A partial synchronization problem in an oscillator network is considered. The concept on a principal quasi-submatrix corresponding to the topology of a cluster is proposed for the first time to study partial synchronization. It is shown that partial synchronization can be realized under the condition depending on the principal quasi-submatrix, but not distinctly depending on the intercluster couplings. Obviously, the dimension of any principal quasi-submatrix is usually far less than the one of the network topology matrix. Therefore, our criterion provides us a novel index of partial synchronizability, which reduces the network size when the network is composed of a great mount of nodes. Numerical simulations are carried out to confirm the validity of the method.
1. Introduction
Since the pioneering work of Pecora and Carroll [1], extensive researches on chaos synchronization have been carried out due to their potential applications in various disciplines such as physics, engineering, and biology [2–5]. For example, wireless sensor networks are widely researched due to the important applications in the real scenarios. One of the biggest challenges in investigating wireless sensor networks is how to obtain higher synchronization accuracy with minimal overhead [6]. And many other network-induced phenomena have also been discussed extensively in engineering [7].
Up to date, there have been many synchronization types being proposed and discussed, including complete synchronization [8, 9], partial synchronization [10], and inner and outer synchronization [11]. Generally speaking, there exists an intimate relationship between the phenomena of synchronization and invariant manifolds of coupled systems [9–12]. Thus, when we carry out researches on partial synchronization of coupled systems, it is always supposed that the corresponding full or partial synchronization manifolds are invariant manifolds. The established tools for the stability of invariant synchronization manifolds mainly consist of local linearization method and Lyapunov function method. One of the prominent results of the former is the master stability function method [9]. The method has obtained two significant factors determining the local stability of the synchronous state, that is, the maximum Lyapunov exponent of the node dynamics and the eigenvalues of the topology matrix. One of the prominent results of the latter is the study of the global attractiveness of the synchronization manifold. A crucial requirement for the method is the condition of the node dynamics satisfyingf∈QUAD(Δ,P,Ω)[10]. In some sense, the condition means that the system can synchronize when the coupling is made sufficiently large. Recently, some new conditions have been obtained for synchronization of networks without Lipschitz condition or QUAD condition [12].
The type of synchronization concerned in this paper is partial synchronization. It means that the coupled oscillators split into subgroups called clusters, and all the oscillators in the same cluster behave in the same fashion. Many relative studies have been carried out [13–18]. By using pinning control strategy, partial synchronization of coupled stochastic delayed neural networks was discussed in [14]. Similar control strategy is also proposed to select controlled communities by analyzing the information of each community such as in-degrees and out-degrees [15]. Afterwards, some simple intermittent pinning controls and centralized adaptive intermittent controls are proposed [16]. However, a suitable control law must be presented in order to use pinning control strategy. Some other researches focused on partial synchronization induced by the coupling configuration. An arbitrarily selected partial synchronization manifold was constructed for a network with cooperative and competitive couplings [17]. Recently, a sufficient and necessary condition for partial synchronization manifolds being invariant manifolds was obtained in networks coupled linearly and symmetrically [10]. More significantly, some sufficient conditions for the global attractiveness of the partial synchronization manifold were derived by decomposing the whole space into a direct sum of the synchronization manifold and the transverse space. The results are meaningful and interesting. Based on the method in [10], partial synchronization bifurcations [19] were analyzed for a globally coupled network with a parameter, which topology is not complex. Nevertheless, all the eigenvalues of the topology matrix are essential for that method, which needs a large quantity of computation when the network size is very large.
In this paper, a novel criterion on partial synchronization is proposed through the analysis of principal quasi-submatrices corresponding to the clusters. Previous researches have obtained several criteria on partial synchronization [10]. However, these criteria depend heavily on the topology matrix of the whole network. For many complex networks in real world, it is tedious to obtain the eigenvalues and eigenvectors of the topology matrix of the whole network. Therefore, this paper aims to propose a novel criterion, which is not distinctly dependent on the intercluster couplings and the topology matrix of the whole network. Instead, it is sufficient for partial synchronization to verify that the conditions on the intracluster couplings are satisfied. Therefore, it will be advantageous to discuss partial synchronization in the complex network with a large number of nodes because the dimension of any principal quasi-submatrix is usually far less than the one of the network topology matrix. At first, the tedious work of solving the eigenvectors corresponding to the second-largest eigenvalue is avoided. Secondly, partial synchronization is studied by the inner topologies of the individual clusters. Obviously, the network size reduction provides convenience for the study of partial synchronization in networks with great mounts of oscillators. In order to confirm its effectiveness, some numerical simulations are carried out to study partial synchronization in a star-global network. The numerical simulations are in good agreement with the theoretical analysis.
The present paper is built up as follows. Some concerned concepts and a lemma for the invariance of the partial synchronization manifold are introduced in Section 2. Principal quasi-submatrices corresponding to the clusters are proposed, and our main results are then established in Section 3. Numerical examples are presented to confirm the effectiveness of the results in Section 4. Finally, a brief discussion of the obtained results is given in Section 5.
2. Preliminaries
In this section, we introduce some basic concepts of invariant synchronization manifolds and a related lemma, which are required throughout the paper.
In past years, many studies [9, 10, 12] of synchronization phenomena focused on oscillator networks coupled linearly and symmetrically. The system can be described by the following ordinary differential equations:
(1)x˙i=f(xi,t)+ɛ∑j=1maijΓxj,i=1,…,m,
wherexi=(xi1,…,xin)⊤is then-dimensional state variable of theith oscillator,m>1is the network size,t∈[0,+∞) is a continuous time,f:Rn×[0,+∞)→Rnis a continuous map,ɛ>0is the coupling strength, andΓ=diag{γ1,…,γn}is a nonnegative matrix determining the interaction of variables. The coupling weight matrixA=(aij)m×mis assumed to satisfy thataij=aji≥0, fori≠j, and∑j=1maij=sfori=1,…,m.
In order to study partial synchronization of the system (1), the set of nodes{1,…,m}is divided intodnonempty subsets (clusters). LetG={G1,…,Gd}be the partition, and denoteK=(K1,K2,…,Kd), whereKk≥1is the cardinal number of the clusterGk,k=1,…,d. Without loss of generality, suppose thatG1={1,…,K1},…,Gd={∑p=1d-1Kp+1,…,m}.Based on the partitionG, we rewrite the coupling matrixAas a block matrix,
(2)A=[A11A12⋯A1dA21A22⋯A2d⋯⋯⋯⋯Ad1Ad2⋯Add],
where Akl∈RKk×Kl,k,l=1,…,d.
We will discuss sufficient conditions for theKknodes in the clusterGkto synchronize with each other,k=1,…,d. Before that, the concepts of partial synchronization manifolds and transverse spaces introduced in the following [10]:
(3)MK={∑(x1⊤,…,xm⊤)⊤∣(x∑p=0k-1Kp+1⊤,…,x∑p=0kKp⊤)⊤∈MKk,k=1,…,d∑},
where K0=0,
(4)MKk={∑(x∑p=0k-1Kp+1⊤,…,x∑p=0kKp⊤)⊤∈RnKk∣x∑p=0k-1Kp+1=⋯=x∑p=0kKp∑},
are called a partial synchronization manifold. We also call
(5)LK={∑(x1⊤,…,xm⊤)⊤∣(x∑p=0k-1Kp+1⊤,…,x∑p=0kKp⊤)⊤∈LKk,k=1,…,d∑},
where
(6)LKk={∑i=1Kk(x∑p=0k-1Kp+1⊤,…,x∑p=0kKp⊤)⊤∈RnKk∣∑i=1Kkx∑p=0k-1Kp+i=0},
a transverse space forMK.
In casen=1, the four sets mentioned above are denoted by MK, MKk, LK, and LKk, respectively. In cased=1, the synchronization manifoldMKis called a full synchronization manifold. For simplicity, we denote the full synchronization manifold MK by M, and the corresponding transverse spaceLK by L. Sometimes, the manifoldMKis also denoted byMK(G)to emphasize its partitionG.
Definition 1.
The synchronization manifoldMKis said to be globally attractive for the system (1), or the system (1) is said to realize partial synchronization with the partitionG={G1,G2,…,Gd}if, for any initial condition(x1⊤(0),x2⊤(0),…,xm⊤(0))⊤∈Rmn,
(7)limt→+∞∑k=1d∑i∈Gk∥xi(t)-x∑p=0k-1Kp+1(t)∥=0,
where∥·∥denotes 2-norm of vectors. In case ofd=1, the system (1) is said to realize full synchronization, if the full synchronization manifold M is globally attractive.
Synchronization manifolds are always supposed to be invariant in order to discuss the attractiveness. Therefore, it is necessary to recall the definition of an invariant manifold [20] for the ordinary differential equations,
(8)x˙=X(x,t),x∈RN,X:RN×[0,+∞)⟶RN.
Denote M as the manifold of codimensionpdefined by a vector equationH(x)=0, H=(h1,h2,…,hp), 1≤p≤N-1. M is called an invariant manifold of the system (8) if
(9)(gradH·X)|H=0≡0,
which implies that the vector field (8) is tangent to M. For more details of the existence of invariant manifolds, one can refer to the papers by Golubitsky and coworkers [21, 22]. The following lemma gives a sufficient and necessary condition for a partial synchronization manifold being an invariant manifold.
Lemma 2 (see [10]).
LetK=(K1,K2,…,Kd). The synchronization manifold MK is an invariant manifold of the system (1) if and only if the coupling matrixAhas the form (2) with everyAijbeing equal row sum.
Remark 3.
Based on Lemma 2, we can find all invariant synchronization manifolds for a given coupling matrix. As we know, each diagonal block reveals the intracluster information communication, and each nondiagonal block represents the information communication among different clusters. By the condition that everyAijis equal row sum, we mean that every node in the same cluster receive an equal amount of information communication from every other cluster.
3. Main Results
Noticing the sufficient and necessary condition in Lemma 2, we assume that every submatrixAkl∈RKk×Klis an equal row sum matrix with row sumskl,k,l=1,…,d. SinceAsis an equal row sum matrix, that is, ∑j=1maij=∑k=1dsi^k=s, where the denotationi^representsk, for alli∈Gk, we define the matrices A¯kk=(a¯ij)Kk×Kk as
(10)A¯kk=Akk+(s-skk)EKk=Akk+∑l=1,l≠kdsklEKk,
whereEKk∈RKk×Kkis the identity matrix,k=1,…,d. It is easy to conclude from (10) that ∑j∈Gi^a¯ij=s and
(11)a¯ij={aij,i≠j;aij+∑l=1,l≠i^dsi^l,i=j.
Noticing thatAkkis a principal submatrix ofA, we call the matrixA¯kka principal quasi-submatrix corresponding to the clusterGk,k=1,…,d.
Now, we are now in a position to carry out the following theorem with the help of Lyapunov function method.
Theorem 4.
LetK=(K1,K2,…,Kd),P=diag(p1,…,pn)be a positive-definite diagonal matrix, and letΔ=diag(δ1,…,δn)be a diagonal matrix. Supposeδj≤0ifj∉J, whereJ={j:1≤j≤n,γj≠0}. Then under the following three conditions.
Every submatrixAklin the block matrix (2) has equal row sumskl, and every principal submatrixAkkis irreducible.
There exists a constantϵ>0such that, for anyu,v∈Rn and all t≥0,
(12)(u-v)⊤P{[f(u,t)-f(v,t)]-Δ(u-v)}≤-ϵ(u-v)⊤(u-v).
For allj=1,…,n, k=1,…,d, the matricesɛγjA¯kk+δjEKkare negative semidefinite in the transverse spaceLKk, that is,
(13)z⊤(ɛγjA¯kk+δjEKk)z≤0,z∈LKk,
or, in particular,
(14)ɛγjλk(2)+δj≤0,
where λk(2) is the second-largest eigenvalue ofA¯kk.
The synchronization manifoldMKis globally attractive for the system (1).
For convenience of the proof, the following notations are introduced.
(15)x¯k(t)=1Kk∑i∈Gkxi(t),x¯(t)=[x¯1^⊤(t),…,x¯m^⊤(t)]⊤,δxi(t)=xi(t)-x¯i^(t),δx(t)=[δx1⊤(t),…,δxm⊤(t)]⊤,δx~ks(t)=[δx∑p=0k-1Kp+1s(t),…,δx∑p=0kKps(t)]⊤,δx~k(t)=[δx~k1⊤(t),…,x~kn⊤(t)]⊤,
wherek=1,…,d,i=1,…,m,s=1,…,n.
Then any vectorx=(x1⊤,…,xm⊤)⊤∈Rmncan be decomposed into
(16)x=x¯⊕δx,x¯∈MK,δx∈LK.
Therefore, the attractiveness of the invariant synchronization manifoldMKis equivalent toδx→0whent→+∞; that is, the dynamical flow in the transverse subspaceLKconverges to zero.
Proof.
Noticing the condition (i), one gets
(17)∑j=1maijΓxj(t)=∑l=1d∑j∈GlaijΓ[δxj(t)+x¯l(t)]=∑l=1maijΓδxj(t)+∑l=1dsi^lΓx¯l(t).
Therefore,
(18)dδxi(t)dt=dxi(t)dt-1Ki^∑p∈Gi^dxp(t)dt=f(xi(t),t)-f(x¯i^(t),t)+ɛ∑j=1maijΓδxj(t)+Ji^,
where
(19)Ji^=f(x¯i^(t),t)-1Ki^∑p∈Gi^[f(xp(t),t)+ɛ∑q=1mapqΓxq(t)]+ɛ∑l=1dsi^lΓx¯l(t).
In order to utilize the QUAD(Δ,P,Rn) condition, a Lyapunov function is defined as follows:
(20)V(δx(t))=12∑j=1mδxi⊤(t)Pδxi(t).
One can conclude from∑i∈Gkδxi(t)=0that
(21)∑i=1mδxi⊤(t)PJi^=∑k=1d[∑i∈Gkδxi⊤(t)]PJi^=0,
and then,
(22)dV(δxi(t))dt=∑i=1mδxi⊤(t)P[f(xi(t),t)-f(x¯i^(t),t)+ɛ∑j=1maijΓδxj(t)]≤-ϵ∑i=1mδxi⊤(t)δxi(t)+∑i=1mδxi⊤(t)P[ɛ∑j=1maijΓδxj(t)+Δδxi(t)].
Denote the second term in the right hand of (22) asSfor convenience; then,
(23)S=∑k=1d∑i∈Gkδxi⊤(t)P[ɛ∑l=1d∑j∈GlaijΓδxj(t)+Δδxi(t)]=∑k=1d∑i∈Gkδxi⊤(t)P[ɛ∑j∈GkaijΓδxj(t)+Δδxi(t)+ɛ∑l=1,l≠kd∑j∈GlaijΓδxj(t)].
Sincei,j∈Gkholds for the first term in the right hand above, replacingaijwitha¯ijaccording to (11) gives rise to that
(24)S=∑k=1d∑i∈Gkδxi⊤(t)P[ɛ∑j∈Gka¯ijΓδxj(t)+Δδxi(t)]+ɛ∑k=1d∑i∈Gkδxi⊤(t)P∑l=1,l≠i^d[-si^lΓδxi(t)+∑j∈GlaijΓδxj(t)]=S1+S2.
As a result of the condition (13), one obtains that
(25)S1=∑k=1d∑i∈Gkδxi⊤(t)P[ɛ∑j∈Gka¯ijΓδxj(t)+Δδxi(t)]=∑k=1d∑s=1npsδx~ks⊤(t)(ɛγsA¯kk+δsEKk)δx~ks(t)≤0.
Sincesi^l=∑j∈Glaij, the sumS2can be decomposed into
(26)S2=ɛ∑k=1d∑i∈Gkδxi⊤(t)P∑l=1,l≠kd∑j∈GlaijΓ(δxj(t)-δxi(t))=ɛ∑k=1d-1∑l=k+1d∑i∈Gk∑j∈Glaijδxi⊤(t)PΓ(δxj(t)-δxi(t))+ɛ∑l=1d-1∑k=l+1d∑i∈Gk∑j∈Glaijδxi⊤(t)PΓ(δxj(t)-δxi(t)).
Renaming in the second termkbyl,ibyj, and vice versa [23] and utilizing the symmetry of aij, or Alk⊤=Akl, one gets
(27)S2=ɛ∑k=1d-1∑l=k+1d∑i∈Gk∑j∈Glaijδxi⊤(t)PΓ(δxj(t)-δxi(t))+ɛ∑k=1d-1∑l=k+1d∑j∈Gl∑i∈Gkajiδxj⊤(t)PΓ(δxi(t)-δxj(t))=-ɛ∑k=1d-1∑l=k+1d∑i∈Gk∑j∈Glaij(δxj(t)-δxi(t))⊤PΓ×(δxj(t)-δxi(t))≤0.
Therefore, one obtains thatS≤0and
(28)dV(δxi(t))dt≤-ϵ∑i=1mδxi⊤(t)δxi(t)≤-2ϵV(δxi(t))maxipi,
which implies that the partial synchronization manifoldMKis globally attractive for the system (1).
The remainder of the proof is to show that condition (14) is also sufficient forS1≤0.
It is well known that a symmetric matrixA¯kkhas the decompositionA¯kk=UkΛkUk⊤, whereΛk=diag{λk(1),…,λk(Kk)}is a real diagonal matrix andUk∈RKk×Kkis an unitary matrix,thatis,UkUk⊤=EKk. The diagonal elements ofΛkare the eigenvalues ofA¯kksatisfyings=λk(1)>λk(2)≥⋯≥λk(Kk). Theith column ofUkis the eigenvector ofA¯kkcorresponding to the eigenvalueλk(i),i=1,…,Kk. By changes of variablesδx~ks(t)=Uηks(t), the quadratic form (25) can be diagonalized as follows:
(29)S1=∑k=1d∑s=1npsδx~ks⊤(t)(ɛγsA¯kk+δsEKk)δx~ks(t)=∑k=1d∑s=1npsηks⊤(t)(ɛγsΛk+δsEKk)ηks(t).
Noticing that the first column ofUkis[1,…,1]⊤, one can conclude from∑i∈Gkδxi(t)=0andηks(t)=U⊤δx~ks(t)thatηk1(t)=0. Therefore, condition (14) is sufficient forS1≤0.
The proof is completed.
Compared with the previous results [10], the conditions in Theorem 4 are not dependent on the intercluster couplings, which are eliminated in the proof through a set of mathematical skills in inequalities (25) and (27). The obtained results greatly reduce the network sizes theoretically. However, another question arises naturally: how to implement the proposed condition in reality? The following remarks might answer these questions.
Remark 5.
In case that the network size is not very large and the coupling matrix is given, it is easy to verify all the conditions in Theorem 4. In case that the network consists of great mounts of oscillators, it should still be full of challenges to implement though our result reduces the network size greatly. And it might be hard to implement.
In order to make clear the implications of Theorem 4, several remarks on the conditions are given as follows.
Remark 6.
(1)Many chaotic oscillators have been proved to satisfy condition (ii), such as Chua circuits [24] and standard Hopfield neural networks [25]. However, many other systems are not the case such as a lattice ofx-coupled Rössler systems, in which the stability of synchronization regime is lost with the increasing of coupling [26].
(2)Providing that the trajectories of the uncoupled systemsxi=f(xi,t),i=1,…,mare eventually dissipative, that is, the trajectories will be in the absorbing domain B eventually, it has been proved that each trajectory of the coupled system (1) is also eventually dissipative and will be in the absorbing domainB×⋯×B︸m[17] eventually. Therefore, condition (ii) holds when the timetis large enough. For example, the uncoupled Lorenz system
(30)u˙=σ(v-u),σ=10,v˙=ru-v-uw,r=28,w˙=-bw+uv,b=83,
is eventually dissipative, where
(31)B={(u)2+(v)2+(w-σ-r)2<b2(σ+r)24(b-1)=B}.
It has been proved thatx-coupled [27] or y-coupled Lorenz systems [28] satisfy condition (ii) when the timetis large enough.
(3)
If there exists a k0∈{1,2,…,d} such that Kk0=1, which implies that the subset Gk0 contains only one element, which does not synchronize with any other node. Without loss of generality, suppose thatKk=1fork>d¯. In this case, the corresponding principal quasi-submatrixA¯kk=s and the constantscan be regarded as the single eigenvalue ofA¯kksinceA¯kkv→=sv→for anyv→∈R. Therefore, the second-largest eigenvalueλk(2)does not exist. But notice the definition of partial synchronization in Section 2, synchronization in the clusterGkalways occurs, and conditions (13) and (14) should be regarded to hold for anyɛ>0,k>d¯.
(4)SinceAkkis irreducible and diffusive, the largest eigenvalue ofAkkis zero, which is simple. Therefore, the largest eigenvalue ofA¯kkiss, and it is simple also. In cases=0, it can be seen thatλk(2)<0and condition (14) is equivalent to
(32)ɛ≥maxj∈J{0,δj/γj}min1≤k≤d¯|λk(2)|.
Condition (14) provides us a novel index of partial synchronizability.
4. Numerical Examples
Consider the system (1) with2my-coupled Lorenz systems
(33)x˙i=f(xi,t)+ɛ∑j=12maij(θ)Γxj,i=1,…,2m,
wheref(xi,t),xi=(ui,vi,wi)⊤defined by the system (30),Γ=diag(0,1,0). LetP=diag(B/2(b-ϵ)σ+ϵ/σ,1,1), andΔ=diag(0,2b-ϵ-1+[B-2(b-ϵ)(σ-ϵ)]2/2B(b-ϵ),0), and thefin (27) has been proved to satisfy condition (ii) when the timetis large enough. Through translating time, the network system (27) satisfies condition (ii) in Theorem 4.
4.1. A Star-Global Network
Design the following topology matrix of the system (33) as
(34)A(θ)=[A11(θ)θEmθEmA22(θ)]2m×2m,
whereθ≥0is the coupling weight parameter of couplings between the two clusters, and the identity matrixEmimplies that theith (1≤i≤m) oscillator in the first cluster is coupled with the(m+i)th oscillator in the second cluster. As a special case, the submatricesA11(θ)andA22(θ)are taken as follows:
(35)A11(θ)=[1-m-θ1⋯111-m-θ⋯1⋯⋯11⋯1-m-θ],A22(θ)=[1-m-θ11⋯11-1-θ0⋯0⋯⋯100⋯-1-θ].
Obviously,A11(θ)andA22(θ)are the topology matrices of a globally coupled network and a star-coupled network, respectively. Therefore, we call the network a star-global (coupled) network.
4.2. Numerical Simulations
As an example, the topology structure of a star-global network with m=4 is shown in Figure 1. Due to the specific topological structure of the network, the following partitions satisfy condition (i),
(36)G1={1,2,3,4,5,6,7,8},G2={1,2,3,4;5,6,7,8},G3={1,5;2,3,4,6,7,8}.
Topology structure of a star-global network composed of8oscillators. The solid lines represent the inner couplings of each clusters; the dashed lines represent the outer couplings between different clusters.
Remark 7.
According to the criterion in the previous researches [10], we should firstly find the eigenvalues and the corresponding eigenvectors of the8×8matrix A(θ). As we know, it should be of a vast amount of calculations. However, based on Theorem 4, it is enough for partial synchronization with the partition G2(G3)if we can find the eigenvalues of two 4×4matrices (four2×2matrices). Obviously, Theorem 4 reduces the calculations greatly.
The principal quasi-submatrices corresponding to the clusters in the partitionsG1orG2, areA(θ)orA11(θ)|θ=0andA22(θ)|θ=0, respectively. And denote the ones in the partitionG3asA¯11(θ)andA¯22(θ). Further analysis gives rise to the eigenvalues sets of the quasi-submatrices,
(37)σ(A(θ))={0,-4,-2θ,-4-2θ,λ±,λ±},σ(A11(θ)|θ=0)={0,-4,-4,-4},σ(A22(θ)|θ=0)={0,-1,-1,-4},σ(A¯11(θ))={0,-2θ},σ(A¯22(θ))={0,-2θ,λ′±,λ′±},
whereλ±=-(2θ+5±9+4θ2)/2,λ′±=-(2θ+3±9+4θ2)/2. Therefore, partial synchronization with the partitionG2occurs ifɛ≥δ; partial synchronization with the partitionG3occurs ifɛ≥δ/|λ′-(θ)|; and full synchronization with the partitionG1occurs ifɛ≥max{δ/2θ,δ/|λ-(θ)|}. These are seen much more clearly in Figure 2.
Threshold of the coupling strengthɛand the coupling weightθfor partial synchronization with different partitions.
By fixingθ∈(0,0.5]and increasingɛgradually, partial synchronization with the partitionG2will firstly occur; then, the one withG3and full synchronization occurs at the same time. Figure 2 also implies that the threshold for partial synchronization with the partitionG3is sufficient for full synchronization.
In order to validate the effectiveness of Figure 2 numerically, the following average cluster errors of the system (33) are defined to measure partial synchronization with partitionsG1,G2, andG3, respectively,
(38)e1=18∑i=18∥xi(t0)-x1(t0)∥,e2=18∑i=14∥xi(t0)-x1(t0)∥+18∑i=58∥xi(t0)-x5(t0)∥,e3=18∥x5(t0)-x1(t0)∥+18∑i=2,i≠58∥xi(t0)-x2(t0)∥.
Choose the initial conditions(ui(0),vi(0), and wi(0))randomly on[-1,1]×[-1,1]×[-1,1], and pickt0=500. Fix the coupling weight θ at 0.2, and5; the dependence of the cluster errors onɛis shown, respectively, in Figure 3. As can be seen, there is a good agreement between Figures 3 and 2.
Dependence of the average cluster errors on the coupling strength ɛ for the system (33) with the topology matrix (34). The errorse1(t0),e2(t0),ande3(t0)measure the attractiveness of the manifoldsMK(G1), MK(G2), and MK(G3), respectively.
θ=0.2
θ=5
5. Conclusions
In summary, this paper introduced a novel index of partial synchronizability of a network. It is shown that partial synchronization can be ensured by the conditions merely on the quasi-submatrices corresponding to the clusters. If a network is composed of a great mount of nodes, the enormous amount of calculation can be reduced by replacing the coupling matrix with several quasi-submatrices. Numerically, different types of partial synchronization occur in a star-global network when the coupling strength is increased, the order of which is forecasted accurately by our result. It should be a meaningful and effective method to study partial synchronization with different partitions.
In the past decades, the networked control systems have attracted much attention due to their applications covering a wide range of industries. And the network-induced phenomena under consideration in engineering have been discussed widely, including missing measurements [29], fading measurements [30], and probabilistic sensor delays [31]. Therefore, the obtained approach for partial synchronization in this paper might be applicable to the complex networks with networked induced phenomena. The related studies should be one of the future research topics.
Conflict of Interests
The authors declare that they have no conflict of interests regarding the publication of this paper.
Acknowledgments
The authors gratefully acknowledge the support of the Natural Science Foundation of Zhejiang Province (no. LQ12A01003), the Natural Science Foundation of Hebei Province in China (no. A2014205152), the Natural Science Foundation of Guangxi Province (no. 2013GXNSFAA019006), and the National Natural Science Foundation of China (nos. 11162004 and 11171084).
PecoraL. M.CarrollT. L.Synchronization in chaotic systems199064882182410.1103/PhysRevLett.64.821MR1038263ZBL0938.370192-s2.0-0343689904ArenasA.GuileraA. D.KurthsJ.MorenoY.ZhouC.Synchronization in complex networks200846939315310.1016/j.physrep.2008.09.002MR24770972-s2.0-56549084695YuW.ChenG.LüJ.On pinning synchronization of complex dynamical networks200945242943510.1016/j.automatica.2008.07.016MR25273392-s2.0-58349088745ZhangJ. B.MaZ. J.ZhangG.Cluster synchronization induced by one-node clusters in networks with asymmetric negative couplings20132347043128ZhangJ.MaZ.CaoJ.Full synchronization studied by a set of partitions connected together201320138352826MR311180610.1155/2013/352826QianX.ShenX.DaiG.ZhangJ.LvC.Clapping and broadcasting synchronization in wireless sensor networkProceedings of the 6th International Conference on Mobile Ad-hoc and Sensor Networks (MSN '10)December 2010Hangzhou, China14014510.1109/MSN.2010.282-s2.0-79952541272HuJ.WangZ. D.DongH. L.GaoH. J.Recent advances on recursive filtering and sliding mode design for networked nonlinear stochastic systems: a survey2013201312646059MR314844410.1155/2013/646059ZhangJ.LiuZ.XuJ.Synchronization in oscillator networks with coupling balance200939255656610.1016/j.chaos.2007.01.0842-s2.0-62549133190PecoraL. M.CarrollT. L.Master stability functions for synchronized coupled systems19988010210921122-s2.0-4243241129WuW.ChenT.Partial synchronization in linearly and symmetrically coupled ordinary differential systems2009238435536410.1016/j.physd.2008.10.012MR25896242-s2.0-58849165655SunW.WangR.WangW.CaoJ.Analyzing inner and outer synchronization between two coupled discrete-time networks with time delays20104322523110.1007/s11571-010-9118-92-s2.0-77956060087LiuB.LuW.ChenT.New conditions on synchronization of networks of linearly coupled dynamical systems with non-Lipschitz right-hand sides20122551310.1016/j.neunet.2011.07.007ZBL1258.930072-s2.0-82355169673ZhangJ. B.MaZ. J.ChenG. R.Robustness of cluster synchronous patterns in small-world networks with inter-cluster co-competition balance20142428023111LiL. L.CaoJ. D.Cluster synchronization in an array of coupled stochastic delayed neural networks via pinning control201174584685610.1016/j.neucom.2010.11.0062-s2.0-78650751468HuC.JiangH.Cluster synchronization for directed community networks via pinning partial schemes201245111368137710.1016/j.chaos.2012.06.015MR29902512-s2.0-84866507396LiuX.ChenT.Cluster synchronization in directed networks via intermittent pinning control20112271009102010.1109/TNN.2011.21392242-s2.0-79960118796MaZ.LiuZ.ZhangG.A new method to realize cluster synchronization in connected chaotic networks2006162902310310.1063/1.2184948MR22438052-s2.0-33748745031ZhangJ.MaZ.CaoJ.Hub-induced synchronization in scale-free networks with cluster structure20142014714945310.1155/2014/149453MR3193487MaZ.ZhangS.JiangG.LiK.Effect of the coupling matrix with a weight parameter on synchronization pattern in a globally coupled network2013741-2556410.1007/s11071-013-0948-yMR31051752-s2.0-84877776100BelykhV. N.BelykhI. V.HaslerM.Hierarchy and stability of partially synchronous oscillations of diffusively coupled dynamical systems2000625, part A6332634510.1103/PhysRevE.62.6332MR17964422-s2.0-0034317183GolubitskyM.StewartI.TörökA.Patterns of synchrony in coupled cell networks with multiple arrows2005417810010.1137/040612634MR21365192-s2.0-11144319940GolubitskyM.StewartI.Nonlinear dynamics of networks: the groupoid formalism200643330536410.1090/S0273-0979-06-01108-6MR22230102-s2.0-33744777430BelykhV. N.BelykhI. V.HaslerM.Connection graph stability method for synchronized coupled chaotic systems20041951-215918710.1016/j.physd.2004.03.012MR20746412-s2.0-3042858242WuC. W.ChuaL. O.Synchronization in an array of linearly coupled dynamical systems199542843044710.1109/81.404047MR13518792-s2.0-0029358909LuW. L.ChenT. P.New approach to synchronization analysis of linearly coupled ordinary differential systems2006213221423010.1016/j.physd.2005.11.009MR22012002-s2.0-30144443190HeagyJ. F.PecoraL. M.CarrollT. L.Short wavelength bifurcations and size instabilities in coupled oscillator systems199574214185418810.1103/PhysRevLett.74.41852-s2.0-11944273756BelykhI.BelykhV.NevidinK.HaslerM.Persistent clusters in lattices of coupled nonidentical chaotic systems200313116517810.1063/1.1514202MR19649702-s2.0-0242668717LiuX. W.ChenT. P.Boundedness and synchronization of y-coupled Lorenz systems with or without controllers2008237563063910.1016/j.physd.2007.10.006MR24548852-s2.0-41449092277HuJ.WangZ.ShenB.GaoH.Quantised recursive filtering for a class of nonlinear systems with multiplicative noises and missing measurements201386465066310.1080/00207179.2012.756149MR3046931ZBL1278.932692-s2.0-84876142734HuJ.WangZ. D.GaoH. J.Recursive filtering with random parameter matrices, multiple fading measurements and correlated noises201349113440344810.1016/j.automatica.2013.08.021MR3115817HuJ.WangZ.ShenB.GaoH.Gain-constrained recursive filtering with stochastic nonlinearities and probabilistic sensor delays20136151230123810.1109/TSP.2012.2232660MR30288472-s2.0-84873894174