1: \documentclass[preprint,info]{epl}
2: %\documentclass[published]{epl}
3: \usepackage{epsf}
4:
5: \title{Dynamic scaling and universality in evolution\\
6: of fluctuating random networks}
7:
8: \author{Miroslav Kotrla\inst{1}
9: \thanks{E-mail: \email{kotrla@fzu.cz}}
10: \and Franti\v{s}ek Slanina\inst{1}
11: \and Jakub Steiner\inst{2}}
12: \shortauthor{Miroslav Kotrla \etal}
13: \shorttitle{Dynamic scaling of fluctuating networks}
14: \institute{
15: \inst{1}Institute of Physics, Academy of Sciences
16: of the Czech Republic, Na Slovance 2, \\
17: 182~21 Prague 8, Czech Republic \\
18: \inst{2} Charles University at Prague, Department
19: of Theoretical Physics,
20: V Hole\v{s}ovi\v{c}k\'ach 2,
21: 180~00 Prague 8, Czech Republic
22: }
23:
24: \pacs{05.40.-a}{Fluctuation phenomena, random
25: processes, noise, Brownian motion}
26: \pacs{89.75.Fb}{Structure and organization in complex systems}
27: \pacs{05.10.-a}{Computational methods in statistical physics
28: and nonlinear dynamics}
29:
30:
31: \begin{document}
32:
33: \maketitle
34:
35: \begin{abstract}
36: %VERSION OF JULY 3.
37: We found that models of evolving random networks
38: exhibit dynamic scaling similar to scaling of growing surfaces.
39: It is demonstrated by numerical simulations of two variants of
40: the model in which nodes are added as well as removed
41: {[}{\it Phys. Rev. Lett.} {\bf 83}, 5587 (1999){]}.
42: The averaged size and connectivity of the network
43: increase as power-laws in early times
44: but later saturate.
45: Saturated values and times of saturation change with paramaters
46: controlling the local evolution of the network topology.
47: Both saturated values and times of saturation
48: obey also power-law dependences on controlling parameters.
49: Scaling exponents are calculated and universal features are discussed.
50: \end{abstract}
51:
52: \section{Introduction}
53: In a couple of last years, there has been an increasing activity
54: in the study of
55: the structure and the time evolution of complex random
56: networks \cite{review1,review2,review3}.
57: Real data have been analyzed and at the same time
58: many simplified models have been formulated and investigated.
59: The applications in different contexts are
60: ranging from evolution of WWW and Internet, over
61: evolution of metabolic networks to the structure of social networks,
62: or linguistic networks.
63: Detailed mechanisms of evolution differ from one problem
64: to another, nevertheless, some common
65: features have been already identified.
66: For example, it was revealed that many networks are
67: {\it scale free}, i.e., the distribution of the connectivity
68: of nodes (vertices) is a power-law.
69: It was shown that the
70: scale free structure can be reproduced
71: in the case of growing systems by models
72: with the preferential attachment for adding new nodes \cite{prefAtt}.
73:
74: The challenging open problem is to clarify
75: what are the universal feature of network dynamics, if any.
76: The concept of scale free networks does not provide
77: a classification into universality classes. A wide range of exponents
78: was found by analysis of the real data,
79: and the exponents in models with preferential attachment
80: can take an arbitrary value larger than two
81: \cite{dorogovtsev00,krapivsky00}.
82: Moreover, this concept deals with the internal structure
83: of the networks rather than with possible types of dynamics.
84:
85: It is of interest to explore the time dependence of global
86: characteristics of networks.
87: Two such global quantities are the size of the network $N(t)$
88: and the mean connectivity $\overline{k}(t)$.
89: The size of the network is the total number of nodes and the mean
90: connectivity is defined as the average of the degrees
91: of nodes (the local connectivities) over all nodes in the network.
92: Different types of the network evolution can be encountered.
93: In some cases, one observe a rapid increase of the network size,
94: the typical example being the Internet.
95: Another systems, e.g. linguistic networks,
96: seem to be in a stationary state with an approximately constant size.
97: Similarly the mean connectivity can also increase or saturate.
98: The situation clearly depends on the time scale considered.
99: A network which is rapidly growing in a certain time interval may
100: after some transient reach a stationary state.
