assumption, i.e. all information is transmitted continu-
ously among agents. However, information may not be
transmitted continuously due to the unreliability of
communication channels, the limitations of sensing
ability of agents and the constraints of total cost.
Therefore, it is more practical to consider the case of
intermittent information transmission for agents with
continuous-time dynamics. Consensus problems of
continuous-time second-order agents were studied in
Hayakawa, Matsuzawa, and Hara (2006) and Ren and
Cao (2008), where each agent obtains the measurements
of relative states only at discrete times. Moreover, it was
assumed that the discrete times of all agents are the
same, namely, all agents obtain information in a
synchronous manner. On the other hand, it is difficult
to guarantee that all agents obtain information syn-
chronously because of technology limitations and
environment disturbances, and thus, the asynchronous
case also deserves to be studied. Asynchronous consen-
sus problems of agents with discrete-time dynamics
have been investigated extensively (see e.g. Cao et al.
2008; Fang and Antsaklis 2008). However, there has
been a little literature on asynchronous consensus
problems of agents with continuous-time dynamics,
such as Xiao and Wang (2008), where each agent,
described by first-order dynamics, obtains its neigh-
bour’s states only at discrete times and the discrete times
of each agent are independent of the others’. By
applying nonnegative matrix theory, the asynchronous
consensus problem with switching topology and time-
varying delays was addressed in Xiao and Wang (2008).
To the authors’ best knowledge, there are few research
results on asynchronous consensus problems of second-
order agents. Based on the above considerations, we
investigate asynchronous consensus problems of con-
tinuous-time second-order agents with fixed topology
and time-varying delays. Note that the main analysis
tool in Xiao and Wang (2008), nonnegative matrix
theory, is not applicable to the asynchronous case of
continuous-time second-order agents.
This article is organised as follows. Section 2
presents some concepts in graph theory and formulates
the model to be studied. Convergence analysis is
provided in Section 3. Simulation examples are
presented in Section 4. Finally, conclusion remarks
are stated in Section 5.
Notations: Let I
n
2R
nn
be an identity matrix and
1
n
¼[1 1]
T
2R
n
; 0 is an all-zero vector or matrix with
compatible dimension; lim
t!t*
(t) ¼* means lim
t!t*
jj(t) *jj¼ 0, where (t), * 2R
n
; for any complex
number s, Re(s), Im(s) and jsj denote its real part,
imaginary part and modulus, respectively; for any
symmetric matrix A, A 5 0 (resp., A 4 0) means that
A is a negative definite (resp., positive definite) matrix;
for any square matrix H, H(i, j ) denotes the element in
the i-th row and j-th column of H and (H ) stands for
the set of all eigenvalues of H; * denotes a term which
is induced by symmetry and diag{} represents a
block-diagonal matrix.
2. Preliminaries
2.1 Graph theory
Graph plays a key role in modelling the interaction
topology among agents. We first introduce some basic
definitions in graph theory (Godsil and Royal 2001).
A directed graph G consists of a vertex set V(G) and
an edge set E(G), where V(G) ¼{v
1
, ...,v
n
} and
E(G) {(v
j
,v
i
):v
j
,v
i
2V(G)}. For edge (v
j
,v
i
), v
j
is
called the parent vertex of v
i
and v
i
is called the child
vertex of v
j
. If two ends of an edge are the same vertex,
then such an edge is called loop. The set of neighbours
of vertex v
i
is defined by N(G,v
i
) ¼{v
j
:(v
j
,v
i
) 2E(G)
and j 6¼i}, and the associated index set is denoted by
N(G, i) ¼{ j :v
j
2N(G,v
i
)}. A (directed) path from v
i
1
to
v
i
k
is a sequence, v
i
1
, ...,v
i
k
, of distinct vertices such
that (v
i
j
,v
i
jþ1
) 2E(G) for any j ¼1, ..., k 1. A directed
graph G is strongly connected if there is a path from
every vertex to every other vertex. A directed tree is a
directed graph, where every vertex except one special
vertex has exactly one parent vertex, and the special
vertex, called root vertex, has no parent vertices and
can be connected to any other vertices via paths.
A subgraph G
s
of G is a graph such that V(G
s
) V(G)
and E(G
s
) E(G). G
s
is said to be a spanning subgraph if
V(G
s
) ¼V(G). For any v
i
,v
j
2V(G
s
), if (v
i
,v
j
) 2E(G
s
) ,
(v
i
,v
j
) 2E(G), then G
s
is said to be an induced subgraph
of G, and G
s
is also said to be induced by V(G
s
).
A spanning tree of G is a directed tree which is a
spanning subgraph of G. G is said to have a spanning
tree if some edges form a spanning tree of G.
A matrix is called nonnegative if each of its
elements is nonnegative. A weighted directed graph
G(A) is a directed graph G plus a nonnegative matrix
A ¼[a
ij
] 2R
nn
, where a
ij
4 0 ,(v
j
,v
i
) 2E(G), and a
ij
is
called the weight of edge (v
j
,v
i
). It is assumed that
a
ii
¼0, i ¼1, ..., n. The Laplacian matrix
L ¼[l
ij
] 2R
nn
of G(A) is defined as
l
ij
¼
a
ij
, i 6¼ j
P
n
s¼1,s6¼i
a
is
, i ¼ j
8
>
<
>
:
,
and it has some properties as follows.
Lemma 2.1 (Ren and Beard 2005):
(i) Zero is an eigenvalue of L, and 1
n
is the
associated right eigenvector.
International Journal of Control 553
Downloaded by [Academy of Mathematics and System Sciences] at 18:30 17 February 2012