- 1 -

http://www.sj-ce.org
Scientific Journal of Control Engineering
February 2014, Volume 4, Issue 1, PP.1-7
Dichotomy and Absolute Stability Analysis of a
Class of Complex Dynamical Networks
Haijing Liu, Dun Ao
#
, Lili Xia, Di Liang
College of Electronic Information and Control Engineering, Beijing University of Technology, Beijing 100124, China
#
Email: aodun@bjut.edu.cn
Abstract
This paper deals with dichotomy and absolute stability problem for a class of complex network systems with each node be a
general Lure system which has sector constraint functions. The interconnections of complex network systems referred in this
paper conclude not only the linear interconnection of states, but also the interconnection of the sector nonlinearities. Besides,
bilinear matrix inequality (BMI) conditions of dichotomy and absolute stability of complex network systems are derived.
Moreover, controller design methods of dichotomy and absolute stability of complex system are given in this paper and the
related controller can be constructed via feasible solutions of a certain set of bilinear matrix inequalities (BMIs). Finally,
examples are given to illustrate the effectiveness of the proposed methods.
Keywords: Complex Network System; Lure System; Property of Dichotomy; Absolute Stability; BMI (Bilinear Matrix
I nequality)
1 INTRODUCTION
Complex dynamic networks have been extensively studied in the recent years by physicists, biologists, social
scientists and control scientists [1]. For dynamic network systems, stability criterion of large-scale system is given
by Moylan and Hill [2], and the formation problem of airplane using a distributed control method is derived by
Wolfe [3]. With the concept of the S-hull, a multi-variable stability criterion similar is given by Lestas and
Vinnicombe [4]. The control problems for a spatially invariant system are studied in [5]. The systems considered
above are linear systems, while the nonlinearity cant be avoided in practical systems.
Lure system is one of the classic nonlinear systems, which can be regarded as a feedback connection of linear
system and nonlinear element satisfying some sector conditions [6], which can portray a lot of nonlinear dynamical
systems in practical engineering. Nowadays, the study of global properties and robust stability of nonlinear network
systems have attracted a lot of attentions of researchers[7-10]. In [7], the stability of distributed heterogeneous
systems with static nonlinear interconnections is considered. Considering the interconnection matrix of the network
systems, the synchronization conditions

[8] and stability conditions [9]

are given. The consensus region of multi-
agent systems with nonlinear dynamics is investigated in [10]. Reference [11] presents the formation controller for
the second-order multiagent systems with time-varying delay and nonlinear dynamics. But the interconnection
considered above is only one kind of interconnections.
In this paper, we consider the dichotomy and absolute stability of a class of nonlinear complex dynamical networks
with two kinds of interconnections. The complex network system is composed of subsystems, which are a kind of
nonlinear system, i.e., Lure systems. Besides, the interconnections of the complex network systems referred in this
paper conclude not only the linear interconnection of states, but also the interconnection of the sector nonlinearities.
2 DESCRIPTION OF THE PROBLEM
Suppose that every node in a complex network is a Lure systems described by

#
This work was supported by National Science Foundation of China under Grants 61004012.

- 2 -
http://www.sj-ce.org
i i
i i
i i i
x Ax Bu
y Cx
u f (y )
= +
=
=
1,..., i N = (1)
where
1 1 1 1
, , , ( ,..., ) , ( ,..., ) , ( ) ( ( ),..., ( ))
n n n m m n T T T
i i in i i im i i i i im im
A R B R C R x x x y y y f y f y f y

e e e = = = , and the
nonlinear function
i
f satisfying
2 2
( ) ,
i i i i i
f t t t o t o s s s .
The corresponding transfer function is defined as
1
( ) ( ) K s C sI A B

= .
Consider a class of dynamic complex network with each node being a general Lure system
12 12
1, 1,
( ) ( )
( )
N N
i i ij j i i ij j i
j j i j j i
i i
i i i
x Ax a A x x Bu a B u u
y Cx
u f y
= = = =
= + + +
=
=

(2)
Where , , , ,
n m
i i
x R y R A B C e e have the same meaning as those in (1).
12
n n
A R

e ,
12
n n
B R

e are the inner
coupling matrix describing the interconnection among components of ( )
j i
x x and ( )
j i
u u , , 1,..., i j N =
respectively. And ( )
ij N N
L a

= denotes the connection topology and coupling strength ,
ij
a i j = are given as in (2)
and
1,
N
ii ij j j i
a a
= =
=

which are referred to the outer coupling matrix. By the Kronecker product, network (2) can be
rewritten as
12 12
( ) ( )
( )
( )
N N
N
x I A L A x I B L B u
y I C x
u f y
= + + +
=
=
(3)
Lemma 1[6] Suppose that A has no eigenvalues on the imaginary axis and ( , ) A B is controllable and ( , ) A C is
observable. And suppose that nonlinear function
i
f ( 1, , ) i m = is piecewise continuously differentiable and
'( )
i
f t is bounded. If system (1) has isolated equilibriums and there exist diagonal matrices P and Q with Q>0, a
scalar 0 c > and a symmetric matrix W such that the following LMI is feasible
11 12
22
0
*
H H (
s
(
H

(4)
Where
1 1
( , , ), ( , , )
m m
diag diag o o I = A = ,
11
T T T T
WA A W C QC A C CA c H = + IA + ,
22
T T T T
PCB B C P Q B C CB c H = + + ,
12
1
( )
2
T T T T T
WB A C P A C CB C Q c H = + + I + A , then system (1) is dichotomous for all | | , f e I A .
Lemma 2[12]
1 1 2 3
, ,
T
T T T T = are compatible dimension matrix. The following conditions are equivalent
(1)
2
1 2 3
( ) 0, :
T
T He T T I + A < A A A s ; (2) there exists 0 q > satisfying
2
1 2 2 3 3
1
0
T T
T T T T T q
q
+ + < ;
(3) there exists 0 q > satisfying
2
1 2 2 3
3
0
T T
T T T T
T I
q
q
(
+
<
(

(

;(4) there exists 0 q > satisfying
2
1 3 3 2
0
T
T T T T
I
q
q
(
+
<
(
- (

.
Lemma 3[9]Suppose
i
are distinct eignvalues of L, then
12 N
I A L A + is stable if and only if
12 i
A A + (i=1,...,N)
are stable simultaneously.
Proof Let T be a non-singular matrix such that
1
T LT J

= , where J is the diagonal form of L. Then the similarity

transformation, i.e.,
1
12 12
( )( )( )
N N
T I I A L A T I I A J A

+ = + completes the proof .

Lemma 4[6] System (1) is absolutely stable for all | | f e I A , if A B C I is stable and there exist diagonal
matrices P and Q with 0 Q > , and a symmetric matrix W such that the following LMI is feasible

- 3 -
http://www.sj-ce.org
11 12
0
T T
PCB B C P Q
(
<
(
- + (

(5)
Where
11 12
1
, ( )
2
T T T T T
WA A W C QC WB A C P C Q = + IA = + I + A .
3 MAIN RESULTS
Firstly, BMI criterion about dichotomy of nonlinear network system is given. Then the problem of the absolute
stability of nonlinear network systems is considered, and the method of designing the related controllers is obtained.
Theorem 1 Suppose L is a matrix with eigenvalues , 1,... ,
i
i N = different from each other. If
12
i
i
A A A

= + ,
1,... i N = have no eigenvalues on the imaginary axis. ( , )
i I
A B

is controllable, and( , )
i
A C

( , )
i
A C

is observable.
If system (3) has isolated equilibrium and there exist diagonal matrices P and Q with Q>0, a scalar 0 c > and a
symmetric matrix W such that the following LMI is feasible
11 12
22
0
*
(
s
(

(6)
where
11 12
1
, (
2
i i i i I i i I
T T T T T T T T T
WA A W C QC A C JA WB A C P A C CB J Q

c c = + IA + = + + I + A ,
22
I I I I
T T T T
PCB B C P Q B C CB

c = + + then systems (3) is dichotomous for all f satisfying the sector
condition.
Proof: By lemma 1, if system (3) has isolated equilibrium and there exist diagonal matrices P and Q with Q>0, a
scalar 0 c > and a symmetric matrix ,
N N
W I W Q I Q ' ' = = such that the following LMI is feasible
11 12
22
0
*
A A (
s
(
A

(7)
where
11 12 12 12 12
( ) ( ) ( ) ( ) ( )
T T T T
N N N N N
W I A L A I A L A W I C QC I A L A C C I A L A c ' ' ' A = + + + IA + + +

12 12 12 12
( ) ( ) ( )
T T T T
N N N
I A L A C P I A L A C C I B L B c A = + + + +
12
1
( ) ( ) ( )
2
T
N N
I C Q W I B L B ' ' I + A + +
22 12 12 12 12
( ) ( ) ( ) ( )
T T T T
N N N N
PC I B L B I B L B C P Q I B L B C C I B L B c ' A = + + + + + +
Suppose U is an orthogonal matrix such that
T
U LU = A, where
1
( ,..., )
N
diag A = , and
i
are eigenvalues of L.
Let { , }
n n
X diag U I U I = , and multiplying (7) on the left and on the right by
T
X and X respectively. Than we
can have (6), and this completes the proof.
In this section, network (3) with state feedback controller and uncertainties can be given as follows
12 12 1
( ( ) ) ( ( ) )
( )
( )
N N k
N
x I A A L A x I B B L B u B u
z I C x
u f y
= + A + + + A + +
=
=
(8)
where the uncertainties satisfy | | | |
1 2
A B HF E E A A = , and
1 2
, ,
n i j n j m
H R E R E R

e e e ,
2 T
F F I s .The
feedback controller chosen here is ( )
ki i
u Kx t = .
Theorem 2 Suppose that L is a matrix with , 1,... ,
i
i N = are eigenvalues of L, different each other. If
12
i
i
A A A

= + , 1,... i N = have no eigenvalues on the imaginary axis, ( , )

i I
A B

is controllable, and ( , )
i
A C

( , )
i
A C

is observable. And f satisfy the sector condition in (2). If system (8) has isolated equilibrium and there exist
diagonal matrices P, Q>0,
1
0, 0, c q > >
2
0 q > ,a invertible matrix X=X
T
and matrix Y , such that the following LMI is
feasible

- 4 -
http://www.sj-ce.org
11 12 13 1 1 16 17
22 1 2
1
1
2
2
* 0 0
* * 0 0 0
0
* * * 0 0 0
* * * * 0 0
* * * * * 0
* * * * *
T
T T T
H XE
B C PCH E
I CH
I
I
I
I
q
c q
c c
q
q
q
q
(
M M M M M
(
(
M
(

(
s (

(
(

(
- (

(9)
where
11 1 1 12 1
1
, ( )
2
i i i i
T T T T T T T T T
A X XA BY Y B B XA C P Y B C P XC Q

M = + + + M = + + I + A ,
13 1
i
T T T T T
XA C Y B C

c c M = + ,
16
1
2
T
XC Q M = ,
17 2 22
,
i i
T T T
XC PCB B C P Q

q M = IA M = + . Then system (8) is dichotomy. And
1
K YX

= is the state feedback

parameter matrix.
Proof: according to the theorem 1, system (8) is dichotomy, if the inequality (9) satisfy, then inequality (9) can be
rewritten as
| |
1 2
M 0 0
WH
He PCH F E E
CH c
(

(
+ s
`
(

(
)
(10)
where
11 12 12
22
( )
M *
* *
T T
i
T T
A A C
B C
I
c
c
c
(
u u +
(
(
= u
(

(

,
11 12 12
( ) ( )
T T
i i
W A A A A W C QC u = + + + IA ,
12 12 12
1
( ) ( ) ( )
2
T T T
i i
W B B A A C P C Q u = + + + I + A ,
22 12 12
( ) ( )
T T
i i
PC B B B B C P Q u = + + + .
According to the lemma 2 and Schur complement lemma, the inequality (10) is satisfied, for
1
0 q > ,
2
0 q > if and
only if the following inequality is satisfied
11 12 13 1 1 16 17
22 1 2
1
1
2
2
* 0 0
* * 0 0 0
0
* * * 0 0 0
* * * * 0 0
* * * * * 0
* * * * *
T
T T T
WH E
B C PCH E
I CH
I
I
I
I
q
c q
c c
q
q
q
q
(
T T T T T
(
(
T
(

(
s (

(
(

(
- (

(11)
where
11 12 12 1 1 12 12 12
( ) ( ) , ( ) ( )
T T T T T
i i i i
W A A A A W WB K K B W W B B A A C P T = + + + + + T = + + +
1
1
( ),
2
T T T T
K B C P C Q + I + A
13 12 1
( )
T T T T T
i
A A C K B C c c T = + + ,
16 17 2
1
,
2
T T
C Q C q T = T = IA ,
22 12 12
( ) ( )
T T
i i
PC B B B B C P Q T = + + + .
Let
1
{ , , , } diag W I I

Z = , and multiply inequality (11) on left and on right by Z. let

1
X W

= ,
1
Y KW

= . Then the
proof is completed.
Remark1: Inequality (9) is a nonlinear matrix inequality which involves
1 2
, , , , , , Y P X Q c q q . If , P Q,
1 2
, , c q q are
fixed, then inequality (9) is a linear matrix inequality. And if , X Y are fixed, inequality (9) is a linear matrix
inequality which involves , P Q,
1 2
, , c q q . Then we use the alternating algorithm until the inequality feasible solution
can be obtained.

- 5 -
http://www.sj-ce.org
Theorem 3 Suppose L is a matrix with , 1, ,
i
i N = , which are eigenvalues of L, and are different from each other.
Network (3) is absolutely stable for all f satisfying the sector condition in (2), if
12
, 1, ,
i
A A i N + = are stable and
there exist diagonal matrices P and Q with 0 Q > , and symmetric matrices W , such that the following LMI is
feasible
11 12
22
0
( A A
<
(
- A

(12)
where
( )
11 12 12 12 12 12
1
( ) ( ) , ( ) ( ) ,
2
T T T T T
i i i i
W A A A A W C QC W B B A A C P C Q A = + + + IA A = + + + I + A
22 12 12
( ) ( )
T T
i i
PC B B B B C P Q A = + + + .
Proof: By lemma 4, network (3) is absolutely stable, if
12 N
I A L A + is stable and there exist diagonal matrices
P and
N
Q I Q ' = with 0 Q > , and symmetric matrices
N
W I W ' = such that 0
( I +
<
(
- u

, where
1 12 12
( ) ( ) ( ) ( )
T T
N N N N
W I A L A I A L A W I C Q I C ' ' ' I = + + + IA ,
2 12 2 12
( ) ( )
T T
N N
PC I B L B I B L B C P Q' u = + + + ,
2 12 12
1
( ) ( ) ( ) ( )
2
T T T
N N N
W I B L B I A L A C P I C Q ' ' + = + + + I + A .
Let U be an orthogonal matrix such that
T
U LU = A, where
1
( , , )
N
diag A = and , 1, ,
i
i N = are eigenvalues
of L. Take ( , ) X diag U I U I = . Multiplying on the left and the right of the above inequality by
T
X and X ,
respectively, and this completes the proof.
Theorem 4 Suppose that L is a matrix, and , 1, ,
i
i N = are eigenvalues of L, different each other. Network (3) is
absolutely stable for all f satisfying the sector condition in (2) , if
12
, 1, ,
i
A A i N + = are stable and there exist
diagonal matrices P and Q with 0 Q > , and matriceY ,
T
X X = , such that the following inequality (13) satisfied.
Where
11 12 12 1 1 22 12 12
( ) ( ) , ( ) ( )
T T T T T
i i i i
A A X X A A Y B BY PC B B B B C P Q H = + + + + + H = + + + .
Then system (3) is absolutely stable.
1
K YX

= is the controller matrix.

Proof: according to theorem 3, if the system (3) is absolute stable with controller ( )
ki i
u Kx t = , then the inequality
11 12
22
0
A A | |
<
|
- A
\ .
is satisfied, where
11 12 1 12 1
( ) ( )
T T
i i
W A A B K A A B K W C QC A = + + + + + IA ,
22 12 12
( ) ( )
T T
i i
PC B B B B C P Q A = + + + , and
12 12 12 1
1
( ) ( ) ( )
2
T T T
i i
W B B A A B K C P C Q A = + + + + I + A .
Let
1
{ , , , } diag W I I

Z = . Multiply the inequality

11 12
22
0
A A | |
<
|
- A
\ .
on the left and on the right by Z,
let
1
X W

= ,
1
Y KW

= , and using lemma 2, then we can get the inequality (13).

1 11 12 1 12
2 3 4
22
1
1
2
2
3
3
4
4
1
0 0 0.5 0 ( )
2
0 0 Q( )
* 0 0 0
0 0 0 0 0
* * 0 0
0 0 0 0 0
* * * 0
0 0 0 0 0
* * * *
0 0 0 0
* * * * *
* 0 0 0
* * * * *
* * 0 0
* * * * *
* * * 0
* * * * *
* * * *
* * * * *
T T T T T T
i i
T T
Y B XC B B XC Q XC X A A
P C P C
I
I
I
I
I
I
I
I
q
q q q
q
q
q
q
q
q
q
q

(
H + IA +

I + A
H

0
(
(
(
(
(
(
<
(
(
(
(
(
(
(
(13)

- 6 -
http://www.sj-ce.org
Remark 2: The dynamic consensus of system (3) is lim ( ) ( ) 0 , , 1, 2...,
i j
t
x t x t i j N

= = , then theorem 4
gives the special results of dynamic consensus for network system (3) with lim ( ) lim ( ) 0 ,
i j
t t
x t x t

= =
, 1, 2..., i j N = .
4 SIMULATION
In this section, the related examples are provided to validate the effectiveness of the results obtained above.
Example 1, for the network system (8) , which is composed by three subsystems, the related matrix of the network
is given as
0.4 3
1 0.5
A
(
=
(


,
1 0.5
2.4 1
B
(
=
(

,
1
0 1 1 0
,
1 0 0 1
C B
( (
= =
( (

,
1 0
0 1
(
I =
(

,
1 0
0 1
(
A =
(

,
12
0.1 0.1
0.1 0.1
A
(
=
(

,
1
0.1 0.2
0.2 0.1
E
(
=
(

,
2
0.8 0
0.2 0.5
E
(
=
(

,
0.1 0
0 0.1
H
(
=
(

,
12
0.1 0.1
0.1 0.1
B
(
=
(

,
( ) sin( ) , 1,..., 3
i i i
f y y i = = ,
2 1 1
0 1 1
1 1 2
L
(
(
=
(
(

. We can find that the eigenvalues of L are 0, 2, 3.
When
0
(0.1, 0.2, 0.3, 0.1, 0.2, 0.3) x = , the states diagram of uncertain network (8) without controller are shown
in fig.1(a), and the phase diagram of (x
1
, x
2
) of subsystem 1 is shown in fig.1(b). We can find that the network
system (8) without controller contains limit circle. Using theorem 2, the controller K
1
=[1830.1 25.0;416.9 984.6] is
obtained. The states diagram of network system (8) with the controller K
1
shown in fig.1(c), we can find that the
system is dichotomous.
0 10 20 30 40 50
-10
-5
0
5
10
t/s
s
t
a
t
e
s
-10 -5 0 5 10
-4
-2
0
2
4
x1
x
2
0 0.002 0.004 0.006 0.008 0.01
-4
-2
0
2
4
x 10
7
t/s
s
t
a
t
e
s

(a) (b) (c)
FIG.1:(A) STATE DIAGRAM OF NETWORK SYSTEM, (B) PHASE DIAGRAM OF SUBSYSTEM 1, (C) STATE DIAGRAM OF NETWORK
SYSTEM (8) WITH CONTROLLER K
1

0 10 20 30 40 50
-6
-4
-2
0
2
4
6
t/s
s
t
a
t
e
s

0 0.2 0.4 0.6 0.8 1
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
t/s
s
t
a
t
e
s

(a) (b)
FIG.2: (A) STATE DIAGRAM OF NETWORK SYSTEM WITH CONTROLLER K
1
, (B) STATE DIAGRAM OF NETWORK SYSTEM (3) WITH
CONTROLLER K
2
Example 2, for the network system (3), the nonlinear functions are chosen as ( ) 0.5(1 sin( )) , 1,...,3
i i i
f y y i = + = ,
then
0 0
0 0
(
I =
(

,
1 0
0 1
(
A =
(

and the other parameter are chosen the same as example 1.

- 7 -
http://www.sj-ce.org
When
0
(1, 2, 3, 1, 2,3) x = , the states diagram of network system (3) without controller are shown in fig.2(a),
Using theorem 4, we have K
2
=[-182.1389 68.7286; -75.1388 20.2610]. The states diagram of network system (3)
with the controller K
2
shown in fig.2(b), we can find that the system is absolute stability.
5 CONCLUSIONS
In this paper, we consider the dichotomy and absolute stability of a class of nonlinear complex dynamical networks.
The complex network system is composed of many subsystems by a certain interconnection relationship. The
subsystems are a kind of nonlinear system, i.e., Lure systems. And the interconnection relationship of complex
network systems referred in this paper not only concludes the interconnection of states, but also the interconnection
of the sector nonlinear. Some necessary criterions are derived in this paper. Then, controller design methods of
dichotomy and absolute stability of complex system are given and the related controller can be constructed via
feasible solutions of a certain set of blinear matrix. At last, examples are given to illustrate the effectiveness of the
proposed methods.
REFERENCES
[1] Wang X. F., Complex networks: Topology, dynamics, and synchronization. International Journal of Bifurcation and Chaos, 2002,
vol.5, pp. 885-916
[2] MOYLAN JP and HILL J D, Stability criteria for large-scale systems, IEEE Transaction on Automatic Control, 1978, 23(2): 143-
149
[3] WOLFE J D,CHICHKA D F, and SPEYER J L, Decentralized controller for unmanned aerial vehicle formation flight, USA:
American Institute of Aeronautics and Astronautics, 1996
[4] I. Lestas, G. Vinnicombe, Scalable decentralized robust stability certificates for networks of interconnected heterogeneous
dynamical systems,IEEE Trans, Automat. Control, 2006, vol. 51, pp. 1613-1635
[5] B.Bamieh, F.Paganini, M.Dahleh, Distributed control of spatially invariant systems, IEEE Trans. Automat. Control 2002, vol. 47,
pp. 1091-1118
[6] Jinzhi Wang, Zhisheng Duan, Ying Yang, and Lin Huang, Analysis and control of Nonlinear systems with stationary sets: time-
domain and frequency domain methods, Singapore: World Scientific Publisher, 2009
[7] Kai Wen, Zhiyong Geng, Stability of distributed heterogeneous systems with static nonlinear interconnections, System& Control ,
2010, vol. 59, pp. 680- 686
[8] Liu X, Wang J. Z., Huang L., Global synchronization for a class of dynamical complex networks, Physica A, 2007, 386:543-556
[9] Zhisheng Duan, Jinzhi Wang. Stability analysis and decentralized control of a class of complex dynamical networks, Automatica,
2008, vol. 22, pp. 1028-1035
[10] Li Z. K., Duan Z. S., and G. R. Chen, Global consensus regions of multi-agent systems with nonlinear dynamics, Proceedings of
WCICA, 2010, 580-585
[11] Li W., Chen Z., Liu Z., Leader-following formation control for second-order multiagent systems with time-varying delay and
nonlinear dynamics, Nonlinear Dynamics, 2013:1-10
AUTHORS
1
Haijing Liu was born in Hebei, China, in
1988. She received the bachelors degree in
electronic information engineering from
Hengshui University, Hengshui, China. She
is a master in the college electronic
information and control engineering in
Beijing University of Technology. Her
research interests include complex dynamical network systems,
and robust control.

2#
Dun Ao was born in Inner Mongolia,
China, in 1978. She received the Doctorate
in general mechanics and mechanic
foundation from Peking University, Beijing,
China. She is the lecturer and master
instructor in the college electronic
information and control engineering in
Beijing University of Technology. Her research interests include
analysis and synthesis of complex systems, fuzzy control
systems, and robust control. Email: aodun@bjut.edu.cn