Automatica 74 (2016) 10–15

Contents lists available at ScienceDirect

Automatica
journal homepage: www.elsevier.com/locate/automatica

Technical communique

An improved summation inequality to discrete-time systems with

time-varying delay✩
Chuan-Ke Zhang a,b , Yong He a,1 , Lin Jiang b , Min Wu a
a
School of Automation, China University of Geosciences, Wuhan 430074, China
b
Department of Electrical Engineering & Electronics, University of Liverpool, Liverpool, L69 3GJ, United Kingdom

article info abstract

Article history: The summation inequality plays an important role in developing delay-dependent criteria for discrete-
Received 5 November 2015 time systems with time-varying delay. This note proposes an improved summation inequality to estimate
Received in revised form the summation terms appearing in the forward difference of Lyapunov–Krasovskii functional. Compared
16 April 2016
with the inequality recently developed by the Wirtinger-based summation inequality and the reciprocally
Accepted 15 July 2016
convex lemma, the proposed one reduces the estimation gap while requires the same number of decision
variables. A relaxed stability criterion of a linear discrete-time system with a time-varying delay is
Keywords:
established by using such novel inequality. Two numerical examples are given to demonstrate the
Discrete-time system advantages of the proposed method.
Time-varying delay © 2016 Elsevier Ltd. All rights reserved.
Summation inequality
Linear matrix inequality

1. Introduction such admissible region. The following double summation term

is frequently applied during the construction of LKF to obtain
In the last few years, the stability analysis of discrete-time delay-dependent criterion (Seuret et al., 2015):
systems with time-varying delays has become a hot topic in the −
h 1 −1 
k−1
field of control theory (Feng, Lam, & Yang, 2015; Gao & Chen, Vr (k) = ηT (j)Rη(j) (1)
2007; He, Wu, Liu, & She, 2008; Huang & Feng, 2010; Kim, 2015; i=−h2 j=k+i
Kwon, Park, Park, Lee, & Cha, 2012, 2013; Meng, Lam, Du, & Gao,
2010; Nam, Pathirana, & Trinh, 2015; Nam, Trinh, & Pathirana, where h1 and h2 are respectively the lower and the upper bounds
2015; Peng, 2012; Seuret, Gouaisbaut, & Fridman, 2015; Shao & of a time-varying delay (i.e., h1 ≤ d(k) ≤ h2 ), R ≥ 0, and
Han, 2011; Xu, Lam, Zhang, & Zou, 2014; Zhang & Han, 2015; η(k) = x(k + 1) − x(k) with x(k) being the system state. Then
Zhang, Peng, & Zheng, 2016; Zhang, Xu, & Zou, 2008). An important the following term will appear in the forward difference of Vr (k):
objective of stability analysis is to find the admissible delay region k−h1 −1 k−d(k)−1
 
such that time-delay systems remain stable for the time-varying S (k) := ηT (i)Rη(i) + ηT (i)Rη(i) (2)
delay within this region (Zhang, He, Jiang, Wu, & Zeng, 2016). i=k−d(k) i=k−h2
Delay-dependent stability criteria developed in the framework of
the Lyapunov–Krasovskii functional (LKF) and the linear matrix During the development of stability criteria, a challenging prob-
inequality (LMI) are the most effective criteria to determine lem is how to estimate the lower bound of the above sum-
mation term (Zhang & Han, 2015). Obtaining tighter bound of
summation term (i.e., reducing the estimation gap) plays a key
role in reducing the conservatism. In the early literature, the
✩ This work is supported partially by the National Natural Science Foundation free-weighting matrix (FWM) approach (He et al., 2008) and the
of China under Grant 61503351, 51428702, 61573325, and 61210011, and the Jensen-based inequality (JBI) (Shao & Han, 2011) were two impor-
Hubei Provincial Natural Science Foundation of China under Grant 2015CFA010. tant methods for this issue. By relaxing the JBI, Wirtinger-based
The material in this paper was not presented at any conference. This paper was
inequalities (WBIs) were simultaneously reported in Nam, Pathi-
recommended for publication in revised form by Associate Editor Chunjiang Qian
under the direction of Editor André L. Tits. rana, and Trinh (2015) and Seuret et al. (2015) and later in Zhang
E-mail addresses: [email protected] (C.-K. Zhang), [email protected] and Han (2015). Very recently, an auxiliary function based inequal-
(Y. He), [email protected] (L. Jiang), [email protected] (M. Wu). ity (AFBI) (Nam, Trinh, & Pathirana, 2015) and a free-matrix-based
1 Fax: +86 27 87175080. summation inequality (FMBI) (Chen, Lu, & Xu, 2016) inspired by the
http://dx.doi.org/10.1016/j.automatica.2016.07.040
0005-1098/© 2016 Elsevier Ltd. All rights reserved.
 C.-K. Zhang et al. / Automatica 74 (2016) 10–15 11

research of Zeng, He, Wu, and She (2015, 2016) were developed by The following RCL is usually applied to combine with the WBI
further improving the WBI. in the literature.
Those inequality-based estimation methods include two key
Lemma 2 (Reciprocally Convex Lemma (RCL) Park et al., 2011). For
steps to estimate S (k): (1) applying the JBI/WBI/AFBI to estimate
a given scalar α in the interval (0, 1), symmetric
 positive
 definite
two summation terms in S (k), respectively; and (2) using the U1 X
reciprocally convex lemma (RCL) (Park, Ko, & Jeong, 2011) to matrices U1 and U2 , and any matrix X such that ∗ U2
≥ 0, the
handle the d(k) appearing in the denominator. The recently following inequality holds
developed techniques (the WBIs, the AFBI, and the FMIB) focus on
1
 
the first step. To the best of the authors’ knowledge, there is no U1 0  
research that discusses the tighter estimation of S (k) considering α U1 X
≥ ∗ . (7)

 1 U2
two steps together. This note aims to fill this research gap. ∗ U2
This note proposes an improved summation inequality by 1−α
considering two terms of S (k) together. It is tighter than the one
The estimation of the S (k) via the WBI and the RCL leads to the
obtained by combining the WBI and the RCL but keeps the same
following lemma.
number of decision variables. A new stability criterion for a linear
discrete-time system with a time-varying delay is established by Lemma 3. For asymmetric positive definite matrix R, any matrix X
applying the proposed inequality. Finally, two numerical examples R̃ X
satisfying ≥ 0 with R̃ = diag{R, 3R}, the S (k) defined in (2) is
are given to illustrate the effectiveness of the proposed inequality ∗ R̃
and the corresponding criterion. estimated as
Throughout this note, the superscripts T and −1 mean the  T   
1 E1 R̃ X E1
transpose and the inverse of a matrix, respectively; Rn denotes the S (k) ≥ ζ T (k) ζ (k) (8)
h21 E2 ∗ R̃ E2
n-dimensional Euclidean space; ∥ · ∥ refers to the Euclidean vector
norm; P > 0 (≥ 0) means that P is a symmetric positive-definite where
(semi-positive-definite) matrix; diag{·} denotes a block-diagonal
matrix; Sym{X } = X + X T ; and the symmetric term in a symmetric h21 = h2 − h1 , d = d(k) (9)
ζ (k) = x (k), x (k − h1 ), x (k − d), x (k − h2 ),
T T T T

matrix is denoted by ∗. Matrices, if their dimensions are not explic-
itly stated, are assumed to be compatible for algebraic operations. T
v1T (k), v2T (k), v3T (k) (10)
k−h1
x(i) x(i)
2. Problem formulation and preliminaries k 
v1 (k) = , v2 (k) = (11)
i=k−h1
h1 + 1 i=k−d
d − h1 + 1
Consider the following linear discrete-time system with a
time-varying delay: k−d
 x(i)
v3 (k) = (12)
x(k + 1) = Ax(k) + Ad x(k − d(k)),

k≥0 i=k−h2
h2 − d + 1
(3)
x(k) = φ(k), k ∈ [−h2 , 0]    
e2 − e3 e3 − e4
E1 = , E2 = (13)
where x(k) ∈ Rn and φ(k) are the system state and the initial e2 + e3 − 2e6 e3 + e4 − 2e7
condition, respectively; A and Ad are the system matrices; and d(k)
ei = 0n×(i−1)n , In×n , 0n×(7−i)n , i = 1, 2, . . . , 7.
 
(14)
is a positive integer which is time-varying and satisfies
1 ≤ h1 ≤ d(k) ≤ h2 . (4) Proof. Using WBI (6) and RCL (7) to estimate the S (k) yields
k−h1 −1 k−d−1
This note is concerned with the stability of system (3). As men-  
S (k) = ηT (i)Rη(i) + ηT (i)Rη(i)
tioned in Section 1, a challenging problem is how to estimate the
i=k−d i=k−h2
summation term S (k). Therefore, the first aim of this note is to de-
velop a more effective estimation method for this task. Then, this 1 1
≥ ζ T (k)E1T R̃E1 ζ (k) + ζ T (k)E2T R̃E2 ζ (k)
note will apply the proposed method to derive a new stability cri- d − h1 h2 − d
terion for judging the influence of the time-varying delay on the  T   
1 E1 R̃ X E1
stability of system. ≥ ζ T (k) ζ (k).
h21 E2 ∗ R̃ E2
Several WBIs with different forms were simultaneously re-
ported in Nam, Pathirana, and Trinh (2015), Seuret et al. (2015) and
later in Zhang and Han (2015). The ones to be applied in this note 3. A relaxed summation inequality
are recalled from Seuret et al. (2015).
This section develops an improved summation inequality for
Lemma 1 (Wirtinger-based Inequality Seuret et al., 2015). For a estimating S (k), shown in the following lemma.
given symmetric positive definite matrix R, integers b > a, any
sequence of discrete-time variable x: Z[a, b] → Rn , the following Lemma 4. For a symmetric positive definite matrix R, any matrix X ,
inequalities hold the S (k) defined in (2) is estimated as
 h − d 
 T R 2
0 T 0
   T
b−1
ϑ1 ϑ1
 
 1 1 E1  R̃ X  h21 1
η (i)Rη(i) ≥
T
S (k) ≥ ζ T (k)
b − a + 1
(5) +

b − a ϑ2 0 3 R ϑ2 h21 E2 ∗ R̃ d − h1
 
i=a b−a−1 0 T2
 T  h21
ϑ1 ϑ1
 
1 R 0  
≥ (6) E1
b − a ϑ2 0 3R ϑ2 × ζ (k) (15)
E2
x(i)
where ϑ1 = x(b) − x(a), and ϑ2 = x(b) + x(a) − 2
b
i=a b−a+1 . where R̃ = diag{R, 3R}, T1 = R̃ − X R̃−1 X T and T2 = R̃ − X T R̃−1 X .
12 C.-K. Zhang et al. / Automatica 74 (2016) 10–15

Proof. Firstly, for the symmetric matrix R > 0 and any matrices, Thirdly, calculating Πi , i = 1, 2 based on (18)–(25). It follows
Mi , i = 1, 2, 3, 4, with appropriate dimension, the following holds from (20), (22), (23) and (25) and d−h1 +1 < 1 that
d−h −1
1
based on Schur complement:
   T  
M2i−1 −1 M2i−1 M2i−1 k−h1 −1
  g (k) T M R−1 M T M1 R−1 M2T g (k)
 
 M2i R M2i M2i  ≥ 0, i = 1, 2. 1 1
f1 (i)g (k) ∗ M2 R−1 M2T f1 (i)g (k)
∗ R i=k−d

Then, for any vector ςj (k, i), j = 1, 2, the following is true: = (d − h1 )g T (k)M1 R−1 M1T g (k)
+ 2 × 0 × g T (k)M1 R−1 M2T g (k)
k−h1 −1
(d − h1 )(d − h1 − 1) T

Π1 = ς1T (k, i) + g (k)M2 R−1 M2T g (k)
i=k−d
  3(d − h1 + 1)
M1 R−1 M1T M1 R−1 M2T M1 < (d − h1 )g T (k)M1 R−1 M1T g (k)
× ∗ M2 R−1 M2T M2  ς1 (k, i) ≥ 0 (16)
+ (d − h1 )g T (k)M2 (3R)−1 M2T g (k)
∗ ∗ R  T
d − h1 T E1
k−d−1 = ζ (k )
h221 E2

Π2 = ς2T (k, i)
 
i=k−h2 R 0 LT1  
E1
 
M3 R−1 M3T M3 R−1 M4T M3 ×  0 0 0  ζ (k)
E2
× ∗ M4 R−1 M4T M4  ς2 (k, i) ≥ 0. (17) L1 0 L1 R−1 LT1
∗ ∗ R  T
d − h1 E1
Secondly, letting f (i, a, b) = 2i−b−a+1
yields the following + ζ T (k)
b −a +1 h221 E2
equalities  
0 0 0  
E1
k−h1 −1
× 0 3R T
L2  ζ (k)
E2

η(i) = (e2 − e3 )ζ (k), 0 L2 L2 (3R)−1 LT2
i=k−d
(18) d − h1
 T
E1
k−h1 −1
 = ζ T (k)
f1 (i)η(i) = (e2 + e3 − 2e6 )ζ (k) h221 E2
i=k−d
 
R 0 LT1  
E1
k−d−1 × 0 3R L2T  ζ (k)
E2

η(i) = (e3 − e4 )ζ (k), L1 L2 L1 R−1 LT1 + L2 (3R)−1 LT2
i=k−h2
(19) d − h1
 T   
k−d−1 E1 R̃ X E1
 = ζ T (k) ζ (k). (26)
f2 (i)η(i) = (e3 + e4 − 2e7 )ζ (k) h221 E2 ∗ X T R̃−1 X E2
i=k−h2

k−h1 −1
 k−h1 −1
 It follows from (18), (22), (23), (25) that
1 = d − h1 , f1 (i) = 0,
i=k−d i=k−d 
(20) k−h1 −1
 g (k)
T 
M1

k−h1 −1
(d − h1 )(d − h1 − 1) η(i)
f1 (i)g (k)

f12 (i) = M2
i=k−d
3(d − h1 + 1) i=k−d
T  
g (k)

M
k−d−1 k−d−1 + η (i) 1
T
M2 f1 (i)g (k)
 
1 = h2 − d , f2 (i) = 0,
i=k−h2 i=k−h2 k−h1 −1
(21)  
g T (k)M1 η(i) + f1 (i)g T (k)M2 η(i)

k−d−1
 (h2 − d)(h2 − d − 1) =2
f22 (i) = i=k−d
i=k−h2
3(h2 − d + 1)  
k−h1 −1

where f1 (i) = f (i, k − d, k − h1 ) and f2 (i) = f (i, k − h2 , k − d).  η(i) 
Moreover, redefine vector ςj (k, i), j = 1, 2 and matrices Mi , i = = 2g T (k) M1 M2 
   i=k−d 

k−
h1 −1
1, 2, 3, 4, in Πi as follows

f1 (i)η(i)
 
T
ςj (k, i) = g T (k), fj (i)g T (k), ηT (i) ,

i=k−d
T T  R 0

g (k) = E1T , E2T ζ (k)
 
(22) 2 E1
=− ζ (k)
T
0 3R E1 ζ (k)
1  T 1  T h21 E2
L1 L2
M1 = − R, 0, LT1 , M2 = − 0, 3R, LT2 (23)
h21 h21
 
 T 2R 0 LT1  
1  T 1  T 1 E1 E1
M3 = − LT3 , R, 0 , M4 = − LT4 , 0, 3R (24) =− ζ T (k)  0 6R L2T 
ζ (k)
h21 E2 E2
h21 h21 L1 L2 0
X = [L1 , L2 ]T = [L3 , L4 ], R̃ = diag{R, 3R} (25)  T   
1E 2R̃ X E1
where Li , i = 1, 2, 3, 4 are any matrices. = − ζ (k) 1 T
ζ (k). (27)
h21 E2 ∗ 0 E2
 C.-K. Zhang et al. / Automatica 74 (2016) 10–15 13

Similarly, based on (19), (21), (22), (24), (25), and

h2 −d−1
h2 −d+1
< 1, the corresponding new integral inequalities for continuous-time
following holds systems would be developed based on the similar idea of deriving
of summation inequality (15).
  g (k) T M R−1 M T M R−1 M T   g (k) 
k−d−1
3 3 3 4
f2 (i)g (k) ∗ M4 R−1 M4T f2 (i)g (k)
i=k−h2
4. A novel stability criterion
 T
h2 − d T E1
< ζ ( k)
h221 E2 By using the summation inequality (15) and the LKF taken from
L3 R L3 + L4 (3R)−1 LT4 Seuret et al. (2015), the following stability criterion for system (3)
 −1 T

L3 L4  
E1 is established.
×  T
L3 R 0  ζ (k)
E2
LT4 0 3R
 T  Theorem 1. For given integers h1 and h2 , system (3) with a time-
X R̃−1 X T
 
h2 − d E X E1 varying delay satisfying (4) is asymptotically stable if there exist
= ζ (k) 1
T
ζ (k) (28)
h221 E2 ∗ R̃ E2 symmetric positive definite matrices P ∈ R3n×3n , Q1 ∈ Rn×n , Q2 ∈
Rn×n , R1 ∈ Rn×n , R ∈ Rn×n , and any matrix X ∈ R2n×2n , such that
and the following LMIs hold:
 T  T  
k−d−1
g (k) g (k)
 
 M3 M
η(i) + ηT (i) 3 Ψ (h1 ) − Υ5,1 E1T X
 
f2 (i)g (k) M4 M4 f2 (i)g (k) <0 (31)
i=k−h2 ∗ −R̃
 
0 L3 L4
Ψ (h2 ) − Υ5,2
T
E2T X T
    
1 E1 E1
=− ζ (k)
T  LT3 2R 0  ζ (k) <0 (32)
h21 E2 E2 ∗ −R̃
LT4 0 6R
 T    where
1 E1 0 X E1
=− ζ T (k) ζ (k). (29)
Ψ (d) = Υ1 (d) + Υ2 + Υ3 − Υ4
h21 E2 ∗ 2R̃ E2
Υ1 (d) = Sym Γ T (d)P (Γ1 − Γ2 ) + Γ1T P Γ1 − Γ2T P Γ2
 
Thus, combining (16), (17) and (26)–(29) yields
T
Γ (d) = eT1 , (h1 + 1)eT5 , (d − h1 + 1)eT6 + (h2 − d + 1)eT7
 T 
1 E1
Π1 + Π2 < S (k) − ζ (k)
T
T
E2 Γ1 = eTs , − eT2 , − eT3 − eT4

h21
 h − d  T
Γ2 = eT0 , − eT1 , − eT2 − eT3

2
  T1 0  
 R̃ X  E1
× +  h21  E ζ (k). (30) Υ2 = eT1 Q1 e1 − eT2 Q1 e2 + eT2 Q2 e2 − eT4 Q2 e4

∗ R̃ d − h1 2
0 T2
h21 Υ3 = eTs (h21 R1 + h221 R)es
T
Finally, the summation inequality (15) can be obtained based

e1 − e2
on Π1 + Π2 ≥ 0 and (30). This completes the proof.  Υ4 =
e1 + e2 − 2e5
 
Remark 1. On the one hand, it is obvious that Ti ≥ 0, i = 1, 2 R1  0 
 
h1 + 1 e1 − e2
hold during slack matrix X is selected to satisfy
R̃ X
≥ 0(it is ×  (33)
∗ R̃
0 3 R1 e1 + e2 − 2e5
h1 − 1
the requirement of Lemma 3). Thus, Ti -dependent positive term  T   
appearing in (15) reduces the estimation gap between two sides E1 2R̃ X E1
Υ5,1 =
of (8). Thus, the proposed inequality (15) is tighter than (8). On the E2 ∗ R̃ E2
other hand, the slack matrix introduced by inequality (15) (i.e., X )  T   
is the same as the one arising in (8). That is to say, compared E1 R̃ X E1
Υ5,2 =
with inequality (8), the proposed (15) is a variable-increase-free E2 ∗ 2R̃ E2
inequality. Therefore, inequality (15) has the potential to derive 
e2 − e3
 
e3 − e4

new criteria that have less conservatism but require the same E1 = , E2 =
e2 + e3 − 2e6 e3 + e4 − 2e7
number of decision variables.
e0 = 0n×7n , es = (A − I )e1 + Ad e3
Remark 2. The AFBI proposed in Nam, Trinh, and Pathirana (2015)
ei = 0n×(i−1)n , In×n , 0n×(7−i)n , i = 1, 2, . . . , 7.
 
improves the WBI by adding additional terms to reduce the
estimation gap existing in the WBI. The FMBI proposed in Chen
et al. (2016) improves the WBI by introducing many free matrices. Proof. Consider the LKF candidate taken from Seuret et al. (2015):
Both improvements target to a single summation term, i.e., the
first step for handling S (k) as mentioned in Section 1. The idea of k−1
 k−h1 −1

deriving inequality (15) provides a new way to improve the WBI, V (xk ) = ξ T (k)P ξ (k) + xT (i)Q1 x(i) + xT (i)Q2 x(i)
i.e., considering two summation terms of S (k) together. i=k−h1 i=k−h2

−1 
 k−1

Remark 3. The techniques for continuous time-delay system are + h1 ηT (j)R1 η(j)
usually similar to the ones for discrete-time systems with time- i=−h1 j=k+i

varying delay (for example, Wirtinger-based integral inequality −

h1 −1 
k−1
for continuous-time systems and Wirtinger-based summation + h21 ηT (j)Rη(j) (34)
inequality for discrete-time systems). It is expected that the i=−h2 j=k+i
14 C.-K. Zhang et al. / Automatica 74 (2016) 10–15

where P > 0, Qi > 0, i = 1, 2, R1 > 0, R > 0, and Remark 6. Theorem 1 can be further improved by combining
 T the proposed inequality with several existing techniques, such as
k−1 k−h1 −1
  introducing zero-value terms for estimating the forward difference
ξ (k) = xT (k), xT (i), xT (i) , of the LKF (Kwon et al., 2013), and constructing augmented-based
i=k−h1 i=k−h2
and/or delay-partition-based LKF (Nam, Pathirana, & Trinh, 2015;
η(k) = x(k + 1) − x(k). Nam, Trinh, & Pathirana, 2015). The details are omitted in this
note, since the main contribution of this note is to propose a
Calculating the forward difference of V (xk ) yields (Seuret et al., novel way to improve the WBI while keep the same number of
2015): decision variables, and there is not much technical difficulty in the
aforementioned extension.
∆V (xk ) = ζ T (k) Υ1 (d) + Υ2 + Υ3 ζ (k)
 

k−1
 k−h1 −1
 5. Numerical examples
− h1 ηT (i)R1 η(i) − h21 ηT (i)Rη(i) (35)
i=k−h1 i=k−h2 Two numerical examples are used to demonstrate the advan-
tages of the proposed criterion via the comparisons of the calcu-
where ζ (k) is defined in (10). lated maximal admissible delay upper bounds (MAUBs) and of the
Using WBI (5) to estimate R1 -dependent summation term yields number of decision variables (NDVs).

k −1
 Example 1. Consider system (3) with
h1 ηT (i)R1 η(i) ≥ ζ T (k)Υ4 ζ (k). (36)
0.8 −0.1
   
0 0
i=k−h1
A= , Ad = . (39)
0.05 0.9 −0.2 −0.1
And using the proposed summation inequality (15) to estimate
R-dependent summation term (= h21 S (k)) yields This example is widely used for checking the conservatism of
stability criteria. The MAUBs calculated by Theorem 1, together
k−h1 −1
 with the ones reported in several literature, are given in Table 1,
h21 ηT (i)Rη(i) ≥ ζ T (k)Υ̃5 (d)ζ (k) (37) where the NDVs of several criteria are also listed. The following
i=k−h2
observations are summarized from the results listed in table:
where • In the early literature, stability criteria were developed via
 h − d  the FWM approach (Gao & Chen, 2007; He et al., 2008; Meng
2
 T   T 0   et al., 2010; Shao & Han, 2011; Zhang et al., 2008) or the JBI
E1  R̃ X  h21 1  E1
Υ̃5 (d) = + d − h1  E . (Huang & Feng, 2010; Kwon et al., 2012; Peng, 2012), commonly
E2 
∗ R̃ 2 combined with a simple LKF (Gao & Chen, 2007; He et al.,
0 T2
h21 2008; Huang & Feng, 2010; Shao & Han, 2011; Zhang et al.,
2008), and they are very conservative. Then, improved criteria
Thus, based on (35)–(37), the forward difference of the LKF is
were established by constructing LKFs with more general form
estimated as
(delay-partition-based LKF (Feng et al., 2015) and augmented-
∆V (xk ) ≤ ζ T (k) Υ1 (d) + Υ2 + Υ3 − Υ4 − Υ̃5 (d) ζ (k) based LKF (Kwon et al., 2013)) and/or by replacing the JBI with
 
tighter WBI (Nam, Pathirana, & Trinh, 2015; Seuret et al., 2015).
:= ζ T (k)Φ (d)ζ (k). (38) The WBI reduces the conservatism and avoids much increase of
NDVs. (Note that Theorem 4 of Feng et al. (2015) requires tune
It is easy to check that Φ (d) is affine with respect to the three parameters di , i = 1, 2, 3, which is a time-consuming
time-varying delay d ∈ [h1 , h2 ], thus Φ (d) < 0 if and only if procedure, although its NOV is smaller than that of the
Φ (h1 ) < 0 and Φ (h2 ) < 0, which are equivalent to LMIs (31) and WBI-based criteria in Nam, Pathirana, and Trinh (2015) and
(32), respectively, based on Schur complement. Therefore, if LMIs Seuret et al. (2015).)
(31) and (32) hold, then ∆V (xk ) ≤ −ε∥x(k)∥2 for a sufficient small • Very recently, two types of methods were developed to im-
ε > 0, which shows the asymptotical stability of system (3). This prove the WBI, including the AFBI and the FMBI. It can be
completes the proof.  found from the table that both the AFBI-based criterion (Nam,
Trinh, & Pathirana, 2015) and the FMBI-based criterion (Chen
et al., 2016) achieve the reduction of conservatism at the cost
Remark 4. In Seuret et al. (2015), the summation term, S (k), was
of increase of NDVs. On the contrary, Theorem 1 obtained by
estimated by using the WBI and the RCL, while it is handled by
the proposed inequality provides less conservative results but
using a tighter inequality (15) in this note. On the other side, keeps the same NDVs in comparison with the WBI-based crite-
the matrices to be determined in Theorem 1 are the same as the rion (Nam, Pathirana, & Trinh, 2015; Seuret et al., 2015). More-
ones in the criterion in Seuret et al. (2015, Theorem 5 therein). over, Theorem 1 provides better results than the AFBI- and
Thus, Theorem 1 has the potential to provide less conservative FMBI-based criteria (Chen et al., 2016; Nam, Trinh, & Pathirana,
results while requires the same number of decision variables in 2015) but requires smaller NDVs. It clearly shows the advan-
comparison to the one in Seuret et al. (2015). tages of the proposed inequality and the corresponding crite-
rion.
Remark 5. In Nam, Trinh, and Pathirana (2015), the AFBI tighter
Example 2. Consider system (3) with
than the WBI, together with the RCL, was applied to improve the
0.6480 0.0400 −0.1512 −0.0518
   
criterion of Seuret et al. (2015), and in Chen et al. (2016), the FMBI
A= , Ad = . (40)
including the WBI was used to improve the WBI-based criterion 0.1200 0.6540 0.0259 −0.1091
(Seuret et al., 2015). Both improved criteria require the increase
of the number of decision variables. On the contrary, Theorem 1 This example is recalled from the recently published literature
improves the criterion in Seuret et al. (2015) but does not require (Chen et al., 2016). As mentioned in Remarks 1 and 2, the proposed
additional decision variables. inequality and both the AFBI and the FMBI can be considered as
 C.-K. Zhang et al. / Automatica 74 (2016) 10–15 15

Table 1
The MAUBs and the NDVs for different criteria (Example 1).
Methods h1 NDVs (n = 2)
2 4 6 9 11

Gao and Chen (2007), He et al. (2008), Huang and Feng (2010), Kwon et al. (2012), <20 <20 ≤ 20 ≤ 21 <22
Meng et al. (2010), Peng (2012), Shao and Han (2011), Zhang et al. (2008),
Feng et al. (2015, Theorem 4l=3 ) 21 21 21 22 23 7.5n2 + 4.5n
Kwon et al. (2013, Theorem 2) 22 22 22 22 23 27n2 + 9n
Seuret et al. (2015, Theorem 5) 20 21 21 22 23 10.5n2 + 3.5n
Nam, Pathirana, and Trinh (2015, Remark 4) 20 21 21 22 23 10.5n2 + 3.5n
Nam, Trinh, and Pathirana (2015, Remark 6) 20 21 21 22 23 20.5n2 + 5.5n
Nam, Trinh, and Pathirana (2015, Theorem 1) 20 21 21 22 23 29.5n2 + 8.5n
Chen et al. (2016, Theorem 1) 21 22 22 23 23 78.5n2 + 12.5n

Theorem 1 21 22 22 23 24 10.5n2 + 3.5n

Table 2
The MAUBs and the NDVs for different criteria (Example 2).
Methods h1 NDVs (n = 2)
5 7 11 13 20

Seuret et al. (2015, Theorem 5) 20 22 25 27 34 10.5n2 + 3.5n

Nam, Trinh, and Pathirana (2015, Theorem 1) 20 22 26 28 34 29.5n2 + 8.5n
Chen et al. (2016, Theorem 1) 21 22 26 27 34 78.5n2 + 12.5n

Theorem 1 21 22 26 28 35 10.5n2 + 3.5n

improvements of the WBI. In this example, Theorem 1 proposed Huang, H., & Feng, G. (2010). Improved approach to delay-dependent stability
in this note is compared with the criteria obtained through the analysis of discrete-time systems with time-varying delay. IET Control Theory
& Applications, 4(10), 2152–2159.
WBI (Seuret et al., 2015), the AFBI (Nam, Trinh, & Pathirana, 2015), Kim, S. (2015). Further results on stability analysis of discrete-time systems with
and the FMBI (Chen et al., 2016). The MAUBs calculated by those time-varying delays via the use of novel convex combination coefficients.
criteria, together with the corresponding NDVs, are summarized Applied Mathematics and Computation, 261, 104–113.
Kwon, O., Park, M., Park, J., Lee, S., & Cha, E. (2012). Improved robust stability criteria
in Table 2. Theorem 1 provides less conservative results but keeps for uncertain discrete-time systems with interval time-varying delays via new
the same NDVs in comparison with the WBI-based criterion (Seuret zero equalities. IET Control Theory & Applications, 6(16), 2567–2575.
et al., 2015), and it leads to better results than the FMBI-based Kwon, O. M., Park, M. J., Park, J. H., Lee, S. M., & Cha, E. J. (2013). Stability
and stabilization for discrete-time systems with time-varying delays via
criterion (Chen et al., 2016) and the AFBI-based criterion (Nam, augmented Lyapunov-Krasovskii functional. Journal of the Franklin Institute,
Trinh, & Pathirana, 2015) with requiring smaller NDVs. It means 350(3), 521–540.
that Theorem 1 has the potential to reduce the conservatism of Meng, X., Lam, J., Du, B., & Gao, H. (2010). A delay-partitioning approach to the
stability analysis of discrete-time systems. Automatica, 46(3), 610–614.
the WBI-based criteria without introducing much extra decision Nam, P. T., Pathirana, P. N., & Trinh, H. (2015). Discrete wirtinger-based inequality
variables. and its application. Journal of the Franklin Institute, 352(5), 1893–1905.
Nam, P. T., Trinh, H., & Pathirana, P. N. (2015). Discrete inequalities based on
multiple auxiliary functions and their applications to stability analysis of
6. Conclusions time-delay systems. Journal of the Franklin Institute, 352, 5810–5831.
Park, P., Ko, J., & Jeong, C. (2011). Reciprocally convex approach to stability of
This note has proposed a novel summation inequality by systems with time-varying delays. Automatica, 47(1), 235–238.
Peng, C. (2012). Improved delay-dependent stabilisation criteria for discrete
considering two summation terms appearing in the forward systems with a new finite sum inequality. IET Control Theory & Applications, 6(3),
difference of the LKF together. Compared with the recently 448–453.
Seuret, A., Gouaisbaut, F., & Fridman, E. (2015). Stability of discrete-time systems
reported inequality derived by the WBI and the RCL, the proposed with time-varying delays via a novel summation inequality. IEEE Transactions
one reduces the estimation gap while requires the same NDVs. It is on Automatic Control, 60(10), 2740–2745.
a new way to reduce the conservatism caused by the inequality Shao, H., & Han, Q. L. (2011). New stability criteria for linear discrete-time systems
with interval-like time-varying delays. IEEE Transactions on Automatic Control,
based estimation. Application of this inequality to the linear 56(3), 619–625.
discrete-time system with a time-varying delay has lead to a Xu, S., Lam, J., Zhang, B., & Zou, Y. (2014). A new result on the delay-dependent
relaxed stability criterion. Two numerical examples have been stability of discrete systems with time-varying delays. International Journal of
Robust and Nonlinear Control, 24(16), 2512–2521.
given to demonstrate the advantages of the proposed method. Zeng, H. B., He, Y., Wu, M., & She, J. H. (2015). Free-matrix-based integral inequality
for stability analysis of systems with time-varying delay. IEEE Transactions on
Automatic Control, 60, 2768–2772.
References Zeng, H. B., He, Y., Wu, M., & She, J. H. (2016). New results on stability analysis for
systems with discrete distributed delay. Automatica, 60(10), 189–192.
Chen, J., Lu, J., & Xu, S. (2016). Summation inequality and its application to Zhang, X. M., & Han, Q. L. (2015). Abel lemma-based finite-sum inequality and
stability analysis for time-delay systems. IET Control Theory & Applications, its application to stability analysis for linear discrete time-delay systems.
10(4), 391–395. Automatica, 57, 199–202.
Feng, Z., Lam, J., & Yang, G. H. (2015). Optimal partitioning method for stability Zhang, C. K., He, Y., Jiang, L., Wu, M., & Zeng, H. B. (2016). Delay-variation-dependent
analysis of continuous/discrete delay systems. International Journal of Robust stability of delayed discrete-time systems. IEEE Transactions on Automatic
and Nonlinear Control, 25(4), 559–574. Control, http://dx.doi.org/10.1109/TAC.2015.2503047.
Gao, H., & Chen, T. (2007). New results on stability of discrete-time systems Zhang, J., Peng, C., & Zheng, M. (2016). Improved results for linear discrete-time
with time-varying state delay. IEEE Transactions on Automatic Control, 52(2), systems with an interval time-varying input delay. International Journal of
328–334. Systems Science, 47(2), 492–499.
He, Y., Wu, M., Liu, G. P., & She, J. H. (2008). Output feedback stabilization Zhang, B., Xu, S., & Zou, Y. (2008). Improved stability criterion and its applications
for a discrete-time systems with a time-varying delay. IEEE Transactions on in delayed controller design for discrete-time systems. Automatica, 44(11),
Automatic Control, 53(10), 2372–2377. 2963–2967.