1: \documentclass{epl}
2: \usepackage{amssymb}
3:
4: \title{Phase transition in the modified fiber bundle model}
5: \shorttitle{Phase transitions in the modified fiber bundle model}
6:
7: \author{Beom Jun Kim\thanks{E-mail: \email{beomjun@ajou.ac.kr}}}
8: \shortauthor{B.J. Kim}
9: \institute { Department of Molecular Science
10: and Technology, Ajou University, Suwon 442-749, Korea}
11:
12: \pacs{62.20.Mk}{Fatigue, brittleness, fracture, and cracks}
13: \pacs{89.75.Hc}{Networks and genealogical trees}
14: \pacs{05.70.Jk}{Critical point phenomena}
15: \pacs{64.60.Fr}{Equilibrium properties near critical points, critical exponents}
16:
17: \begin{document}
18:
19: \maketitle
20:
21: \begin{abstract}
22:
23: We extend the standard fiber bundle model (FBM) with the local
24: load sharing in such a way that the conservation of the
25: total load is relaxed when an isolated fiber is broken.
26: In this modified FBM in one dimension (1D), it is
27: revealed that the model exhibits a well-defined phase
28: transition at a finite nonzero value of the load, which
29: is in contrast to the standard 1D FBM.
30: The modified FBM defined in the Watts-Strogatz network is also
31: investigated, and found is the
32: existences of two distinct transitions: one discontinuous
33: and the other continuous. The effects of the long-range
34: shortcuts are also discussed.
35:
36:
37: \end{abstract}
38:
39:
40: The fiber bundle model (FBM) was introduced to explain a variety of phenomena
41: related with the problems of damage spread~\cite{classic}. More
42: specifically, consider a system composed of a bundle of mingled fibers exposed
43: to a common stress along the direction of fibers. It is natural to assume that
44: each fiber has a certain value of the capacity, or the threshold value of the
45: load, and if the load on the fiber exceeds the capacity allowed to it, the
46: fiber is broken. Since the broken fiber stops functioning, its neighbor fibers
47: should carry the load which was assigned to the broken one~\cite{lls}. The damage caused
48: by the breaking of a single fiber may propagate over its local neighbors and
49: then stop spreading. Another interesting possibility is the existence of an
50: avalanche or a cascading failure, i.e., the original local damage spreads over
51: the whole system, causing the complete breakdown of the total fibers. Recent
52: blackout in the United States can also be understood as the cascading failures
53: of power cables and power stations caused by the spread of the overload
54: breakdowns of elements in power grid.
55:
56: Recently, the subject of complex networks is the one of the most
57: studied research area, not only
58: for its intrinsic interest but also for its promising
59: broad future applicability~\cite{ref:network}. When a dynamic system
60: is defined on a complex network, the emergent behavior may
61: totally differ from the one on the local regular array which has
62: been the standard arena of statistical physics. For example,
63: the phase transitions of the Ising and the $XY$ models on complex
64: networks have been shown to belong to the mean-field universality
65: class~\cite{isingXY}.
66: The spread of damage on networks, which is not described by
67: the equilibrium Hamiltonian, also draws much interest
68: in relation to the epidemic spread~\cite{epidemic} as well as
69: the overload breakdowns of networks subject to local~\cite{morenoepl}
70: or nonlocal~\cite{holme} loads.
71:
72: The fiber bundle model in this work on a general network structure is defined
73: as follows:
74: \begin{enumerate}
75: \item To each vertex in the network of the size $N$ the capacity
76: $c_i$ for the $i$th vertex is assigned at random following a given distribution function (in most cases we use
77: the uniform distribution with $c_i \in [0,1]$).
78: \item The system is then exposed to a total load $N\sigma$, and thus each
79: vertex should carry the same load $\sigma_i = \sigma$
80: (we in this work $\sigma = 0$ is used as an initial condition).
81: \item Examine the breaking condition $\sigma_i > c_i$ for each $i$
82: and if the inequality is satisfied the vertex $i$ is broken and its load $\sigma_i$ is equally
83: transfered to its surviving neighbor vertices. For example, if the
84: vertex $i$ has $k_i$ surviving neighbors the vertex $j$, which is
85: a neighbor of $i$, has now new load:
86: $\sigma_j \rightarrow \sigma_j + \sigma_i/k_i$. Repeat this step until
87: there is no more breaking.
88: \item
89: Increase the total load $N\sigma \rightarrow N(\sigma + d\sigma)$.
90: The total load should be carried by
91: unbroken surviving vertices, the number of which is $N_s$, and accordingly
92: each vertex is assigned the new load $\sigma_i \rightarrow \sigma_i +
93: (N/N_s)d\sigma$. Note that we denote the local load on the vertex $i$
94: as $\sigma_i$ while $\sigma$ is the total load divided by $N$.
95: \end{enumerate}
96: The above procedures are identical to the standard FBM with the local
97: load sharing (LLS) in the literature~\cite{lls}. However,
98: our modified FBM in the present work differs as follows:
99: \begin{itemize}
100: \item When an isolated vertex, which has no surviving neighbor vertices,
101: is overload broken, we assume that
102: the load which has been carried by this vertex is not transferred
103: to other vertices. On the other hand, in the standard FBM with
104: LLS the load transfers via edges, not via chain of surviving vertices,
105: and consequently the load of the broken isolated fiber can still
106: be transferred to other (suitably defined) closely located vertices.
107:
108: \end{itemize}
109: Our modified FBM appears to be reasonable
110: in itself since the standard FBM assumes that the load can be
111: transferred to vertices which are not connected to the broken vertex
112: through a chain of {\it surviving} vertices.
113: In contrast, our model allows only the load transfer across
114: the actually existing surviving vertices.
115: One disadvantage of our modified FBM
116: is that the total load applied to the system
117: is not constant in time. It might be difficult to find
118: the actual physical realization of our model in the original context of the
119: FBM. However, in abstract networks such as Internet
120: one may not rule out the possibility that when an isolated
121: vertex is broken its load does not spread to other vertices but just disappears.
122: It should be noted that even in our modified FBM, the total load is resumed
123: when each vertex is assigned the new load as
124: $\sigma_i \rightarrow \sigma_i + (N/N_s)d\sigma$ (see step 4) since the
125: breaking of an isolated vertex also decreases $N_s$. Accordingly, the
126: total load conservation condition is relaxed only during the spread of
127: damage in step 3.
128:
129: We repeat the above described procedure for our modified version of
130: FBM and measure
131: the ratio $\rho \equiv N_s/N$ as a function of the
132: load $\sigma$ averaged over $10^4 \sim 10^5$ different capacity
133: distributions and different network structures.
134: In most numerical simulations, we use $d\sigma = 0.001$, which
135: has been numerically confirmed to be small enough, and
136: network sizes are $N=128, 256, \cdots, 524 \; 288$ (or from $N=2^7$ to
137: $2^{19}$).
138:
139:
140: We first consider the local regular one-dimensional (1D) network with
141: the connection range $r$.
142: For various values of $r$, including $r=1$
143: which corresponds to the usual 1D lattice, we observe the
144: qualitatively the same behavior.
145: In Fig.~\ref{fig:p0}(a) for $r=2$, it is shown that
146: at a certain value of the load $\sigma$ the ratio $\rho$
147: between the number of working fibers and the total number of
148: fibers vanishes beyond some critical value $\sigma_c$. As the
149: system size $N$ becomes larger
150: [$N=2^7, 2^8, \cdots, 2^{19} (=524 \; 288)$ in Fig.~\ref{fig:p0}(a)], the phase transition becomes clearer.
151: It is interesting to note that the existence of the phase transition
152: at nonzero $\sigma_c$ is completely different from the
153: well-established result of $\sigma_c = 0$ for the standard FBM
154: in the thermodynamic limit~\cite{lls}. Simple change of the
155: conservation law of the total load yields the dramatic change
156: of the universality class.
