1: \documentclass[prl,aps,preprint,showpacs,showkeys]{revtex4}
2: \topmargin -0.5 cm
3: \textwidth 16.0 cm
4: \oddsidemargin -0.0 cm
5: \linespread{1.6}
6: \renewcommand{\baselinestretch}{1.2}
7:
8: \begin{document}
9: %\setcounter{page}{1}
10: \date{\today}
11: \title{Families and clustering in a natural numbers network}
12:
13: \author{Gilberto Corso }
14:
15: \affiliation{ International Center for Complex Systems and
16: Departamento de Biof{\'\i}sica,
17: Centro de Bioci\^encias,
18: Universidade Federal do Rio Grande do Norte, Campus Universit\'ario
19: 59072 970, Natal, RN, Brazil.}
20:
21:
22: \begin{abstract}
23:
24: We develop a network in which the natural numbers are the vertices. The
25: decomposition of natural numbers by prime numbers is used to establish the
26: connections. We perform data collapse and show that the degree
27: distribution of these networks scales
28: linearly with the number of vertices. We explore th families of vertices
29: in connection with prime numbers decomposition. We compare the average
30: distance of the network and the clustering coefficient with the distance
31: and clustering coefficient of the corresponding random graph. In case
32: we set connections among vertices each time the numbers share a common
33: prime number the network has properties similar to a random graph.
34: If the criterium for establishing
35: links becomes more selective, only prime numbers greater
36: than $p_l$ are used to establish links, the network
37: has high clustering coefficient.
38:
39:
40:
41: \end{abstract}
42: \pacs{89.75.-k, 89.75.Fb, 89.75.Hc, 02.10.De, 89.20.-a }
43:
44: \keywords{networks, complex systems, small-world, clustering coefficient}
45:
46: \maketitle
47:
48:
49: \section{I. Introduction}
50: The new century have started with a strong development in
51: network theory and specially in small-world networks.
52: Three ingredients are necessary to define
53: a small-world network: a sparse network, small distance and
54: high clustering coefficient \cite{WTmodel,barabasi}. Many examples of these networks
55: have been analyzed in diverse fields as
56: computation \cite{computation1,computation2},
57: linguistic \cite{linguistic1,linguistic2}, biology \cite{biology1,biology2},
58: economics \cite{economics,eco2} and social phenomena \cite{new1,sex}.
59: In addition to the interest in the description of these particular
60: phenomena, small-world networks are promising elements to compose
61: a general theory of complex systems.
62:
63: Nowadays the two major research lines in networks are the search for
64: small world networks in nature and the investigation of
65: theoretic models that explain
66: the properties of such networks. The first line was pointed out
67: in the first paragraph. The second line is dominated by the
68: evolving network models \cite{evol1,evol2} and the study of phase order transitions
69: in networks \cite{phase1,phase2}.
70: This work is situated at an intermediary place between both
71: lines. We characterize a network using recent techniques of
72: statistical physics but
73: instead of looking for real data in the world we construct
74: a network using some properties of the set of natural numbers.
75:
76: We use as the keystone in the construction of number networks the
77: fundamental theorem of number theory which says that each natural
78: number has an unique decomposition in prime factors. It means,
79: for a number $a$ and prime numbers $p_j$, there is an unique
80: product:
81: \begin{equation}
82: a = p_1^{\alpha_1} p_2^{\alpha_2}... p_m^{\alpha_m}
83: \label{fund}
84: \end{equation}
85: where the exponent $\alpha_j$ is the multiplicity of the prime $p_j$.
86: The simplest way to define a link between two vertices (numbers) is the
87: following: each time two numbers have a prime in common in equation (\ref{fund})
88: they have a connection. We shall see that only this criterium is not
89: good enough to construct networks with small clustering coefficient.
90:
91:
92:
93: This paper has a twofold objective: present a new network model
94: based on number properties and characterize this network using
95: the approach of modern complex network analyzes.
96: In section $2$ we show the model $M$ and its extensions $M_l$, we
97: drive our attention to the degree distribution of the network, the set of families and
98: the search of invariant quantities. In section $3$ we analyse the distance
99: and the clustering coefficient for these networks.
100: In section $4$ we give our final remarks.
101:
102: \section{II. The Model}
103:
104: In this section we present the standard model $M$ and its extensions
105: $M_l$. We use the set of natural numbers as vertices
106: and an arithmetic property
107: establishes the connections. The connections have no weight
108: neither direction.
109:
110: \subsection{A. The Basic Model $M$}
111:
112: In model $M$ the criterium for the existence of a connection between two
113: vertices is the following: there is a link between two numbers $a$ and $b$ if
114: they share a common divisor. In other words, if $a$ and $b$ have a common
115: prime number $p_j$ in decomposition (\ref{fund}) a link is established.
116:
117: Figure \ref{fig1} shows a simple realization of this rule for number of
118: vertices $N=16$. All the even numbers are interconnected
119: (they share the prime $p=2$). Besides the divisors of $3$, $5$, etc, are
120: also interconnected.
121: The most connected numbers for this $N$ are $6$ and $12$ because they
122: are linked to all the even numbers and all the multiples of $3$.
123:
124: The first point we explore in the network $M$ is the degree
125: distribution, it means the connectivity $k$ of the vertex $i$
126: for vertices ordered according to the connectivity.
127: Figure \ref{fig2} displays $k$ versus $i$. In this
128: figure we have $N=2^{12}=4096$
129: vertices. The vertex $i$ with maximum number of links is $i=2310=2 \times 3
130: \times 5 \times 7 \times 11$ which is
131: connected with all the even numbers
132: and with the multiples of $3$, $5$, $7$, and $11$. In general,
133: the most connected vertex in a network $M$ of $N$ vertices corresponds to
134: the maximum number $i=p_1 p_2 ... p_m \leq N$ where the $p_j$
135: correspond to the first primes (which are the most connected numbers).
136: In the opposite side of the graphic there
137: are the prime numbers $p_j$ that have $k=0$, these numbers
138: satisfy the relation $2 p_j \leq N$ (if $p_j$ do not
139: fulfill this relation it will be connected to
140: the node $2$ and it would not have $k=0$).
141: The general view of the degree distribution shown in
142: figure \ref{fig2} is the following. Half of the vertices, which
143: corresponds to the even numbers,
144: are connected among them, so they have $k \geq N/2$. Otherwise, the odd numbers
145: have $k< N/2$ because they are at least not connected to all the even numbers.
146: The frontier between these two sets is indicated by the large
147: plateau that starts at $i=N/2$.
148:
149: The plateau in the middle of figure \ref{fig2} corresponds
150: to the family of multiples of the number $2$. In the beginning of this plateau we find the numbers
151: $2^{\alpha}$ for $1\leq \alpha \leq 12=log_2 N$. These numbers are connected with all the even
152: numbers and only with them. For these numbers $k = N / 2$. The smooth tail
153: that comes for $i < N / 2$ is formed by the numbers type $2^{\alpha} p_j \leq N$
154: where $p_j$ is a weak connected prime. These numbers are connected with all the
155: even numbers and with the connections of $p_j$ that are just a few.
156:
157: The second largest plateau is related to the multiples of three (indicated
158: $F_3$ in figure \ref{fig2}).
159: One third of the $N$ natural numbers are multiples of three and have connectivity
160: larger that all the other odd numbers.
161: In the beginning of this plateau we find the
162: numbers of the form $3^{\beta}$ where $1 \leq \beta \leq log_3 N$.
163: This plateau is smaller than
164: the former one because there are less multiples of three than two.
165: For these numbers $k = N / 3$, it means, they are connected only
166: with the multiples of three. The tail of this plateau is formed by
167: the numbers $3^{\beta} p_j \leq N$, for $p_j$ weakly connected. There are
168: also plateaus visible in figure \ref{fig2}
169: corresponding to the primes $5$, $7$ and $11$. For these plateaus we have in the
170: coordinate axis $ k =N / 5$,
171: $N / 7$, and $N / 11$, respectively.
