364 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 54, NO.

2, FEBRUARY 2009

Finite-Time Stabilization of Linear this differential inequality. More recently, the results of [14] have been
Time-Varying Continuous Systems extended to the time-varying continuous systems case in [15].
In this technical note, the objective is to derive finite time stabiliza-
Germain Garcia, Sophie Tarbouriech, and Jacques Bernussou tion controllers for linear time-varying continuous systems exploiting
specific properties of the solutions of a Lyapunov differential matrix
equation. It is one of the main difference with the technique developed
in [14] and [15]. In [14] and [15], matrix inequalities are used rather
Abstract—In this technical note, the problem of finite time stabilization of
linear time-varying continuous systems is considered. Necessary and suffi- than matrix equalities. Hence, a necessary and sufficient condition for
cient conditions, based upon the solution to some Lyapunov differential ma- stability analysis of linear time-varying system is first derived. This
trix equations, are proposed for particular cases of interest. From these con- condition expressed through the solution of a Lyapunov differential ma-
ditions, the design of time-varying state feedback controller guaranteeing trix equation is then used to derive a state feedback controller, when it
the finite time closed-loop stability is presented. Numerical experiments il-
lustrate the potentialities of the approach.
exists, which finite time stabilizes the system. This controller is a state
feedback with time-varying gains. Indeed, the controller, according to
Index Terms—Finite-time stability and stabilization, linear time-varying the definition of finite time stability considered, allows one to main-
systems, Lyapunov differential equations, Lyapunov stability.
tain the closed-loop trajectories in a certain domain (of ellipsoidal or
polyhedral shape) during a specified interval of time, provided that the
initial state belongs to a given ellipsoidal set.
I. INTRODUCTION
The technical note is organized as follows. In Section II, the system
In control theory, the concept of Lyapunov stability is very popular, under consideration and the problem we intend to solve are described.
particularly when dealing with systems described in the state space Section III is dedicated to the stability analysis results, whereas Sec-
[1]. Numerous applications in both control design or analysis prob- tion IV addresses the stabilization problem by state feedback. Section V
lems contexts are widely based on the concept of Lyapunov stability presents some numerical experiments to illustrate the potentialities of
because it corresponds in general to a satisfactory behavior in practice, the proposed approach. Finally, the technical note ends by some con-
even if from a performance point of view, other constraints have to be cluding remarks and prospectives.
imposed. Notation: Notation is standard. The euclidian vector norm is de-
However, in many applications, it is necessary to maintain the state noted by k:k. For two symmetric matrices, A and B , A > B means
under some bounds during, at least, a specific time interval. This is that A 0 B is positive definite. When no confusion is possible, iden-
the case, for example, when saturations are present [2], [3], or when a tity and null matrices will be denoted by I and 0, respectively.
is
linear model resulting from a linearization of the nonlinear one around an interval of time, defined by
= [t0 ; T ], where T and t0 are fixed
an equilibrium point is used [4]. The linear model being only valid and such that T > t0 . From this, one can define the set C (
; S ) as
C (
; S ) = ff :
! S ; f is continuous on
g. i (A), min (A)
around the equilibrium point, it is important to be able to avoid large
excursions of the state in order that the occurrence of the nonlinearities and max (A) denote the ith, minimal and maximal eigenvalue of ma-
does not lead the system permanently far away from the equilibrium. In trix A, respectively.
these situations, another interesting concept, like the finite time stability
(FTS), can be advantageously considered. The (FTS) concept has been
introduced in [5]–[8]. Furthermore, a more general concept, so-called II. PROBLEM STATEMENT
the Practical Stability (PS) concept, has been studied in the literature Consider the linear time-varying continuous system described by:
[9], [10]. Some works dealing with analysis and control design of FTS
control problems have been recently published, see for example [11],
[12] in which sufficient conditions are proposed through Linear Matrix
_( ) = ( ) ( ) + ( ) ( )
x t A t x t B t u t ; ( )=
x t0 x0 (1)

where matrices A(t) 2 C (

; n2n ) and B (t) 2 C (
; n2m ).
Inequality (LMI) formulations, or [13] in which sufficient conditions
are proposed through Riccati Differential Equations (RDE). In [14], a
System (1) is supposed to be completely state reachable at time t > t0 .
necessary and sufficient condition for FTS of linear systems has been
A necessary and sufficient condition for complete state reachability at
developed improving in particular the results developed in [12]. Using
time t is that rank of reachability grammian defined by:
classical concepts of operator theory, a necessary and sufficient condi-
tion is formulated through a differential Lyapunov inequality, a sym- t

metric uniformly bounded piecewise continuous differentiable matrix (

W t0 ; t ) = 8( ) ( ) 0 ( )80 ( )
1
t; s B s B s t; s ds (2)
function has to satisfy on a time interval with constraints on the in-
t
terval extremities. One of the main limitation is to obtain a solution to
is equal to n [16], with 8(t; s) the state transition matrix.
In some cases, it can be necessary to maintain the state under some
Manuscript received September 28, 2008; revised December 11, 2006, bounds during a specific time interval [3], [13], [15]. Note that in such
November 28, 2007, and June 19, 2008. Current version published February cases, the FTS concept can be invoked [6]. Let us consider the family
11, 2009. Recommended by Associate Editor D. Nesic.
of compact sets in n , denoted Et and parameterized in t 2
, such
that x = 0 2 Et . Consider also a compact set I0 , containing x = 0
G. Garcia is with LAAS-CNRS, University of Toulouse Toulouse 31077,
France and also with INSA, Toulouse 31077, France (e-mail: [email protected];
[email protected]). and such that I0  Et . We introduce for system (1), the following
S. Tarbouriech and J. Bernussou are with LAAS-CNRS, Univer- concept of stability [17], [18].
sity of Toulouse, Toulouse 31077, France (e-mail: [email protected]; Definition 1: System (1) with u = 0 is said to be (I0 ; Et ;
)-stable
[email protected]).
Color versions of one or more of the figures in this technical note are available if for any x0 2 I0 one gets:
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAC.2008.2008325 (
x t; t0 ; x0 )2 Et f or all t 2
(3)

0018-9286/$25.00 © 2009 IEEE

Authorized licensed use limited to: Wuhan University. Downloaded on May 04,2023 at 07:00:18 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 54, NO. 2, FEBRUARY 2009 365

or equivalently the solution x(t) = 0 of system (1) is said to be Proof: Let us start with the proof of relation (9). Consider the
(I0 ; Et ;
)-stable. following optimization problem:
Definition 1, which is stated in a relatively general way, can be par- 2
max (t) = x(t) R(t)x(t)
0

ticularized to some interesting cases. For example, ellipsoidal and poly- x(t)

hedral domains are often used to bound the state of a system subject to 
x00 S0 x0 20 ; S0 = S00 > 0
constraints [3]. Therefore, one can consider the following ellipsoidal
This problem corresponds to characterize the maximal ellipsoid Et , de-
sets I0 and Et :
fined in (5), containing the state at time t for all initial condition taken
I0 = x 2 n ; x0 S0 x  02 ; S0 = S00 > 0; 0 > 0 (4) in the ellipsoid I0 , defined in (4). Because t0 is fixed, an equivalent for-
mulation only involving the initial condition x0 is obtained as follows:
Et = x 2 n ; x0 R(t)x  (t)2 ; R(t) = R(t)0 > 0;
2
max (t) = x00 8(t; t0 )0 R(t)8(t; t0 )x0
(t) > 0; 8t 2
g (5) x
x00 S0 x0  20 ; S0 = S00 > 0
As mentioned above, other interesting sets Et can be defined as a poly-
hedral set: The Lagrangian of this optimization problem reads:

Et = x 2 n
; fi (t) x  1 for all t 2
; i = 1; 1 1 1 ; N
0
(6) L(x0 ;  ) = x00 8(t; t0 )0 R(t)8(t; t0 )x0 +  x00 S0 x0 0 20
where fi (t) 2 C (
; n21 ), i = 1; 1 1 1 ; N . with   0. Then, by computing the optimality conditions, one gets:
Remark 2: In the case where the sets Et and I0 are defined with @L
R(t) = I , (t) = > 0, 8t, S0 = I , and 0 =  , Definition 1 is = 28(t; t0 ) R(t)8(t; t0 )x0 + 2S0 x0 = 0 ,
0

