2.2. Problem description
Consider the following homogenous singular complex network
E
_
x
i
ðtÞ¼Ax
i
ðtÞþBu
i
ðtÞ;
y
i
ðtÞ¼Cx
i
ðtÞ;
ð1Þ
where i ¼ 1; 2; ...; N; A 2 R
dd
, B 2 R
dp
; C 2 R
qd
, E 2 R
dd
with rankðEÞ¼r 6 d; x
i
ðtÞ; u
i
ðtÞ and y
i
ðtÞ are the state, control input
and measured output of agent i, respectively. It is assumed that
(A1): The pair E; AðÞis regular and impulse-free.
Note that u
i
ðtÞði ¼ 1; 2; ...; NÞ can only use information from neighboring agents of agent i. An undirected graph
G ¼ VðGÞ; EðGÞðÞcan be used to illustrate the neighboring relationship among agents, i.e., the communication topology,
where VðGÞ¼ 1; 2; ...; N
fg
is composed of all agents, j; iðÞ2EðGÞ means that agent j is one of the neighbors of agent i, and
w
ij
denotes the communication strength of the edge j; iðÞ.
The following distributed synchronization protocol is applied:
E
_
t
i
ðtÞ¼ðA þ BK
1
Þ
t
i
ðtÞK
2
C
X
j2N
i
ðtÞ
w
ij
t
j
ðtÞ
t
i
ðtÞ
þ K
2
X
j2N
i
ðtÞ
w
ij
y
j
ðtÞy
i
ðtÞ
;
u
i
ðtÞ¼K
1
t
i
ðtÞ;
8
<
:
ð2Þ
where
t
i
ðtÞ is the protocol state, K
1
and K
2
are gain matrices with compatible dimensions and N
i
ðtÞ represents the time-
varying neighbor set of agent ii2 1; 2; ...; NfgðÞ. The current paper focuses on switching communication topology cases; that
is, the neighbor set of agent ii2 1; 2; ...; N
fg
ðÞmay be time-varying, the communication strength between agents i and j is w
ij
when agent i can receive the information of agent j and the communication strength between agents i and j is 0 when there is
no communication between them.
Let the finite topology set ‘ with an index set w N contain all possible communication topologies for switching, where N
denotes the natural number set and
r
ðtÞ : 0; 1Þ½!w denotes the switching signal, whose value at time t is the index of the
communication topology at time t. It is clear that if the switching signal
r
ðtÞ is identically equal to a constant, then the
neighbor set is time-invariant, which means that the communication topology is fixed. Here, G
r
ðtÞ
denotes the switching
communication topology. G
r
ðtÞ
is said to be connected if each topology in ‘ is connected, and is said to be jointly connected
if some topologies in ‘ may not be connected but the union of communication topologies across a series of time intervals is
connected. Moreover, it is supposed that.
(A2): The switching sequences 0 < t
1
< < t
k
< satisfy inf
k
ðt
kþ1
t
k
Þ¼T
d
> 0.
Let
g
i
ðtÞ¼ x
T
i
ðtÞ;
t
T
i
ðtÞ
T
ði ¼ 1; 2; ...; NÞ, then it can be obtained by (1) and (2) that
E 0
0 E
_
g
i
ðtÞ¼
ABK
1
0 A þ BK
1
g
i
ðtÞ
00
K
2
CK
2
C
X
j2N
i
ðtÞ
w
ij
g
j
ðtÞ
g
i
ðtÞ
:
ð3Þ
Let
g
ðtÞ¼
g
T
1
ðtÞ;
g
T
2
ðtÞ; ...;
g
T
N
ðtÞ
T
and L
r
ðtÞ
denote the Laplacian matrix of the communication topology G
r
ðtÞ
, then from (3),
the dynamics of network (1) with protocol (2) can be written in a compact form as
I
N
E 0
0 E
_
g
ðtÞ¼ I
N
ABK
1
0 A þ BK
1
L
r
ðtÞ
00
K
2
C K
2
C
g
ðtÞ: ð4Þ
Now, the definitions of the stable-protocol admissible synchronization and synchronizability are given respectively as
follows.
Definition 4. Network (4) is said to achieve stable-protocol admissible synchronization if for any admissible bounded initial
state
g
ð0Þ, it is regular and impulse-free and there exists a vector-valued function sðtÞ dependent on
g
ð0Þ such that
lim
t!1
ðx
i
ðtÞsðtÞÞ ¼ 0 and lim
t!1
t
i
ðtÞ¼0 ði ¼ 1; 2; ...; NÞ, where sðtÞ is called a synchronization function.
Definition 5. Network (1) is said to be stable-protocol admissibly synchronizable by protocol (2) if there exist K
1
and K
2
such that it achieves stable-protocol admissible synchronization.
Remark 1. When outputs instead of states are available to neighboring agents, two types of synchronization protocols were
constructed in the literature; that is, static output feedback ones and dynamic output feedback ones. In [32], static output
feedback synchronization protocols were applied and quadratic matrix inequality (QMI) criteria for synchronization were
proposed, where an iterative algorithm was given to determine the gain matrices of synchronization protocols. In [33],
dynamic output feedback synchronization protocols were proposed and LMI criteria for synchronization were presented,
where changing variable methods which may bring in some conservative were used. It should be pointed out that both
QMI and LMI approaches are not analytic; that is, the existence of the solutions of synchronization criteria cannot be guar-
4 X. Yang et al. / Information Sciences 307 (2015) 1–17