:
177: \begin{equation}\label{F2.2}
178: P(k)=ck^{-\alpha}, \ \ k=m, m+1, \ldots, K,
179: \end{equation}
180: where $k=m$ is the minimal connectivity and $k=K$ is an effective
181: connectivity cutoff presented in finite networks. To the power-law
182: degree distribution, the average $\langle k \rangle$ can be given
183: with the usual continuous approximation, this yields
184: \begin{equation}\label{F2.3}
185: %\begin{array}{rcl}
186:  \langle k \rangle = \int_m^K{k\cdot ck^{-\alpha}}dk
187:      =  c\frac{[K^{(2-\alpha)}-m^{(2-\alpha)}]}{2-\alpha}.
188: %\end{array}
189: \end{equation}
190:  From (\ref{F2.2}), $\kappa_0$ can be calculated as
191: \begin{equation}\label{F2.3}
192:   \kappa_0  =  \langle k^2\rangle/\langle k \rangle
193:        =
194:       \frac{2-\alpha}{3-\alpha}\frac{[K^{(2-\alpha)}-m^{(2-\alpha)}]}{[K^{(3-\alpha)}-m^{(3-\alpha)}]}.
195: \end{equation}
196: In a finite network, the largest connectivity $K$ can be estimated
197: from\cite{10}
198: $$
199: \int_K^{\infty}P(k)dk=\frac{1}{N},
200: $$
201: where $N$ is the number of the network nodes. Then we have that
202: \begin{equation}\label{F2.6}
203: \big(\frac{K}{m}\big)^{\alpha-1}=N.
204: \end{equation}
205: To the power-law degree distribution $P(k)$, we have
206: $$
207: \begin{array}{rcl}
208:     1 & = & \int_m^K ck^{-\alpha}dk \\[8pt]
209:       & = & c[K^{(1-\alpha)}-m^{(1-\alpha)}]/(1-\alpha), \\[8pt]
210: \end{array}
211: $$
212: this yields
213: $$
214:   c=\frac{\alpha-1}{m^{(1-\alpha)}-K^{(1-\alpha)}}=\frac{m^{\alpha}(\alpha-1)}{m[1-(\frac{K}{m})^{(1-\alpha)}]}.
215: $$
216: In the real world, there always exists the relation $K\gg m$, so
217: we have
218: \begin{equation}\label{F2.4}
219: c\approx \frac{m^{\alpha}(\alpha-1)}{m}.
220: \end{equation}
221: Combining (\ref{F2.3}) and (\ref{F2.4}), we have that
222: \begin{equation}\label{F2.5}
223: \langle k \rangle\approx
224: \frac{(\alpha-1)}{(2-\alpha)}m\big[\big(\frac{K}{m}\big)^{(2-\alpha)}-1\big].
225: \end{equation}
226: From (\ref{F2.5}), we have the following numerical results.
227: %\begin{table}[p]
228: %\tbl{When $N=10^3$, the relationship of the parameters $\alpha$
229: %and $m$.} {\begin{tabular}{cccccccc@{}} \toprule
230: %   $\langle k\rangle$       & $a(m=1)$  & $a(m=2)$    & $a(m=3)$   &  $a(m=4)$     &  $a(m=5)$ & $a(m=6)$\\
231: %\colrule
232: %     3 & 2.403     &  3.938       &   --       &  --        &  --       &  --     \\ %\hline
233: %     4 & 2.219     &  2.931       & 4.9317     &  --        &  --       &  --     \\ %\hline
234: %     5 & 2.225     &  2.584       & 3.438      &  5.92      &  --       &  --     \\ %\hline
235: %     6 & 2.068     &  2.4037      & 2.93       &  3.938     & 6.905     & --      \\ %\hline
236: %     7 & 2.0427    &  2.2913      & 2.672      &  3.27      & 3.27      & 7.88    \\ \botrule
237: %\end{tabular} }
238: %\end{table}
239: 
240: \vspace{0.3cm}
241:  It can been seen from table 1 that the
242:  distribution exponent $\alpha$ increases when the minimal connectivity $m$ increases.
243: 
244: Combining (\ref{F2.6}) and (\ref{F2.5}), we have that
245: \begin{equation}\label{F2.15}
246: \langle k \rangle=
247: \frac{(\alpha-1)}{(\alpha-2)}m\frac{1-N^{-\frac{\alpha-2}{\alpha-1}}}{1-N^{-1}}.
248: \end{equation}
249: \begin{table}[pt]
250: \tbl{When $N=10^6$, the relationship between $m$ and $p_c^{\rm
251: rand}$.} {\begin{tabular}{cccccccc@{}} \toprule
252:    $\langle k\rangle$       & $a(m=1)$  & $a(m=2)$    & $a(m=3)$   &  $a(m=4)$     &  $a(m=5)$ & $a(m=6)$\\
253: \colrule
254:      3 & 2.492    &  4.000     & --       &  --      &  --    & --    \\
255:      4 & 2.318    &  2.998     & 5.000    &  --      &  --    & --    \\
256:      5 & 2.225    &  2.662     & 3.498    &  6.000   &  --    & --    \\
257:      6 & 2.172    &  2.491     & 2.998    &  4.000   &  7.00  & --    \\
258:      7 & 2.126    &  2.388     & 2.747    &  3.333   &  4.50  & 8.00  \\ \botrule
259: \end{tabular} }
260: \end{table}
261: 
262: From table 1, we can get the following relationship:
263: \begin{description}
264: \item [(1)] When the average connectivity $\langle k \rangle$ per
265: node is constant, the exponent $\alpha$ increases when the minimum
266: connectivity $m$ increases;
267: 
268: \item[(2)] To the minimum connectivity $m=1$, the exponent
269: $\alpha$ decreases when the average connectivity $\langle k
270: \rangle$ of the network increases.
271: \end{description}
272: 
273: Using the results obtained above we construct the following model.
274: \begin{equation}\label{F2.8}
275:   \left\{\begin{array}{rl}
276:     \max & \{1-\frac{1}{\kappa_0-1}\}
277:     \\[10pt]
278:      s.t.
279:     &\frac{(\alpha-1)}{(\alpha-2)}m[1-N^{-\frac{\alpha-2}{\alpha-1}}]=\langle k \rangle\\[10pt]
280:     & m\in Z^{+},
281:   \end{array}
282:   \right.
283: \end{equation}
284: where
285: $\kappa_0=\frac{2-\alpha}{3-\alpha}\frac{[K^{(2-\alpha)}-m^{(2-\alpha)}]}{[K^{(3-\alpha)}-m^{(3-\alpha)}]}$.
286: The numerical results suggest that whether the network size $N$ is
287: very large or not, $p_c$ reaches its maximum value when $m=1$. The
288: numerical results are presented in table 2.
289: 
290: %\setlength{\unitlength}{1mm}
291: %\begin{center}
292: %\begin{picture}(50,30)
293: %\put(0,30){\special{bmp:maxpc.bmp x=4cm y=3cm}}
294: %\end{picture}
295: %\end{center}
296: 
297: \begin{figure}[th]
298: \centerline{\psfig{file=maxpc.eps,width=10cm}} \vspace*{8pt}
299: \caption{The critical percolation thresholds for random failures
300: to different $\langle k \rangle$.}
301: \end{figure}
302: 
303: \begin{table}[pt]
304: \tbl{When $N=10^6$, the relationship between $m$ and $p_c^{\rm
305: rand}$.} {\begin{tabular}{cccccccc@{}} \toprule
306:               & $\langle k \rangle$=3   &  $\langle k \rangle$=4     & $\langle k \rangle$=5    &  $\langle k \rangle$=6   &  $\langle k \rangle$=7 \\
307: \colrule
308:      $N=10^3$ & 0.9766    &  0.9872      & 0.9906     &  0.9917    &  0.9922  \\
309:      $N=10^4$ & 0.9715    &  0.9847      & 0.9888     &  0.9906    &  0.9916  \\
310:      $N=10^6$ & 0.9683    &  0.9824      & 0.9869     &  0.9889    &  0.9903  \\ \botrule
311: \end{tabular} }
312: \end{table}
313: 
314: From table 2, we can get the following three conclusions:
315: \begin{description}
316: \item[{(1)}] If the average connectivity $\langle k \rangle$ per
317: node and the exponent $\alpha$ of the power-law degree
318: distribution is constant, the robustness of the scale-free
319: networks will decrease when the network size becomes larger.
320: 
321: 
322: \item[{(2)}] If the network size $N$ is constant, the robustness
323: of the scale-free networks increase when the average connectivity
324: $\langle k \rangle$ becomes larger.
325: 
326: \item[{(3)}]  To the random failures, we have to take several
327: times cost to increase the robustness of the scale-free networks
328: about 1\%.
329: \end{description}
330: 
331: \section{Discussion and Summary}
332: 
333:  It is well known that the networks with power-law degree distribution
334:  are resilient to random failures. But this conclusion don't answer the
335: following three questions: (i) To a constant average connectivity
336: $\langle k \rangle$, how to determine the distribution exponent of
337: the scale-free networks so that the networks are more robust to the
338: random failures. (ii) To different network size, how many edges we
339: need to add to the network to satisfy the robustness level to random
340: failures.  (iii) To an exist network with power-law degree
341: distribution, what we need to do to improve the network robustness.
342: In this paper, we use the percolation theory and the mathematic
343: programme method to optimize the robustness of the scale-free
344: networks for random failures and give the numerical results.
345: Finally, we give the relationship between the threshold $p_c$ and
346: the network size, the degree distribution exponent $\alpha$ and the
347: average connectivity $\langle k \rangle$ per node.
348: 
349: From Fig. 1, we can get the conclusion that if the scale-free
350: networks size become large, the network robustness to random
351: failures would become weak. To the internet and other growing
352: scale-free networks the designers must add more links to the network
353: to improve the average connectivity per node to random failures.
354: 
355: Subjects for further study include (i) an analysis of the
356: robustness of the scale-free networks to the intentional attack to
357: the highest connectivity nodes. (ii) the optimization of complex
358: network under both random failures and intentional attack. (iii)
359: the topology structure to improving the robustness of existing
360: scale-free networks.
361: 
362: 
363: 
364: 
365: 
366: 
367: 
368: \section*{Acknowledgments}
369: The authors are grateful to Dr. Qiang Guo for her valuable
370: comments and suggestions, which have led to a better presentation
371: of this paper. This research was supported by Chinese Natural
372: Science Foundation Grant Nos. 70431001 and 70271046.
373: 
374: %\appendix
375: 
376: 
377: \begin{thebibliography}{0}
378: \bibitem{1} R. Albert and A.-L. Barab\'{a}si, {\it Rev. Mod. Phys.} {\bf
379: 74}, 47 (2002).
380: 
381: \bibitem{2} M. E. J. Newmann, {\it SIAM Rev.} {\bf 45}, 167 (2003).
382: 
383: \bibitem{3} J. F. F. Mendes, S. N. Dorogovtsev and A. F. Ioffe, {\it
384: Evolution of Networks: From Biological Nets to the Internet and
385: the WWW} (Oxford University Press, Oxford, 2003).
386: 
387: \bibitem{4} R. Pastor-Satorras and A. Vespignani, {\it Evolution and
388: Structure of the Internet: A Statistical Physics Approach}
389: (Cambridge University Press, Cambridge, 2004).
390: 
391: \bibitem{10} R. Cohen, K, Erez, D. ben-Avraham, S. Havlin, {\it Phys.
392: Rev. Lett.} {\bf 85}, 4626 (2000).
393: 
394: \bibitem{9} G. Paul, T. Tanizawa, S. Havlin and H. E. Stanley, {\it Eur.
395: Phys. J. B} {\bf 38}, 187 (2004).
396: 
397: \bibitem{11} R. Cohen, K, Erez, D. ben-Avraham, S. Havlin, {\it Phys.
398: Rev. Lett.} {\bf 86}, 3682 (2001).
399: 
400: 
401: 
402: \bibitem{5} P. Erd\u{o}s and  A. R\'{e}nyi, {\it Publicationes
403: Mathematicae} {\bf 6}, 290 (1959).
404: 
405: 
406: \bibitem{6} P. Erd\u{o}s and  A. R\'{e}nyi, {\it Publicationes
407: Mathematical Inst. of the Hungarian Acad. of Sciences} {\bf 5}, 17
408: (1960).
409: 
410: \bibitem{7} D. J. Watts and S. H. Strogatz, {\it Nature} {\bf 393}, 440
411: (1998).
412: 
413: \bibitem{8} B. Bollobas, {\it Random Graphs} (Academic, London, 1985).
414: 
415: 
416: \bibitem{12} R. Albert, H. Jeong, and A. L. Barab\'{a}si, {\it Nature},
417: {\bf 406}, 6794 (2000).
418: 
419: 
420: \end{thebibliography}
421: 
422: \end{document}
423: