LETTER You may also like

- Collective stochastic coherence and

Role of links on the structural properties of synchronizability in weighted scale-free
networks
different network topologies Pablo Balenzuela, Pau Rué, Stefano
Boccaletti et al.

- Cluster synchronization in multiplex

To cite this article: Motahare Hazrati et al 2021 EPL 133 40001 networks
Sarika Jalan and Aradhana Singh

- The relationship between the topology and

synchronizability of partially
interdependent networks
View the article online for updates and enhancements. Lilan Tu, Shuai Song, Yujuan Wang et al.

This content was downloaded from IP address 115.76.51.1 on 29/04/2024 at 02:01

 February 2021
EPL, 133 (2021) 40001 www.epljournal.org
doi: 10.1209/0295-5075/133/40001

Role of links on the structural properties of different network

topologies
Motahare Hazrati1 , Shirin Panahi1 , Fatemeh Parastesh1 , Sajad Jafari3,1,2 and Dibakar Ghosh4
1
Department of Biomedical Engineering, Amirkabir University of Technology - No. 350, Hafez Ave., Valiasr Square,
Tehran 159163-4311, Iran
2
Health Technology Research Institute, Amirkabir University of Technology - No. 350, Hafez Ave., Valiasr Square,
Tehran 159163-4311, Iran
3
Center for Computational Biology, Chennai Institute of Technology - Chennai, India
4
Physics and Applied Mathematics Unit, Indian Statistical Institute - Kolkata 700108, India

received 1 September 2020; accepted in final form 15 January 2021

published online 30 March 2021

PACS 05.45.-a – Nonlinear dynamics and chaos

PACS 05.45.Xt – Synchronization; coupled oscillators

Abstract – Complex networks are of major importance in many areas of science. The network
property analysis of such networks can help researchers to understand many real-world systems.
Different collective phenomena emerge in complex networks, synchronization is one of the most
interesting states. The number of links plays a major role in synchronizability. In this paper,
the specific properties of the different well-known network topologies are studied, and their syn-
chronizability is compared. Networks with different structures, such as the regular, star, random,
small-world, and scale-free networks, are investigated. For each topology, the clustering coef-
ficient, average and variance of the path length, and the eigenvalues of the Laplacian matrix of
connections are obtained by varying the number of links. The results show some relations between
the network’s properties and synchronizability. One of the obtained results is that the type-one
networks in a small number of links show better synchronization in the lowest average and variance
of path length. However, for a greater number of links, the best synchronizability belongs to the
topologies with lower clustering coefficient.

c 2021 EPLA
Copyright 

Introduction. – Dynamical networks are observed ev- decades [5,6]. Synchronization is known as the strongest
erywhere in nature. They have a wide range of appli- form of the network’s cooperative dynamics [7,8]. Every
cations in many different fields of science. For instance, individual system of the dynamical networks regulates its
networks of the power grid, the World Wide Web, food motion during the synchronization state [9]. Synchronized
chain networks, economic networks between syndicates, behavior of the networks can be observed all around the
brain neurons networks can be mentioned [1,2]. These real world [10,11]. Accordingly, there is plenty of research
networks share the same specific characteristic, such as in the literature that addresses the synchronization of
interaction [3]. This implies that complex networks are complex networks with both analytical and numerical ap-
all composed of different individuals, which are effectively proaches [12–14]. For instance, the Master Stability Func-
interacting with each other as one system. In other words, tion (MSF) method provides the necessary condition for
superposition cannot be established in complex networks, the local stable synchronizability of the network [15]. The
and a small change in each unit affects the entire network global synchronization criteria can be provided with the
nonlinearly [4]. From the mathematical point of view, help of the connection graph stability method [16] and by
a network is considered as a graph G = (V, E), where V appropriate construction of a Lyapunov function [17,18].
is the set of nodes and E is the set of links or edges con- There are many factors involved in the synchroniza-
necting the nodes [1]. Varying the configurations of the tion of coupled oscillators. Three main factors are a) the
links in the networks can lead to various topologies with dynamical properties of each oscillator of the network,
different properties. b) coupling coefficients, and c) the network structure and
Collective behaviors that emerge in interactive com- topology. For instance, the dynamical properties of each
plex networks have drawn lots of attention in the last oscillator can affect the type of synchronization in the

40001-p1
Motahare Hazrati et al.

MSF approach. For example, the Lorenz and Rossler os- the “star” network. In this structure, all of the nodes are
cillators have type I and type II MSFs, respectively [19]. connected to the central node, known as the hub. This
The proper coupling strength is also crucial for the emer- network is used in many engineering applications [32].
gence of the collective behavior of the networks, in partic- “Random” networks are one of the pioneer dynamical
ular the synchronization, chimera state [20–22], and spiral networks that share some similarities with the real-life net-
waves [23]. Furthermore, topology and the structural works. There are different types of random graphs in the
properties of the network play an important role for the literature [34]. In the late 1950s, the random graph was
emerging of synchronization in complex networks [24–26]. studied by Erdős and Rényi (ER), which resulted in intro-
The different structural properties of the network have ducing the ER random graph [35]. About 40 years later,
been investigated in [27]. Some research papers have in- Barabasi and Albert (BA) proposed another random net-
vestigated the effect of topology of networks on the syn- work model, which is known as a scale-free network [36].
chronization [28–31]. However, the novelty of this work However, it was completely different from the ER random
is to study the effect of the number of links on synchro- network since the network is formed by increasing the new
nization for different network topologies. The number of nodes continuously. Also, the degree distribution of each
links is an important issue for understanding the epidemic node conforms to the power law [17]. The scale-free topol-
spreading pattern in epidemic networks. ogy starts with m0 nodes. Then, at each step, every node
In this paper, to comprehend the behavior of complex of the network is connected with m (<m0 ) other nodes.
networks better and to compare the synchronizability of At last, after K steps, a scale-free network is formed with
the networks, different dynamical properties of the net- N = K + m0 nodes, and mK links.
works are studied. We answer the key question “What A significant progress in investigating the attribute of
is the role of the number of links in the network synchro- the dynamical networks together with the advancements
nizability? ” To this end, complex networks with different of computer simulations revealed that neither regular nor
topologies, consisting of N identical oscillators are consid- random networks are as optimum as real networks. Watts
ered as follows: and Strogatz proposed a new type of topology called
N “small-world” network [37]. The Watts-Strogatz small-

ẋi = F (xi ) − ε gij xj , i = 1, 2, . . . , N, (1) world network is neither regular nor random. This model
j=1 is first formed by a regular network. Then, randomly, the
connection of node changes, and each link of the network
where xi and F (·) denote the state vector of the variables rewires with probability p. The small-world network is
and dynamical function of each oscillator, respectively. ε is characterized by two basic features, which are high clus-
the coupling coefficient and G = [gij ]N ×N is the zero-row tering coefficient and low average shortest path length [38].
sum Laplacian matrix which reveals the topology of the The clustering coefficient quantifies the mean degree of the
network. Then, different characteristics of the network, nodes, which tend to be clustered with each other. The
such as average path length, clustering coefficient, vari- real-world networks have low average shortest path length,
ance, and eigenvalues of the connection matrix of the net- which provides efficient communication through all possi-
works, are calculated. ble pairs of nodes.
The rest of this paper is structured as follows: In the Here, the small-world, regular, random, star, and scale-
next section, the different structures are described. In the free structures are considered. It should be noted that
third section, some topological properties of networks are the comparison between the different properties of vari-
investigated and compared in different network topologies. ous structures, the same number of nodes (N = 100) and
The obtained results are discussed in the fourth section. links are considered in all topologies. To this aim, for the
We conclude our results in the last section. star topology, the non-local structure with a central hub
(connected to all the other nodes) is considered.
Network structures. – One of the simple configu-
rations of dynamical networks is “regular network” [32]. In the next section, we discuss the effect of the num-
Every node in the regular networks shares the same de- ber of links on the average path length, clustering coeffi-
gree, which connects it to the same number of neighbors. cient and geodesic distance for different types of networks.
Regular networks with different degrees have been investi- Then, we will investigate the optimal number of links for
gated in the literature [33]. For instance, a fully connected synchronizability.
or global network is a regular network with the maximum
Dynamical properties of the network. –
number of links (i.e., N nodes sharing N (N2−1) number of
edges). There are two other types of regular networks, Average path length. The average path length is the
which are local and non-local. Each node in the local average of the shortest path between all pairs of network
network is connected to its two nearest neighbors on a nodes [39] which can be formulated as
ring. However, for non-local networks, each node is con- 
nected to more than one nearest neighbor nodes on a ring i=j dij
in both directions. Another well-known regular network is L = . (2)
N (N − 1)

40001-p2
 Role of number of links for synchronizability

Fig. 1: The average path length of different topologies of the Fig. 2: The clustering coefficient of different topologies of the
networks vs. the number of links. The number of the nodes networks vs. the number of links. The number of the nodes is
is set to N = 100 in all the topologies. Each topology is set to N = 100 in all the topologies. The random and scale-free
represented with a specific color. The blue and yellow lines networks (the red spectra) are almost increasing linearly. The
correspond to the non-local and star topologies, which show non-local (blue line) and small-world topologies with three dif-
the maximum and minimum average path length in compar- ferent rewiring probabilities (the green spectra) share the same
ison with the other topologies, respectively. The small-world trend, in which a rapid growth appears for a small number of
topologies with three different rewiring probabilities are repre- links. However, they increase slowly at rest. The maximum
sented in the green spectrum. The random and scale-free (with and minimum clustering coefficient in different topologies be-
m0 = 25) topologies are shown in the red spectrum. longs to star (yellow line) and random (dark red line) topolo-
gies, respectively.

A low average path length facilitates the information flow

among the nodes for optimum communication. Figure 1 if the clustering coefficient’s value is near zero, there can
represents the effect of different topologies on the average hardly be any connections in the neighborhood. However,
path length when the number of links increases. if the neighbors are fully connected, the value of the clus-
As is shown in fig. 1, the different topologies have the tering coefficient is 1. Figure 2 shows the degree of the
same general trend of the average path length while in- clustering of each topology according to the number of
creasing the number of the links. However, there are slight the links.
differences in the details. The average path length decline According to fig. 2, unlike the average path length, the
rate of these networks is different, in which the maximum behaviors of the clustering coefficient of different topolo-
and minimum average path length belongs to the non- gies are not the same. They can be divided schemati-
local and the star networks, respectively. Increasing the cally into two groups. The non-local, small-world, and
rewiring probability of the small-world topology leads to star networks are in the same group, and the scale-free
a decrease in average path length. The scale-free and ER and ER random networks are just like each other and be-
random networks share nearly the same trends in red spec- long to the same group. The clustering coefficient dia-
tra. They both have lower average path length than the grams of both groups are strictly increasing. However,
small-word and non-local networks. It is remarkable to the first group generally is more clustered than the second
note that after a critical value of the links (around 2500), group. The clustering coefficients of the first group show
the average path length of different topologies converges the steep slope just at the beginning of the diagram, and
to each other and finally goes to 1 when the graph be- they increase slowly. The star network has a maximum
comes fully connected. This means that the average path clustering coefficient. The non-local and the small-world
length is independent of the network topologies after the networks come after. Increasing the rewiring probability
critical link. of the small-world networks leads to the lower clustering
Clustering coefficient. The clustering coefficient in coefficient. Generally, the scale-free is more clustered than
the network is the mean of the local clustering of each the random network.
node [40] which can be defined as follows:
Variance of path length. One of the properties of the
N
1  ci complex network is the geodesic distance [41]. Geodesic
C= , (3) distance (dij ) is the shortest paths between two nodes i
N i=1 cc
and j in the network. In this subsection, we calculate
where ci and cc represent the number of triangles con- the variance of the geodesic distance and use it as a new
nected to node i and the number of triples centered around feature for complex networks which can be calculated as
node i, respectively. The value of the clustering coefficient follows:
 ¯2
varies in a range of [0, 1] which indicates the nodes’ ten- i=j (dij − d)
dency to form connections and be clustered. Therefore, V = . (4)
N (N − 1)

40001-p3
Motahare Hazrati et al.

connection matrix of the networks can be considered as

the most important factor in the synchronizability of
the network. Xi et al. [43] likened synchronizability in
small-world topology with regular and random topology
using the Master Stability Function (MSF) approach [15].
Using this MSF approach, one can investigate the neces-
sary condition of the stable synchronized manifold s(t) of
the dynamical networks [15]. To this end, the dynami-
cal states of the networks are considered in the synchro-
nization manifold with the presence of small perturbations
(xi = s + δxi ). If the perturbation decays in time, then all
the states of the networks are back to the synchronization
manifold which shows that it is stable; and if it grows to
infinity, the network cannot be synchronized. Therefore,
eq. (1) is rewritten in terms of δxi as a variational equation
in the form
N

δ ẋi = DF (s)δxi − ε gij δxj , (5)
j=1

where D denotes the Jacobian operator and s(t) is the

synchronized solution. If eq. (5) is projected to the space
Fig. 3: The variance of the distance in different topologies of of the eigenvectors of the coupling matrix G = [gij ]N ×N
the networks vs. the number of links. The number of the then it turns into the N eigenmodes form ζi with the cor-
nodes is set to N = 100 in all the topologies. (a) All topolo- responding eigenvalue λi of G as follows:
gies. (b) The enlarged picture of the star, random and scale-
free topologies to better demonstrate the differences between ζ̇i = DF (s)ζi − ελi ζi . (6)
the variance of the distances in these topologies. For a small
number of links, the non-local (blue line) and star (yellow line) The stability of the synchronization manifold can be in-
topologies show the maximum and minimum variance of the vestigated by the Lyapunov exponents diagram of eq. (6)
geodesic distance in comparison with the other topologies. according to smooth changes in coupling strength (ε). The
synchronizability of the dynamical networks can be di-
vided into two different kinds according to their varia-
Figure 3 represents the variance of the distance in different tion of the MSF. The networks with the MSF diagram of
topologies according to the number of links. type I show one zero-crossing point which means that the
network can be synchronized when the coupling strength
The diagrams of the variance of the distance in all
of the networks passes the critical value of the zero-cross
topologies, except for the star topology, share the same
point. However, the networks with the MSF diagram of
patterns and are decreasing by the increment of the links
type II share two limits for the coupling strength of the
number. In comparison, the star topology follows a
network to be synchronized. And the networks can be
parabola pattern (fig. 3(b)). Increasing the rewiring prob-
synchronized whenever the coupling strength remains be-
ability of the small-world topology decreases the variance
tween these two limits.
of the distance. As is shown in fig. 3(a), the non-local
To check the synchronizability of the different topolo-
network has the maximum variance.
gies mentioned in the previous section, the MSF approach
Synchronizability. Synchronization is an emerging is utilized. According to the MSF approach, the differ-
phenomenon in complex networks and it has been explored ence between the synchronizability of the different topolo-
using different coupling topologies. But the role of the gies can be followed by the eigenvalues of the connection
number of links on synchronizability in different networks matrix, especially λ2 and λN . In general, the MSF of a
is not yet well explored. Synchronization in networks usu- network can belong to type I or type II regardless of the
ally depends on three main factors: the dynamical prop- topology. Therefore, to comprehensively compare the syn-
erties of each oscillator in each node, coupling strength, chronizability of the different topologies, we have checked
and connection topological matrix of the network. Duan the synchronizability of both networks with MSF type I
et al. [42] investigated the synchronizability of the two and II in all topologies. In fact, first, we have assumed
simple networks with the same properties, such as the that all the topologies have the MSF of type I, and then we
same average distance, node betweenness centrality, and have checked the synchronizability of different topologies
degree distribution. The different synchronizability of with comparing λ2 in each topology. In the second step,
networks has been reported [28], which claims that the we have assumed that all the topologies have the MSF of

40001-p4
 Role of number of links for synchronizability

Fig. 4: Variation of the second eigenvalue λ2 of the network

connectivity matrix in different topologies by changing the link
numbers which shows the type I synchronizability. The number
of the nodes is set to N = 100 in all the topologies. The results
are shown by different colors for different topologies. The red
spectra show the random and scale-free topologies. The green
spectra belong to small-world topologies with three different
rewiring probabilities. The blue and yellow lines show the non-
local and star topologies, respectively.

type II, and then we have checked the synchronizability of Fig. 5: The value of λλN2 vs. the number of links showing the
different topologies with comparing λλN2 . For the network synchronizability of the different topologies of type II networks.
with MSF of type I, larger λ2 , and for the network with The number of the nodes is set to N = 100 in all the topologies.
MSF type II, lower λλN2 , are desired, respectively, for bet- (a) All topologies. (b) Enlargement of the larger link num-
ter network synchronizability. Figure 4 shows the compar- bers. The maximum and minimum values of λλN2 are observed
ison between the synchronizability of different topologies for star (yellow line) and random (dark red line) topologies,
of type I networks according to the number of links. respectively.
According to fig. 4, different topologies can be divided
into two groups. The star, non-local and small-world
topologies are in the same group. On the other hand, the has the lowest average path length and variance of the dis-
random and scale-free topologies are in another group. For tances, the largest clustering coefficient, and the best syn-
a small number of links, the scale-free topology and for a chronization of the type I networks. Therefore, it can be
large number of links, the random topology have higher λ2 concluded that for a small number of links, the type I net-
values. Decreasing the rewiring probability in the small- works with the best synchronization are accompanied by
world network leads to the larger value of λ2 . The random the topologies with low average path length and low vari-
and scale-free topologies show the larger λ2 , respectively.
ance of the distance. As was shown in fig. 4, the higher λ2
Figure 5(a) shows the synchronizability of the differ- belongs to the star, scale-free, random, small-world, and
ent topologies for the type II network. The maximum non-local topologies, respectively. However, increasing the
and minimum values of the λλN2 occur for random and star number of links alters this order. For instance, increasing
topologies, respectively, in the enlarged part of fig. 5(b).the number of links changes the star topology to the net-
Increasing the rewiring probability of the small-world work with the smallest λ2 and the worst type I network’s
topologies leads to the lower value of the λλN2 . synchronization. Furthermore, according to fig. 2, for a
large number of links, the smaller clustering coefficient
Discussion. – In the previous section, we have dis-
may lead to better synchronization of the networks with
cussed various network properties by considering five net-
the type I MSF diagram. The synchronizability of the
work topologies. For instance, by looking at the schemes
type II networks is more challenging than the type I. Ac-
much more precisely, it can be figured out that despite
cording to fig. 5(b), the lower λλN2 in the middle number of
the fact that both the small-word and non-local topology
links moves inversely with the clustering coefficient.
share the same number of neighbors and links, they have
different structural properties. Generally, increasing the Conclusion. – The number of links in a network plays
rewiring probabilities of small-word networks decrease the a vital role for the study of real-world networks. The
average path length, clustering coefficient, variance and network’s topology can be expressed by the connection
range of the λλN2 in comparison with non-local topology matrix, which represents the connection of the nodes
while the range of λ2 has been increased. It has been re- with each other through links. In this paper, the prop-
vealed that for a small number of links, the star topology erties of the complex networks with different topologies

40001-p5
Motahare Hazrati et al.

were obtained, and their synchronizability was compared. [8] Levanova T. A., Kazakov A. O., Korotkov A. G.
To this aim, different structures such as the regular, star, and Osipov G. V., Izvestiya VUZ. Appl. Nonlinear Dyn.,
small-world with different rewiring probability, scale-free, 26 (2018) 101.
and ER random, were considered with the same number [9] Boccaletti S., Latora V., Moreno Y., Chavez M.
of nodes and by varying the number of links. The struc- and Hwang D.-U., Phys. Rep., 424 (2006) 175.
[10] Graham W., The art, soul, science of beholding nature-
tural properties of the network, such as the average and
patterns in nature, in preparation (2013).
variance of the path length, clustering coefficient, and the [11] Andreev A., Frolov N., Pisarchik A. and Hramov
eigenvalues of the connection matrix, were calculated for A., Phys. Rev. E, 100 (2019) 022224.
each topology. According to the MSF method, the suffi- [12] Bolotov M., Levanova T., Smirnov L. and Pikovsky
cient conditions for synchronization of the oscillators can A., Cybernet. Phys., 8 (2019) 215.
be attained according to the eigenvalues (λ1 , λ2 , . . . , λN ) [13] Pisarchik A. N., Maksimenko V. A., Andreev A. V.,
of the connection matrix. In fact, the higher λ2 and lower Frolov N. S., Makarov V. V., Zhuravlev M. O.,
λN Runnova A. E. and Hramov A. E., Sci. Rep., 9
λ2 represent better synchronization for type I and type II
networks, respectively. The results showed that the star (2019) 1.
topology has the minimum average path length, maximum [14] Korotkov A. G., Kazakov A. O., Levanova T. A.
clustering coefficient, a completely different pattern of the and Osipov G. V., IFAC-PapersOnLine, 51 (2018)
241.
variance of the distance, the best synchronizability for a
[15] Pecora L. M. and Carroll T. L., Phys. Rev. Lett., 80
large number of links in type I networks, and the worst (1998) 2109.
synchronizability in networks with the MSF of type II. [16] Belykh V. N., Belykh I. V. and Hasler M., Phys. D:
The non-local topology has the maximum average path Nonlinear Phenom., 195 (2004) 159.
length. Increasing the rewiring probability of the small- [17] Albert R. and Barabási A.-L., Rev. Mod. Phys., 74
world topology decreases the average path length, cluster- (2002) 47.
ing coefficient, the variance of the distance, the value of [18] Rakshit S., Bera B. K. and Ghosh D., Phys. Rev. E,
the λ2 and λλN2 . Furthermore, it has been revealed that 101 (2020) 012308.
the best synchronizability of type I networks for a small [19] Boccaletti S., Pisarchik A. N., Del Genio C. I.
number of links belongs to the networks with the low- and Amann A., Synchronization: From Coupled Sys-
est average and variance of path length. But for a large tems to Complex Networks (Cambridge University Press)
2018.
number of the links, the topology with a lower clustering
[20] Maksimenko V. A., Makarov V. V., Bera B. K.,
coefficient has better synchronizability. This study may
Ghosh D., Dana S. K., Goremyko M. V., Frolov
be helpful for the understanding of an epidemic’s transi- N. S., Koronovskii A. A. and Hramov A. E., Phys.
tion pattern in spreading dynamics where the number of Rev. E, 94 (2016) 052205.
links is crucial. [21] Bera B. K., Rakshit S., Ghosh D. and Kurths J.,
Chaos, 29 (2019) 053115.
∗∗∗ [22] Majhi S., Bera B. K., Ghosh D. and Perc M., Phys.
Life Rev., 28 (2019) 100.
[23] Sun X., Lei J., Perc M., Kurths J. and Chen G.,
This work is partially funded by Center for Nonlinear Chaos, 21 (2011) 016110.
Systems, Chennai Institute of Technology, India vide fund- [24] Boccaletti S., Hwang D.-U., Chavez M., Amann A.,
ing No. CIT/CNS/2020/RD/065. Kurths J. and Pecora L. M., Phys. Rev. E, 74 (2006)
016102.
[25] Wang X. F. and Chen G., IEEE Circuits Syst. Mag., 3
REFERENCES (2003) 6.
[26] Rakshit S., Bera B. K., Bollt E. M. and Ghosh D.,
[1] Newman M. E., SIAM Rev., 45 (2003) 167. SIAM J. Appl. Dyn. Syst., 19 (2020) 918.
[2] Korotkov A. G., Kazakov A. O., Levanova T. A. [27] Arenas A., Dı́az-Guilera A., Kurths J., Moreno Y.
and Osipov G. V., Commun. Nonlinear Sci. Numer. and Zhou C., Phys. Rep., 469 (2008) 93.
Simul., 71 (2019) 38. [28] Belykh I., Hasler M., Lauret M. and Nijmeijer H.,
[3] Lü J., Chen G., Ogorzalek M. J. and Trajković Int. J. Bifurcat. Chaos, 15 (2005) 3423.
L., Theory and applications of complex networks: Ad- [29] Baptista M. S., De Carvalho J. X. and Hussein
vances and challenges, in Proceedings of 2013 IEEE In- M. S., PLoS ONE, 3 (2008) e3479.
ternational Symposium on Circuits and Systems (ISCAS) [30] Belykh I., Belykh V. and Hasler M., Chaos, 16 (2006)
(IEEE) 2013, pp. 2291–2294. 015102.
[4] Von Bertalanffy L., Acad. Manag. J., 15 (1972) 407. [31] Belykh I., Carter D. and Jeter R., SIAM J. Appl.
[5] Kenett D. Y., Perc M. and Boccaletti S., Chaos, Dyn. Syst., 18 (2019) 2267.
Solitons Fractals, 80 (2015) 1. [32] Chen W.-K., Graph Theory and Its Engineering Applica-
[6] Chowdhury S. N., Majhi S., Ozer M., Ghosh D. and tions, Vol. 5 (World Scientific Publishing Company) 1997.
Perc M., New J. Phys., 21 (2019) 073048. [33] Schöll E., Eur. Phys. J. ST, 225 (2016) 891.
[7] Boccaletti S., Kurths J., Osipov G., Valladares [34] Durrett R., Random Graph Dynamics, Vol. 200 (Cite-
D. and Zhou C., Phys. Rep., 366 (2002) 1. seer) 2007.

40001-p6
 Role of number of links for synchronizability

[35] Erdős P. and Rényi A., Publ. Math. Inst. Hung. Acad. [40] Fronczak A., Holyst J. A., Jedynak M. and
Sci., 5 (1960) 17. Sienkiewicz J., Phys. A: Stat. Mech. Appl., 316 (2002)
[36] Barabási A.-L. and Albert R., Science, 286 (1999) 688.
509. [41] Bouttier J., Di Francesco P. and Guitter E., Nucl.
[37] Watts D. J. and Strogatz S. H., Nature, 393 (1998) Phys. B, 663 (2003) 535.
440.
[38] Newman M. E., Barabási A.-L. E. and Watts D. J., [42] Duan Z., Chen G. and Huang L., Phys. Rev. E, 76
The Structure and Dynamics of Networks (Princeton (2007) 056103.
University Press) 2006. [43] Xi X., Panahi S., Pham V.-T., Wang Z., Jafari
[39] Lovejoy W. S. and Loch C. H., Soc. Netw., 25 (2003) S. and Hussain I., Appl. Math. Comput., 379 (2020)
333. 125226.

40001-p7