101: A system which looks
102: to be stationary or only slowly varying
103: in a limited time interval may exhibit strong fluctuations
104: on a long time scale.
105:
106: The network size and the mean connectivity are
107: expected to fluctuate in real systems.
108: The fluctuation and/or shortage of reliable real data in a
109: sufficiently long time interval can result
110: in difficulty or even impossibility of tracing the
111: general trends.
112: However, the fluctuations can be reduced
113: by averaging over many independent realizations
114: when the evolution is explored by the simulations of
115: simplified models.
116: Then the information about the time
117: behaviour can be extracted, and bias or saturation can be revealed.
118: One can ask what are the time dependences of the global %averaged
119: quantities,
120: whether there are generic scenarios of the network evolution etc.
121:
122: In this Letter, we study the time evolution of global characteristics
123: in the model which has been recently introduced by us in the context
124: of study of the biological evolution \cite{sla_ko_99,sla_ko_00}.
125: We found that the averaged network size and the averaged
126: mean connectivity
127: initially increase as power-laws and later saturate.
128: We show that these quantities
129: exhibit dynamical scaling
130: analogous to the scaling revealed previously in the study of surface growth
131: \cite{fam_vic_85}.
132: The existence of scaling invariance suggests the possible
133: universal behaviour. Therefore, we check the sensitivity to
134: modifications of the model.
135:
136: \section{Model}
137: To fix ideas, we briefly describe our model.
138: The model is based on extremal dynamics \cite{ba_sne_93}
139: with particular rules
140: for adding and removing a node.
141: The system is composed of $N$ nodes, connected by links.
142: The state of each node is described
143: by a single dynamical variable $b\,\epsilon\, (0,1)$, called barrier.
144: In each step, the following dynamical rules \cite{sla_ko_99}
145: are applied:
146:
147:
148: {\it (i) } The node with minimum $b$ is found and mutated.
149: The barrier of the mutated node is replaced by a new random
150: value $b^\prime$.
151:
152: {\it (ii) }
153: The barriers of all nodes linked to
154: the selected unit are also replaced
155: by new random numbers.
156: If $b^\prime$ is larger than barriers of all its linked neighbours,
157: the node gives birth to a new node (speciation). If $b^\prime$
158: is lower than barriers of all linked neighbours, the
159: node dies out (extinction). In remaining cases,
160: the number of nodes remains unchanged.
161:
162: The motivation of these rules is the following.
163: Well-adapted nodes
164: proliferate more rapidly and chance for speciation is higher.
165: However, if the local diversity, measured by connectivity of the node, is
166: bigger, there are fewer available possibilities for the change
167: and the probability of speciation is lower. On the other hand,
168: poorly adapted nodes are more vulnerable to extinction, but at the
169: same time larger diversity (larger connectivity) may favour the survival.
170:
171: {\it (iii) } Speciation means, that a new node is added
172: into the system, with a random barrier.
173:
174: {\it (iv) } The links are established between the new node and the
175: neighbours of the mutated node: each link of the ``mother'' node is
176: inherited with the probability $p$ by the ``daughter'' node.
177: The rule {\it (iv) } reflects the fact that the new node is
178: to a certain extent
179: a copy of the
180: original, so the relations to the environment will be initially
181: similar to the ones the old node has.
182: Moreover, if a node which speciates has only one neighbour, a link between
183: ``mother'' and ``daughter'' is also established.
184: Similar models with the duplication of the local geometry
185: were recently used
186: for modeling of protein interaction network \cite{vazquez01,sole,kim}.
187:
188:
189: {\it (v) }
190: When the node is extinct it means, that the node is removed
191: without any substitution and all links it has, are broken.
192: As a boundary condition, we use the
193: following exception: if the network
194: consists of a single isolated node only, it never dies out.
195: We suppose that a node completely isolated
196: from the rest of the system has lower chance to survive. This
197: leads to the following rule.
198:
199: {\it (vi) }
200: If a node dies out, all its neighbours which
201: are not connected to any other node die out
202: with the probability $p_{\rm sing}$.
203: We call this kind of
204: extinctions {\it singular extinctions}.
205:
206:
207: \begin{figure}
208: \twofigures[scale=0.7]{fig1.ps}{fig2.ps}
209: \caption{Time evolution of the averaged number of nodes
210: $\langle N \rangle$
211: (upper curves)
212: and the averaged mean connectivity
213: $\langle\overline{k}\rangle$ (lower curves)
214: for different values of the parameter $p$.
215: Values of $p$ from bottom to top are
216: $p=0.85$, $0.9$, $0.95$, $0.97$, $1$ for number of nodes
217: and
218: $p=0.85$, $0.9$, $0.95$, $0.97$ for connectivity.
219: Data are averaged over 1000 independent runs.
220: Straight lines are guides for eye with slopes 0.35
221: (dashed)
222: and 0.24 (dot-dashed).
223: }
224: \label{fig:N-c-evol}
225: \caption{Data collapse of time evolution of the number of
226: nodes $\langle N \rangle$ for different
227: values of the parameter $p$ from Fig. \ref{fig:N-c-evol}.
228: Inset: Saturated values of the number of
229: nodes $N^{\rm sat}$ (crosses)
230: and times of saturation $t^{\rm sat}_N$ (plusses) vs. $(1-p)$.
231: The straight lines are guides for eye with slopes 0.8 (full) and 2.3 (dashed).
232: }
233: \label{fig:N-collapse}
234: \end{figure}
235:
236: We previously showed that the model is self-organized critical
237: \cite{sla_ko_99,sla_ko_00}.
238: Here, we focus on scaling with time.
239: In all simulations below, the evolution starts with a single node.
240:
241: \section{Dynamic scaling of evolving network}
242: In this part, we restrict ourself to the situation $p_{\rm sing}=1$
243: (all isolated nodes will die out), but we consider different values of
244: the parametr $p$.
245: The size
246: $N$ as well as the connectivity $\overline{k}$
247: averaged over all nodes
248: strongly fluctuate
249: during the evolution (cf. Fig. 2 in \cite{sla_ko_99}),
250: and no clear tendency can be deduced.
251: However, two different regimes are identified
252: after averaging over many independent runs (Fig. \ref{fig:N-c-evol}).
253: A crossover time $t^{\rm sat}$
254: between two regimes depends on the parameter $p$.
255: In the initial stage $t\ll t^{\rm sat}$,
256: both averaged quantities
257: increase as power-laws
258: \begin{equation}
259: \langle N\rangle \propto t^{\beta_N}, \hspace*{10mm}
260: \langle \overline{k}\rangle \propto t^{\beta_k}
261: \end{equation}
262: with the exponents $\beta_N \simeq 0.35$ and
263: $\beta_k \simeq 0.24$.
264: Here $\langle ... \rangle$ is a statistical average,
265: and the bar denotes an average over nodes.
266: In the second regime, the averaged number of nodes $\langle N\rangle$
267: and therefore also the averaged mean
268: connectivity $\langle \overline{k}\rangle$
269: saturate after times $t^{\rm sat}_N$ and $t^{\rm sat}_k$.
270: The crossover times $t^{\rm sat}_N$ and $t^{\rm sat}_k$
271: are different ($t^{\rm sat}_k$ being smaller that $t^{\rm sat}_N$).
272:
273: \begin{figure}
274: \twofigures[scale=0.7]{fig3.ps}{fig4.ps}
275: \caption{Data collapse of time evolution of the
276: averaged mean connectivity
277: $\langle \overline{k}\rangle$
278: for different
279: values of parameter $p$ from Fig. \ref{fig:N-c-evol}.
280: Inset: Saturated values of the averaged mean connectivity
281: $\overline{k}^{\rm sat}$ (crosses)
282: and time for saturation $t^{\rm sat}_k$ (plusses) vs. $(1-p)$.
283: The straight lines are guides for eye with slopes 0.6 (full) and 2.5 (dashed).
284: }
285: \label{fig:c-collapse}
286: \caption{
287: Time evolution of the averaged number of nodes
288: $\langle N \rangle$
289: in the lattice variant of the model with 8 neighbours
290: and
291: for different values of the parameter $p_{\rm sing}$.
292: Values of $p_{\rm sing}$ from top to bottom are $p_{\rm sing}=$
293: $0.05$, $0.1$, $0.2$, $0.4$, $0.6$, $0.8$, $1$.