172:
173: Another plateau indicated in the figure is formed by the numbers that
174: share the two primes $2$ and $3$, it means the numbers that are multiples of
175: $6$. This plateau starts with numbers of the form: $2^{\alpha} 3^{\beta} \leq N$.
176: If the network were weighted, $k$ of this plateau would be the sum of $k$ of
177: $2$ and $3$ plateaus. As the network does not count multiple links the
178: corresponding value of $k$ is
179: smaller than the cited sum.
180: We call
181: $F_2$ the even plateau, $F_{10}$ the plateau generated by the numbers
182: that are divisible only by $2$ and $5$, and so one.
183: The most important family plateaus $F_2$, $F_3$, $F_5$, $F_7$, $F_{10}$,
184: $F_{14}$ and $F_{15}$
185: are indicated in figure \ref{fig2}.
186:
187: Figure \ref{fig2} shows two distinct regions,
188: above $i = N / 2$ the degree distribution is
189: dominated mainly by plateaus of prime numbers of the form: $p_j^{\alpha_j}$ and
190: their respective tails. Bellow
191: this number there are only plateaus composed by the combination
192: of prime numbers of the form: $p_j^{\alpha_j} p_m^{\alpha_m}$ and their
193: tails. In fact in the
194: limit of $N \rightarrow \infty$ an infinite number of plateaus would appear in
195: the curve whose relative sizes is determined
196: by equation (\ref{fund}). We conjecture that in this limit
197: a fractal distribution will appear.
198:
199:
200:
201: We have $k = N / 2$ for the beginning of the
202: plateau $F_2$, and in general $k = N / p_k$
203: for the families of primes $F_k$. This
204: relation suggests an
205: interesting scale property: the degrees scales linearly with $N$. We verify
206: numerically this fact in figure \ref{fig3} where we plot the normalized connectivity,
207: ($k/N$), versus the normalized index, ($i / N$),
208: for $N=128, 512$ and
209: $2048$. This figure verify by simulation that the connectivity of
210: the model $M$ scales linearly with the number of vertices.
211:
212:
213:
214: \subsection{B. The extension $M_l$ of the Model $M$}
215:
216: In the standard number decomposition, equation (\ref{fund}), there is the
217: possibility of including, or not, the factor $1$ because all the numbers are
218: trivially divisible by the unity. If the number $1$ is
219: included in the decomposition all the numbers would share the same common
220: divisor and, as a consequence, the network will became trivial: all the
221: vertices will be interconnected. In the same way as we excluded the number
222: $1$ as a divisor in the criterium for establishing links, we could also exclude
223: the number $2$. This idea suggests an alternative procedure
224: to define a network model for natural numbers.
225:
226: The network model $M_{p_l}$ is the following. The vertices are again the
227: natural numbers and
228: the connections are set using equation (\ref{fund}), but we take into account
229: only connections of primes $p_j \geq p_l$. In this sense the former model $M$ is
230: in fact $M_2$ because the links are established once there is a common factor
231: $p_j$ such that $p_j \geq 2$. In this way, the network $M_3$ take into account
232: the primes $p_j=3,5,...$ to establish links, but not the prime $2$.
233:
234:
235: Figure \ref{fig4} shows the degree distribution ($k$ versus $i$
236: in order of decreasing connectivity) for
237: $M_2$, $M_3$, $M_5$, and $M_7$ as indicated in the figure.
238: We use in the figure $N=2^{11}$. The curve of $M_2$ is the same curve shown in figure
239: \ref{fig2}. The curve of $M_3$ is similar to $M_2$, but the
240: largest even plateau, $F_2$, is absent. The largest plateau of $M_3$
241: starts at $i=N/3$, because all the one
242: third of the most connected numbers are multiples of $3$. We observe that
243: the largest plateau of $M_3$ has the same $k$ of $F_3$ of network $M_2$, in
244: fact in both cases the plateaus is formed by the $F_3$ family.
245: The curve $M_5$ does
246: not show the plateaus corresponding to the families $F_2$ and $F_3$ as expected;
247: and in this case the largest plateau is formed by the $F_5$ family.
248: The curve of $M_7$ follows the same tendency.
249: As a general trend the curves of degree distribution of
250: $M_l$ become smother for increasing $l$ because they have less connections related to
251: important prime numbers and, as a consequence, they present less plateaus.
252: The criterium for establishing links in the model $M_l$
253: becomes more restrictive as $l$ increases because the number of
254: connections decrease. It is interesting that the average connectivity,
255: $< \hspace{-0.1cm} k \hspace{-0.1cm} >= 2 n / N$, normalized by $N$ tends to a constant as
256: $N \rightarrow \infty$ ($n$ is the number of connections in the
257: network). In Table $I$ we show $< \hspace{-0.1cm} k \hspace{-0.1cm} > / N$ for the networks
258: $M_l$.
259:
260:
261: {\it \centerline{Table I}}
262: \begin{tabbing}
263:
264: \hspace{0.7cm} $M_l$ \hspace{1.8cm} \= $M_2$ \= \hspace{1.2cm} $M_3$ \= \hspace{1.2cm} $M_5$ \= \hspace{1.3cm} $M_7$ \\
265:
266: \hspace{0.1cm} $< \hspace{-0.1cm} k \hspace{-0.1cm} > / N $ \hspace{1.1cm} \= $0.45$ \= \hspace{1.1cm} $0.20$ \= \hspace{1.1cm} $0.09$ \= \hspace{1.1cm} $0.05$\\
267: \hspace{0.7cm} $\bar C$ \hspace{1.8cm} \= $1.78$ \= \hspace{1.1cm} $3.65$ \= \hspace{1.1cm} $8.22$ \= \hspace{1.1cm} $14.5$\\
268:
269: \end{tabbing}
270:
271: The first conclusion we take from the table is that the model $M$ is not sparse,
272: it means, the network does not fulfill the
273: condition $< \hspace{-0.1cm} k \hspace{-0.1cm} > \> \ll N$. Therefore
274: we have to be caution to compare properties of such
275: graphs with usual complex graphs in the physics literature.
276: The reason for $< \hspace{-0.1cm} k \hspace{-0.1cm} > \> \propto N$ is rooted in
277: the way connections are established in the network.
278: Each time a new even number is added it is connected with, at least, $N/2$ vertices,
279: and a multiple of three with $N/3$ vertices. This fact illustrates that in the
280: average $< \hspace{-0.1cm} k \hspace{-0.1cm} >$ increases linearly with $N$.
281:
282: We also analyze the scaling properties of the model $M_l$.
283: Figure \ref{fig5} shows the normalized connectivity,
284: ($ k / N $), versus the normalized index, ($i / N$), for
285: the model $M_5$. This curve is similar to the
286: one of figure \ref{fig3}, here we also use $N=128,512$ and $2048$.
287: The data collapse performed in the figure indicates that
288: the network $M_5$ scales with $N$. In fact, this same tendency is
289: observed for all networks $M_l$ analyzed. This behavior is related, as before,
290: with the plateaus of prime numbers $p_j$ whose connectivity
291: scales with $N$.
292: The fact that the degree distribution of the networks $M_l$
293: scales with $N$ suggests the existence of magnitudes that are independent
294: of $N$. This is the case of $< \hspace{-0.1cm} k \hspace{-0.1cm} > / N$ and
295: this is also the case of the clustering coefficient as we shall see in the
296: next section.
297:
298:
299:
300:
301: \section{III. Clustering coefficient and network distance}
302:
303: In this section we characterize the network models $M_l$ using the
304: distance, $d$, and the clustering coefficient, $C$. One of our objectives in
305: this work is to differentiate $M_l$ from random networks, it means,
306: networks whose distribution of links among vertices follow
307: a Poisson distribution. Therefore we compare
308: $d$ and $C$ of $M_l$ with $d$ and $C$ of the random network associated, it means,
309: the random network with the same number of vertices and connections.
310:
311:
312: The distance of a network is defined as the average distance between all the
313: two vertices of the network. The clustering coefficient is a global parameter
314: $C = \frac{\sum_N c_i}{N}$ which is based on the local clustering coefficient
315: $c_i$. For each vertex $i$ the respective $c_i$, is defined as the
316: normalized number of connection among its first neighbors. The parameter $C$
317: measures the average interconnection of the network. Using
318: the example of social networks of acquaintances, $c_i$ measures how much
319: the friends of a person (vertex $i$) are friends among them.
320: For a major treatment on this topic see \cite{barabasi}.
321:
322:
323: The distance of the random network associated, $d_{rand}$, is
324: estimated by simulation. We note that, because the graph is not sparse,
325: it is not valid that $d \propto ln(N)$. In fact, for
326: non-sparse graphs the distance is almost always two
327: \cite{mathbook}. Otherwise, the clustering
328: coefficient of the random graph associated, $C_{rand}$,
329: is analytically estimated \cite{barabasi}
330: and depend only on the number of
331: vertices $N$ and the number of connections $n$. For the random graph
332: the clustering coefficient is
333: $C_{rand} = < \hspace{-0.1cm} k \hspace{-0.1cm} > / N$.
334: We call $\bar d \equiv \frac{d}{d_{rand}}$
335: and $\bar C \equiv \frac{C}{C_{rand}}$ as the normalized distance and clustering
336: coefficient.
337:
338:
339: Figure \ref{fig6} shows the distance $d$ against
340: the network size $N$ for the data of the models $M_2$, $M_3$, $M_5$, and $M_7$.
341: The graphic is in log-linear form because of the large interval used in $N$.
342: The data confirms the prevision for non-sparse graphs that the
343: distance is around $2$. We estimate $\bar d$ for
344: the networks $M_2$, $M_3$, $M_5$, and $M_7$ in the range $2^5 \leq N \leq 2^{12}$
345: and find that $0.5 < \bar d < 0.9$. The general tendency is:
346: $\bar d$ slowly increases with $N$. In addition, $l$ large in models $M_l$ implies
347: smaller $\bar d$. This last fact is expected since for large $l$ the connections
348: of the network are more selective and organized.
349:
350:
351:
352:
353: The analysis of $\bar C$ for $M_2$, $M_3$, $M_5$ and $M_7$ shows
354: that it increases with $N$ until $N \simeq 2^6$ and stabilize
355: around a constant value. In Table $II$ it is shown $\bar C$ for
356: several $N$ for the model $M_5$, the other models $M_l$ follow a similar trend.
357:
358:
359: {\it \centerline{Table II}}
360: \begin{tabbing}
361: $\hspace{0.2cm} N$ \hspace{1.2cm} \= 32 \= \hspace{1.5cm} 64 \= \hspace{1.3cm} 128 \= \hspace{1.2cm} 256 \= \hspace{1.1cm} 512 \= \hspace{1.1cm} 1024 \\
362: $\hspace{0.2cm} \bar C$ \hspace{1.2cm} \= $6.95$ \= \hspace{1.1cm} $8.33$ \= \hspace{1.1cm} $8.18$ \= \hspace{1.1cm} $8.19$ \= \hspace{1.1cm}$8.28$ \= \hspace{1.1cm} $8.36$ \\
363:
364: \end{tabbing}
365:
366: The data points to a constant $\bar C$ in the limit $N
367: \rightarrow \infty$. The size invariance of $\bar C$ is compatible
368: with the size invariance of the degree distribution.
369: The best values of $\bar C$ for $M_2$, $M_3$, $M_5$ and $M_7$ are shown
370: in Table $I$. We observe in this table that $\bar C$ for the
371: network $M_l$ increases with $l$. The clustering
372: coefficient increases as the criterium
373: for establishing connections in the network becomes more selective.
374: In fact, for a constant $N$,
375: the number of connection, $n$, in the $M_l$ model decreases
376: for high $l$ (see Table $I$). As a result
377: $C_{rand} = < \hspace{-0.1cm} k \hspace{-0.1cm} > / N = 2 n / N^2$ decreases
378: in contrast with estimated $C$ that remain almost constant.
379:
380:
381:
382:
383: \section{IV. Final Remarks}
384: In this work we propose a new network model in which the natural numbers
385: are the vertices and the connections are based on their decomposition
386: by prime numbers. Using this criterium we develop a non-sparse
387: network ($< \hspace{-0.1cm} k \hspace{-0.1cm}> \sim \ll N$) which
388: have a distance of the order of $2$.
389: If we consider that all the prime numbers
390: in the decomposition set a link the network formed is similar to a random graph
391: because the high number of connections implies in a
392: small clustering coefficient. If the criterium for establishing
393: links becomes more selective, only prime numbers greater than $3$, or $5$, are used
394: to establish links the network has
395: a large clustering coefficient.
396:
397: We perform data collapse on the data and verify that the networks
398: studied have a degree distribution that is invariant with the number
399: of vertices $N$. The general view of the degree distribution is
400: a funny discontinuous curve with plateaus of all sizes.
401: These plateaus are generated by the families of numbers that share
402: the same prime numbers in their decomposition.
403:
404: An important class of networks are the scale-free ones. This concept is
405: mainly used to distinguish between networks that have exponential and power-law
406: degree distributions. In the exponential case
407: most of the vertices have a typical connectivity inside a range defined by
408: the exponent of the exponential function.
409: In the other side, a power-law case has
410: connections of all orders, it does not have a set of vertices with
411: a typical connectivity. The
412: number network $M_l$ do not have a smooth degree distribution
413: because of the plateaus formed by the families of prime numbers $F_k$,
414: therefore it is not possible to fit the degree by
415: a smooth curve. On the other side, due to families $F_k$, this network has
416: vertices with all orders of connectivity corresponding to all sizes
417: of primes and
418: their combinations. In this broad sense the network $M_l$ can be called a
419: scale-free network.
420:
421:
422:
423: This work unfolds an alternative perspective in the study of complex networks. Instead
424: of search for real networks in nature we explore deterministic
425: mathematical networks that
426: show small distance and high clustering coefficient. Despite the present network
427: is non-sparse it is a promising laboratory in the study of degree
428: distribution and cluster families.
429: In a future work we intend to explore some theoretical developments of this
430: problem: an evolving network algorithm for $M_l$ and an analysis of phase
431: transition in this model.
432:
433: \vspace{0.5cm}
434:
435:
436:
437: \centerline{\bf Acknowledgments}
438:
439: The author thanks Marcos Vinicius C. Henriques, Sidney Redner, Liacir S. Lucena,
440: and the anonymous referee for the fertile exchange of ideas.
441: The author also acknowledges the financial support of Conselho Nacional
442: de Desenvolvimento Cient{\'\i}fico e Tecnol{\'o}gico (CNPq)-Brazil and the
443: use of the following software for network analysis
444: http://vlado.fmf.uni-lj.si/pub/networks/pajek/.
445:
446:
447: %\vspace{1cm}
448:
449: \begin{thebibliography}{9}
450: %
451: %
452:
453: \bibitem{WTmodel} D. J. Watts and S. H. Strogatz, Nature {\bf 393} 440 (1998).
454:
455:
456: \bibitem{barabasi} R. Albert and A-L Barab\'asi, Rev. Mod. Phy. {\bf 74} 47
457: (2002).
458:
459: \bibitem{computation1} M. Faloutsos, P. Faloutsos and C. Faloutsos, Comput. Commun. Rev.
460: {\bf 29}, 251 (1999).
461:
462: \bibitem{computation2} R. Albert, H. Jeong, and A.-L. Barab\'asi,
463: Nature {\bf 401} 378 (1999).
464:
465: \bibitem{linguistic1} A. E. Motter, A. P. S. de Moura, Y-C Lai, and P. Dasgupta,
466: Phys. Rev. E {\bf 65} 065102(R) (2002).
467:
468: \bibitem{linguistic2} R. Ferrer i Cancho and R. V. Sol\'e, Proceedings
469: of the Royal Society of London B {\bf 268} 2261 (2001).
470:
471: \bibitem{biology1} H. Jeong, S. P. Mason, Z. N. Oltvai, and A.-L. Barab\'asi,
472: Nature {\bf 411} 41 (2001).
473:
474: \bibitem{biology2} J. M. Montoya and R. V. Sol\'e,
475: Journal of Theoretical Biology 214(3) (2002).
476:
477: \bibitem{economics} G. Corso, L.S. Lucena, and Z. D. Thom\'e,
478: Physica A, {\bf 324} 430 (2003).
479:
480: \bibitem{eco2} W. Souma, Y. Fujiwara and H. Aoyama,
481: Physica A, {\bf 324} 396 (2003).
482:
483: \bibitem{new1} M. E. J. Newman, Phys. Rev E. {\bf 64} 016131 (2001).
484:
485: \bibitem{sex} F. Liljeros, C. R. Edling, Luis A. Nunes Amaral, H. E. Stanley, and Y. Aberg,
486: Nature {\bf 411} 907 (2001).
487:
488: \bibitem{evol1} P. L. Krapivsky and S. Redner,
489: Computer Networks 39, 261-276 (2002)
490:
491: \bibitem{evol2} S. N. Dorogovtsev and
492: J. F. F. Mendes, Adv. Phys. {\bf 51 } 1079 (2002).
493:
494:
495: \bibitem{phase1} R. Cohen, K. Erez, D. Ben-Avraham, and S. Havlin,
496: PRL {\bf 85} 4626 (2000).
497:
498: \bibitem{phase2} D. S. Callaway, M. E. J. Newman, S. H. Strogatz, and D. J. Watts,
499: PRL {\bf 85} 5468 (2000).
500:
501:
502:
503: %\bibitem{att} S. N. Dorogovtsev,
504: %J. F. F. Mendes, and A. N. Samukhin, PRL {\bf 85} 4633 (2000).
505:
506: \bibitem{mathbook} B. Bollobas, {\it Random Graphs}
507: (Academic Press, London, 1985).
508:
509:
510: \end{thebibliography}
511:
512:
513: \centerline{FIGURE LEGENDS}
514:
515:
516: \begin{figure}[ht]
517: \begin{center}
518: \caption{ The network of the model $M$ for number of vertices $N=16$.
519: The natural numbers are connected according to the prime number decomposition. Two
520: numbers have a connection if they share a common prime in the decomposition. }
521: \label{fig1}
522: \end{center}
523: \end{figure}
524:
525: \begin{figure}[ht]
526: \begin{center}
527: \caption{ The connectivity $k$ versus index $i$ for the data of
528: network of model $M$. It is
529: used $N=4096$. The main families $F_k$ are indicated in the figure. }
530: \label{fig2}
531: \end{center}
532: \end{figure}
533:
534:
535: \begin{figure}[ht]
536: \begin{center}
537: \caption{ The normalized connectivity $k/N$ versus
538: normalized index $i/N$ for data of model $M$.
539: It is used $N=128$, $512$ and $2048$, as indicated in the figure. }
540: \label{fig3}
541: \end{center}
542: \end{figure}
543:
544:
545: \begin{figure}[ht]
546: \begin{center}
547: \caption{ The connectivity $k$ versus index $i$ for the models $M_2$, $M_3$,
548: $M_5$ and $M_7$ as indicated in the figure. It is used $N=2048$.}
549: \label{fig4}
550: \end{center}
551: \end{figure}
552:
553:
554: \begin{figure}[ht]
555: \begin{center}
556: \caption{ The normalized connectivity $k/N$ versus
557: normalized index $i/N$ for the data of model $M_5$. It is used
558: $N=128$, $512$ and $2048$ as indicated in the figure.}
559: \label{fig5}
560: \end{center}
561: \end{figure}
562:
563:
564: \begin{figure}[ht]
565: \begin{center}
566: \caption{The distance $d$ against network size $N$, for
567: the models $M_2$, $M_3$, $M_5$, and $M_7$ as indicated in the figure. }
568: \label{fig6}
569: \end{center}
570: \end{figure}
571:
572:
573: \end{document}
574:
575:
576:
577:
578:
579: