Cluster synchronization of community network with distributed time delays via impulsive control
Leng Hui, Wu Zhao-Yan†,
College of Mathematics and Information Science, Jiangxi Normal University, Nanchang 330022, China

 

† Corresponding author. E-mail: zhywu@jxnu.edu.cn

Project supported by the National Natural Science Foundation of China (Grant No. 61463022), the Natural Science Foundation of Jiangxi Province, China (Grant No. 20161BAB201021), and the Natural Science Foundation of Jiangxi Educational Committee, China (Grant No. GJJ14273).

Abstract
Abstract

Cluster synchronization is an important dynamical behavior in community networks and deserves further investigations. A community network with distributed time delays is investigated in this paper. For achieving cluster synchronization, an impulsive control scheme is introduced to design proper controllers and an adaptive strategy is adopted to make the impulsive controllers unified for different networks. Through taking advantage of the linear matrix inequality technique and constructing Lyapunov functions, some synchronization criteria with respect to the impulsive gains, instants, and system parameters without adaptive strategy are obtained and generalized to the adaptive case. Finally, numerical examples are presented to demonstrate the effectiveness of the theoretical results.

1. Introduction

In the real world, community structure has been discovered in many large-scale physical and social systems such as school friendship networks,[1] biological networks,[2] congressional cosponsorship networks,[3] protein–protein interaction networks,[4] World Wide Web,[5] and so on. They are usually modeled by a community network, in which the nodes in the same community have the identical function or property but in different communities have nonidentical ones. For example, in protein–protein interaction networks, communities correspond to functional groups/clusters, i.e., to proteins having the same or similar functions, which are expected to be involved in the same processes.[6] A great number of phenomena show that the current state of a node is affected by not only the current states but also the states in the previous period of its neighbor nodes. For example, in biological neural networks, it usually has a spatial extent due to the presence of parallel pathways with a variety of axon sizes and lengths.[7,8] Thus, there is a distribution of transmission delays and it cannot be modeled with discrete time delays. In such a case, a more appropriate way is to incorporate the distributed time delays in the models.[7,8] Until now, many valuable results about dynamical networks with distributed delays have been obtained.[911] However, community networks with distributed delays have seldom been considered, thus they deserve further investigations.

In recent years, synchronization has gained extensive attention from various fields on account of its broad practical applications, such as regulation of power grid, parallel image processing, the operation of no-man air vehicle, the realization of chain detonation, etc.[12,13] Cluster synchronization,[1422] a special synchronization of complex dynamical networks, denotes that synchronization of the nodes in the same community can be achieved, but synchronization of the nodes in different communities cannot be achieved. That is, the nodes in the same or different communities have identical or nonidentical goals. Further, many studies show that cluster synchronization has increasing applications in different fields.[23,24] Kaneko pointed out that cluster synchronization is an important phenomenon in communication engineering and biological science.[23] Yoshioka showed that neurons are divided into several clusters in neurons networks interconnected with chemical synapses and synchronization of the neurons is achieved in each cluster. From the viewpoint of real life, cluster synchronization by inner adjustment of communities and influence of different communities without external control cannot be achieved. An effective method of achieving synchronization, i.e., adding proper controllers, has been found from lots of obtained results. Thus, how to design proper controllers is a key issue.

Impulsive control is a representative control strategy and its main characteristic is working on the nodes only at inconsecutive instants.[21,22,2531] Due to the characteristic, researchers have been studying impulsive control deeply for a long time. Cai et al. considered the exponential cluster synchronization of hybrid-coupled delayed dynamical networks via impulsive control.[21] Fan et al. considered the cluster synchronisation in non-linearly coupled networks with non-identical nodes and time-varying delays via impulsive control.[22] As we know, the impulsive gains and intervals related to the cost and efficiency are important parameters of impulsive controllers. Therefore, designing proper impulsive controllers means choosing appropriate impulsive gains and intervals in terms of synchronization criteria. However, for different community networks with nonidentical system parameters, the needed impulsive gains and intervals need to be chosen repetitively, which undoubtedly increases workloads and reduces efficiency. In order to overcome the shortcoming, we introduce an adaptive strategy to design unified impulsive controllers. A community network combined with adaptive impulsive controllers can adjust impulsive gains or intervals adaptively and achieve cluster synchronization.

Motivated by the above discussion, the cluster synchronization of the community network with distributed time delays is investigated. On one hand, proper impulsive controllers are added to the community networks for achieving cluster synchronization, and simple criteria about the system parameters, the impulsive gains, and intervals are obtained on the basis of the Lyapunov function method and linear matrix inequality technique. On the other hand, proper controllers are improved by introducing an adaptive strategy and criteria are obtained as well. Noticeably, the adaptive law of the impulsive gain and the method for estimating the impulsive instants are provided in terms of synchronization criteria. In Section 2, the network model is introduced and some preliminaries are given. In Section 3, the impulsive controllers for achieving synchronization are designed and the synchronization criteria are provided. In Section 4, several numerical simulations are performed to verify the results. In Section 5, the conclusion for this paper is given.

2. Model and preliminaries

Consider a community network consisting of k (2 ≤ kN) communities and N nodes with distributed time delays. As we know, in many real networks with community structure, such as neural, social, biological, and communication networks, the individual nodes in the same community can be viewed as the identical functional units, while any pair of nodes in different communities are essentially different according to their functions.[21] Therefore, we assume that the node dynamics and the distributed delays in the same community are identical, while those in different communities are nonidentical. Further, the distributed delays between different pairs of communities are nonidentical as well. According to these settings, the community network can be described as

where yi(t) = (yi1(t),yi2(t),…,yin(t))TRn is the state variable of node i, and fπi : RnRn is a vector function in the πi-th community. In different communities, the node dynamics are supposed to be nonidentical, i.e., if πiπj, then fπifπj. is the inner coupling matrix, Gm (m = 1,2,…,k) are sets of all nodes belonging to the m-th community, if node jGm, then πj = m. B = (bij) ∈ RN × N is the zero-row-sum outer coupling matrix, which denotes the network topology and coupling strength and is defined as follows: if node j has influence on node i (ij), then bij ≠ 0; otherwise, bij = 0. ξπiπj > 0 are time delays. ϕπiπj:[0,τπiπj] → [0,+∞] is the continuous density function and satisfies .

Let ηπi(t) be a solution of an isolated node in the πi-th community, i.e., . The objective here is to achieve synchronization of the network (1) by designing proper controllers, i.e.,

where ‖·‖ denotes the Euclidean norm.

For achieving the synchronization, proper impulsive controllers are designed and applied onto the network (1). The controlled network is

where i = 1,2, …,N, υ = 1,2,…, the impulsive time instants tυ satisfy 0 = t0 < t1 < t2 < ··· < tυ < ···, and tυ → ∞ as υ → ∞. . Any solution of Eq. (2) is assumed to be left continuous at each tυ, i.e., . φ(tυ) is the impulsive gain at t = tυ and φ(t) = 0 for ttυ.

Let ei(t) = yi(t) − ηπi(t) be the synchronization errors. Then the error systems are

Let Gm = {γm−1 + 1,…,γm} (m = 1,2,…,k) with γ0 = 0 and γk = N. suppose that BRN × N can be divided into the following block form:

where BpmR(γpγp−1)×(γmγm−1), p,m = 1,2,…,k.

Definition 1 If Bpp is a zero-row-sum, symmetric, and irreducible matrix with non-negative off-diagonal elements, we say that Bpp belongs to class 𝐶1 and denote Bpp ∈ 𝐶1, p = 1,2,…,k.

Definition 2 If Bpm is a zero-row-sum matrix, we say that Bpm belongs to class 𝐶2 and denote Bpm ∈ 𝐶2, p,m = 1,2,…,k.

Definition 3 If Bpp ∈ 𝐶1, Bpm ∈ 𝐶2, p,m = 1,2,…,k (pm), we say that B belongs to class 𝐶3 and denote B ∈ 𝐶3.

Assumption 1 Suppose that there exists a positive constant M such that

holds for any x(t),y(t) ∈ Rn, p = 1,2,…,k, and t > 0.

Lemma 1 (see Ref. [7]) Let A be a symmetric n × n real matrix and A > 0. Then

where ξ > 0, ϕ(θ) ≥ 0, and .

3. Main results

Let , be the impulsive intervals,, and be the minimum and the maximum of , λ be the largest eigenvalue of , μ(t) = (1 + φ(t))2. From the definition of φ(t), one has μ(t) = 1 for ttυ. The outer coupling matrix B considered in this paper is supposed to belong to 𝐶3.

Theorem 1 Suppose that Assumption 1 holds. If there exists a constant δ > 0 such that

hold, then the cluster synchronization of network (2) is achieved.

Proof Consider the following Lyapunov function candidate:

for t ∈ (tυ−1,tυ], υ = 1,2,…

When t ∈ (tυ−1,tυ),

and the derivative of V(t) is

which gives

When t = tυ,

When υ = 1, from inequalities (5) and (6),

When υ = 2,

By mathematical induction, for any positive integer υ,

If condition (4) holds,

which implies .

Then, for t ∈ (tυ,tυ+1],

which implies V(t) → 0 for t → ∞, i.e., the cluster synchronization is achieved. This completes the proof.

Remark 1 In particular, if the set is empty set, the Lyapunov function is

if the set only has an element d, i.e., ,

Remark 2 From condition (4), for any given community network, the needed impulsive gains and intervals for achieving cluster synchronization can be estimated by some calculations. Similar to many existing results about the impulsive control, when the impulsive gains are fixed, the upper bound of the impulsive intervals should be smaller than a certain positive constant. On the other hand, for different community networks, the needed values are different. That is, condition (4) is conservative. For solving this problem, an adaptive strategy is adopted to design adaptive impulsive controllers.

Theorem 2 Suppose that Assumption 1 holds. If there exists a constant δ > 0 such that

hold, where and ω > 0 is an adaptive gain, then the synchronization of network (2) is achieved.

Proof Consider the following Lyapunov function candidate:

for t ∈ (tυ−1,tυ],υ = 1,2,…

When t ∈ (tυ−1,tυ),

and the derivative of V(t) is

which gives

When t = tυ,

Thus, similar to the proof of Theorem 1, the proof is obtained.

Remark 3 Generally, for any given ρυ and δ, the conditions (7) hold when choosing

where α > 0 is an arbitrary small constant. In practical applications, we can choose with small positive constants α and δ, and as a small value. The definition of implies that is a monotone increasing function. When the cluster synchronization is achieved, converges to a positive constant and the impulsive gain φ(tυ) converges to a needed value.

Remark 4 In practical application, when φ(tυ) and δ are fixed, the control intervals should be chosen to be as large as possible to reduce the cost. Thus, the impulsive instants tυ can be chosen by finding the maximum value of tυ subject to with υ = 1,2,….

Remark 5 From conditions (7), the constant λ in condition (4) with respect to the system parameters does not need to be calculated beforehand and can be estimated by when the cluster synchronization is achieved. Remarks 3 and 4 show that the impulsive gains or instants can be estimated adaptively through choosing proper parameters. Specially, from Remark 4, the upper bound of the impulsive intervals does not need to be calculated beforehand. That is, the conditions (7) are not conservative to some extent.

4. Numerical illustrations

In this section, we use three numerical examples to demonstrate the effectiveness of the theoretical results. Choose the density function as ϕpm(θ) = 1/ξpm, the inner coupling matrix as the identity matrix, and ξ11 = 0.2, ξ12 = 0.1, ξ13 = 0.2, ξ21 = 0.4, ξ22 = 0.4, ξ23 = 0.1, ξ31 = 0.2, ξ32 = 0.4, ξ33 = 0.1 in the following examples. In addition, choose the initial values of yi(t) and ηπi(t) randomly.

Example 1 Consider a community network consisting of 20 nodes in 3 communities, whose topology is shown in Fig. 1. Choose the node dynamics of the first community as the Lorenz system[32]

the second community as the Chen system[33]

and the third community as the Lü system[34]

Fig. 1. A community network consisting of 20 nodes in 3 communities. The top, bottom-left, and bottom-right are the first, the second, and the third communities, respectively. The outer coupling matrix B is defined as: if nodes i and j (ji) are connected by a solid line, then bij = 0.001; if nodes i and j (ji) are connected by a dashed line, then bij = −0.001.

According to the discussion in Ref. [35], we can choose M = 53 such that Assumption 1 holds. Choose the impulsive gains φ(tυ) = −0.999 and intervals ρυ = 0.12. Then the maximum eigenvalue of is λ = 109.0001, and ln μ(tυ) + δ + λρυ = −0.7345 < 0 with δ = 0.001, i.e., the condition 4 holds. Figure 2 shows the orbits of synchronization errors in 3 communities.

Fig. 2. The orbits of synchronization errors in 3 communities.

Example 2 Consider the same community network in Example 1 via adaptive impulsive control.

First, choose ρυ = 0.12, δ = 0.001, ω = 0.0003, and . According to Remark 3, choose with α = 0.001. Figure 3 shows the orbits of synchronization errors in 3 communities. Figure 4 shows the impulsive gain φ(tυ).

Fig. 3. The orbits of synchronization errors in 3 communities.
Fig. 4. The impulsive gain φ(tυ).

Second, choose the impulsive gain φ(tυ) = −0.999, ω = 0.0003, , δ = 0.001. According to Remark 4, we can estimate the impulsive instants. Figure 5 shows the orbits of synchronization errors in 3 communities. Figure 6 shows the impulsive interval ρυ.

Fig. 5. The orbits of synchronization errors in 3 communities.
Fig. 6. The impulsive interval ρυ.

In Example 1, the impulsive interval is ρυ = 0.12. Figure 6 shows that the network achieves synchronization when the impulsive interval converges to 0.848. Obviously, the impulsive interval with the adaptive impulsive strategy is larger than that without the strategy.

Example 3 We generate the first community network topology as the WS small-world network[36] with N1 = 50, K = 5, and p = 0.2, the second as the WS small-world network[36] with N2 = 50, K = 6, and p = 0.1, and the third as the NW small-world network[37] with N3 = 50, K = 2, and p = 0.1. Further, we add a few connections among the three communities. Choose the same node dynamics as Example 1.

First, choose ρυ = 0.2, δ = 0.001, ω = 0.00005, and . According to Remark 3, choose with α = 0.001. Figure 7 shows the orbits of synchronization errors in 3 communities. Figure 8 shows the impulsive gain φ(tυ).

Fig. 7. The orbits of synchronization errors in 3 communities.
Fig. 8. The impulsive gain φ(tυ).

Second, choose the impulsive gain φ(tυ) = −0.9, ω = 0.00005, , δ = 0.001. According to Remark 4, we can estimate the impulsive instants. Figure 9 shows the orbits of synchronization errors in 3 communities. Figure 10 shows the impulsive interval ρυ.

Fig. 9. The orbits of synchronization errors in 3 communities.
Fig. 10. The impulsive interval ρυ.

Examples 2 and 3 indicate that the cluster synchronization of networks with nonidentical node dynamics and topological structures can be achieved without calculating the constants with respect to the system parameters in advance. Thus, the adaptive impulsive controllers are unified for different community networks.

5. Conclusion

Community networks with distributed time delays and nonidentical node dynamics are investigated. Further, the coupling delays in different communities are also assumed to be nonidentical. Through designing proper impulsive controllers, the nodes in the different communities are synchronized onto different orbits and the nodes in the same community are synchronized with each other. That is, the cluster synchronization of the community network is achieved. With the linear matrix inequality technique and Lyapunov function method, some synchronization criteria with respect to the impulsive gains, instants, and system parameters are obtained. Noticeably, an adaptive strategy is adopted to make the impulsive controllers possess universality and the corresponding synchronization criteria are obtained as well. When the impulsive intervals are fixed, the impulsive gain can adaptively approach to the proper value according to the updating law. When the impulsive gains are fixed, the impulsive instants can be estimated by solving a sequence of maximum value problems. Finally, the effectiveness of the theoretical results are demonstrated by three numerical examples.

Reference
1Gonzalez M CHerrmann H JKertesz JVicsek T 2007 Physica 379 307
2Girvan MNewman M E J 2002 Proc. Nat. Acad. Sci. USA 99 7821
3Zhang YFriend A JTraud A LPorter M AFowler J HMucha P J 2008 Physica 387 1705
4Jonsson P FCavanna TZicha DBates P A 2006 BMC Bioinf. 7 2
5Flake G WLawrence SGiles C LCoetzee F M 2002 IEEE Comput. 35 66
6Fortunato S 2010 Phys. Rep. 486 75
7Li C HYang S Y 2008 Int. J. Bifur. Chaos 18 2039
8Gopalsamy KHe X Z 1994 Physcia 76 344
9Kyrychko Y NBlyuss K BSchoell E 2014 Chaos 24 043117
10Punetha NRamaswamy RAtay F M 2015 Phys. Rev. 91 042906
11Hu J BLu G PZhao L D 2016 Nonlinear Dyn. 83 1101
12Andrew L Y TLi X FChu Y DZhang H 2015 Chin. Phys. 24 100502
13Chen GWang XLi X2012Introduction to Complex Networks: Models, Structure and DynamicsBeijingHigh Education Press
14Yao H XWang S G 2012 Chin. Phys. 21 110506
15Gan L Y NWu Z YGong X L 2015 Chin. Phys. 24 040503
16Cai G LJiang S QCai S MTian L X 2014 Chin. Phys. 23 120505
17Hou H ZZhang Q LZheng M 2012 Nonlinear Dyn. 83 739
18Jalan SSingh AAcharyya SKurths J 2015 Phys. Rev. 91 022901
19Ji PPeron T K D MRodrigues F AKurths J 2014 Phys. Rev. 90 062810
20Juang JLiang Y H 2014 Chaos 24 013110
21Cai S MLi X JJia QLiu Z R 2016 Nonlinear Dyn. 85 2405
22Fan H GZhao YFeng J W 2016 IET Control Theory Appl. 10 762
23Kaneko K 1994 Physica 75 55
24Yoshioka M 2005 Phys. Rev. 71 061914
25Zhao HLi L XPeng H PXiao J HYang Y XZheng M W 2016 Nonlinear Dyn. 83 1437
26Liu D FWu Z YYe Q L 2014 Chin. Phys. 23 040504
27Sun WLu J HChen S HYu X H 2014 Chaos 24 013141
28Yang X SCao J DLu J Q 2012 IEEE Trans. Circuits Syst. I 59 371
29Wu Z Y 2014 Commun. Theor. Phys. 61 590
30Cai S MZhou P PLiu Z R 2014 Nonlinear Dyn. 76 1677
31Cai S MZhou P PLiu Z R 2015 J. Franklin Inst. 352 2065
32Lorenz E N 1963 J. Atmos. Sci. 20 130
33Chen G RUeta T 1999 Int. J. Bifur. Chaos 9 1465
34 J HChen G R 2002 Int. J. Bifur. Chaos 12 659
35Yu WChen GLu J 2009 Automatica 45 429
36Watts D JStrogatz S H 1998 Nature 393 440
37Newman M E JWatts D J 1999 Phys. Lett. 263 341