@x0
directly related to the definition of Finite Time Stability introduced in
[13], [14]. See also [15] in the case where S0 = R(t0 ), (t) = > 0,  8(t; t0 )0 R(t)1=2
8t, and 0 =  . 2 R(t) 1=2
0
8(t0 ; t)
0
S0 8(t0 ; t)R(t)01=2 +  01 I
The problem we intend to solve in this technical note can be sum-
marized as follows. R(t)1=2 8(t; t0 )x0 = 0 (12)
Problem 1: Given an interval
and sets I0 and Et . Determine a
gain K (t) 2 C (
; m2n ) such that with the state feedback con-
6 0 to these two conditions, it is necessary to
To obtain a solution x0 =
have
troller u(t) = K (t)x(t) the closed-loop system x_ (t) = (A(t) +
B (t)K (t))x(t) is (I0 ; Et ;
)-stable. det R(t)01=2 8(t0 ; t)0 S0 8(t0 ; t)R(t)01=2 +  01 I =0 (13)
Problem 1 is referred to as the (I0 ; Et ;
)-Stabilization Problem.
Multiplying condition (12) on the left by x00 , it follows:
III. PRELIMINARY RESULTS: STABILITY ANALYSIS PROBLEM
In this section, we address the problem of stability analysis of system x00 8(t; t0 )0 R(t)8(t; t0 )x0 + x00 S0 x0 = 0
(1) with u = 0, which can be defined by:
and therefore one can deduce that (t)2 = 020  . Moreover,
x_ (t) = A(t)x(t); x(t0 ) = x0 (7) from (13), it follows that 0 01 is an eigenvalue of matrix
R(t)01=2 8(t0 ; t)0 S0 8(t0 ; t)R(t)01=2 . Then it follows:
where A(t) 2 C (
; n2n ). In other words, we want to study the
3 (t)2 = max (t)2
sub-problem of Problem 1, described as follows.
2 01
Problem 2: Given an interval
and two sets I0 and Et . Determine = 0 min R(t)01=2 8(t0 ; t)0 S0 8(t0 ; t)R(t)01=2
conditions such that system (7) is (I0 ; Et ;
)-stable. 2
According to Definition 1, suppose that the sets I0 and Et are given = 0 max R(t)1=2 X (t; t0 )R(t)1=2
and defined by (4) and (5). Indeed, the symmetric positive definite ma-
trices S0 and R(t), and the positive scalars 0 and (t), are known, for If (t)  3 (t) for all t 2
, then system (7) is (I0 ; Et ;
)-stable and
all t 2
. The following theorem can be stated. the sufficiency part of relation (9) is proven. To prove the necessity part,
Theorem 3: Suppose that I0  Et . System (7) is (I0 ; Et ;
)-stable suppose that system (7) is (I0 ; Et ;
)-stable and (t) < 3 (t) for t 2

. Then, this implies that there exist both x0 2 I0 and t 2
such that
if and only if 0   3 and (t)  3 (t) for all t 2
, with  3 and
3 (t) defined by x(t) does not belong to the ellipsoid Et , defined in (5). Consequently,
from Definition 1, system (7) is not (I0 ; Et ;
)-stable, which leads to
 3 = min (t)1min
=2
R(t)01=2 Y (t0 ; t)R(t)01=2 a contradiction. This proves the necessity part.
t2
Now consider relation (8). Let us then define the following function:
01=2 1=2
= min (t)max R(t) X (t; t0 )R(t)1=2 (8) 01=2
t2
(t) =  (t)min R(t)01=2 Y (t0 ; t)R(t)01=2
1=2
3 (t) = 0 0
min R(t)01=2 Y (t0 ; t)R(t)01=2
with  (t) > 0 and the sets:
1=2
= 0 max R(t)1=2 X (t; t0 )R(t)1=2 (9)

where matrices X (t; t0 ) and Y (t0 ; t) are respectively the solutions to x2 n

;x
0
S0 x   (t)2 ; S0 = S00 > 0
Lyapunov differential matrix equations:
From the above part of the proof, it follows that each set represents the
X (t; t0 ) = A(t)X (t; t0 ) + X (t; t0 )A(t)0 set of initial conditions such that x(t) 2 Et , defined in (5), at time
@
@t (10)
X (t0 ; t0 ) = S001 t 2
. Then for all t 2
, if the initial condition x0 belongs to the set
\t
fx 2
2 ; x S0 x
n 0
 (t)2 ; S0 = S0 > 0g =fx 2
0 n
;x
0
S0 x 
and  3
; S0 = S0 > 0 with
0
g
@
Y (t0 ; t) = 0A(t)0 Y (t0 ; t) 0 Y (t0 ; t)A(t) 1=2
@t
Y (t0 ; t0 ) = S0
(11) 3 = min (t)min R(t)01=2 Y (t0 ; t)R(t)01=2
t2

Authorized licensed use limited to: Wuhan University. Downloaded on May 04,2023 at 07:00:18 UTC from IEEE Xplore. Restrictions apply.
366 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 54, NO. 2, FEBRUARY 2009

then it follows that x(t) 2 Et , defined in (5), for all t 2

. The represents the set of initial conditions x0 such that jfi (t)0 x(t)j  1
necessity of condition (8) can be proved by contradiction as in the above at time t for all i = 1; 1 1 1 ; N . The set of initial conditions such that
part. jfi (t)0 x(t)j  1 for all i = 1; 1 1 1 ; N and t 2
is then defined as:
Some particular cases to Theorem 3 can be considered, depending \t2
fx 2 0
; x S0 x
n
 (t)2 ; S0 = S00 > 0g= fx 2 n 0
; x S0 x 
on the choice of matrices S0 , R(t) and scalars 0 , (t) when the sets  3
; S0 = S00 > 0 withg
I0 and Et are defined by (4) and (5): for example, by considering the 1
case invoked in Remark 2. Another interesting extension to Theorem  3 = min  (t) = min min
3 corresponds to the case where Et is a polyhedral set defined by (6).
t2
t2
i=1;111;N fi (t)0 X (t; t0 )fi (t)
Thus, suppose that the symmetric positive definite matrix S0 , the pos-
Hence, if 0   3 , system (7) is (I0 ; Et ;
)-stable, with Et defined
itive scalar 0 and the matrices fi (t), i = 1; 1 1 1 ; N , are known for all
in (6). The necessity can be proven by contradiction as in the proof of
t 2
. The following theorem can then be stated. Theorem 3.
Theorem 4: Suppose that I0  Et . System (7) is (I0 ; Et ;
)-stable
if and only if 0   3 , with  3 defined by
IV. MAIN RESULTS: STABILIZATION PROBLEM
1
3 = min min In this section, we propose solution to Problem 1 in both cases where
t2
i=1;111;N fi (t)0 X (t; t0 )fi (t)
Et is an ellipsoidal and a polyhedral set. We first investigate some
1
= min min (14) properties relative to the solution of the following -parameterized Lya-
t2
i=1;111;N
fi (t)0 [Y (t0 ; t)]01 fi (t) punov differential matrix equation:

where matrices X (t; t0 ) and Y (t0 ; t) are respectively the solutions of

@
@t
Z (t; t0 ) = A(t)Z (t; t0 ) + Z (t; t0 )A(t)0 0 B (t)B 0 (t)
Lyapunov differential matrix (11). Z (t0 ; t0 ) = S001 ;  > 0 and S0 = S00 > 0
(16)
Proof: Consider the following family of convex optimization
Lemma 5: The solution Z (t0 ; t) to (16) satisfies:
problems, i = 1; 1 1 1 ; N :
1) For all 1 and 2 , 0 < 1 < 2 , it follows that Z (t; t0 ) 
0
max fi (t)
x(t)
x(t) Z (t; t0 ) < X (t; t0 ) for all t 2
, where X (t; t0 ) is the solution
x00 S0 x0  02 to (11).
2) For T > 0, there exists a value of , denoted (T ) and defined by
which is equivalent, because t0 is fixed, to: T
0
max fi (t) 8(t; t0 )x0 (T ) = 01
max S01=2 8(t0 ; s)B (s)B 0 (s)80 (t0 ;s)S01=2 ds (17)
x

x00 S0 x0 20 t

The Lagrangian is then given by L(x0;  ) = fi (t)0 8(t; t0 )x0 + such that for all  2 (0; (T )[, one gets Z (t; t0 ) > 0, for all
 [x00 S0 x0 0 20 ] with   0. Then, by computing the optimality t2
.
conditions, one gets: Proof: The solution to (16) can be written:
@L Z (t; t0 ) = X (t; t0 ) 0 W (t0 ; t)
@x0
0
= 2S0 x0 + 8(t; t0 ) fi (t) ) x0
where X (t; t0 ) is the solution to (11) and W (t0 ; t) is defined in (2). By
= 0 21 S001  01 8(t; t0 )0 fi (t) (15)
definition, X (t; t0 ) and W (t0 ; t) are symmetric and positive definite
matrices for all t. Hence, for all  > 0, it follows that Z (t; t0 ) <
Using the expression of x0 in the constraint expression leads to:
X (t; t0 ). From the definition of W (t0 ; t), it is clear that for 0 < 1 <
2
 02 =
40 2 one gets Z (t; t0 )  Z (t; t0 ). The point 1 is then proven. To
01
fi (t) 8(t; t0 )S0 8(t; t0 )0 fi (t)
0 prove the point 2, remark that

and Z (t; t0 ) = 8(t; t0 )S001=2

0 S001 8(t; t0 )0 fi (t) 2 01 I 0 S01=2 8(t0 ; t)W (t0 ; t)80 (t0 ; t)S01=2 S001=2 80 (t; t0 )
x0 =
fi (t)0 8(t; t0 )S0018(t; t0 )0 fi (t)
Thus, one gets that Z (t; t0 ) is a positive definite matrix for a given t
if the following inequality is satisfied:
Then max fi (t)0 8(t; t0 )x0 = 0 fi (t)0 8(t; t0 )S0018(t; t0 )0 fi (t).
x
Define the following function:
1
01 I > S01=2 8(t0 ; t)W (t0 ; t)80 (t0 ; t)S01=2
i (t) = t

fi (t)0 8(t; t0 )S0018(t; t0 )0 fi (t) = S01=2 8(t0 ; s)B (s)B 0 (s)80 (t0 ; s)S01=2 ds
t
and the set fx 2 n ; x0 S0 x  i (t)2 ; S0 = S00 > 0g. Each set repre-
sents the set of initial conditions x0 such that jfi (t)0 x(t)j  1, at time or equivalently, one gets Z (t; t0 ) > 0 for all t 2
, if  satisfies for
t 2
. We can conclude that the set fx 2 n ; x0 S0 x   (t)2 ; S0 = all t 2
:
S00 > 0g with 01
t

I < S01=2 8(t0 ; s)B (s)B 0 (s)80 (t0 ;s)S01=2 ds

 (t) = min i (t)
i=1;111;N t

1
= min Noting that the right term increases with t, this right term allows to
i=1;111;N
fi (t)0 8(t; t0 )S0018(t; t0 )0 fi (t) define the value of (T ) for t = T .

Authorized licensed use limited to: Wuhan University. Downloaded on May 04,2023 at 07:00:18 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 54, NO. 2, FEBRUARY 2009 367

A practical way for computing (T ) is proposed in the following

lemma.
Lemma 6: The value of (T ) is given by:
01
(T ) = max 0S01=2 8(t0 ; T )S(T )80 (t0 ; T )S01=2 (18)

where 8(t0 ; T ) is the solution to

@ 0
@t 8(t0 ; t) = 8(t0 ; t)A(t)
8(t0 ; t0 ) = I for all t
2 (19)

and S(t) is the solution to the following Lyapunov differential matrix

equation:
_
S(t) = A(t)S(t) + S(t)A(t)
0 0 B(t)B0 (t)
S(t0 ) = 0; for all t 2
(20)

Proof: The solution to (20) evaluated at t = T reads:

T
S(T ) = 0 0 0
8(T ; s)B(s)B (s)8 (T; s)ds
t Fig. 1. Function  (t) for R(t) = 1,  = 1 and S = I .

From the properties of 8(t0 ; t), the result follows.

Based on the use of Lemmas 5 and 6, the control law investigated to
solve Problem 1 is defined by

u(t) = K(t)x(t) = 0 2 B(t)Z01 (t; t0 )x(t) (21)

with  2 (0; (T )[ and Z (t; t0 ) the solution of Lyapunov differential

matrix (16). Note that considering the point 2 of Lemma 5, the family
of controllers is well-defined.

A. Et is a Ellipsoidal Set
In this subsection, the sets I0 and Et are ellipsoidal shape and there-
fore defined by (4) and (5), respectively. Indeed, the symmetric positive
definite matrices S0 and R(t), and the positive scalars 0 and (t), are
known, for all t 2
. A solution to Problem 1 is then stated in the fol-
lowing theorem.
Theorem 7: Suppose that I0  Et . Consider  3 and 3 (t) defined
in relations (8) and (9). If there exists 3 2 (0; (T )[ such that 0 
3 3 3 3 3 3
K ( ) and (t)  K (t;  ) for all t 2
, where K () and K (t; )
are defined as follows:
Fig. 2.  (t;  ) and  (t;  ) for  = 1 and  = 1:803.
3 01=2
K () = min (t)max R(t)
1=2
Z (t; t0 )R(t)
1=2
>
3
t2

(22)
3 1=2 1=2 1=2
3 () <  3 and 3K (t; 2 ) 
Then for all  2 (0; (T )[, it follows that K
K (t; ) = 0 max R(t) Z (t; t0 )R(t) (23) 3 3
K (t; 1 )   (t) for all t 2
and 1 < 2 . Combining these facts

and 3K (t; ) satisfies, for all 1 and 2 2 (0; (T )[, 1 < 2 : with the results of Theorem 4, the result of Theorem 7 follows.
Remark 8: The limit of the control gain is attained for values of 
3 3 3
 (t; 2 )   (t; 1 )   (t) for all t 2
close to (T ). For  = (T ), note that the resulting matrix Z (t; t0 ) is
K K
singular implying that the control gain is also singular.
then the control (21) solves Problem 1 for all  2 [3 ; (T )[. As in the context of the stability analysis problem (Theorem 3), the
Proof: From (16), we can write: particular case invoked in Remark 2 can be considered.
@  0 0 01
@t Z (t; t0 ) = A(t) 2 B(t)B(t) Z (t; t0 ) Z (t; t0 )
0
+ Z (t; t0 ) A(t)  0 0 01 B. Et is a Polyhedral Set
2 B(t)B(t) Z (t; t0 )
01
Z (t0 ; t0 ) = S0 ;  2
(0; (T )[ In this subsection, the sets I0 and Et are supposed to be defined by
From the point 1 in Lemma 5, it follows that for all  2 (0; (T )[, one
(4) and (6), respectively. Thus, the symmetric positive definite matrix
S0 , the positive scalar 0 and the matrices fi (t), i = 1; 1 1 1 ; N , are
gets Z (t; t0 ) < X(t; t0 ) for all t 2
. By using the Weyl’s mono-
known for all t 2
. A solution to Problem 1 is then stated in the
tonicity principle(see [19] Corollary IV.4.9), one gets:
following theorem.
i [Z (t; t0 )] < i [X(t; t0 )] ; i = 1; 1 1 1 ; n; for all t 2
Theorem 9: Suppose that I0  Et . If there exists 3 2 (0; (T )[
such that 0  K 3 (3 ) where K3 () is defined by:
In particular, the following inequality is satisfied:
3 1 3
max [Z (t; t0 )] < max [X(t; t0 )] ; for all t 2
K () = min min
t2
i=1;111;N fi (t)0 Z (t; t0 )fi (t)
>

Authorized licensed use limited to: Wuhan University. Downloaded on May 04,2023 at 07:00:18 UTC from IEEE Xplore. Restrictions apply.
368 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 54, NO. 2, FEBRUARY 2009

Fig. 5.  (t; ), for  = 0( (t)), 0.3, 0.4, 0.5.

Fig. 3. Variation of the control gains for  = 1 (top) and  = 1:803

(bottom).

Fig. 6. Variation of the control gains for  = 0:3 (top) and  = 0:4 (bottom).
Fig. 4. Evolution of the norm of state for 50 initial conditions of norm equal to
1 randomly generated.
Remark 10: The results of Theorems 7 or 9 can be used in order to
improve the property of (I0 ; Et ;
)-stability, in the sense that a con-
with  3 defined in relation (14), then the control (21) solves Problem 1 troller (21) can be computed in order to increase the size of admissible
for all  2 [3 ; (T )[. sets I0 and Et for which the closed-loop system is (I0 ; Et ;
)-stable,
Proof: The proof is similar to the one of Theorem 7. with respect to the sets obtained in the open-loop case.

Authorized licensed use limited to: Wuhan University. Downloaded on May 04,2023 at 07:00:18 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 54, NO. 2, FEBRUARY 2009 369

conditions taken in I0 and generated randomly. There is no gap

with the bound 3 (t).
• Control design case. The value of (T ) is equal to 0.5216. By
computing K 3
() for three values of  :  = 0:3,  = 0:4 and
 3= 0:5 one obtains K3 (0:3) = 0:5612, K3 (0:4) = 0:7106 and
K (0:5) = 0:8346. It appears (according to Remark 10) that the
control significantly improves the open-loop behavior, that is the
size of the initial conditions domain. The functions 3K (t; 0:3),
3K (t; 0:4) and 3K (t; 0:5) are represented in Fig. 5 and the corre-
sponding controls in Figs. 6 and 7.

VI. CONCLUSION
In this technical note, the problem of finite time stabilization of linear
time-varying continuous systems is considered. After deriving a neces-
sary and sufficient condition of finite time stability expressed through
the solution of a Lyapunov differential matrix equation, a time-varying
state feedback controller is proposed. In all cases, this controller allows
one to improve the open-loop finite time stability properties. Numer-
Fig. 7. Variation of the control gains for  = 0:5. ical examples illustrate the potentialities of the proposed theory. Some
interesting points not solved in this technical note concern the problem
of finite time stabilization of uncertain systems and the extension of the
V. NUMERICAL ISSUES proposed approach to the case of time-varying discrete systems. These
points will be investigated in a near future.
In this section, two examples are proposed to illustrate how the re-
sults of both stability analysis and control design can be used.
Example 1: Consider the linear time-varying continuous system (1) REFERENCES
borrowed from [20] and described by the following data: [1] R. Kalman, “Lyapunov functions for the problem of lur’e in automatic
control,” Proc. Nat. Acad. Sci., vol. 49, no. 2, pp. 201–205, 1963.
A(t) = 2t 1
B (t) = 1 [2] , S. Tarbouriech and G. Garcia, Eds., Control of Uncertain Systems With
0 t+1 ; 2t+2
2t+3
Bounded Inputs. New York: Springer-Verlag, 1997, vol. 227.
[3] , S. Tarbouriech, G. Garcia, and A. G. , Eds., Advanced Strategies
with
= [0; 1] (i.e. t0 = 0 and T = 1). The resulting system (1) is
in Control Systems With Input and Output Constraints. New York:
Springer Verlag, 2007, vol. 346.
completely state reachable [20]. In the sequel we consider two cases in [4] H. K. Khalil, Nonlinear Systems. New York: MacMillan, 1992.
which R(t) = I and S0 = I . [5] H. D’Angelo, Linear Time-Varying Systems: Analysis and Synthesis.
• Stability analysis case. Let us consider the open-loop behavior White Plains, NY: Allyn and Bacon, 1970.
[6] P. Dorato, “Short time stability in linear time-varying systems,” in
(i.e. the behavior of the associated system (7)) and apply the re-
sults of Theorem 3. By considering 0 = 1, one obtains 3 (t)
Proc. IRE Int. Convention Record Part.4, 1961, pp. 83–87.
[7] W. Hahn, Stability of Motion. New York: Springer Verlag, 1967.
which is depicted in Fig. 1. By considering (t) = 1, one gets [8] L. Weiss and E. Infante, “Finite time stability under perturbing forces
3 = 0:1616 , whereas by considering (t) = exp(02t), one and on product spaces,” IEEE Trans. Automat. Control, vol. AC-12, no.
obtains  3 = 0:0219. As expected, the domain of initial con- 1, pp. 54–59, Feb. 1967.
[9] A. Michel and D. Porter, “Practical stability and finite time stability of
ditions is drastically reduced in order to satisfy the property of discontinous systems,” IEEE Trans. Circuits Theory, vol. 19, no. CT-2,
(I0 ; Et ;
)-stability. The size of the sets Et , defined by (5) de- pp. 123–129, Mar. 1972.
creases with time and this reduced the domain of initial conditions [10] L. Grujic, “On practical stability,” Int. J. Control, vol. 17, no. 4, pp.
I0 which ensure that trajectories of system (7) belong to sets Et . 881–887, 1973.
[11] C. Abdallah, F. Amato, M. Ariola, P. Dorato, and V. Koltchinsky, “Sta-
• Control design case. In order to apply the results of Theorem 7, tistical learning methods in linear algebra and control problems: The
let us compute the value of (T ) : (T ) = 1:8036. With  = example of finite-time control of uncertain linear systems,” Linear Al-
1 = 1 and (t) = 1, one obtains K3 (1 ) = 0:201, 3whereas with gebra Appl., vol. 351–352, pp. 11–26, 2002.
 = 2 = 1:803, one gets K (2 ) = 0:253 > K (1 ). When
3 [12] F. Amato, M. Ariola, and P. Dorato, “Finite-time control of linear sys-
 increases, the size of the initial conditions domain increases as tems subject to parametric uncertainties and disturbances,” Automatica,
expected, according to Remark 10. For 1 and 2 , Figs. 2 and 3
vol. 37, no. 9, pp. 1459–1463, 2001.
[13] G. Garcia and S. Tarbouriech, “Finite time stabilization of linear
represent the functions 3K (t; 1), 3K (t; 1:803) and the variation systems,” in Proc. IFAC Workshop Control Appl. Optim. (CAO’06),
of control gains on the interval
, respectively. Cachan, France, 2006, [CD ROM].
[14] F. Amato, M. Ariola, C. Cosentino, C. Abdallah, and P. Dorato,
Moreover, as predicted by the theory, the monotony properties of
the functions K 3 () and 3 (t; ) can be observed. Note also that “Necessary and sufficient condition for finite-time stability of linear
K systems,” in Proc. Amer. Control Conf., Denver, CO, 2003, pp.
when the value of  approaches (T ), the control gains increase 4452–4456.
drastically as predicted. [15] F. Amato, M. Ariola, and C. Cosentino, “Finite-time control of linear
Example 2: Consider the linear time-varying continuous system (1) time-varying systems via output feedback,” in Proc. Amer. Control
Conf., Portland, OR, 2005, pp. 4722–4726.
described by the following data: [16] P. Antsaklis and A. Michel, Linear Systems. New York: McGraw-
Hill, 1997.
A(t) = 02 cos p 1
t sin 10t ; B = 1
0 1
6t
[17] F. Garashchenko, “Study of practical stability problems by numerical
methods and optimization of beam dynamics,” Pricl. Matem. Mekhan,
vol. 51, no. 6, pp. 559–564, 1987.
with
= [0; 2] (i.e. t0 = 0 and T = 2). Consider R(t) = I , 0 = 1 [18] V. Lakshmikantham, S. Leela, and A. Martynyuk, Practical Stability
and S0 = I . of Nonlnear Systems. Singapore: World Scientific, 1990.
[19] G. Stewart and J. Sun, Matrix Perturbation Theory. Boston, MA:
• Stability analysis case. The open-loop behavior is characterized
by the function 3 (t) represented in Fig. 4 ( = 0) and one obtains
Academic Press, 1990.
[20] P. Lu, “Closed-form control laws for linear time-varying systems,”
3 = 0:3822. Fig. 4 depicts the norm of the state for 50 initial IEEE Trans. Automat. Control, vol. 45, no. 3, pp. 537–542, Mar. 2000.

Authorized licensed use limited to: Wuhan University. Downloaded on May 04,2023 at 07:00:18 UTC from IEEE Xplore. Restrictions apply.