157:
158: We in Fig.~\ref{fig:p0} (b) show $\rho$ as a function of $N$
159: in log scales at various values of
160: $\sigma = 0.2812, 0.2814, 0.2816, 0.2818$, and 0.2820 (from top to bottom).
161: For $\sigma > \sigma_c \approx 0.2817$, $\rho (N)$ shows a downward
162: curvature, implying $\rho \rightarrow 0$ in the thermodynamic limit, while
163: $\rho(N)$ bends upward for $\sigma < \sigma_c$, suggesting that $\rho(N)$
164: saturates to a finite value as $N$ becomes larger.
165: Separating these behaviors, there exists a well-defined $\sigma_c$ at which
166: $\rho$ as a function of $N$ decays algebraically, and the system does
167: not possess any length scale except $N$ at $\sigma = \sigma_c$. This
168: is commonly found behavior in the system with a continuous phase transition.
169:
170: \begin{figure}
171: \twofigures[height=4.8cm]{p0.eps}{p0gauss.eps}
172: \caption{The modified fiber bundle model on the 1D regular lattice
173: with couplings up to the next nearest neighbors with the
174: uniform distribution of the capacity $c_i \in [0,1]$.
175: (a) Density $\rho$ of unbroken fibers versus the load $\sigma$
176: for $N=2^7, 2^8, \cdots, 2^{19}$ (from top to bottom).
177: As $\sigma$ is increased, the network begins to break up, resulting in
178: $\rho = 0$ at sufficiently large values of $\sigma$.
179: (b) $\rho$ versus $N$ for $\sigma = 0.2812, 0.2814, 0.2816, 0.2818$,
180: and 0.2820 (from top to bottom). Around $\sigma \approx 0.2817$, $\rho$
181: decays algebraically with $N$, manifesting the existence of the
182: continuous phase transition. (c) Precise determination of $\sigma_c$
183: from the finite-size scaling form~(\ref{eq:scaling}). Determined are
184: $a=0.50(1)$ and $\sigma_c = 0.28165(2)$.
185: (Curves correspond to $N = 2^{12}, 2^{13}, \cdots, 2^{19}$, respectively,
186: from bottom to top on the left-hand side of the crossing point).
187: (d) Scaling collapse of all data points in (c) through the use
188: of the finite-size scaling~(\ref{eq:scaling}); $\nu = 2.0(1)$ is obtained.
189: }
190: \label{fig:p0}
191: \caption{The FBM on the 1D regular lattice with connections
192: up to the next nearest neighbors with the Gaussian distribution
193: of the capacity (the mean and the variance are fixed the same as in Fig.~\ref{fig:p0}).
194: Finite-size scaling (compare with Fig.~\ref{fig:p0} for the
195: uniform distribution) yields $\sigma_c = 0.27363(2)$ and critical
196: exponents $a=\beta/\nu = 0.50(1)$ and $\nu = 2.0(1)$, identical
197: to the uniform distribution.
198: }
199: \label{fig:p0gauss}
200: \end{figure}
201:
202: For a more precise determination of $\sigma_c$, we use the standard
203: finite-size scaling method with the scaling assumption:
204: \begin{equation} \label{eq:scaling}
205: \rho (\sigma, N) = N^{-a} f( \bigl (\sigma - \sigma_c) N^{1/\nu} \bigr ) ,
206: \end{equation}
207: where $f(x)$ is the scaling function and the exponent $\nu$
208: describes the divergence of the correlation length $\xi$ [or more generally,
209: the correlation volume for the case when the length is ill-defined (see
210: discussions in, e.g., Ref.~\cite{isingXY})]
211: near the critical point, i.e., $\xi \sim |\sigma - \sigma_c|^{-\nu}$.
212: From the definition of the critical exponent $\beta$,
213: $\rho \sim (\sigma_c - \sigma)^\beta$ for $\sigma < \sigma_c$
214: in the thermodynamic limit,
215: we obtain the relation $a = \beta/\nu$.
216: Figure~\ref{fig:p0}(c) is for a precise determination of $\sigma_c$
217: from the finite-size scaling~(\ref{eq:scaling}): $\rho N^a$ is shown
218: to have a unique crossing point at $\sigma = \sigma_c = 0.28165(2)$
219: with the exponent $a=\beta/\nu = 0.50(1)$.
220: Once $\sigma_c$ and the exponent $a$ are determined, one can again
221: use the finite-size scaling~(\ref{eq:scaling}) to make the all data
222: points in Fig.~\ref{fig:p0}(c) collapse to a single smooth curve as displayed
223: in Fig.~\ref{fig:p0}(d) with the determination of $\nu = 2.0(1)$.
224:
225:
226: The above results of the existence of the continuous phase transition
227: with critical exponents $\nu \approx 2$ and $\beta \approx 1$ is
228: expected to be universal regardless of the detailed form of the
229: distribution function of the capacity. We in Fig.~\ref{fig:p0gauss} show
230: the finite-size scaling plot for the same FBM but with
231: the Gaussian distribution of the capacity instead. For comparisons,
232: the same values of the mean and the variance as ones with
233: the uniform distribution in Fig.~\ref{fig:p0} are used.
234: Confirmed is the existence of the continuous phase transition
235: with the identical critical exponents.
236: Although we have
237: not used the commonly used Weibull distribution, we believe
238: that the universality class should be identical for the
239: various capacity distribution functions.
240:
241: \begin{figure}
242: \onefigure{pall.eps}
243: \caption{The FBM on the Watts-Strogatz small-world networks
244: with the connection range $r=2$ and at various rewiring probabilities
245: $P=$ (a) 0.04, (b) 0.1, (c) 0.2, and (d) 0.5. Except (a), all
246: show interesting transition behaviors: the density of working
247: fibers $\rho$ drops down abruptly at a certain value of the
248: load $\sigma_1$ and then approaches $\rho = 0$ continuously
249: at $\sigma_2$. Networks with the sizes $N = 2^7, 2^8, \cdots 2^{19}$
250: (from top to bottom in each plot)
251: and the uniform capacity distribution are used.
252: }
253: \label{fig:pall}
254: \end{figure}
255:
256: \begin{figure}
257: \twofigures[height=4.8cm]{p0.5disc.eps}{p0.5cont.eps}
258: \caption{The FBM on the WS network with the rewiring probability $P = 0.5$. (a)
259: The maximum of the derivative of $\rho(\sigma)$ with respect to $\sigma$ versus
260: the system size $N$. The divergence of the derivative implies the discontinuous
261: transition nature in the thermodynamic limit. $\sigma_1$ is defined as the
262: value of $\sigma$ when this maximum occurs. (b) The location $\sigma_1$ of the
263: derivative maximum versus $1/N$. In the thermodynamic limit $\sigma_1 \approx
264: 0.207$ is estimated.
265: }
266: \label{fig:p0.5disc}
267: \caption{
268: The continuous phase transition in the
269: FBM on the WS network with $P=0.5$. (a) $\rho N^a$ versus $\sigma$
270: at various sizes $N=2^{13}, \cdots 2^{19}$ with $a=0.5$ crosses
271: at a unique crossing point, resulting in the estimation $\sigma_2
272: = 0.22376(2)$ as the critical point of the second phase transition
273: of the continuous nature. (b) Finite-size scaling collapse~(\ref{eq:scaling})
274: of data points in (a). The critical exponents $a = \beta/\nu = 0.5,
275: \nu = 2.0$ (and thus $\beta = 1.0$) result in the scaling
276: collapse.
277: }
278: \label{fig:p0.5cont}
279: \end{figure}
280:
281:
282: We next study the FBM on the Watts-Strogatz (WS) networks, which is generated
283: following the standard procedure in Ref.~\cite{WS}. The two important
284: parameters in the WS network are the connection range $r$, ($r=2$ is used here)
285: and the rewiring probability $P$ controlling the number of long-range
286: shortcuts. As soon as $P$ has a nonzero value, it is well-known that the
287: network undergoes so-called small-world transition that the network diameter
288: increases logarithmically with the network size $N$. In Fig.~\ref{fig:pall}
289: $\rho$ versus $\sigma$ is shown at $P=$ (a) 0.04, (b) 0.1, (c) 0.2, and (d)
290: 0.5. At sufficiently large values of $P$, it is clearly demonstrated that the
291: system has two phase transitions at $\sigma_1$ and $\sigma_2$ ($\sigma_1 <
292: \sigma_2$): At $\sigma_1$, $\rho$ appears to drop abruptly to a nonzero finite
293: value and as $\sigma$ is further increased, $\rho$ eventually vanishes beyond
294: $\sigma_2$.
295:
296:
297: To investigate the nature of the first transition at $\sigma_1$,
298: we compute $d\rho / d\sigma$ as a function of $\sigma$ and then measure
299: at which value of $\sigma$ the derivative takes the maximum value.
300: In Fig.~\ref{fig:p0.5disc}(a), the maximum value of the derivative
301: $(d\rho / d\sigma)_{\rm max}$ is plotted as a function of the system
302: size $N$: The power-law divergence
303: $(d\rho / d\sigma)_{\rm max} \sim N^{0.4}$ clearly suggests that the
304: transition nature is indeed discontinuous in the thermodynamic limit
305: of $N \rightarrow \infty$. The location of the
306: derivative maximum, $\sigma_1$, is displayed as a function of $1/N$
307: in Fig.~\ref{fig:p0.5disc}(b); $\sigma_1 \approx 0.207$ is concluded
308: in the thermodynamic limit.
309: We then apply the finite-size scaling analysis to the regime
310: where $\rho$ approaches zero continuously (see Fig.~\ref{fig:p0.5cont}).
311: The unique crossing point in the $\rho N^a$ versus $\sigma$ at
312: various $N$ in Fig.~\ref{fig:p0.5cont}(a) as well as the smooth
313: scaling collapse in Fig.~\ref{fig:p0.5cont}(b) is again observed
314: with the identical values of critical exponents, i.e, $a \approx 0.5$,
315: $\nu \approx 2.0$, and $\beta \approx 1.0$.
316: It is noteworthy that the FBM on the scale-free network of Barab{\'a}si and
317: Albert (BA)~\cite{BA} has shown to have a different transition
318: nature~\cite{morenoepl}, i.e., a discontinuous phase transition
319: without the subsequent continuous transition
320: as in the WS network considered here. Consequently,
321: as soon as the network undergoes the global spread of the damage, no
322: fiber survives the disaster in the BA network, while in the WS network
323: fibers which survived the discontinuous transition are subject to the second
324: continuous transition at a higher load.
325:
326: In the context of complex
327: networks, the size of giant component has been more frequently measured
328: since it can detect the global breakdown of the information flow between
329: two arbitrarily chosen vertices. In the viewpoint of the universality,
330: the use of the different order parameters is expected to
331: change neither the nature of the phase transition nor the values of critical
332: exponents, which are the main interest in this work. The density of
333: surviving fibers in the WS network can also have a practical meaning
334: if one considers the real fiber bundles in which most inter-fiber
335: couplings are local while some of them are long-ranged.
336:
337: Figure~\ref{fig:phd} summarizes the phase diagram obtained in the
338: present work. There exist in general two phase transition
339: lines separating three different phases (two unbroken phases UI and
340: UII with $\rho \neq 0$ and the broken phase B with $\rho = 0$):
341: $\sigma_1$ is estimated from the maximum of $d\rho/d\sigma$
342: while $\sigma_2$ is obtained from the standard finite-size
343: scaling analysis applied for the continuous phase transition.
344: As the load $\sigma$ is increased from zero, the system only suffers from
345: locally isolated damage spreading (the phase UI in Fig.~\ref{fig:phd})
346: until the first discontinuous transition
347: point $\sigma_1$ is reached.
348: As $\sigma$ crosses $\sigma_1$ from below, the system
349: undergoes abrupt change, probably in the form of the cascading failures
350: or an avalanche, and the number of fibers still alive decays discontinuously.
351: Interestingly,
352: the existence of the long-range shortcuts not only makes the discontinuous
353: transition happen, but also help the system to avoid
354: complete disaster, i.e, the global spread of damage does not destroy
355: all fibers in the system but there still remain a finite fraction of fibers which
356: survived the first transition at $\sigma_1$ (the phase UII in Fig.~\ref{fig:phd}).
357: Further increase of the load then eventually destroys
358: all fibers resulting in the $\rho = 0$ state (phase B in Fig.~\ref{fig:phd}).
359: At sufficiently small values of $P$, it appears that
360: the system has only one phase transition which is continuous in
361: nature: At $P \lesssim 0.04$, it is found that the maximum
362: of $d\rho/d\sigma$ and the continuous phase transition occur simultaneously.
363: At $P \gtrsim 0.06$, the system starts to have well-separated two
364: phase transitions, and the increase of $P$ appears to make the system more
365: strong (both $\sigma_1$ and $\sigma_2$ increase with $P$ in the
366: intermediate region).
367:
368: \begin{figure}
369: \onefigure{phd.eps}
370: \caption{Phase diagram of the FBM on the WS network with the
371: connection range $r=2$. The points (filled squares) on the
372: full line are obtained
373: from the finite-size scaling analysis for the continuous phase
374: transition (see Fig.~\ref{fig:p0.5cont}) and the points (empty squares)
375: on the dotted line are from the maximum
376: position of $d\rho/d\sigma$ extrapolated to $N \rightarrow \infty$
377: (see Fig.~\ref{fig:p0.5disc}). There exist three phases: two unbroken
378: phases with $\rho \neq 0$ (UI and UII), and completely broken phase
379: with $\rho =0$ (B). In UI, most fibers are still working with higher
380: value of $\rho \gtrsim 0.7$, while in UII only small fraction (but
381: still nonzero) of fibers hold together to endure the external load.
382: The transition between UI and UII is discontinuous
383: while between UII and B the nature of the transition is continuous.
384: For sufficiently small values of $P (\lesssim 0.04)$, it appears that there exists
385: only one transition from UI to B.
386: }
387: \label{fig:phd}
388: \end{figure}
389: In summary, we in this work have extended the existing model of fiber bundles
390: in such a way that the total load conservation condition is relaxed. For a pure
391: 1D regular network, our modified FBM has been found to show well-defined
392: continuous phase transition at a finite nonzero value of the load $\sigma_c$
393: with the critical exponents $\beta \approx 1.0$ and $\nu \approx 2.0$. This is
394: in a sharp contrast to the standard FBM where $\sigma_c = 0$ has been
395: confirmed in the thermodynamic limit. It has also been verified that the
396: universality class of the phase transition does not change when different
397: distribution function for the capacity is tried. For the Watts-Strogatz
398: network, as the rewiring probability $P$ is increased, the system undergoes two
399: phase transitions at $\sigma_1$ and $\sigma_2$. At $\sigma_1$ the density
400: $\rho$ of working fibers decreases discontinuously to a nonzero finite value.
401: At $\sigma_2 (> \sigma_1)$, $\rho$ approaches continuously zero, with the phase
402: transition with the identical universality class ($\beta \approx 1.0$ and $\nu
403: \approx 2.0$). The roles played by the existence of long-range shortcuts have
404: been shown to be two-fold: On the one hand, long-range shortcuts make the fiber
405: breaking transition abrupt, helping the damage spread across the system more
406: easily. On the other hand, the long-range shortcuts prohibit the system from
407: being completely destroyed, and even after the global propagation of damage
408: (at $\sigma_1$), there still remain a finite fraction of fibers still working.
409:
410: The author thanks D.H. Kim for useful discussions.
411: This work has been supported by the Korea Science
412: and Engineering Foundation through Grant
413: No. R14-2002-062-01000-0 and Hwang-Pil-Sang research
414: fund in Ajou University.
415: Numerical works have been
416: performed on the computer cluster Iceberg at Ajou University.
417:
418:
419:
420: \begin{thebibliography}{99}
421: \bibitem{classic} Daniels H.E., Proc. R. Soc. London A, {\bf 183} (1945) 405;
422: Coleman B.D., J. Appl. Phys., {\bf 28} (1957) 1058;
423: Sen P.K., J. Appl. Proc., {\bf 10} (1973) 586.
424: Phoenix S.L. and Taylor H.M., Adv. Appl. Prob., {\bf 5} (1973) 200;
425: Hemmer P.C. and Hansen A., J. Appl. Mech., {\bf 59} (1992) 909; Sornette D.,
426: J. Phys. A, {\bf 22} (1989) L243; J. Phys. I (France), {\bf 2} (1992) 2089.
427:
428:
429: \bibitem{lls} Smith R.L., Proc. R. Soc. London A, {\bf 372} (1980) 539;
430: Harlow D.G. and Phoenix S.L., Int. J. Fracture, {\bf 17} (1981) 601;
431: G\'omez J.B., I{\~n}iquez D. and Pacheco A.F., Phys. Rev. Lett., {\bf 71} (1993)
432: 380; Duxbury P.M and Leath P.L., Phys. Rev. B, {\bf 49} (1994) 12676;
433: V\'azqeuz-Prada, G\'omez, Moreno Y. and Pacheco A.F., Phys. Rev. E, {\bf 60} (1999) 2581;
434: Hidalgo R.C., Moreno Y. Kun F. and Herrmann H.J., Phys. Rev. E, {\bf 65}
435: (2002) 046148;
436: Pradhan S. and Chakrabarti, Int. J. Mod. Phys. B, {\bf 17} (2003) 5565.
437:
438: \bibitem{ref:network}For general reviews, see, e.g.,
439: Watts D.J., {\em Small Worlds} (Princeton University Press, Princeton, 1999);
440: Albert R. and Barab{\'a}si A.-L., Rev. Mod. Phys., {\bf 74}, (2002) 47;
441: Dorogovtsev S.N. and Mendes J.F.F., {\it Evolution of Networks}
442: (Oxford University Press Inc., New York, 2003).
443:
444: \bibitem{isingXY} Gitterman M., J. Phys. A: Math. Gen., {\bf 33} (2000) 8373;
445: Barrat A. and Weigt M., Eur. Phys. J. B, {\bf 13} (2000) 547;
446: Hong H., Kim B.J. and M.Y. Choi, {\it ibid.}, {\bf 66} (2002) 018101;
447: Kim B.J., Hong H., Holme P., Jeon G.S., Minnhagen P. and Choi M.Y.,
448: Phys. Rev. E, {\bf 64} (2001) 056135.
449:
450:
451: \bibitem{epidemic}
452: Kuperman M. and Abramson G., Phys. Rev. Lett., {\bf 86} (2001) 2909; Moore C. and Newman M.E.J., Phys. Rev. E, {\bf 61} (2000) 5678 (2000); Pastor-Satorras R. and Vespignani A., Phys. Rev. Lett., {\bf 86} (2001) 3200;
453: Moreno Y. and V{\'a}zquez A., Eur. Phys. J. B, {\bf 31} (2003) 265.
454:
455: \bibitem{morenoepl}
456: Moreno Y., G{\'o}mez J.B. and Pacheco A.F., Europhys. Lett., {\bf 58} (2002) 630 (2002).
457:
458:
459: \bibitem{holme}
460: Holme P. and Kim B.J., Phys. Rev. E, {\bf 65} (2002) 066109.
461:
462:
463:
464: \bibitem{WS}
465: Watts D.J. and Strogatz S.H., Nature (London), {\bf 393} (1998) 440.
466:
467: \bibitem{BA}
468: Barab{\'a}si A.-L. and Albert R., Science, {\bf 286} (1999) 509;
469: Barab{\'a}si A.-L., Albert R. and Jeong H., Physica A, {\bf 272} (1999) 173.
470:
471:
472: \end{thebibliography}
473:
474:
475: \end{document}
476: