Automatica 71 (2016) 281–291

Contents lists available at ScienceDirect

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

Brief paper

Razumikhin and Krasovskii stability theorems for time-varying

time-delay systems✩
Bin Zhou a,1 , Alexey V. Egorov b
a
Center for Control Theory and Guidance Technology, Harbin Institute of Technology, P.O. Box 416, Harbin, 150001, China
b
Department of Control Theory, St. Petersburg State University, 7/9 Universitetskaya nab., St. Petersburg, 199034, Russia

article info abstract

Article history: The main results of the paper are generalizations of the Razumikhin and of the Krasovskii classical stability
Received 1 August 2015 theorems for stability analysis of time-varying time-delay systems. The condition of negativity of the time-
Received in revised form derivative of Razumikhin functions and Krasovskii functionals is weakened. This is achieved by using the
20 January 2016
notion and properties of uniformly stable functions. We also show how to apply the results to the stability
Accepted 16 April 2016
Available online 3 June 2016
analysis of linear time-varying time-delay systems of retarded type. Both the system matrices and time-
delays are allowed to be time-varying. Some constructive sufficient stability conditions are obtained and
Keywords:
their effectiveness is demonstrated by some examples.
Time-varying systems © 2016 Elsevier Ltd. All rights reserved.
Time-varying delays
Krasovskii theorems
Razumikhin theorems
Uniformly stable functions

1. Introduction solutions of the system. By the Razumikhin theorem we need a

It is well known that there are two ways to extend the Lyapunov
direct method to time-delay systems. The first way is the Razu-
mikhin approach, the second one is the Krasovskii approach (Hale,
1977). Both of them have been successfully used for the stabil-
ity analysis and stabilization of time-invariant time-delay sys-
tems (see, for example, Hale, 1977; Kharitonov, 2013; Kojima,
Uchida, Shimemura, & Ishijima, 1994; Zhang, Liu, & Feng, 2013),
impulse time-delay systems (Chen & Zheng, 2011), functional dif-
ference systems (Melchor-Aguilar, Kharitonov, & Lozano, 2010;
Pepe, 2014), and also allow to obtain some sufficient stability
conditions for time-delay systems with time-varying coefficients
and/or time-varying delays (see, for instance, Cacace, Conte, & Ger-
mani, 2016; Egorov & Mondié, 2015; Fridman & Orlov, 2009; Gu,
Kharitonov, & Chen, 2003; Kharitonov & Niculescu, 2003; Mazenc
& Malisoff, 2016). To guarantee asymptotic stability of a system by
the Lyapunov–Krasovskii theorem we have to find a positive defi-
nite functional with a negative definite time derivative along the
✩ The material in this paper was presented at the 28th Chinese Control
and Decision Conference, May 28–30, 2016, Yinchuan, China. This paper was
recommended for publication in revised form by Associate Editor Emilia Fridman
under the direction of Editor Ian R. Petersen.
E-mail addresses: binzhoulee@163.com, binzhou@hit.edu.cn (B. Zhou),
alexey.egorov@spbu.ru (A.V. Egorov).
1 Corresponding author.
http://dx.doi.org/10.1016/j.automatica.2016.04.048
0005-1098/© 2016 Elsevier Ltd. All rights reserved.
positive definite function whose time-derivative is also negative
definite under the Razumikhin condition (Hale, 1977).
The asymptotic stability analysis of time-varying system, for
instance, for linear time-varying (LTV) system, is challenging. To
see this, we notice that the stability analysis of LTV system has
been listed as the first open problem in mathematical systems and
control theory (see Aeyels & Peuteman, 1999). Even for a particular
LTV system with periodic varying coefficients, the analysis and
design are far from trivial, and thus some fundamental problems
such as controllability (Savkin, 1997), stability analysis (Zhou,
2016) and stabilization (Savkin & Petersen, 2000), have received
considerable attention in the literature (see Rugh, 1996; Zhou,
2016 and the references therein). In recent years there has
been an increasing interest on the analysis and control of LTV
systems with (time-varying) delays. For example, the classical
predictor feedback for linear systems with constant delays has
been extended to systems with time-varying delays and/or time-
varying coefficients in Krstic (2010), Mazenc and Malisoff (2016)
and Mazenc, Malisoff, and Niculescu (2014, 2015), and a finite
dimensional predictor feedback referred to as truncated predictor
feedback initially proposed in Zhou, Lin, and Duan (2012) has
been extended to LTV systems with (time-varying) delays in Zhou
(2014).
However, as was shown in Zhou (2016) for LTV systems
described by delay free differential equations, the condition of
negativity of the time-derivative of the Lyapunov function is quite
282 B. Zhou, A.V. Egorov / Automatica 71 (2016) 281–291
conservative and should be weakened. This is also the case for
time-delay systems. An intuitive understanding is this: a scalar
LTV (time-delay) system can still be asymptotically stable even
if it is unstable in a (short) period within which a time-invariant
quadratic Lyapunov function(al), whose time-derivative is strictly
decreasing, may not exist. To overcome such a shortcoming, one
may use the results in Malisoff and Mazenc (2009) to relax
the condition of negativity of the time-derivative of Lyapunov
functions. Very recently, by using the notion of stable functions,
it was able to weaken the condition of negativity of the time-
derivative of the Lyapunov function in Zhou (2016) for delay free
LTV system (one of the main results related with this paper is cited
as Lemma 5 later).
In this contribution, motivated by the first author’s work (Zhou,
2016), we show how to relax the condition of negativity of the
time-derivative of Lyapunov function(al)s for time-varying time-
delay systems. The main results of the paper are generalizations
of the Razumikhin and Krasovskii stability theorems to the time-
varying setting. In the proposed generalized Razumikhin and
Krasovskii stability theorems, we no longer require that the
time-derivatives of the Razumikhin functions and the Krasovskii
functionals are negative definite for all t. It is shown that the
obtained theorems allow us to deduce some constructive delay-
independent and delay-dependent stability conditions for LTV
systems with both time-varying coefficients and time-delays.
Another motivating paper for our research is Mazenc et al.
(2015), where some stability conditions for LTV time-delay
systems were obtained. The idea of the approach in Mazenc et al.
(2015) is to extend the Halanay lemma (Halanay, 1966) to the
time-varying case. The proposed conditions do not impose any
constraints on the maximal value of the system parameters such as
the delay and the norms of system coefficients. Instead, these are
with some constraints on the integrals of the system parameters
(functions). As noticed in Mazenc et al. (2015), these results cannot
be obtained by using the classical Razumikhin and Krasovskii
approaches. In this paper we show by examples that our proposed
new Razumikhin and Krasovskii stability theorems can obtain
some similar stability conditions that are less conservative than
the results of Mazenc et al. (2015). To the best of our knowledge,
for the proposed Razumikhin theorem, the only comparable result
allowing indefinite time-derivatives of Razumikhin functions was
given in Ning, He, Wu, and She (2014), which generalizes the
results (Ning, He, Wu, Liu, & She, 2012) to time-delay systems.
It will be clear that the result is more conservative than the one
presented in our paper (see Remark 3 given later for details).
The paper is divided into five sections. Section 2 is devoted to
the stability analysis of general nonlinear time-varying time-delay
systems of retarded type by providing generalized Razumikhin and
Krasovskii theorems. In Section 3 we show how to apply the results
to the stability analysis of linear systems with both time-varying
coefficients and time-varying delays. In Section 4 we give three
numerical examples. Section 5 concludes the paper.
2. Razumikhin and Krasovskii stability theorems
2.1. Preliminaries and motivation
Throughout this paper, if not specified, the symbol J denotes
the set [t # , ∞) with t # being some constant. Consider the time-
varying (TV) time-delay system
ẋ(t ) = f (t , xt ) , xt0 = φ, t , t0 ∈ J , t ≥ t0 ,
where f (t , φ) : J × C([−τ , 0], Rn ) → Rn is a known function that
is piecewise continuous with respect to t, locally Lipschitz in φ , and
f (t , 0) = 0. The restriction of the solution x(t ) of system (1) to the
interval [t − τ , t ] is denoted by xt : θ → x(t + θ ), θ ∈ [−τ , 0],
where τ is the delay. In this paper, PC(L, Rn ) is the space of Rn -
valued piecewise continuous functions defined on L, and Jτ =
[t # − τ , ∞).
The TV time-delay system (1) is said to be globally uniformly
asymptotically stable (GUAS), if there exists a KL-function κ such
that
|x(t )| ≤ κ xt0  , t − t0 , t , t0 ∈ J , t ≥ t0 ,
  
where | · | refers to the usual Euclidean norm and ∥xt ∥ =
sups∈[−τ , 0] |x(t + s)|. It is said to be globally uniformly exponentially
stable (GUES) if there exist two constants α > 0 and β > 0 such
that
|x(t )| ≤ β e−α(t −t0 ) xt0  , t , t0 ∈ J , t ≥ t0 .
 
The following two theorems are well known.
Lemma 1 (Razumikhin Theorem, Th. 4.2 in Hale, 1977). The TV time-
delay system (1) is GUAS if there exist a continuous function V (t , x),
t ∈ Jτ , x ∈ Rn , three K∞ -functions u, v , w and a continuous
nondecreasing function q(s) > s for s > 0, such that the following
two conditions are met for all t ∈ J:
u (|x|) ≤ V (t , x) ≤ v (|x|) , x ∈ Rn ,
V̇ (t , x(t )) ≤ −w (|x(t )|)
if V (t + s, x(t + s)) ≤ q (V (t , x(t ))) , ∀ s ∈ [−τ , 0] . (2)
Lemma 2 (Krasovskii Theorem, Th. 2.1 in Hale, 1977). The TV time-
delay system (1) is GUAS if there exist a continuous functional V (t , φ),
t ∈ J, φ ∈ PC([−τ , 0], Rn ), and three K∞ -functions u, v , w such
that the following two conditions are met for all t ∈ J:
u (|φ(0)|) ≤ V (t , φ) ≤ v (∥φ∥) , φ ∈ PC([−τ , 0], Rn ),
V̇ (t , xt ) ≤ −w (|x(t )|) .
We notice that in the Krasovskii stability theorem the time-
derivative of the Krasovskii functional V (t , xt ) is required to
be negative definite, and in the Razumikhin stability theorem
the time-derivative of the Razumikhin function V (t , x(t )) is also
required to be negative definite under the Razumikhin condition
V (t + s, x(t + s)) ≤ q (V (t , x(t ))) , ∀ s ∈ [−τ , 0]. As noticed
in Mazenc et al. (2015), these conditions are either very strong or
difficult to meet in practice (this is also the case for delay free TV
systems as pointed out in Zhou (2016)). In the following we use an
example to demonstrate the situation.
Example 1. Consider the scalar time-delay system
ẋ(t ) = −x(t ) + b(t )x(t − h(t )), (3)
where b(t ) and h(t ) ≥ 0 are scalar functions. Such a system has
been previously studied in many references (see, for example, p.
129 in Hale (1977) and Mazenc et al. (2015)). Let V (x(t )) = x2 (t ).
Then it was shown in Hale (1977) that, by using the classical
Razumikhin theorem (Lemma 1), condition (2) is satisfied (or the
system is GUAS) if |b(t )| < 1. Now consider a periodic function
b(t ) with period ω = 1, and
0, t ∈ [0, c ),

b(t ) = (4)
d, t ∈ [c , 1),
(1)
where c ∈ (0, 1) and d > 0 is some constant (Mazenc et al., 2015).
If the classical Razumikhin theorem is applied, it follows from the
above discussion that the system is GUAS if
d < 1. (5)
 B. Zhou, A.V. Egorov / Automatica 71 (2016) 281–291 283

This condition is very restrictive. Recently, if h is a constant and (3) If µ(t ) is periodic, namely, µ(t ) = µ(t + ω) where ω > 0 is a
h = 1, by extending the Halanay lemma to time-varying systems, constant, then ω ∈ Ωµ .
it was shown in Mazenc et al. (2015) that the system is GUAS if (4) Ωµ = (0, ∞) if and only if µ(t ) < 0 for almost all t ∈ J.
Proof. Proof of Item 1: For any t ∈ J, we have
1 − e− 1
(1 − c )d < ≈ 0.3161, (6)  t +kT k−1  ti + T
2 
µ(s)ds = µ(s)ds < 0,
which is better than (5) in the sense that d can be arbitrarily large, t i=0 ti
as long as 1 − c is sufficiently small.
where ti = t + iT ∈ J. Hence, by definition, kT ∈ Ωµ .
Motivated by this example, as well as the existing results Proof of Item 2: This is a consequence of Lemma 3.
in Mazenc et al. (2015) and Zhou (2016), in this paper we are Proof of Item 3: Let k∗ > 0 be an integer such that −ε k∗ ω + δ < 0,
interested in providing new Razumikhin and Krasovskii stability where ε > 0 and δ ≥ 0 satisfy (8). Then
theorems for TV time-delay systems by weakening the strong ∗
 t0 +ω k −1  t0 +(i+1)ω
conditions on the time-derivatives of Razumikhin functions and 1 
µ(s)ds = µ(s)ds
Krasovskii functionals. The obtained results can be seen as t0 k∗ i=0 t0 +iω
extensions of those in Zhou (2016) to time-delay systems.  t0 +k∗ ω
1 1 
µ(s)ds ≤ −εk∗ ω + δ < 0,

=
2.2. Uniformly stable functions k∗ t0 k∗
which implies that ω ∈ Ωµ .
We first recall the following result from Zhou (2016) regarding Proof of Item 4: If µ(t ) < 0 for almost all t ∈ J we can show that
uniformly stable functions. ε ∈ Ωµ for any ε > 0, i.e., Ωµ = (0, ∞). To show the converse,
we assume that there exists an interval [t1 , t2 ] ⊂ J, t2 > t1 , such
Definition 1. A function µ ∈ PC(J , R) is said to be a uniformly that µ(t ) ≥ 0, t ∈ [t1 , t2 ]. Then we set T = t2 − t1 to get
 t1 + T
stable function (USF) if the following linear time-varying (LTV) µ(s)ds ≥ 0, which implies T ̸∈ Ωµ , i.e., Ωµ ̸= (0, ∞). 
t 1
equation is GUAS:
The following result is then a consequence of Item 3 of
ẏ(t ) = µ(t )y(t ), t ∈ J. Proposition 1.

It is well known that an LTV system is GUAS if and only if it Corollary 1. Assume that µ ∈ PC(J , R) is a periodic function,
is GUES (see, for example, Theorem 6.13 in Rugh, 1996). Hence, it namely, µ(t ) = µ(t + ω), where ω > 0 is some constant. Then
follows from µ(t ) is a USF if and only if
 ω
t
χµ , µ(s)ds < 0.
 
y(t ) = exp µ(s)ds y(t0 ), t , t0 ∈ J , t ≥ t0 , (7) 0
t0

that µ(t ) is a USF if and only if there exist two constants ε > 0 and It is worthy of mention that Ωµ may be not a connected set. An
δ ≥ 0 such that example is given below.
 t Example 2. Consider a periodic function µ(t ) = µ(t + 1) defined
µ(s)ds ≤ −ε(t − t0 ) + δ, t , t0 ∈ J , t ≥ t0 . (8) below
t0
3l, t ∈ [0, 1/2) ,

µ(t ) =
−4l, t ∈ [1/2, 1) ,
Lemma 3 (Zhou, 2016). The function µ(t ) is a USF if and only if, for 1
any given c > 0, there exists T ∗ (c ) > 0 such that where l > 0 is a constant. This function is a USF since 0
µ(s)ds =
 t +T  t +θ
− 12 l < 0 (by Corollary 1). By Proposition 1, {1, 2, . . .} ⊂ Ωµ .
3
µ(s)ds ≤ −c , µ(s)ds ≤ d, θ ∈ [0, T ], (9) However, as 02 µ(s)ds = l > 0, we know that 23 ̸∈ Ωµ . Hence,
t t
Ωµ is not a connected set.
hold for any finite T ≥ T ∗ (c ), t ∈ J, and some d ∈ R.
We next give another definition.
Clearly, if the integral of µ(t ) is bounded, then the second
Definition 3. Let µ(t ) be a USF. Then for any given T ≥ 0 the
inequality in condition (9) can be removed. Motivated by the first
overshoot ϕµ (T ) of µ(t ) is defined as
inequality in condition (9), we give the following definition.
t +θ
  
Definition 2. Let µ(t ) be a USF. Then the set ϕµ (T ) = sup max µ(s)ds .
t ∈J θ∈[0,T ] t
t +T
   
Ωµ = T > 0 : sup µ(s)ds < 0 , If we let t0 ∈ J and θ0 ∈ [0, T ] be such that ϕµ (T ) =
t ∈J t  t0 +θ0
t0
µ(θ )dθ , then it follows from (7) that
is said to be the uniform convergence set (UCS) of µ(t ).   t +θ 
 0 0 µ(s)ds
|y(t0 + θ0 )| = e y(t0 ) = eϕµ (T ) |y(t0 )| ,

t0
We have the following properties of Ωµ .


which just indicates that eϕµ (T ) is exactly the overshoot of the

Proposition 1. Let µ(t ) be a USF and Ωµ be the associate UCS. Then
solution y(t ) (note that the notation here is slightly abused, as we
(1) If T ∈ Ωµ , then kT ∈ Ωµ for any positive integer k. use both eϕµ (T ) and ϕµ (T ) to denote the overshoot). We next give
(2) There exists a positive T ∗ such that (T ∗ , ∞) ⊂ Ωµ . some properties of ϕµ (T ).
284 B. Zhou, A.V. Egorov / Automatica 71 (2016) 281–291

Proposition 2. Let µ(t ) be a USF and ϕµ (T ) be the overshoot for a Proof. We consider the inequality
given T ≥ 0. Then
y(s) ≥ ψ (z (s)) , (15)
(1) ϕµ (T ) is a nondecreasing function of T .
(2) 0 ≤ ϕµ (T ) ≤ δ < ∞, where δ is from inequality (8). and for any t ∈ J we consider the following two cases:
(3) ϕµ (T ) = 0 if and only if µ(t ) ≤ 0 holds true for almost all t ∈ J . Case (A): Inequality (15) holds for all s ∈ [t − T , t ].
(4) ϕµ (T ) can be bounded by Case (B): Inequality (15) does not hold for some s ∈ [t − T , t ].
 t +T  In the Case (A), by (13) we know that
ϕµ (T ) ≤ sup max {µ(s), 0} ds
t
 
t ∈J t
 t +T
 y(t ) ≤ y (t − T ) exp µ(s)ds . (16)
≤ sup |µ(s)| ds . (10)
t −T

t ∈J t Now consider Case (B), which implies that a measure of the set
(5) If µ(t ) is periodic, namely, µ(t ) = µ(t + ω) where ω > 0, then {s ∈ [t − T , t] : y(s) < ψ (z (s))} is positive. Denote t ∗ =
ϕµ (ω) can be computed as sup {s ∈ [t − T , t ] : y(s) < ψ(z (s))}. Then we have either t ∗ < t
  or t ∗ = t. If t ∗ < t, then for all s ∈ [t ∗ , t ] we have y(s) ≥ ψ (z (s)),
which implies ẏ(s) ≤ µ(s)y(s), from which it follows that
I0 + Iω
 t2   t1
ϕµ (ω) = max I t1 , ,

max max t2 (11)
0≤t1 ≤t2 ≤ω 0≤t1 ≤t2 ≤ω  t

y(t ) ≤ y t µ(s)ds
 ∗
t  t2 exp
where It21 = t1
µ(s)ds. t∗
 t 
Proof. The proofs for Items 1–4 are simple. We only need to prove = ψ z t ∗ exp µ(s)ds
  
Item 5. By definition and the periodicity of µ(t ), we have t∗

≤ sup {ψ (z (s))} exp ϕµ (T ) .

 
(17)
  
t +θ
ϕµ (ω) = sup max It11 s∈[t −T ,t ]
t1 ∈[0,ω] θ∈[0,ω]
     If t ∗ = t, then by the definition of t ∗ we have
t +θ t +θ
= sup max max It11 , max It11
y(t ) = y t ∗ ≤ ψ z t ∗
    
≤ sup {ψ (z (s))}
t1 ∈[0,ω] θ∈[0, ω−t1 ] θ∈[ω−t1 , ω]
 s∈[t −T ,t ]
 
≤ sup {ψ (z (s))} exp ϕµ (T ) .
 t2  t +ω
 
= sup max max It1 , max It21 (18)
t1 ∈[0,ω] t2 ∈[t1 , ω] t2 ∈[0, t1 ] s∈[t −T ,t ]
 
Combining (16)–(18) gives the result. 
Iω2 +ω + Iω
 t2 
It1 , max
t 
= sup max max t1
t1 ∈[0,ω] t2 ∈[t1 , ω] t2 ∈[0, t1 ]
  The idea in the proof of Lemma 4 is rather standard (see, for
 t2   t2 ω example, the proof of Lemma IV.2 of Angeli, Sontag, and Wang
I t1 , .

= max max max I 0 + I t1
0≤t1 ≤t2 ≤ω 0≤t2 ≤t1 ≤ω (2000), the proof of Theorem 1 in Hou and Qian (1998), and the
proof of Lemma 1 in Ning et al. (2012)). For this lemma, we mention
Thus, we get (11).  that, differently from the conventional comparison principle (see,
We can get immediately the following corollary from Item 5 of for example, Lemma IV.2 in Angeli et al. (2000) and Lemma 1 in
Proposition 2. Teel (1998)) where µ(t ) is required to be a negative constant, and
also differently from the comparison principle in Ning et al. (2012)
Corollary 2. If µ(t ) is periodic with period ω and there is an interval (Lemma 1), where the variable y on the right hand side of (14)
(t1 , t2 ) ⊂ [0, ω] such that µ(t ) ≥ 0, t ∈ (t1 , t2 ), and µ(t ) ≤ 0, is only a function of t0 , which is independent of t, we here allow
∀t ∈ [0, t1 ] ∪ [t2 , ω], then µ(t ) to be any USF and allow the variable y on the right hand side
 t2 of (14) to be a function of t, which is essential in our consequent
development (see Eq. (20).).
ϕµ (ω) = µ(s)ds. (12)
t1 With the help of the above lemma, we can prove the following
result which is the TV version of the Razumikhin-type stability
theorems.
2.3. The Razumikhin stability theorem
Theorem 1. The TV time-delay system (1) is GUAS if there exist a
To present the Razumikhin theorem for TV time-delay systems, continuous function V (t , x), t ∈ Jτ , x ∈ Rn , a USF µ(t ), three K∞ -
we need the following lemma, which can be regarded as the TV functions u, v, q, a number T ∈ Ωµ , and a constant ρ ∈ (0, 1)
comparison principle. satisfying

Lemma 4. Let y(t ) and z (t ) : J → R+ be continuous functions such

 
ρ
that q s ≥ s, s ≥ 0, (19)
exp ϕµ (T )
 
ẏ(t ) ≤ µ(t )y(t ) whenever y(t ) ≥ ψ (z (t )) , (13)
such that the following conditions are met for all t ∈ J:
where µ(t ) is a USF and ψ(·) is a K -function. Then, for any given
constant T > 0 and any t ≥ T + t # , there holds (A) u (|x|) ≤ V (t , x) ≤ v (|x|) , x ∈ Rn ,
   t (B) V̇ (t , x(t )) ≤ µ(t )V (t , x(t ))
y(t ) ≤ max y(t − T ) exp µ(s)ds ,
t −T
if V (t + s, x(t + s)) ≤ q (V (t , x(t ))) , ∀s ∈ [−τ , 0] .
If, in addition, there exist three positive numbers η and αi , i = 1, 2

sup {ψ (z (s))} exp ϕµ (T ) .
 
(14)
t −T ≤s≤t
such that u(s) = α1 sη and v(s) = α2 sη , then the system is GUES.
 B. Zhou, A.V. Egorov / Automatica 71 (2016) 281–291 285

Proof. We choose y(t ) = V (t , x(t )) , V (t ), ψ(s) = q−1 (s) and know from Item 3 of Proposition 2 that ϕµ (T ) = 0 for any T .
z (t ) = sups∈[−τ ,0] {V (t + s)}. Then condition (B) is just in the form Consequently, condition (19) becomes q (ρ s) ≥ s, s ≥ 0, which
of (13) and it follows from Lemma 4 that, for any constant T > 0 is equivalent to q(s) ≥ ρ1 s. If q(s) is a linear function (this is
and t ≥ T + t # , commonly used in practice), q(s) ≥ ρ1 s is equivalent to q(s) > s,
t
  
which is exactly the condition required by the classical Razumikhin
V (t ) ≤ max V (t − T ) exp µ(s)ds , theorem in Lemma 1.
t −T
   
q−1 sup sup {V (θ + s)} exp(ϕµ (T )) Remark 2. If we set the function q(s) as a linear function, say,
t −T ≤θ≤t s∈[−τ ,0] q(s) = q s, where q is a constant, then condition (19) becomes
  t 
q≥
(
exp ϕµ (T ) ) , or equivalently,
ρ
≤ max V (t − T ) exp µ(s)ds ,
t −T
q > exp ϕµ (T ) ,
 
   (22)
q−1
sup {V (t + s)} exp(ϕµ (T ))
−T ∗ ≤s≤0 which is very easy to use in practice. Moreover, if the estimate (8)
is known, according to the discussion in the above subsection, we
where T ∗ = T + τ . It follows from (19) that ρ s ≥ q−1 (s) can choose T such that T > δ/ε and ϕµ (T ) = δ . However, in some
t
exp(ϕµ (T )). Hence, by letting fµ (t ) = e t −T µ(s)ds , the above cases, ϕµ (T ) may be easier to compute than δ (see the examples in
inequality can be continued as Section 4).
 
V (t ) ≤ max V (t − T ) fµ (t ), sup {V (t + s)} ρ Remark 3. To the best of our knowledge, the only comparable
−T ∗ ≤s≤0 result in the literature was given in Ning et al. (2014) where a
{V (t + s)} max fµ (t ), ρ . similar condition as (B) was used in its new Razumikhin stability
 
≤ sup
−T ∗ ≤s≤0 theorem (see Corollary 2 in Ning et al. (2014)). However, the results
  in Ning et al. (2014) require
t
Since T ∈ Ωµ , we know that fµ (t ) = exp t −T
µ(s)ds ≤ ρT < 1,  ∞
#
∀t ≥ T + t . Then, for all t ≥ T + t , # max{µ(s), 0}ds < ∞,
0
V (t ) ≤ sup {V (t + s)} max {ρT , ρ} which cannot be satisfied even if µ(t ) is periodic and µ(t ) >
−T ∗ ≤s≤0
0, ∀t ∈ [a, b] ⊆ [0, ω] where ω is the period and b > a (see the
=ρ ∗
sup {V (t + s)} , (20) numerical examples in Mazenc et al. (2015) and also in Sections 4.1
−T ∗ ≤s≤0
and 4.2). Finally, we mention that, as done in Ning et al. (2014)
where ρ ∗ = max{ρ, ρT } < 1. Hence, by Lemma 1 in Mazenc et al. and Teel (1998), Theorem 1 can also be generalized to the input-
(2015), we obtain from (20) that, for all t ≥ t0 ≥ T + t # , to-state stability framework for nonlinear (functional) differential
systems, which will be reported elsewhere.
ln ρ ∗
 
V (t ) ≤ exp (t − t0 ) sup {V (t0 + s)}
T∗ −T ∗ ≤s≤0
2.4. The Krasovskii stability theorem
ln ρ ∗
 
≤ exp (t − t0 ) sup {v (|x(t0 + s)|)}
T∗ −T ∗ ≤s≤0 The following theorem generalizes a result in Zhou (2016) to

ln ρ ∗
 time-delay systems, and can be regarded as the TV version of the
(t − t0 ) v xt0  ,
 
≤ exp (21) Krasovskii stability theorem for time-delay systems.
T∗
which, in view of (A), implies Theorem 2. The TV time-delay system (1) is GUAS if there exist a
continuous functional V (t , φ), t ∈ J, φ ∈ PC([−τ , 0], Rn ), a USF
ln ρ ∗
  
|x(t )| ≤ u−1 v xt0  exp ( ) , µ(t ), and two K∞ -functions u and v such that, for all t ∈ J ,
 
∗
t − t0
T
(a) u (|φ(0)|) ≤ V (t , φ) ≤ v(∥φ∥), φ ∈ PC([−τ , 0], Rn ),
#
where t ≥ t0 ≥ T + t . Hence, the system is GUAS.
If, in addition, u(s) = α1 sη and v(s) = α2 sη , we have from (21) (b) V̇ (t , xt ) ≤ µ(t )V (t , xt ) .
that, for all t ≥ t0 ≥ T + t # , If, in addition, there exist three positive numbers η and αi , i = 1, 2,
1 1 1

1 ln ρ ∗
 1 such that u(s) = α1 sη and v(s) = α2 sη , then the system is GUES.
|x(t )| ≤ V η (t ) ≤ (t − t0 ) α2η xt0 
 
exp
η
1 1
η T∗ Proof. By assumption, there exist ε > 0 and δ ≥ 0 such that
α α1η t
1
 η1 t0
µ(s ) ds ≤ −ε(t − t0 ) + δ, ∀t , t0 ∈ J , t ≥ t0 . Then it follows
α2 xt  exp 1 ln ρ (t − t0 ) ,
∗
  
  from (b) that
=
α1 0
η T∗
V (t , xt ) t t
V̇ (s, xs )
 
which implies that the system is GUES. ln  = d ln V (s, xs ) = ds
 V t0 , xt0 V (s, xs )

t0 t0
 t
Remark 1. We give an explanation of the above theorem. If µ(t ) is ≤ µ(s)ds ≤ −ε(t − t0 ) + δ, ∀t , t0 ∈ J , t ≥ t0 ,
not negative all the time, then for any T > 0, the overshoot ϕµ (T ) t0
is positive (by Item 3 of Proposition 2). As a result, V (t , x(t )) may
be increasing during the interval within which µ(t ) is nonnegative. which, in view of (a), implies
Hence, the scalar q should satisfy (19) so that the overshoot ϕµ (T ) V (t , xt ) ≤ exp (−ε(t − t0 ) + δ) V t0 , xt0
 
is taken into consideration. In the classical Razumikhin theorem
≤ exp (−ε(t − t0 ) + δ) v xt0  .
 
(Lemma 1) µ(t ) is a negative constant and, in this case, we
286 B. Zhou, A.V. Egorov / Automatica 71 (2016) 281–291

Therefore, we can obtain Proof. Proof of Item 1. We choose the function V1 (t , x(t )) =
xT (t )P (t )x(t ) , V1 (t ) whose derivative

|x(t )| ≤ u −1
(V (t , xt ))
exp (−ε(t − t0 ) + δ) v xt0  , 1
−1
  
≤u (23)
V̇1 (t ) ≤

α(t )V12 (t ) + 2xT (t )P (t )B(t )x(t − h)

2V1 (t )
which implies that the system is GUAS. If, in addition, u(s) = α1 sη
and v(s) = α2 sη , we get from (23) that 1   1 
≤ α(t )V12 (t ) + 2V1 (t ) P 2 (t )B(t )x(t − h)
 
2V1 (t )
α2 δ 

 − ε (t −t )
|x(t )| ≤ η
e xt0  e η 0
, ∀t , t0 ∈ J , t ≥ t0 , 1 √
α1 ≤ α(t )V1 (t ) + p2 |B(t )x(t − h)| .
2
namely, system (1) is GUES. The proof is finished. 
Now choose another nonnegative functional
Clearly, Theorem 2 is a generalization of the classical Lya-
punov–Krasovskii stability theorem in Lemma 2 as V̇ (t , xt ) is t

no longer required to be negative definite, which is required in V2 (t , xt ) = f0 (t , θ ) |B(θ + h)x(θ )| dθ , (30)
t −h
Lemma 2.
where f0 (t , θ ) = w2 + θ−
h
t
(w2 − w1 ) in which w1 and w2 are
3. Applications to stability analysis of linear systems some nonnegative numbers. The time-derivative of V2 (t , xt ) can
be evaluated as
In this section, we apply the obtained results to the stability
analysis of LTV time-delay systems. To this end, we recall the V̇2 (t , xt ) = w2 |B(t + h)x(t )| − w1 |B(t )x(t − h)| + f1 (t ),
following result from Zhou (2016), which will be frequently used t
in the sequel, regarding the GUES of the following LTV system where f1 (t ) = − 1h (w2 − w1 ) t −h |B(θ + h)x(θ )| dθ . We choose
√ √
without delay V (t , xt ) = V1 (t , x(t )) + V2 (t , xt ), and w1 = p2 , w2 = p2 /ρ1 .
ẋ(t ) = A(t )x(t ), t ∈ J, (24) Then

where A ∈ PC(J , Rn×n ). 1

V̇ (t , xt ) ≤ α(t )V1 (t ) + w2 |B(t + h)x(t )| + f1 (t )
2
Lemma 5 (Zhou, 2016). The LTV system (24) is GUES if there exist 1 w2 |B(t + h)| √
(α(t ) + η1 ) V1 (t ) + p1 x(t ) + f1 (t )

P (t ) = P T (t ) ∈ C1 (J , Rn×n ), a USF α(t ), and constants p2 ≥ p1 > 0 ≤ √
such that, for all t ∈ J , 2 p1
1 1
AT (t )P (t ) + P (t )A(t ) + Ṗ (t ) ≤ α(t )P (t ), (25) ≤ µ1 (t ) (V1 (t ) + V2 (t , xt )) − µ1 (t )V2 (t , xt ) + f1 (t )
2 2
p1 In ≤ P (t ) ≤ p2 In . (26)  t
1
= µ1 (t )V (t , xt ) − q1 (t , θ) |B(θ + h)x(θ )| dθ .
We mention that, for LTV system, providing a step by step gen- 2 t −h
eral procedure for finding P (t ), is a very difficult problem (Kalman
For t ∈ J, θ ∈ [t − h, t ], the function q1 (t , θ) satisfies
& Bertram, 1960) and, to the best of our knowledge, a satisfactory
solution is not available by now (see Zhou, 2016 for more discus- 1 1
sions). q1 (t , θ) = (w2 − w1 ) + µ1 (t )f0 (t , θ )
h 2
θ −t
 
1 1
3.1. Stability analysis by the Krasovskii theorem ≥ (w2 − w1 ) + (α0 + η1 ) w2 + (w2 − w1 )
h 2 h
We consider in this subsection the following TV time-delay 1 ρ1 − 1
≥ (w2 − w1 ) + w2 = 0.
system h h
ẋ(t ) = A(t )x(t ) + B(t )x(t − h), t ∈ J, (27) Therefore,
where A, B ∈ PC(J , Rn×n ), and h ≥ 0 is a constant. 1
V̇ (t , xt ) ≤ µ1 (t )V (t , xt ) .
Theorem 3. Assume that conditions (25)–(26) in Lemma 5 are 2
satisfied, where A(t ) is defined in system (27). Let α0 = inft ∈J α(t ). As µ1 (t ) is a USF and V (t , xt ) satisfies all the conditions of
Then (27) is GUES if one of the following conditions holds: Theorem 2, system (27) is GUES.
(1) There exists ρ1 ∈ (0, 1) such that
Proof of Item 2. We choose the quadratic function V1 (t , x(t )) =
ρ1 − 1 xT (t )P (t )x(t ) , V1 (t ). Then
η1 , 2 − α0 ≥ 0 (28)
h
V̇1 (t ) ≤ α(t )V1 (t ) + 2xT (t )P (t )B(t )x(t − h)
is satisfied and the function µ1 (t ) defined below is a USF p2
 ≤ α(t )V1 (t ) + η2 xT (t )P (t )x(t ) + |B(t )x(t − h)|2
µ1 (t ) = α(t ) + η1 +
2 p2
|B(t + h)| . η2
ρ1 p1 p2
= (α(t ) + η2 ) V1 (t ) + |B(t )x(t − h)|2 .
(2) There exists ρ2 ∈ (0, 1) such that η2
ρ2 − 1 Now choose another Lyapunov–Krasovskii functional
η2 , − α0 ≥ 0 (29)
h  t

is satisfied and the function µ2 (t ) defined below is a USF V2 (t , xt ) = f0 (t , θ ) |B(θ + h)x(θ )|2 dθ ,
t −h
1 p2
µ2 (t ) = α(t ) + η2 + |B(t + h)|2 . where f0 (t , θ ) is defined as in (30). By choosing V (t , xt ) =
ρ2 η2 p1
V1 (t , x(t )) + V2 (t , xt ) , V (t ), and w1 = p2 /η2 , w2 = p2 /(ρ2 η2 ),
 B. Zhou, A.V. Egorov / Automatica 71 (2016) 281–291 287

we can compute where we have used (25) and (26). Under the condition that V (t +
s) ≤ q3 V (t ), s ∈ [−τ , 0], we obtain
V̇ (t ) ≤ (α(t ) + η2 ) V1 (t ) + w2 |B(t + h)x(t )| + f2 (t ) 2

|B(t + h)|2
  
1 p2
≤ (α(t ) + η2 ) V1 (t ) + w2 V1 (t ) + f2 (t ) V̇ (t ) ≤ α(t ) + q3 |B(t )| V (t ) = µ3 (t )V (t ).
p1 2 p1
 t
= µ2 (t )V (t ) − q2 (t , θ) |B(θ + h)x(θ )|2 dθ , Since µ3 (t ) is a USF and V (t ) satisfies all the conditions in
t −h Theorem 1, system (31) is GUES.
t
where f2 (t ) = − 1h (w2 − w1 ) |B(θ + h)x(θ )|2 dθ and q2 (t , θ)
t −h Proof of Item 2. Consider the function V (t , x(t )) = xT (t )P (t )x(t ) ,
= (w2 − w1 ) + µ2 (t )f0 (t , θ ). Similarly to the proof of Item 1 of
1
h V (t ) whose time-derivative satisfies
this theorem, q2 (t , θ) ≥ 0, as µ2 (t ) ≥ (ρ2 − 1)/h. Hence, we
obtain V̇ (t ) ≤ α(t )V (t ) + 2xT (t )P (t )B(t )x(t − h(t ))
V̇ (t ) ≤ µ2 (t )V (t ). p2
≤ (α(t ) + ρ4 ) V (t ) + |B(t )|2 V (t − h(t )),
System (27) is GUES by Theorem 2. 
ρ4 p1
where we have used (25) and (26). Under the condition that V (t +
Remark 4. The main difference between Item 1 and Item 2 of
s) ≤ q4 V (t ), s ∈ [−τ , 0], we obtain
Theorem 3 is that the function µ1 (t ) contains |B(t + h)| while the
function µ2 (t ) contains |B(t + h)|2 . Hence, if |B(t )| > 1, then Item  
q4 p2
1 may be less conservative than Item 2; otherwise, Item 2 may be V̇ (t ) ≤ α(t ) + ρ4 + |B(t )|2 V (t ) = µ4 (t )V (t ).
more effective than Item 1. ρ 4 p1
Remark 5. Assume that α(t ) = α0 < 0, ∀t ∈ J. Then it is easy to The result then again follows from Theorem 1. 
see that for i = 1, 2 the function µi (t ) is USF if and only if
t
We mention that Remarks 4 and 5 are also applicable to

|B(s)|i ds ≤ βi (t − t0 ) + γi Theorem 4.
t0

hold true for some sufficiently small number βi ≥ 0, yet |B(t )|

itself can be very large during some small time interval (see the
3.2.2. Delay-dependent stability conditions
examples in Section 4).
In this subsection, we will present delay-dependent stability
3.2. Stability analysis by the Razumikhin theorem conditions for system (31). To this end, we consider a more general
class of LTV with distributed delay:
3.2.1. Delay-independent stability conditions  t
In this subsection we consider the following LTV time-delay ẋ(t ) = C (t )x(t ) + (D1 (t , s)x(s)
system t −h(t )

ẋ(t ) = A(t )x(t ) + B(t )x(t − h(t )), t ∈ J, (31) + D2 (t , s)x (s − l(s))) ds, t ∈ J, (32)
where A, B ∈ PC(J , R ), and h : J → [0, τ ] is a scalar piecewise
n×n

continuous function, where τ may be unknown. We can state the where h : J → [0, τ1 ], and l : J → [0, τ2 ] are some scalar
following result regarding delay-independent stability conditions piecewise continuous functions, C ∈ PC(J , Rn×n ), Di ∈ PC(J ×
for system (31). Jτi , Rn×n ), i = 1, 2, where τ1 and τ2 are some constants that may
be unknown, τ = τ1 + τ2 is the delay of the system. We assume
Theorem 4. Assume that conditions (25)–(26) in Lemma 5 are
that the delay free part of (32) is GUES:
satisfied, where A(t ) is defined in system (31). Then (31) is GUES if
one of the following conditions is satisfied:
ẋ(t ) = C (t )x(t ), t ∈ J.
(1) There exists q3 such that µ3 (t ) defined below is a USF and q3 >
exp ϕµ3 (T ) with T ∈ Ωµ3 ,
 Theorem 5. Assume that conditions (25)–(26) in Lemma 5 are
1 p2
µ3 ( t ) = α(t ) + q3 |B(t )| . satisfied, where A(t ) is replaced by C (t ). Then (32) is GUES if one of
2 p1 the following conditions is satisfied:
 ρ4 > 0 such that µ4 (t ) defined below is a USF
(2) There exist q4 and
and q4 > exp ϕµ4 (T ) with T ∈ Ωµ4 ,
 exists q5 such that µ5 (t ) defined below is a USF and q5 >
(1) There
q4 p2 exp ϕµ5 (T ) with T ∈ Ωµ5 ,
µ4 (t ) = α(t ) + ρ4 + |B(t )|2 .
ρ 4 p1 1

p2
 t
µ5 ( t ) = α(t ) + q5 d5 (t , s)ds,
 Proof of Item 1. Consider the nonnegative function V (t , x(t ))
Proof. 2 p1 t −h(t )
= xT (t )P (t )x(t ) , V (t ):
where d5 (t , s) = |D1 (t , s)| + |D2 (t , s)|.
1
V̇ (t ) ≤ α(t )V 2 (t ) + 2xT (t )P (t )B(t )x(t − h(t )) (2) There exist q6 , ρ6 > 0 and η6 >
 0 such that µ6 (t ) defined below
 
2V (t ) is a USF and q6 > exp ϕµ6 (T ) with T ∈ Ωµ6 ,

1 p2 √
≤ α(t )V (t ) + |B(t )|  p1 x (t − h(t ))
  t
q6 p2
2 p1 µ6 (t ) = α(t ) + ρ6 + h( t ) d6 (t , s)ds,
 ρ6 p1 t −h(t )
1 p2
≤ α(t )V (t ) + |B(t )| V (t − h(t )) ,
2 p1 where d6 (t , s) = (1 + η6 ) |D1 (t , s)|2 + (1 + η1 ) |D2 (t , s)|2 .
6
288 B. Zhou, A.V. Egorov / Automatica 71 (2016) 281–291

 Proof of Item 1. Consider the Razumikhin function V (t , x(t ))

Proof.  exists q7 such that µ7 (t ) defined below is a USF and q7 >
(1) There
= xT (t )P (t )x(t ) , V (t ) whose derivative can be estimated as exp ϕµ7 (T ) with T ∈ Ωµ7 ,
√  t
1 1 p2
V̇ (t ) = xT (t ) C T (t )P (t ) + P (t )C (t ) + Ṗ (t ) x(t ) µ7 (t ) = α(t ) + q7 √ |B(t )| d7 (s)ds,
 
2V (t ) 2 p1 t −h(t )

xT (t )P (t ) t
where d7 (s) = |A(s)| + |B(s)|.

+ (D1 x(s) + D2 x (s − l(s))) ds
V (t ) t −h(t )
(2) There exist q8 , ρ8 > 0 and η8 > 0 such that µ8 (t ) defined below
 t is a USF and q8 > exp ϕµ8 (T ) with T ∈ Ωµ8 ,
1 √
≤ α(t )V (t ) + p2 k(t , x(s))ds  t
2 t −h(t )
q8 p2
µ8 (t ) = α(t ) + ρ8 + h(t ) |B(t )|2 d8 (s)ds,
1

p2
 t ρ8 p1 t −h(t )
≤ α(t )V (t ) + k(t , V (s))ds,  
2 p1 t −h(t ) where d8 (s) = (1 + η8 ) |A(s)|2 + 1 + η1 |B(s)|2 .
8
where D1 = D1 (t , s), D2 = D2 (t , s) and k(t , x(s)) = |D1 (t , s)|
|x(s)| + |D2 (t , s)| |x (s − l(s))|. Under the condition V (t + s) ≤
4. Numerical examples
q5 V (t ), s ∈ [−τ , 0], we can derive

V̇ (t ) ≤ µ5 (t )V (t ). 4.1. A system with time-varying coefficients

If µ5 (t ) is a USF, the result follows from Theorem 1.
Consider the time-delay system (3) where b(t ) is a periodic
Proof of Item 2. Consider the Razumikhin function V (t , x(t )) = function given by (4). We use different approaches developed in
xT (t )P (t )x(t ) , V (t ) whose time-derivative can be estimated as this paper to test the stability of this system. Notice that conditions
(25)–(26) are satisfied with P (t ) = 1, p1 = p2 = 1, and α(t ) = −2.
V̇ (t ) = xT (t ) C T (t )P (t ) + P (t )C (t ) + Ṗ (t ) x(t )
 
We first assume that h(t ) is time-varying and derive delay-
 t
independent stability conditions by using Theorem 4. Let
+ 2xT (t )P (t ) (D1 x(s) + D2 x (s − l(s))) ds
t −h(t ) µ3 (t ) = −1 + q3 |b(t )| .
1
≤ (α(t ) + ρ6 ) V (t ) + ψ(t ), (33) This function is USF if and only if
ρ6  1
where D1 = D1 (t , s), D2 = D2 (t , s), and µ3 (t )dt = −1 + q3 (1 − c )d < 0.
 t 0

ψ(t ) ≤ p2 h(t ) |D1 x(s) + D2 x (s − l(s))|2 ds If d < 1, then there exists a q3 > 1 such that µ3 (t ) < 0.
t −h(t ) Consequently, µ3 (t ) is USF for any h and, in view of Remark 1, the
p2 h(t ) t

system is GUES. Hence, without loss of generality, we assume that
≤ (d1 V (s) + d2 V (s − l(s))) ds, d ≥ 1. By (12) we can compute
p1 t −h(t )
   1
where d1 = (1 + η6 ) |D1 |2 and d2 = 1 + η1 |D2 |2 . Under the ϕµ3 (ω) = µ3 (t )dt = −1 + q3 (1 − c )d + c .
6
c
condition V (t + s) ≤ q6 V (t ), s ∈ [−τ , 0], we obtain
Hence, conditions of Item 1 of Theorem 4 hold if and only if there
p2 h(t ) t
exist three numbers d ≥ 1, c ∈ (0, 1), and q3 such that

ψ(t ) ≤ q6 V (t ) (d1 + d2 ) ds,
p1 t −h(t )
0 > −1 + q3 (1 − c )d,


by which the time-derivative of V (t ) in (33) can be simplified as q3 > exp(−1 + q3 (1 − c )d + c ).

V̇ (t ) ≤ µ6 (t )V (t ). One can show that the above condition is equivalent to the

inequality
The result then follows from Theorem 1. 
1
Now we consider the TV time-delay system (31), which can be (1 − c )d < . (35)
exp(c )
written as
 t Since c < 1, we have 1/ exp(c ) > 1/e ≈ 0.3679. Hence, (35)
ẋ(t ) = (A(t ) + B(t )) x(t ) − B(t ) ẋ(s)ds is better than (6). Moreover, if c → 0, condition (35) approaches
t −h(t ) d < 1, which cannot be improved since system (3) is not GUES if
= (A(t ) + B(t )) x(t ) d ≥ 1 and h = 0.
 t We next assume that h(t ) is a constant and use Theorem 3
− B(t ) (A(s)x(s) + B(s)x (s − h(s))) ds, (34) to derive delay-dependent stability conditions. For the considered
t −h(t ) system, we can obtain
which is exactly of the form (32) with C (t ) = A(t ) + B(t ), 2
D1 (t , s) = −B(t )A(s) and D2 (t , s) = −B(t )B(s). Hence, applying
µ1 (t ) = −2 + η1 + |b(t + h)| ,
ρ1
Theorem 5 to system (34) immediately gives the following 1
corollary. µ2 (t ) = −2 + η2 + |b(t + h)|2 ,
ρ2 η2
Corollary 3. Consider system (31). Assume that conditions (25)–(26) where ρ1 , ρ2 ∈ (0, 1),
in Lemma 5 are satisfied, where A(t ) is replaced by A(t ) + B(t ). ρ1 − 1 ρ2 − 1
Then (31) is GUES if one of the following conditions is satisfied: η1 = 2 + 2, and η2 = + 2.
h h
 B. Zhou, A.V. Egorov / Automatica 71 (2016) 281–291 289

The system is GUES if one of the following conditions is met which is periodic with period ω = 2π . Hence, µ7 (t ) is USF if and
only if the following condition is met
1
2(1 − c )d

µ1 (t )dt = −2 + η1 +
0 ρ1  2π  2π

ρ1 − 1 2(1 − c )d µ7 (t )dt = −2π + q7 h(t )dt < 0. (40)

=2 + < 0, (36) 0 0
h ρ1
If l < 1, then h(s) < 1 and thus there exists q7 > 1 such that
(1 − c )d2
 1
µ2 (t )dt = −2 + η2 + µ7 (t ) < 0. Hence, the system is GUES. We assume that l ≥ 1. By
0 ρ2 η2 (10) we have
ρ2 − 1 h(1 − c )d2 2π
< 0,

= + (37)
h ρ2 (ρ2 − 1 + 2h) ϕµ7 (2π ) ≤ |−1 + q7 h(t )| dt
0
and one of the conditions (28) and (29), which are equivalent to  2π
ρ1 ≥ 1 − h, and ρ2 > 1 − 2h, holds. Eqs. (36)–(37) can be ≤ 2π + q7 h(t )dt < 4π ,
respectively rewritten as 0

(1 − c )d h < ρ1 (1 − ρ1 ), where we have used (40). Hence, we can choose q7 = exp(4π ) to

satisfy (22) and, consequently, (40) is equivalent to
(1 − c )d h < ρ2 (1 − ρ2 )(ρ2 − 1 + 2h).
2 2

2π
2π

It can be shown that the maximal admissible value for the right
h(t )dt < . (41)
hand side of the first inequality is 0 exp(4π )

1 1
 , h≥ , If k = 2m where m ≥ 1 is an integer. Then


4 2
f1 =
h(t ) = max{0, l sin2m (t )} = l sin2m (t ),

1
(1 − h)h, h ∈ 0, ,


2
and straightforward computation gives that
whereas the maximal value for the second inequality is  2π  2π
1 h(t )dt = l sin2m (t )dt
f2 = (2h + 1 − δ) (4h − 1 + δ) (δ + 2 − 2h) , 0 0
27  
√ 2m − 1 2m − 3 1
where δ = 4h2 − 2h + 1. =l ··· 2π
22m 2m − 2
To summarize, system (3) is GUES if one of the following  
conditions is met k−1k−3 1
= 2l ··· π.
k k−2 2
f1 f2
(1 − c )d < , (1 − c )d2 < . (38)
If k = 2m + 1 where m ≥ 0 is an integer. Then
h h2
We notice that f1 , f2 > 0 for any h > 0 and limh→0
f1
= h(t ) = max{0, l sin2m+1 (t )}
h
f
= 1. Conditions in (38) are thus less conservative than l sin2m+1 (t ), t ∈ [2jπ , 2jπ + π ] ,

limh→0 h22
=
(35) and (6) for small h. 0, t ∈ [2jπ + π , 2jπ + 2π ] , j ∈ Z,
Finally, we mention that these results cannot be obtained by
the approach in Ning et al. (2014) since the corresponding function from which it follows that
µ(t ) is periodic and has positive values in non-empty interval (see  2π  π
the explanation in Remark 3). h(t )dt = l sin(2m+1) (t )dt
0 0
2m 2m − 2 2 k−1k−3 2
4.2. A system with time-varying time delay = 2l ··· = 2l ··· .
2m + 1 2m − 1 3 k k−2 3
We consider the following scalar time-delay system Hence, condition (41) is equivalent to
ẋ(t ) = −x(t − h(t )), (39) 
k−1k−3 1 1
 ··· < , k is even,
which corresponds to system (31) with A(t ) = 0 and B(t ) = −1. l exp(4π )

k k−2 2

(42)
Let 0 ≤ h(t ) ≤ l where l is a constant. This system has been k−1k−3 2 π
··· < ,

well understood in the literature. It is known that this system is
 k is odd.
l exp(4π )

k k−2 3
asymptotically stable for any h(t ) satisfying l < 23 and, if l > 32 ,
there exists h(t ) such that it is unstable (see p. 138 of Hale (1977) Since the left hand sides of (42) approach zero as k → ∞, for
and the references cited there). any constant l that can be arbitrarily large there exists an integer
Let h(t ) = max{0, l sink (t )} where l > 0 and k is a positive inte- k∗ such that for any k ≥ k∗ system (39) is GUES. We mention
ger (Mazenc et al., 2015). It is noticed in Mazenc et al. (2015) that that no explicit condition such as (42) was provided in Mazenc
Razumikhin theorem is not applicable to (39) and no Lyapunov– et al. (2015) to ensure asymptotic stability of system (39). One may
Krasovskii functional construction technique in the literature has use Item 2 of Corollary 3 to obtain a similar result and the details
been applied to this system. We next use Item 1 of Corollary 3 to de- are not included here to save space. Finally, for this example, we
rive conditions guaranteeing the GUES. By choosing P (t ) = 1, we also mention that the results in the above cannot be obtained by
have p1 = p2 = 1 and α = −2. It follows that µ7 (t ) is defined as
the approach in Ning et al. (2014) (see again the explanation in
µ7 (t ) = −1 + q7 h(t ), Remark 3).
290 B. Zhou, A.V. Egorov / Automatica 71 (2016) 281–291

4.3. A non-periodic time-varying system
We consider the following LTV time-delay system
ẋ(t ) = A(t )x(t ) + B(t )x(t − h(t ))
= t cos t − λ x(t ) + l sin (t )x(t − h(t )),
2 2k
 
where λ > 21π and l > 0 are some constants, k ≥ 1 is an
integer, t ∈ J = [0, ∞), h : J → [0, τ ] is any function, τ is
an arbitrary positive number. For this system we notice that A(t ),
which is taken from Example 3 in Zhou (2016), is non-periodic and
even unbounded. We claim that this system is GUES for an arbitrary
l > 0, if k is sufficiently large.
We choose P (t ) =
 1. Then we have α1 = α2 = 1 and
α(t ) = 2 t cos t 2 − λ which is a USF as

 t
α(s)ds = −2λ(t − t0 ) + sin t 2 − sin t02
t0
≤ −2λ(t − t0 ) + 2.
Hence, µ3 (t ) = t cos t − λ + q3 l sin t (see Theorem 4), where q3
2 2k
is a constant to be specified. It is easy to see from (43) that
 t
µ3 (s)ds = −λ(t − t0 )
t0
1 1
+ sin t 2 −
2 2
Similarly to the discussion in Section 4.2, since 2λπ − 1 > 0, for
any l > 0, there exists a k∗ such that (47) is satisfied for any k ≥ k∗ .
5. Concluding remarks
This paper establishes generalized Razumikhin and Krasovskii
stability theorems for general time-varying time-delay systems. By
using the notion and properties of uniformly stable functions, the
condition of negativity of the derivative of Razumikhin functions
and Krasovskii functionals is weakened. The proposed Razumikhin
and Krasovskii stability theorems were then applied to the stability
analysis of some linear time-delay systems with both time-varying
coefficients and time-varying delays. Some constructive sufficient
stability conditions were obtained and their effectiveness was
demonstrated by some numerical examples. The obtained results
can be seen as extensions of those in Zhou (2016) to time-
delay systems. We mention that the presented methodology does
(43) not provide any advantages for time-invariant systems since
the corresponding Lyapunov functions (functionals) are generally
time-invariant and, consequently, their time-derivatives are also
time-invariant, which cannot be treated by the methods presented
here. The results in this paper have been extended to the discrete-
time setting and the stochastic setting, which will be reported
elsewhere.
 t
sin t02 + q3 l sin2k s ds,
t0
Acknowledgments

from which it follows that, for any t ∈ J, The authors would like to thank the Associate Editor who
 t +2 π  t +2π helped them draw attention to a stability result for system (39),
µ3 (s)ds ≤ −2λπ + 1 + q3 l sin 2k
s ds and would like to thank all the anonymous reviewers for their
t t
helpful comments and suggestion that have helped to improve
 2π
= −2λπ + 1 + q3 l sin2k s ds. (44) the quality of the paper a lot. The authors also thank Professor F.
0 Mazenc for his comments on an early version of this paper.
This work was supported in part by the National Natural
We choose q3 = exp (2λπ ) and assume that the following
Science Foundation of China under grant numbers 61273028,
inequality is true
61322305 and 61573120, by the Natural Science Foundation of
2π
2λπ − 1 2λπ − 1

Heilongjiang Province of China under grant number F2015007,
sin2k s ds < = . (45)
0 q3 l l exp (2λπ ) by the Foundation for the Author of National Excellent Doctoral
Dissertation of China under Grant 201343, by the Fok Ying Tung
Then we get from (44) that, for any t ∈ J,
Education Foundation under grant number 151060, and by the
 t +2 π  2π
Russian Foundation for Basic Research under Grant 15-58-53017 a.
µ3 (s)ds ≤ −2λπ + 1 + q3 l sin2k s ds < 0, (46)
t 0
References
which implies that 2π ∈ Ωµ3 . Now it follows from (46) and (43)
that Aeyels, D., & Peuteman, J. (1999). Uniform asymptotic stability of linear time-
t +θ varying systems. In V. Blondel, et al. (Eds.), Open problems in mathematical
  
ϕµ3 (2π) = sup max µ3 (s) ds systems and control theory (pp. 1–5). London: Springer-Verlag.
t ∈J θ∈[0,2π] t Angeli, D., Sontag, E. D., & Wang, Y. (2000). A characterization of integral input-to-
  state stability. IEEE Transactions on Automatic Control, 45(6), 1082–1097.
1 1 Cacace, F., Conte, F., & Germani, A. (2016). Memoryless approach to the LQ and
= sup max −λθ + sin(t + θ )2 − sin t 2 LQG problems with variable input delay. IEEE Transactions on Automatic Control,
t ∈J θ∈[0,2π ] 2 2 61(1), 216–221.
 t +θ
 Chen, W.-H., & Zheng, W. X. (2011). Exponential stability of nonlinear time-delay
+ q3 l sin2k s ds systems with delayed impulse effects. Automatica, 47(5), 1075–1083.
Egorov, A. V., & Mondié, S. (2015). The delay Lyapunov matrix in robust stability
t
analysis of time-delay systems. In Proceedings of the 12th IFAC workshop on time
t +θ
  
2k delay systems (pp. 245–250). Ann Arbor, USA.
≤ 1 + q3 l sup max sin s ds Fridman, E., & Orlov, Y. (2009). Exponential stability of linear distributed parameter
t ∈J θ∈[0,2π] t systems with time-varying delays. Automatica, 45(1), 194–201.
 2π Gu, K., Kharitonov, V. L., & Chen, J. (2003). Stability of time-delay systems. Boston:
Birkhäuser.
= q3 l sin2k s ds + 1 Halanay, A. (1966). Differential equations: stability, oscillations, time lags. New York:
0 Academic Press.
< 2λπ − 1 + 1 = 2λπ . Hale, J. K. (1977). Theory of functional differential equations. New York: Springer-
Verlag.
This indicates that q3 = exp (2λπ ) > exp ϕµ3 (2π ) . Then by Item
 
Hou, C., & Qian, J. (1998). On the decay estimate for Razumikhin-type theorems.
Acta Automatica Sinica, 24(5), 699–701. (in Chinese).
1 of Theorem 4, this system is GUES if (45) is satisfied. Notice that Kalman, R. E., & Bertram, J. E. (1960). Control system analysis and design via the
(45) is equivalent to ‘‘second method’’ of Lyapunov: I–Continuous-time systems. Transactions of the
ASME–Journal of Basic Engineering, 82(2), 371–393.
2k − 1 2k − 3 1 2λπ − 1 Kharitonov, V. L. (2013). Time-delay systems: Lyapunov functionals and matrices.
··· < . (47)
2k 2k − 2 2 2π l exp (2λπ ) Basel: Birkhäuser.
 B. Zhou, A.V. Egorov / Automatica 71 (2016) 281–291
Kharitonov, V. L., & Niculescu, S.-I. (2003). On the stability of linear systems with
uncertain delay. IEEE Transactions on Automatic Control, 48(1), 127–132.
Kojima, A., Uchida, K., Shimemura, E., & Ishijima, S. (1994). Robust stabilization of
a system with delays in control. IEEE Transactions on Automatic Control, 39(8),
1694–1698.
Krstic, M. (2010). Lyapunov stability of linear predictor feedback for time-varying
input delay. IEEE Transactions on Automatic Control, 55(2), 554–559.
Malisoff, M., & Mazenc, F. (2009). Constructions of strict Lyapunov functions. London:
Spinger-Verlag.
Mazenc, F., & Malisoff, M. (2016). Stability analysis for time-varying systems with
delay using linear Lyapunov functionals and a positive systems approach. IEEE
Transactions on Automatic Control, 61(3), 771–776.
Mazenc, F., Malisoff, M., & Niculescu, S.-I. (2014). Reduction model approach for
linear time-varying systems with delays. IEEE Transactions on Automatic Control,
59(8), 2068–2082.
Mazenc, F., Malisoff, M., & Niculescu, S.-I. (2015). Stability analysis for systems with
time-varying delay: Trajectory based approach. In Proceedings of the 54th IEEE
conference on decision and control (pp. 1811–1816). Osaka, Japan.
Melchor-Aguilar, D., Kharitonov, V., & Lozano, R. (2010). Stability conditions for
integral delay systems. International Journal of Robust and Nonlinear Control, 20,
1–15.
Ning, C., He, Y., Wu, M., Liu, Q., & She, J. (2012). Input-to-state stability of nonlinear
systems based on an indefinite Lyapunov function. Systems & Control Letters,
61(12), 1254–1259.
Ning, C., He, Y., Wu, M., & She, J. (2014). Improved Razumikhin-type theorem for
input-to-state stability of nonlinear time-delay systems. IEEE Transactions on
Automatic Control, 59(7), 1983–1988.
Pepe, P. (2014). Direct and converse Lyapunov theorems for functional difference
systems. Automatica, 50(12), 3054–3066.
Rugh, W. J. (1996). Linear system theory (2nd ed). Upper Saddle River, NJ: Prentice
Hall.
Savkin, A. V. (1997). Robust output feedback constrained controllability of
uncertain linear time-varying systems. Journal of Mathematical Analysis and
Applications, 215(2), 376–387.
Savkin, A. V., & Petersen, I. R. (2000). A method for simultaneous strong stabilization
of linear time-varying systems. International Journal of Systems Science, 31(6),
685–689.
Teel, A. R. (1998). Connections between Razumikhin-type theorems and the ISS
nonlinear small gain theorem. IEEE Transactions on Automatic Control, 43(7),
960–964.
Zhang, X., Liu, L., Feng, G., & Zhang, C. (2013). Output feedback control of large-scale
nonlinear time-delay systems in lower triangular form. Automatica, 49(11),
3476–3483.
291
Zhou, B., Lin, Z., & Duan, G. R. (2012). Truncated predictor feedback for
linear systems with long time-varying input delays. Automatica, 48(10),
2387–2399.
Zhou, B. (2016). On asymptotic stability of linear time-varying systems. Automatica,
68, 266–276.
Zhou, B. (2014). Truncated predictor feedback for time-delay systems, XIX. (p. 480).
Heidelberg: Springer-Verlag.
Bin Zhou is a Professor of the Department of Control
Science and Engineering at the Harbin Institute of Tech-
nology. He received the Bachelor’s degree, the Master’s
Degree and the Ph.D. degree from the Department of
Control Science and Engineering at Harbin Institute of
Technology, Harbin, China in 2004, 2006 and 2010, respec-
tively. His current research interests include constrained
control, time-delay systems, time-varying systems, non-
linear control, multi-agent systems, and control applica-
tions in astronautic engineering. In these areas, he has
published over 90 papers in archival journals. He is the
author of the book Truncated Predictor Feedback for Time-Delay Systems (Springer-
Verlag, Berlin Heidelberg, 2014). He is a reviewer for American Mathematical
Review and is an active reviewer for many journals. He received the ‘‘National Excel-
lent Doctoral Dissertation Award’’ in 2012 from the Academic Degrees Committee
of the State Council and the Ministry of Education of P.R. China. He is currently an
associate editor on the Conference Editorial Board of the IEEE Control Systems So-
ciety and an associate editor of Asian Journal of Control and Journal of System Science
and Mathematical Science.
Alexey V. Egorov graduated in 2010 and received his
Ph.D. degree in 2013, both from St. Petersburg State
University (SPbSU). Dr. Egorov is an Associate Professor
of Department of Control Theory at the Faculty of Applied
Mathematics and Control Processes of this university. His
scientific interests include time-delay systems, control
theory and stability analysis.