to understand an unknown dynamic system from one or more
well-known dynamic systems, which means two or more systems
asymptotically share a common dynamic behavior. Secondly,
because naturally occurring GRNs are generally much more
complicated, the study of synchronization among GRNs is impor-
tant for the understanding of the rhythmic phenomena of living
organisms at both molecular and cellular levels [3]. Thirdly, in
cellular physiology, we need to focus on how proteins produce and
how gene networks are regulated, which could be better under-
stood by considering the synchronization of GRNs. We hope to
synchronize the complex GRNs by some relatively simple systems
which nevertheless display rich dynamical behaviors and provides
some opportunity to test theoretical results of genetic regulation.
Generally, synchronization can be induced by e xternal forcing or by
coupling, and many types of synchronization have been presented in
the past decades [17]. Similarly , there have a large number of
experimental and theoretical works studying the synchronization in
genetic networks [1 8,19]. Unfortunately , almost all the discussions in
the existing literature regarding the converg ence of synchronization
error do not consider the convergence speed, even though we eagerly
want to synchronize network states as quickly as possible in practical
applications. In order to achieve faster synchronization and to realize
synchronization in finite time rather than merely asympto tically [20],
an effective method is to use finite-time techniq ues, which are
demonstrated to ha ve better robustness and disturbance rejection
properties [21].
Recently, many kinds of finite-time issues have attracted
particular research interests, and there have been some results
on finite-time stabilization, convergence, synchronization, con-
sensus [21–30]. Normally, the term of uðtÞ¼ h signð
δðtÞÞj δðtÞj
α
,
0r
αo 1 was introduced in the above references, where δðtÞ
denotes the error and h the gain. For the different values of
parameter
α
, such techniques can generally be divided into two
types: (i) continuous (when 0o
αo 1) [31–33] and (ii) discontin-
uous (when
α ¼ 0) [34–36]. In this paper, the first type of u(t) will
be introduced into the design of the finite-time stochastic synchro-
nization (FTSS) for GRNs, and another one will be discussed to
optimize the synchronization speed.
Motivated by the above questions, in this paper, in order to
realize the FTS of the stochastic GRNs, a continuous controller is
addressed. Compared with [31–33], the difference of this paper
lies in the following three aspects. First, based on the finite-time
stability theorem of stochastic nonlinear systems [22–24], a new
finite-time controller is proposed for GRNs with noise perturba-
tions. Moreover, in contrast to [31–33], the FTS in this paper is
guaranteed by constructing a suitable Lyapunov functional and the
obtained conditions are easier to be satisfied. Second, the gain
parameters of controller can be designed by solving a linear matrix
inequality and the robust finite-time stochastic synchronization
(RFTSS) for GRNs with parameter uncertainties can be realized as
well. Finally, in order to explore the upper bound of the settling
time as small as possible, we further discuss the relationship
between the settling time and the parameter
α
, including the
situation of
α ¼ 0.
The notations in this paper are quite standard. R
n
and R
nm
denote, respectively, the n dimensional Euclidean space and the
set of all n m real matrices. The superscript “T” denotes the
transpose and the notation X Z Y (respectively, X 4 Y) where X and
Y are symmetric matrices, mean that X Y is positive semi-definite
(respectively, positive definite).
λ
max
ðMÞ and λ
min
ðMÞ denote the
maximal and minimal eigenvalues of real matrix M respectively.
Let ð
Ω; F ; F
t
fg
t Z 0
; PÞ be a complete probability space with a
filtration fF
t
g
t Z 0
satisfying the usual conditions (that is, it is right
continuous and contains all P -null sets). Efxg stands for the
expectation of the stochastic variable x with respect to the given
probability measure P. I and 0 represent the identity matrix and
the zero matrix, respectively. diagð Þ stands for a block-diagonal
matrix; matrices, if their dimensions are not explicitly stated, are
assumed to be compatible for algebraic operations.
2. Model formulation and preliminaries
2.1. Network description
Consider the following genetic network, which is established as
follows:
dmðtÞ
dt
¼AmðtÞþBf ðpðtÞÞþJ;
dpðtÞ
dt
¼CpðtÞþDmðtÞ;
8
>
>
<
>
>
:
ð1Þ
where mðtÞ¼ðm
1
ðtÞ; m
2
ðtÞ; …; m
n
ðtÞÞ
T
, pðtÞ¼ðp
1
ðtÞ; p
2
ðtÞ; …; p
n
ðtÞÞÞ
T
denote, respectively, the concentrations of mRNA and protein of
the gene at time t, A ¼ diagða
1
; a
2
; …; a
n
Þ and C ¼ diagðc
1
; c
2
; …; c
n
Þ
represent the degradation rates of mRNA and protein, respectively,
and D ¼ diagðd
1
; d
2
; …; d
n
Þ is the translation rate.
The nonlinear function f ðpðtÞÞ ¼ ½f
1
ðp
1
ðtÞÞ; f
2
ðp
2
ðtÞÞ; …; f
n
ðp
n
ðtÞÞ
T
,
where
f
j
ðp
j
ðtÞÞ ¼
p
j
ðtÞ=β
j
1þðp
j
ðtÞ=β
j
Þ
H
j
;
with H
j
being the Hill coefficient and
β
j
being a positive scalar. The
matrix B ¼ðb
ij
Þ
nn
is the coupling matrix of the genetic networ k
defined as follows: if transcription factor j is an activ at o r of gene i,
then b
ij
¼ a
ij
; if there is no connection between j and i,thenb
ij
¼ 0; if
transcrip tion factor j is a repressor of gene i,thenb
ij
¼a
ij
. Here, a
ij
is a positive scalar that denotes the transcriptional rate of transcrip-
tion factor j to gene i. J ¼½J
1
; J
2
; …; J
n
T
is defined as a basal rate by
J
i
¼
P
j A V
i
a
ij
,whereV
i
is the set of repressor of gene i.
Since f
i
(i ¼ 1; 2; …; n) is a monotonically increasing and differ-
entiable function with saturation, it satisfies 0r df
i
ðsÞ=dsr
~
m
f
i
,
which is equivalent to
0r
f
i
ðs
1
Þf
i
ðs
2
Þ
s
1
s
2
r
~
m
f
i
; 8s
1
; s
2
A R: ð2Þ
For simplicity, let xðtÞ¼½m
T
ðtÞ; p
T
ðtÞ
T
. Accordingly, the system
(1) becomes
dxðtÞ¼½
~
AxðtÞþ
~
B
~
f ðxðtÞÞ þ
^
IJ dt; ð3Þ
where
~
A ¼
A 0
D C
;
~
B ¼
B
0
;
~
I ¼
I
0
and
~
f ðxðtÞÞ ¼ f ðpðtÞÞ:
From (2), we know that the nonlinear function
~
f ðxðtÞÞ satisfies
~
f ðxðtÞÞð
~
f ðxðtÞÞ
~
M
f
xðtÞÞr 0; ð4Þ
where
~
M
f
¼½0; M
f
with M
f
¼ diagð
~
m
f
1
;
~
m
f
2
; …;
~
m
f
n
Þ.
In this paper, we consider model (1) or (3) as the master
system. The response system is
dyðtÞ¼½
~
AyðtÞþ
~
B
~
f ðyðtÞÞ þ
^
IJþ uðtÞ dt þ
ρðδðtÞÞ dωðtÞ; ð5Þ
where yðtÞ¼½
^
m
T
ðtÞ;
^
p
T
ðtÞ
T
and δðtÞ¼yðtÞxðtÞ is the error state, u
(t) is the controller,
ωðtÞ is one-dimensional Brownian motion
defined on the probability space ð
Ω; F ; fF
t
g
t Z 0
; PÞ, and the inten-
sity function
ρðÞ is the noise intensity vector satisfying the
following condition:
trace½
ρ
T
ðδðtÞÞρðδðtÞÞr ‖MδðtÞ‖
2
; ð6Þ
where M is a matrix with appropriate dimensions.
N. Jiang et al. / Neurocomputing 167 (2015) 314–321 315