294: Data are averaged over 1000 to 10000 independent runs.
295: The straight line is guide for eye with slope 0.41.
296: Inset:
297: Time evolution of the averaged number of nodes
298: $\langle N \rangle$
299: in the lattice variant of the model with 4 neighbours.
300: Values of $p_{\rm sing}$ from top to bottom are $p_{\rm sing}=$
301: $0.05$, $0.1$, $0.2$, $0.4$.
302: The straight line are guide for eye with slope 0.41.
303: }
304: \label{fig:NevolLat}
305: \end{figure}
306:
307: The saturated values $N^{\rm sat}$, $\overline{k}^{\rm sat}$ and
308: the times of saturation $t^{\rm sat}_N$ and $t^{\rm sat}_k$ increase
309: with $p$.
310: The parameter $p$ controls correlations between
311: the structure of the neighbourhood
312: of the ``mother'' node and the ``daughter'' node.
313: We define $\xi =1/(1-p)$ as an analog of the correlation
314: length. We have found that the saturated values $N^{\rm sat}$,
315: $\overline{k}^{\rm sat}$ depend on $\xi$
316: as powers (insets in Figs. \ref{fig:N-collapse}
317: and \ref{fig:c-collapse})
318: \begin{equation}
319: N^{\rm sat} \propto \xi^{\alpha_N}\,,\hspace*{10mm}
320: \overline{k}^{\rm sat} \propto \xi^{\alpha_k}\,\hspace*{10mm}
321: \end{equation}
322: with the exponents
323: $\alpha_N =0.8$, $\alpha_k =0.6$.
324: The exponents $\alpha_N$ and $\alpha_k$ approximately agree
325: with the exponents obtained in \cite{sla_ko_00}
326: by the calculation of the mean values using the stationary distribution
327: ${\cal P}(k)$.
328: Both crossover times also fulfil power-law
329: \begin{equation}
330: t^{\rm sat}_N \propto \xi^{z_N}\,, \hspace*{10mm}
331: t^{\rm sat}_k \propto \xi^{z_k}\,.
332: \end{equation}
333: The exponents are $z_N=2.3$ and $z_k=2.5$
334: (insets in Figs. \ref{fig:N-collapse} and \ref{fig:c-collapse}).
335: The behaviour for $p = 1$ is different.
336: There is no saturation for $p = 1$ on the scale of
337: our simulations and the model is critical \cite{sla_ko_99,sla_ko_00}.
338:
339:
340: This scaling behaviour is similar to the well
341: known dynamic scaling for kinetic roughening
342: during surface growth \cite{fam_vic_85}.
343: Data for $\langle N\rangle (t,\xi)$ as well as for
344: $\langle \overline{k}\rangle (t,\xi)$
345: can be rescaled to a single curve (Figs. \ref{fig:N-collapse}
346: and \ref{fig:c-collapse}).
347: Hence, both $\langle N\rangle$ and $\langle \overline{k}\rangle$
348: fulfil the following scaling relations
349: \begin{equation}
350: \langle N\rangle \propto \xi^{\alpha_N} f_N(t/\xi^{z_N}), \hspace{10mm}
351: \langle \overline{k}\rangle \propto \xi^{\alpha_k} f_k(t/\xi^{z_k}),
352: \end{equation}
353: where $f_N(x)$ and $f_k(x)$ are two different
354: scaling functions with properties:
355: $f_N(x) = const_1$, $x \gg 1$, $f_N(x) \propto x^{\beta_N}$,
356: $x \ll 1$ and
357: $f_k(x) = const_2$, $x \gg 1$, $f_k(x) \propto x^{\beta_k}$, $x \ll 1$.
358: Exponents $\beta_N$, $\alpha_N$, $z_N$
359: ($\beta_k$, $\alpha_k$, $z_k$) are not independent but obey the relation
360: $\beta_N$$=$$\alpha_N/z_N$
361: ($\beta_k$$=$$\alpha_k/z_k$)
362: (for reviews on kinetic roughening see
363: \cite{kr1,kr2}).
364: The relations between exponents are well satisfied for the
365: above obtained values.
366:
367: \section{Lattice variant of the model}
368: If $p_{\rm sing} < 1$ then
369: more and more isolated nodes will be generated.
370: The above rules do not allow to connect
371: an isolated node to any other node.
372: It will remain isolated forever.
373: This may be the unrealistic situation.
374: One rather expects that isolated nodes can be with a certain probability
375: reconnected. In this case an additional rule for reconnection would be needed,
376: e.g., based on the list of links of extinct nodes.
377: This would imply the need of complicated book-keeping.
378: In order to assess the effect of singular
379: extinctions, we use a different
380: variant of the model which allows to avoid the above
381: unrealistic situation.
382:
383: We suppose that nodes are now living on a regular lattice.
384: It means that only some sites of the given lattice are occupied by
385: nodes at the given time. The occupation may change during evolution.
386: The lattice sites can represent the different positions in the
387: Euclidean space or in some abstract space.
388: The extremal dynamic is again emploied,
389: nodes are added or removed using the rule {\it (ii)}.
390: However, a node can speciate only if there is an unoccupied site
391: in the neighbourhood of the selected extremal node.
392: A new ``daughter'' node is placed on a randomly selected
393: unoccupied site within the neighbourhood.
394:
395: The structure of links of the ``daughter'' node
396: is now not inherited but it is automatically given by
397: existing nodes in
398: her neighbourhood.
399: The links are established from a given node to all
400: neighbouring nodes (occupied sites within a given neighbourhood).
401: Hence, the above rule {\it (iv)}
402: does not apply and is replaced be automatic connection within the given
403: neighbourhood.
404: The network is formed by clusters of connected nodes
405: on the lattice.
406: All links of an extinct node are broken.
407: We apply the rule {\it (vi)}
408: for singular extinction with $p_{\rm sing}\,\epsilon\, <0,1>$.
409: Therefore, isolated nodes can be created and can be reconnected.
410: The rules also allow reconnection of previously disconnected clusters.
411: The lattice variant of the model has similar self-organization properties
412: as the above described off-lattice variant.
413: In particular, it exhibits power-law
414: in the distribution of forward avalanches \cite{steiner00}.
415:
416: We consider for simplicity in the following
417: the simple square lattice of the size $L\times L$ with the periodic boundary
418: conditions.
419: The size $L$ has been chosen so large that the finite size effects
420: have no influence ($L$ = 40--160 depending on $p_{\rm sing}$).
421: We performed simulations for two definition of
422: the neighbourhood.
423: The neighbourhood composed of
424: 4 nearest neighbours (4n-model) and
425: 4 nearest neighbours plus 4 next nearest neighbours (8n-model).
426:
427: We found that the time behaviour of the network size and the
428: averaged mean connectivity is qualitatively the same as in the off-lattice model.
429: Both quantities
430: are fluctuating, however, after averaging (over 1000 or more independent
431: runs) two regimes can be clearly seen: the
432: initial power-law increase and the saturation at the late times.
433: The time evolution of network size for
434: two modifications with 4 and 8 neighbours is
435: displayed in Fig. \ref{fig:NevolLat}.
436: The initial power-law increase
437: $\langle N\rangle \propto t^{\beta_N} $
438: with exponent $\beta_N = 0.41$ is observed in both cases.
439: $N_{\rm sat}$ increases with decreasing $p_{\rm sing}$.
440: Saturated values are substantially lower for 4n-model than for 8n-model.
441:
442: The dependence of saturated values and times of saturation
443: on $p_{\rm sing}$ is shown in Fig. \ref{fig:NsatLat}.
444: There are again power-laws with more or less the same exponents
445: for both considered neighbourhoods.
446: Similarly as above, we define an analog of the correlation length
447: $\tilde{\xi} = 1/p_{\rm sing}.$
448: Then we have
449: \begin{equation}
450: N^{\rm sat} \propto \tilde{\xi}^{\alpha_N}\,,\hspace*{10mm}
451: t^{\rm sat}_N \propto \tilde{\xi}^{z_N}\,\hspace*{10mm}
452: \end{equation}
453: with the exponents
454: $\alpha_N = 1.1$ and $z_N = 2.7$.
455: The relation
456: $\beta_N$$=$$\alpha_N/z_N$ is well satisfied.
457: We show the corresponding data collapse
458: supporting the scaling relation
459: $
460: \langle N\rangle \propto \tilde{\xi}^{\alpha_N} f_N(t/\xi^{z_N})
461: $
462: in Fig. \ref{fig:N-collapse2}.
463: When $p_{\rm sing} = 0$ the network grows until the evolution will
464: be affected by the given finite lattice size.
465: \begin{figure}
466: \twofigures[scale=0.7]{fig5.ps}{fig6.ps}
467: \caption{Saturated values of the number of nodes
468: $N^{\rm sat}$ for the model with 4 neighbours (crosses)
469: and with 8 neighbours (plusses) and time of saturation
470: $t^{\rm sat}_N$
471: for the model with 4 neighbours (squares)
472: and with 8 neighbours (stars)
473: vs. $p_{\rm sing}$.
474: The straight lines are guides for eye with slopes 1.1 (full) and 2.7 (dashed).
475: }
476: \label{fig:NsatLat}
477: \caption{Data collapse of time evolution of the number of
478: nodes $\langle N \rangle$ in the lattice variant of the model
479: with 8 neighbours for different
480: values of parameter $p_{\rm sing}$ from Fig. \ref{fig:NevolLat}.
481: Inset: Data collapse of time evolution of the number of
482: nodes $\langle N \rangle$ in the lattice variant of the model
483: with 4 neighbours.
484: }
485: \label{fig:N-collapse2}
486: \end{figure}
487:
488: Due to low maximal possible value,
489: the connectivity saturates very soon
490: and good scaling of connectivity with
491: $p_{\rm sing}$ has not been obtained.
492: Therefore, data for time evolution of connectivity are not
493: presented.
494: In order to study carefully scaling of connectivity
495: in the lattice model, one should consider
496: a larger size of neighbourhood - include also more distant
497: sites or consider model on the hypercube.
498:
499:
500:
501: An external condition limiting the size of the network or the connectivity
502: will lead trivially to the saturation of these quantities. One
503: expects that there will be scaling with the limiting value.
504: We verified that indeed there is dynamic
505: scaling
506: of the form
507: \begin{equation}
508: \langle N\rangle \propto L^{\alpha_N} g(t/L^{z_N})\,
509: \end{equation}
510: with the time and the external parameter $L$.
511: Here, $g(x)$ is again
512: scaling function with the properties:
513: $g(x)=const_3$ for $x \gg 1$
514: $g(x)$$\propto$$x^{\beta_N}$ for
515: $x$$\ll$$1$.
516: In the case of 4n-model,
517: we measured the exponents
518: $\alpha_N =2.07$, $\beta_N =0.44$, $z_N=4.8$
519: fulfilling well the relation $\beta_N$$=$$\alpha_N/z_N$.
520:
521:
522: \section{Conclusion}
523: We have studied the time dependence of global
524: characteristics of evolving random networks.
525: We used the network model with
526: the extremal dynamics and with the
527: variable rules for evolution of the local
528: geometry.
529: We have found that the averaged network size and the averaged
530: mean connectivity
531: exhibit dynamic scaling with time and with
532: the parameter controlling the form of dynamical rules
533: which in turn affects spreading of the correlations during evolution.
534: We considered two different variant of the model:
535: the off-lattice variant with the varying reproduction
536: of the local topology
537: and the lattice variant with the variable amount of
538: removal of isolated nodes.
539: We have found that
540: dynamic scaling is fulfiled in both variants of the model.
541: However, the measured exponents are different.
542: We attribute this variance to the difference
543: in mechanisms of evolution.
544: On the other hand,
545: we verified that there are universal features.
546: The exponents in the lattice variant are
547: robust to the change of
548: the size of neighbourhood.
549:
550: We expect that dynamic scaling can be observed also in other models
551: of evolving networks with the tunable internal dynamics.
552: In general, dynamic scaling of evolving networks
553: can be characterized by four independent exponents, two for the
554: time dependence of the averaged size, and two for the evolution
555: of the averaged mean connectivity.
556: The exponents for evolution of the network size may be trivial
557: or not defined in models in which the number of nodes
558: is only and permanently increasing.
559: Nevertheless, the connectivity may saturate and exhibit nontrivial
560: scaling.
