Uniform Asymptotic Stability of Linear Time-Varying Systems

1
Uniform asymptotic stability of linear

time-varying systems
Dirk Aeyels and Joan Peuteman
Universiteit Gent
SYSTeMS
Technologiepark-Zwijnaarde 9
9052 Gent
BELGIUM
dirk.aeyels~rug.ac.be
joan.peuteman~rug.ac.be
1 Description of the problem

Let x(t) = A(t)x(t), with x: R --t Rn and A: R --t Rnxn, be a linear time-
varying system. We consider the case where the eigenvalues of the system
matrix as functions of the time t E R may have positive real parts. Con-
sider the following examples: the matrix Al(t) := diag(-l + 2cost, -1 -
2cost) and the matrix A2(t) := diag(-l + max{2cost,-2cost},-1 -
max{2cost,-2cost}). Notice that Al(t) and A2(t) have the same eigen-
values Vt E R. The system x( t) = Al (t )x(t) is uniformly asymptotically
stable and the system ±(t) = A2(t)X(t) is unstable. This illustrates that the
stability properties do not only depend on the evolution of the eigenval-
ues. The evolution of the corresponding eigenvectors will also playa crucial
role.
We are interested in formulating a sufficient condition (preferably easily
verifiable) for asymptotic stability of ±(t) = A(t)x(t). Since this problem
turns out to be non-trivial and since specific versions of this problem are
currently being examined within the framework of complexity theory [9], we
will restrict our attention to a particular class ±(t) = - f(t)m(t)mT(t)x(t)
with m : R -+ R n and f : R --t R taking both negative and positive values.
Even this problem turns out to be non-trivial.
V. Blondel et al. (eds.), Open Problems in Mathematical Systems and Control Theory
© Springer-Verlag London Limited 1999
2 Uniform asymptotic stability of linear time-varying systems
2 Available results and desired extensions

1. A fast time-varying system x(t) = A(o:t)x(t) with 0: sufficiently large
and with the additional properties that A(t) and its derivatives up to
the second order are continuous and bounded, is exponentially sta-
ble when the averaged system x(t) = Ax(t) is exponentially stable
[2, 4, 10]. When the system matrix A(t) is periodic with period T,
then A = ~ JOT A(t)dt.
When the system matrix A(t) is periodic with period T, then the
slowly time-varying system x(t) = A(~)x(t) with 0: sufficiently large
is exponentially stable when the average of the real part of the instan-
taneous eigenvalue Amax(t) is negative Le. ~ J{ Amax(t)dt < 0 [11].
Here, Amax(t) denotes the instantaneous eigenvalue of A(t) with the
largest real part.
The study of the stability of a system x(t) = A(t)x(t) with a system
matrix A(t) which is neither fast time-varying nor slowly time-varying
is an ambitious task. There are some results available in the litera-
ture [16, 6] which will be discussed. New contributions in the field are
welcome.
2. Wazewski has shown that when ¢(t, to, xo) denotes the solution of
x{t) = A(t)x(t) with initial condition Xo at to that 11¢(t, to, xo)11 ::;
Ilxoller(t,to) with r(t, to) := !J t: Amax{a)da. Here Amax(a) is the
largest eigenvalue of A(a)+AT (a). For a proof based on Liapunov the-
ory, we refer the reader to [16]. Notice that the system x = A(t)x(t)
with A(t):= diag(-1 +2cost,-I- 2cost) is exponentially stable
without satisfying the condition of WazewskL This example illus-
trates that it is a sufficient condition which is also for systems with
a symmetric system matrix, rather conservative. Therefore, it would
be useful to obtain a less conservative condition e.g. by an explicit
use of the evolution of the eigenvectors.
3. Consider a class of n-dimensional systems, each system matrix in

companion form with coefficients Pi (t) and with the additional prop-
erty that 'Vt E R : 0 ::; al ::; Pi(t) ::; bi (i = 1, ... , n). A necessary
and sufficient condition that all solutions of an arbitrary equation of
this class of systems is nonoscillatory (every nontrivial solution has
only a finite number of zeros on [to, oo[ ) and tends to zero as t ---+ 00
is that the polynomials un + alun - 1 + b2u n - 2 + a3Un-3 + ... and
un +b1u n - 1 +a2 Un - 2 +b3u n - 3 + ... have only real roots. These roots
are negative since the coefficients of the polynomials are nonnega-
tive. Levin [6] shows that this statement implies that the instanta-
neous eigenvalues of each system matrix are negative and real. The
requirement that the roots are negative is a necessary condition for
asymptotic stability of each system belonging to the class (without

Uniform Asymptotic Stability of Linear Time-Varying Systems

Uploaded by

Uniform Asymptotic Stability of Linear Time-Varying Systems

Uploaded by

1

Uniform asymptotic stability of linear

1 Description of the problem

2 Available results and desired extensions

3. Consider a class of n-dimensional systems, each system matrix in

You might also like