561: Varying the external parameters like the maximal size leads to a
562: qualitatively different scaling.
563: We believe that the study of dynamic scaling of evolving networks
564: might provide a clue to identification of
565: universal features of network evolution.
566: It is an open question if dynamic scaling
567: will allow the classification of types of network evolution
568: into different universal classes
569: as in the case of the surface growth, or if evolution of global
570: characteristics turns to be unique for
571: each individual network as it has been found for
572: exponents describing the distribution of connectivities in scale
573: free networks.
574:
575:
576:
577:
578: \acknowledgments
579: This work was supported by grant No. 202/01/1091 of the GA \v{C}R
580:
581: \begin{thebibliography}{0}
582:
583: \bibitem{review1}
584: \Name{Strogatz S. H.}
585: \Review{Nature} \Vol{410} \Year{2001} \Page{131}.
586:
587: \bibitem{review2}
588: \Name{Albert R.} \And \Name{Barab\'asi A.-B}
589: \Review{Reviews of Modern Physics} \Vol{74} \Year{2002} \Page{47}.
590:
591: \bibitem{review3}
592: \Name{Dorogovtsev S.N.} \And \Name{Mendes J.F.F.}
593: cond-mat/0106144.
594:
595: \bibitem{prefAtt}
596: \Name{Barab\'asi A.-L.} \And \Name{Albert R.} \Review{Science}
597: \Vol{286} \Year{1999} \Page{509}.
598:
599: \bibitem{dorogovtsev00}
600: \Name{Dorogovtsev S. N.} \Name{Mendes J. F. F.}
601: \And \Name{Samukhin A. N.}
602: \Review{Phys. Rev. Lett.} \Vol{85} \Year{2000}
603: \Page{4633}.
604:
605: \bibitem{krapivsky00}
606: \Name{Krapivsky P. L.} \Name{Redner S}
607: \And \Name{Leyvraz F.}
608: \Review{Phys. Rev. Lett.} \Vol{85} \Year{2000} \Page{4629}.
609:
610: \bibitem{sla_ko_99}
611: \Name{Slanina F.} \And \Name{Kotrla M.}
612: \Review{Phys. Rev. Lett.} \Vol{83} \Year{1999} \Page{5587}.
613:
614: \bibitem{sla_ko_00}
615: \Name{Slanina F.} \And \Name{Kotrla M.} \Review{Phys. Rev. E}
616: \Vol{62} \Year{2000} \Page{6170}.
617:
618: \bibitem{fam_vic_85}
619: \Name{Family F.} \And \Name{Vicsek T.} \Review{J. Phys A}
620: \Vol{18} \Year{1985} \Page{L75}.
621:
622: \bibitem{ba_sne_93}
623: \Name{Bak P.} \And \Name{Sneppen K.}
624: \Review{Phys. Rev. Lett.}
625: \Vol{71} \Year{1993} \Page{4083}.
626:
627: \bibitem{vazquez01}\Name{Vazquez A.} \Name{Flammini A.}
628: \Name{Maritan A.} \And \Name{Vespignani A.}
629: cond-mat0108043.
630:
631: \bibitem{sole}\Name{Sol\'e R. V} \Name{Pastor-Satorrasf R.}
632: \Name{Smith E.} \And \Name{Kepler T. B.}
633: \Review{Adv. in Complex Systems} \Vol{5} \Year{2002} \Page{43}.
634:
635: \bibitem{kim}\Name{Kim J.} \Name{Krapivsky P. L.} \Name{Kahng B.}
636: \And \Name{Redner S.}
637: cond-mat/0203167.
638:
639:
640: \bibitem{kr1}
641: \Name{Barab\'asi A.-L.} \And \Name{Stanley H. E.}
642: \Book{Fractal concepts in Surface Growth}
643: (Cambridge University Press, Cambridge 1995).
644: \bibitem{kr2}
645: \Name{Halpin-Healy T.} \And \Name{Zhang, Y.-C.}
646: \Review{Phys. Rep.} \Vol{254} \Year{1995} \Page{215}.
647:
648: \bibitem{steiner00}\Name{Steiner J.} thesis, Charles University,
649: Prague, 2000.
650:
651:
652: \end{thebibliography}
653:
654: \end{document}
655:
656:
657: