See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/266002797

Singular control systems

Article  in  Dynamics of Continuous, Discrete and Impulsive Systems Series A: Mathematical Analysis · October 2004

CITATIONS READS
187 1,045

1 author:

Dragutin Lj. Debeljkovic

University of Belgrade
258 PUBLICATIONS   1,311 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

OI 174001 View project

Dynamics of hybrid systems with complex structures. Mechanics of materials View project

All content following this page was uploaded by Dragutin Lj. Debeljkovic on 22 September 2014.

The user has requested enhancement of the downloaded file.

Dynamics of Continuous, Discrete and Impulsive Systems
Series A: Mathematical Analysis 11 (2004) 691-705
c
Copyright °2004 Watam Press

SINGULAR CONTROL SYSTEMS

D. Lj. Debeljkovic1
1
University of Belgrade, Faculty of Mechanical Eng,
Department of Control Engineering, 11000 Belgrade, Yugoslavia

Abstract. Singular systems are those the dynamics of which are governed by a mixture of
algebraic and differential equations. The complex nature of singular systems causes many
difficulties in the analytical and numerical treatment of such systems, particularly when
there is a need for their control. In that sense the question of their stability deserves great
attention. A brief survey of the results concerning the Lyapunov and finite time stability
of a particular class of these systems, are presented as the basis for their high quality
dynamical investigation.
Keywords. Singular systems, Lyapunov stability, Lyapunov matrix equation, Finite
and practical stability

1 Introduction
Singular systems are those the dynamics of which are governed by a mix-
ture of algebraic and differential equations.In that sense the algebrac equa-
tions represent the constraints to the solution of the differential part.
These systems are also known as descriptor, semi-state and generalized
systems arise naturally as a linear approximation of systems models, or lin-
ear system models in many applications such as electrical networks, aircraft
dynamics, neutral delay systems, chemical, thermal and diffusion processes,
large-scale systems, interconnected systems, economics, optimization prob-
lems, feedback systems, robotics, biology, etc..Although singular systems are
mostly present in electric and electro-magnetic circuits, in the sequel, will be
shown their application in chemical and process technology.

2 Notations and preliminaries

Consider linear singular systems represented, by

E ẋ(t) = Ax(t) , x(t0 ) = x0 (2.1)

y(t) = Cx(t) , (2.2)

or
E ẋ(t) = Ax(t) + Bu(t) , x(t0 ) = x0 (2.3)
y(t) = Cx(t) , (2.4)
692 D. Lj. Debeljkovic

with the matrix E possibly singular, where x(t) ∈ <n is a generalized state-
space vector and u(t) ∈ <m is a control variable.
Matrices A, B and C are of the appropriate dimensions and are defined
over the field of real numbers. System given by eq. (2.1) is operatinig in a
free and system given by eq. (2.2) is operating in a forced regime, i.e. some
external force is applied on it. It should be stressed that, in the general case,
the initial conditions for an autonomus and a system operating in the forced
regime need not be the same.
System models of this form have some important advantages in compar-
ison with models in the normal form, e.g. when E = I and an appropriate
discussion can be found in [1].
The complex nature of singular systems causes many difficultes in an-
alytical and numerical treatment that do not appear when systems in the
normal form are considered. In this sense questions of existence, solvability,
uniqueness, and smothness are present which must be solved in satisfactory
manner.
The survey of updated results for singular systems and a broad bibliog-
raphy can be found in [1], [4], [5], [15], [16], and in the two special issues of
the [8], [9].

3 Stability in the sense of Lyapunov

Stability plays a central role in the theory of systems and control en-
gineering. There are different kinds of stability problems that arise in the
study of dynamic systems, such as Lyapunov stability, finite time stabilty,
practical stability, technical stabilty and BIBO stability. The first part of
this section is concerned with the stability of the equlibrium points in the
sense of Lyapunov stability of linear autonomous singular systems. As we
treat the linear systems this is equivalent to the study of the stabilty of the
systems.The Lyapunov direct method is well exposed in a number of very
well known references.
Here we present some different and interesting approaches to this prob-
lem, including the contributions of the author of this paper.

Stability definitions

Definition 3.1 Eq. (2.1) is exponentially stable if one can find two positive
constants α, β such that

kx(t)k 6 β exp(−αt) kx0 k (3.1)

for every solution of Eq. (2.1), [21].

Definition 3.2 The system given by Eq. (2.1) will be termed asymptotically
stable iff, for all consistent initial conditions x0 , x(t) → 0 as t → ∞, [19].
 Singular control systems 693

Definition 3.3 We call system given by Eq. (2.1) asymptotically stable if all
roots of det(sE −A), i.e. all finite eigenvalues of this matrix pencil, are in the
open left - half complex plane, and system under consideration is impulsive
free if there is no x0 such that x(t) exibits discontinuous behavior in the free
regime, [15].

Definition 3.4 The system given by Eq. (2.1) is called asymptotically stable
iff all finite eigenvalues λi , i = 1,...,n1 , of the matrix pencil (λE − A) have
negative parts, [18].

Definition 3.5 The equilibrium x = 0 of system given by Eq. (2.1) is

said to be stable if for every ε > 0, and for any t0 ∈ J, there exists a
δ = δ(ε, t0 ) > 0, such that kx(t, t0 , x0 )k < ε hold for all t > t0 , when-
ever x0 ∈ Wk and kx0 k < δ, where J denotes time interval such that
J = [t0 , +∞) , t0 > 0, [7].

Definition 3.6 The equilibrium x = 0 of a system given by Eq. 2.1 is said

to be unstable if there exist a ε > 0, and t0 ∈ J, for any δ > 0, such that
there exists a t∗ > t0 , for which kx(t∗ , t0 , x0 )k < ε holds, although x0 ∈ Wk
and kx0 k < δ, [7].

Definition 3.7 The equilibrium x = 0 of a system given by Eq. (2.1) is said

to be attractive if for every t0 ∈ J, there exists an η = η(t0 ) > 0, such that
lim x(t, t0 , x0 ) = 0, whenever x0 ∈ Wk and kx0 k < η, [7].
t→∞

Definition 3.8 The equilibrium x = 0 of a singular system given by Eq.

(2.1) is said to be asymptotically stable if it is stable and attractive, [7].

Lemma 3.1 The equilibrium x = 0 of a linear singular system given by

Eq. 2.1 is asymptotically stable if and only if it is impulsive-free, and
σ(E, A) ∈ C − , [7].

Lemma 3.2 The equilibrium x = 0 of a system given by Eq. (2.1) is asymp-

totically stable if and only if it is impulsive-free and lim x(t, t0 , x0 ) = 0, [7].
t→∞

Stability theorems

Theorem 3.1 Eq. (2.1), with A = I, I being the identity matrix, is expo-
nentially stable if and only if the eigenvalues of E have nonpositive real parts,
[21].

Theorem 3.2 Let IΩ be the matrix which represents the operator on <n
which is the identity on Ω and the zero operator on Λ. Eq. (2.1), with
A = I, is stable if an n × n matrix P exist, which is the solution of the
matrix equation
E T P + P E = −IΩ (3.2)
with the following properties
P = PT (3.3)
694 D. Lj. Debeljkovic

P q = 0, q∈Λ (3.4)

qT P q > 0, q 6= 0, q ∈ Ω (3.5)

where
Ω = Wk = ℵ(I − EE D ) (3.6)

Λ = ℵ(EE D ) (3.7)

where Wk is the subspace of consistent initial conditions.

ℵ denotes the kernel or null space of the matrix ( · ), [21].

Theorem 3.3 The system given by Eq. (2.1) is asymptotically stable if and
only if

• A is invertible

• a positive-definite, self-adjoint operator P on <n exist, such that

AT P E + E T P A = −Q (3.8)

where Q is self-adjoint and positive in the sense that

xT (t)Qx(t) > 0, ∀ x ∈ Wk∗ \ {0} (3.9)

[19].

Theorem 3.4 The system given by Eq. (2.1) is asymptotically stable if and
only if

• A is invertible

• a positive-definite, self-adjoint operator P on <n exist, such that

¡ ¢
xT (t) AT P E + E T P A x(t) = −xT (t)Ix(t), ∀x ∈ Wk∗ (3.10)

[19].

Theorem 3.5 Let (E, A) be regular and (E, A, C) be observable.

Then (E, A) is impulsive free and asymptotically stable if and only if a

positive definite solution P to

AT P E + E T P A + E T C T CE = 0 (3.11)

exist and if P1 and P2 are two such solutions, then E T P1 E = E T P2 E, [15].

 Singular control systems 695

Theorem 3.6 If there are symmetric matrices P, Q satisfying

AT P E + E T P A = −Q (3.12)

and if
xT E T P Ex > 0, ∀x = S1 y1 6= 0 (3.13)
T
x Qx > 0, ∀x = S1 y1 (3.14)
then the system described by Eq. (2.1) is asymptotically stable if
· ¸
sE − A
rank = n, ∀s ∈ C (3.15)
S1T Q

and marginally stable if the condition given by Eq. (3.15) does not hold, [18].

Theorem 3.7 The equilibrium x = 0 of a system given by Eq. (2.1) is

asymptotically stable, if an n × n symmetric positive definite matrix P exist,
such that along the solutions of system, given by Eq. (2.1), the derivative
T
of function V (Ex) = (Ex) P (Ex), is a negative definite for the variates of
Ex, [7].

Theorem 3.8 If an n × n symmetric, positive definite matrix P exists, such

that along with the solutions of system, given by Eq. (2.1), the derivative of
T
the function V (Ex) = (Ex) P (Ex), i.e. V̇ (Ex) is a positive definite for all
variates of Ex, then the equilibrium x = 0 of the system given by Eq. (2.1)
is unstable, [7].

Theorem 3.9 If an n × n symmetric, positive definite matrix P exists, such

that along with the solutions of system, given by Eq. (2.1), the derivative of
T
the function V (Ex) = (Ex) P (Ex), i.e. V̇ (Ex) is negative semi definite
for all variates of Ex, then the equilibrium x = 0 of the system, given by Eq.
(2.1), is stable, [7].

Theorem 3.10 Let (E, A) be regular and (E, A, C) be impulse observable

and finite dynamics detectable. Then (E, A) is stable and impulse-free if and
only if a solution (P, H) to the generalized Lyapunov equations (GLE) exists

AT P + H T A + C T C = 0 (3.16)

H T E = ET P > 0 (3.17)
in [22].
The system, given by Eq. (2.1), is equivalent to

E1 ẋ1 (t) = A1 x1 (t) + B1 u(t) (3.18)

E2 ẋ2 (t) = A2 x2 (t) + B2 u(t) (3.19)

£ ¤
where xT (t) = xT1 (t) xT2 (t) .
696 D. Lj. Debeljkovic

Lemma 3.3 The system, given by Eq. (2.1), is asymptotically stable if and
only if the ”slow” sub - system, Eq. (3.18) is asymptotically stable, [25].

Lemma 3.4 x1 6= 0 is equivalent to E h+1 x 6= 0, [25].

Theorem 3.11 The system, given by Eq. (2.1), is asymptotically stable if

and only if for any matrix W > 0, the following equation

(E h )T AT P E h+1 + (E h+1 )T P AE h = −(E h+1 )T W E h+1 (3.20)

has a solution P > 0 with a positive external exponent r, [25].

Theorem 3.12 The system, given by Eq. (2.1), is asymptotically stable if

and only if for any given W > 0 the Lyapunov Eq. (3.20) has the solution
P > 0, [25].

For the sake of breavity all proofs are ommited here and can be found in
cited papers.
It should be noted that the results of the preceeding theorems are very
similar in some way and are derived only for regular linear singular systems.
In order to investigate the stability of irregular singular systems, the fol-
lowing results can be used, [2].
For this case, the linear singular system is used in the suitable canonical
form, i.e.
ẋ1 (t) = A1 x1 (t) + A2 x2 (t) (3.21)
0 = A3 x1 (t) + A4 x2 (t) (3.22)
Herewith, we examine the problem of the existence of solutions which
converge toward the origin of the systems phase-space for regular and ir-
regular singular linear systems. By a suitable nonsingular transformation,
the original system is transformed to a convenient form. This form of sys-
tem equations enables development and easy application of Lyapunov’s direct
method (LDM ) for the intended existence analysis for a subclass of solutions.
In the case when the existence of such solutions is established, an understi-
mation of the weak domain of the attraction of the origin is obtained on
the basis of symmetric positive definite solutions of a reduced order matrix
Lyapunov equation.
The estimated weak domain of attraction consists of points of the phase
space, which generate at least one solution convergent to the origin.
Let, the set of the consistent initial values of Eqs. (3.21-3.22) be denoted
by Wk∗ .
n×n
Also,
¡£ consider the
¤¢ manifold M ⊆ < , determined by Eq. 3.22 as
M=ℵ A3 A4 .
For the system governed by Eq. (3.21-3.22) the set Wk∗ of the consistent
initial values is equal to the manifold M, that is Wk∗ = M.
It is easy to see, that the convergence of the solutions of system given
by Eq. (2.1) and system, given by Eqs. (3.21-3.22), toward the origin is an
equivalent problem, since the proposed transformation is nonsingular.
 Singular control systems 697

Thus, for the null solution of Eqs. (3.21-3.22), the weak domain of at-
traction is going to be estimated.
The weak domain of attraction of the null solution x(t) ≡ 0 of system
given by Eqs. (3.21-3.22), is defined by
½ ¾
S , x0 ∈ <n : x0 ∈ M, ∃x(t, x0 ), lim kx(t, x0 )k → 0 (3.23)
t→∞

The term weak is used because solutions of Eqs. (3.21-3.22), need not to
be unique, and thus for every x0 ∈ S there may also exist solutions which do
not converge toward the origin.
In our case S = M = Wk∗ , and the weak domain of attraction may be
thought of as the weak global domain of attraction.
Note that this concept of global domain of attraction used in the paper,
differs considerably with respect to the global attraction concept known for
state variable systems, [2].
Our task is to estimate the set S.
We will use LDM to obtain the underestimate Se of the set S (i.e.Se ⊆ S).
Our development will not require the regularity condition of the matrix
pencil(sE − A).
For the systems in the form of Eqs. (3.21-3.22) the Lyapunov-like function
can be selected as

V (x(t)) = xT1 (t)P x1 (t), P = P T (3.24)

where P will be assumed to be a positive definite and real matrix.

The total time derivative of V along the solutions of Eqs. (3.21-3.22), is
then

V̇ (x(t)) = xT1 (t)(AT1 P + P A1 )x1 (t)

+ xT1 (t)P A2 x2 (t) + xT2 (t)AT2 P x1 (t) (3.25)

A brief consideration of the attraction problem shows that if Eq. (3.25)

is negative definite, then for every x0 ∈ Wk∗ we have kx1 (t)k → 0 as t → ∞.
Then kx2 (t)k → 0 as t → ∞, for all those solutions for which the following
connection between x1 (t) and x2 (t) holds

x2 (t) = Lx1 (t), ∀t ∈ < (3.26)

The main question is if the relation Eq. (3.26) can be established in a way
so as not to contradict the constraints. Since it is not possible for irregular
singular linear system, then we have to reformulate our task to establish
the relation Eq. (3.26) so that it does not pose to many additional novel
constraints to Eq. (3.22).
In order to efficiently use this fact for the analysis of the attraction prob-
lem, we introduce the following consideration that also proposes a construc-
tion of the matrix L.
698 D. Lj. Debeljkovic

Let Eq. (3.26) hold. Assume that the rank condition

£ ¤
rank A3 A4 = rank A4 = r 6 n2 (3.27)

is satisfied.
Then a matrix L exist, [23], being any solution of the matrix equation

0 = A3 + A4 L (3.28)

where 0 is a null matrix of dimensions the same as A3 .

On the basis of Eq. (3.26), eq. (3.28) and Eq. (3.22), it becomes evident
that whenever a solution x(t) fufills Eq. (3.26), then it has also has to fulfill
Eq. (3.22).
One can investigate in more detail the implications of the introduced
equations. When they hold, then all solutions of the system Eqs. (3.21-
3.22), which satisfy Eq. (3.26), are subject to algebraic constraints
· ¸
A3 A4
F x(t) = x(t) = 0 (3.29)
L −I

Assuming that V̇ (x(t)) determined by Eq. (3.25) is a negative definite,

the following conclusions are important
1. The solution of Eqs. (3.21-3.22) has to belong to set
¡£ ¤¢ ¡£ ¤¢
ℵ A3 A4 ∩ℵ L −I .

2. If rank F = n then judgement on the domain of attraction of the null

solutions not possible on the basis of the adopted approach, or more precisely,
in this case the estimate of the weak domain S of attraction is a singleton
© ¡£ ¤¢ ª
x(t) ∈ ℵ A3 A4 : x(t) ≡ 0

3. If rank F < n, then the estimates of the weak domain of attraction

needs to be a singleton and it is estimated as
½ ¾
x(t) ∈ <n ,
Se = ⊆S (3.30)
x(t) ∈ ℵ ([A3 A4 ]) ∩ ℵ ([L − I])

Now Eq. (3.25) and Eq. (3.26) are employed to obtain

V̇ (x(t)) = xT1 (t)((A1 + A2 L)T P + P (A1 + A2 L))x1 (t) (3.31)

which is a negative definite with respect to x1 (t) if and only if

ΩT P + P Ω = −Q, Ω = A1 + A2 L (3.32)

where Q is real a symmetric positive definite matrix.

We are now in the position to state the following result.
 Singular control systems 699

Theorem 3.13 Let the rank condition Eq. (3.27) hold and let rank F < n,
where the matrix F is defined in Eq. (3.29). Then, the underestimate Se of
the weak domain S of the attraction of the null solution of system given by
Eqs. (3.21-3.22), is determinated by Eq. (3.30), providing (A1 + A2 L) is a
Hurwitz matrix, [2].
If Se is not a singleton, then there are solutions of Eq. (3.21-3.22) different
form null solution, x(t) ≡ 0, which converge toward the origin as time t → ∞.

4 Finite time and practical stability

In practice one is not only interested in system stability (e.g. in the sense
of Lyapunov), but also in bounds of system trajectories. A system could be
stable but still completely useless because it possesses undesirable transient
performances. Thus, it may be useful to consider the stability of such systems
with respect to certain subsets of state-space which are defined a priori in
a given problem. Besides that, it is of particular significance to concern the
behavior of dynamical systems only over a finite time interval.
These boundedness properties of system responses, i.e. the solution of
system models, are very important from the engineering point of view. Re-
alizing this fact, numerous definitions of the so-called technical and practical
stability were introduced. Roughly speaking, these definitions are essentially
based on the predefined boundaries for the perturbation of initial conditions
and allowable perturbation of system response. In the engineering applica-
tions of control systems, this fact becomes very important and sometimes
crucial, for the purpose of characterizing in advance, in quantitative manner,
possible deviations of system response. Thus, the analysis of these particu-
lar boundedness properties of solutions is an important step, which precedes
the design of control signals, when finite time or practical stability control is
concern.
The linear singular system that should be treated in this section are given
with
E ẋ(t) = Ax(t) , x(0) = x0 , (4.1)
or with
ẋ1 (t) = A1 x1 (t) + A2 x2 (t) , (4.2)
0 = A3 x1 (t) + A4 x2 (t) . (4.3)
The boundedness properties of solutions of (4.1) can be expressed in the
equivalent form as constraints on solutions of (4.2-4.3), if the transformation
from (4.1) to (4.2-4.3) is nonsingular. Thus, we will present our problem for
the singular systems given in the both forms.
Our primary interest is to investigate boundedness properties, and in this
connection, the potential (weak) domain of practical or finite time stability.
Before formulating the problem we first introduce the set
© ª
SG (ρ) = x(t) ∈ <n : ||x(t)||2G < ρ , G = GT > 0 (4.4)
700 D. Lj. Debeljkovic

Systems governed by (4.1) or by (4.2-4.3) will be considered over time

interval T = [0, τ ], where quantity τ may be either a positive real num-
ber or symbol +∞, so that the finite and practical stability can be treated
simultaneously. It is obvious that T ∈ <.
Time invariant sets, used as bounds of system trajectories, are assumed
to be open, connected and bounded.
Let index ”a” stands for the set of all allowable states of system and index
”i” for the set of all allowable initial states of the system, such that Si ⊂ Sa .
As the system considered is time invariant it is sufficient to consider its
solutions x(t) as functions of only current time t and initial value x0 at the
initial moment t0 = 0, i.e. the adopted notation is as x(t, x0 ).
In an abbreviated form the value of solution x at the moment t will be
written as x(t).
[11] and [20] derived, for the first time, some new results in the area
of practical and finite time stability and unstability for time invariant linear
singular systems. These results represent the sufficient conditions for stability
of such class of systems and are based on Lyapunov - like functions and their
properties on subspace of consistent initial conditions. In particular, these
functions need not have properties of positivity in whole of state space and
negative derivatives along the system trajectories.
The geometric approach, used in that papers, gives more insight in the
structural properties of singular systems and problems of consistency of ini-
tial conditions and make possible a basis-free description of their dynamical
characteristics.

Stability definitions

Definition 4.1 System 4.1 is finite time stable w.r.t. (T, i, a, G) if and only
if x0 ∈ Wk∗ , satisfying ||x0 ||2G < i, implies, [11].

Here G = E T P E with P = P T > 0, is an arbitrary specified matrix and

Wk∗ is subspace of consistent initial conditions.

Definition 4.2 A solution x(t, x0 ) of the system (4.2-4.3 is (T, i, a, G) -

bounded if and only if x0 ∈ M ∧ ||x0 ||2G < i, implies ||x(t, x0 )||2G < a on T ,
[12], [13].

Definition 4.3 A solution x(t, x0 ) of the system (4.2-4.3) is x(t, x0 ) - un-

bounded if and only if there exists a (t∗, x0 ) ∈ T × M, such that ||x0 ||2G < i,
implies ||x1 (t∗, x0 )||2G > a, [12], [13].

Definition 4.4 A solution x(t, x0 ) of the system (4.2-4.3) is x(t, x0 ) - bounded

if and only if x0 ∈ M ∧ ||x10 ||2 < i, ||x20 ||2 < ia2 /a1 , implies
||x1 (t, x0 )||2 < a1 and ||x2 (t, x0 )||2 < a2 on T , [12], [13].

Definition 4.5 A solution x(t, x0 ) of the system (4.2-4.3) is x(t, x0 ) - un-

bounded if and only if there exists a (t∗, x0 ) ∈ T × M, such that ||x10 ||2 < i
 Singular control systems 701

and ||x20 ||2 < ia2 /a1 , implies ||x1 (t∗, x0 )||2 > a1 or ||x2 (t∗, x0 )||2 > a2 ,
[12], [13].

Two comments are necessary at this stage.

First note that if all solutions starting from all points of M ∩ SG (i) are
x(t, x0 ) - bounded then the system considered is practically stable with re-
spect to x(t, x0 ).
The second comment is that if there is any one solution which is x(t, x0 )
- unbounded, then the system considered is x(t, x0 ) - practically unstable.
In the sequel some new results, concerning non-Lyapunov stability of both
regular and irregular singular systems are presented.

Stability theorems

Theorem 4.1 The system governed by (4.1) is finite time stable w.r.t. x(t, x0 )
if the following condition is satisfied
a
ln > Λmax (G)t , ∀t ∈ T (4.5)
i
where
¡ ¢
Λmax (G) = max xT G x: x ∈ Wk \ {0} , xT E T P Ex = 1 . (4.6)

G = AT P E + E T P A (4.7)
in [11].

There is also a concept which is of interest in the analysis of finite and

practical stability, that is the domain of practical stability. However, we
will use the term ”potential” or ”weak” domain because for each x0 in this
domain, we will guarantee only that there exists at least one solution with
specific ”practical stability” characterization.
We will not prove that all solutions emanating from the concerned pointsx0
possess the required property.
Let e = (T, i, a, G).
The potential (weak) domain of practical stability with respect to (T, i, a, G)
of (4.2 - 4.3) is defined by
½ ¾
x0 ∈ <n : x0 ∈ M ∩ SG (i),
S(e) = (4.8)
∃x(t, x0 ) : ||x(t, x0 )||2G < a , t ∈ T

The potential domain of (T, i, a1 , a2 ) - practical stability is defined in an

analogous manner and will be presented later.
Our task also will be to estimate the set S(e).
We will use the LDM to obtain the underestimate Su (e) of the set S(e)
i.e. Su (e) ⊆ S(e).
As will be seen later, our development will not require the regularity
condition of the matrix pencil (sE − A).
702 D. Lj. Debeljkovic

For the systems in the form (4.2 - 4.3) the Lyapunov-like function can be
selected as
V (x(t)) = xT1 (t)P x1 (t) , P = P T > 0, (4.9)
where P will be assumed to be a symmetric, positive definite real matrix.
The total time derivative of V (x(t)) along the solutions of (4.2 - 4.3) is
then

V̇ (x(t)) = xT1 (t)(AT1 P + P A1 ) x1 (t)

+ xT1 (t)P A2 x2 (t) + xT2 (t)AT2 P x1 (t) (4.10)

For all solutions for which the following connection between x1 (t) and
x2 (t) holds
x2 (t) = Lx1 (t), ∀t ∈ T (4.11)
we will show the specific boundedness characteristics.
In order to efficiently use the relation (4.11) for the boundedness analysis,
we introduce again the rank condition, given (3.27).
On the basis of (3.27), (4.11) and (4.3) it becomes evident that whenever
solution x(t) fulfills (4.11) then it has to fulfill (4.3).
When rank condition holds then all solutions of system (4.2 - 4.3), which
satisfy (4.11) are subjected to algebraic constraints given with (3.29).
But, now, another discussion can be given. (i) These solutions of (4.2 -
4.3) have to belong to the set M ∩ ℵ([L − I]); (ii) If rank F < n then the
potential (weak) domain of (T, i, a1 , a2 ) - practical stability of (4.2 - 4.3) is
underestimated as
½ ¾
x(t) ∈ <n :
Su (e) = ⊆ Se (4.12)
x(t) ∈ M ∩ SG (i) ∩ ℵ([L − I])

The last fact will be shown in Theorem 4.4 for the case of (T, i, a1 , a2 )
- boundedness and the associated practical stability. Employing (4.10) and
(4.11) one obtains
¡ ¢
V̇ (x(t)) = xT1 (t) (A1 + A2 L)T P + P (A1 + A2 L) x1 (t)
= xT1 (t)Z x1 (t) (4.13)

where Z is a real symmetric matrix.

In comparison to the updated results, these enable one to examine the
system stability properties on finite and infinite time interval using a suitable
simple procedure in the circumstances when the proposed rank condition
(3.27) is valid.

Theorem 4.2 Suppose the rank condition (3.27) holds and let rank F 6 n,
where the matrix F is defined in (3.29). Then, the system governed by (4.2
- 4.3) has solutions which are (T, i, a1 , a2 ) - bounded, i 6 a1 , if the following
conditions are satisfied
a1
(i) ln > Λmax t, ∀t ∈ T (4.14)
i
 Singular control systems 703

a22
(ii) kLk 6 (4.15)
a1
where Λmax = Λmax (Z), and where the matrix Z is defined in (4.13), [12],
[13].
Theorem 4.3 Suppose rank condition (3.27) holds and let rank F 6 n,
where the matrix F is defined in (3.29). Then, the system governed by
(4.2 - 4.3) has solutions which are (T, i, a1 , a2 ) - unbounded, i 6 a1 if ∃∆,
0 < ∆ < i ∧ ∃τ ∈ T, τ 6 +∞ and ln ai1 6 λmin τ , where λmin = λmin (Z)
and where the matrix Z is defined in (4.13), [12], [13].
Before we give next Theorem on the underestimate of the potential do-
main of (T, i, a1 , a2 ) - practical stability, we use the notation s = (i, a1 , a2 ),
and a slightly modified definition of S and Su as
½ ¾
x0¡∈ <n : x0 ∈ M ∩ Ss , (∃x(t, x0 )) (∀t ∈¢T )
S(T, s) = (4.16)
||x1 (t, x0 )||2 < a1 , ||x2 (t, x0 )||2 < a2
Su (T, s) = {x(t) ∈ <n : x(t) ∈ M ∩ Ss ∩ ℵ ([L − I])} (4.17)
where Ss is defined by
© ª
SS = x(t) ∈ <n : ||x1 (t)||2 < i, ||x2 (t)||2 < i a2 /a1 , (4.18)
s = (i, a1 , a2 ) (4.19)
Theorem 4.4 Suppose rank condition (3.27) hold and let rank F < n, where
the matrix F is defined in (3.29). Then, providing the conditions of Theorem
4.2 hold, the underestimate Su (T, s) of the potential (weak) domain S (T, s)
of (T, i, a1 , a2 ) - practical stability of (4.2 - 4.3) is determined by (4.16),
where Ss is defined by (4.19), [12], [13].
The proofs and other details can be found in cited references and are
omitted for the sake of brevity.

5 Conclusion
Some linear systems not possessing a state variable representation of such
kind can admit a semi state (descriptor) equations of the form (2.1), where
E and A are constant matrices, with E possibly singular. In mathematical
sense, these models are presented in the form of mixture of differential and
algebraic equations. These systems are also known as semi state, descriptor,
singular as well as generalized state-space systems. They naturally arise
in many practical engineering disciplines and applications such as feedback
control systems, optimization problems, large scale systems, and generally as
linear approximation of system models.
Central part of this paper is devoted to some stability questions from the
viewpoint of Lyapunov and non-Lyapunov stability theory. Some original
results, derived earlier by the authors, have been presented.
Identical results, but for linear discrete-time singular systems, have been
derived in [14] and [3].
704 D. Lj. Debeljkovic

6 References

[1] Bajic, V. B., Lyapunov’s Direct Method in The Analysis of Singular Systems and
Networks, Shades Technical Publications, Hillcrest, Natal, RSA, 1992.

[2] Bajic, V. B., D. Debeljkovic, Z. Gajic, B. Petrovic, ”Weak Domain of Attraction

and Existence of Solutions Convergent to the Origin of the Phase Space of Singu-
lar Linear Systems”, University of Belgrade, ETF, Series: Automatic Control, (1)
(1992), 53 - 62.

[3] Bajic, V. B., D. Lj. Debeljkovic, B. B. Bogicevic, M. B. Jovanovic, ”Non -Lyapunov

Stability Robustness Consideration for Discrete Desriptor Linear Systems”, IMA J.
Math. Control and Information, (1998), Vol.15, No. 2, 105 - 115.

[4] Campbell, S. L., Singular Systems of Differential Equations, Pitman, Marshfield,

Mass., 1980.

[5] Campbell, S. L., Singular Systems of Differential Equations II, Pitman, Marshfield,
Mass., 1982.

[6] Campbell, S. L., C. D. Meyer, N. J. Rose, ”Application of Drazin Inverse to Linear

Systems of Differential Equations”, SIAM J. Appl. Math., 31 (1976) 411 - 425.

[7] Chen,C., Y. Liu, ”Lyapunov Stability Analysis of Linear Singular Dynamical Sys-
tems”, Proc. Int. Conference on Inteliligent Processing Systems, Beijing, (China),
October 28 - 31, (1997) 635 - 639.

[8] Circuits, Systems and Signal Processing, Special Issue on Semistate Systems, 5 (1)
(1986).

[9] Circuits, Systems and Signal Processing, Special Issue: Recent Advances in Singular
Systems, 8 (3) (1989).

[10] Dai, L., Singular Control Systems, Springer Verlag, Berlin, 1989.
[11] Debeljkovic, D. Lj., D. H. Owens, ”On Practical Stability”, Proc. MELECON
Conference, Madrid (Spain), October (1985), 103 - 105.

[12] Debeljkovic, D. Lj., V. Bajic, Z. Gajic, ”Further Results in Non-Lyapunov Stability

and Unstability of Regular and irregular Generalized State Space Systems”, Proc.
SAUM IV, June (1992), Kragujevac (YU), PP. 316-333.

[13] Debeljkovic, D. Lj., V. Bajic, B. Petrovic, ”Boundedness and Existence of Solutions

of Regular and Irregular Singular Systems”, Publications of the Faculty of Elec.
Eng., Belgrade, Series: Automatic Control, No.1 (1993), pp. 69-78.

[14] Debeljkovic, D. Lj., V. B. Bajic, T. Eric, S. A. Milinkovic, ”Lyapunov Stability

Robustness Consideration for Discrete Descriptor Systems”, IMA J. Math. Control
and Information, (1998), Vol. 15, No.1, 52 - 63.
View publication stats

Singular control systems 705

[15] Lewis, F. L., ”A Survey of Linear Singular Systems”, Circ. Syst. Sig. Proc., 5 (1)
(1986) 3-36.

[16] Lewis, F. L., ”Recent Work in Singular Systems”, Proc. Int. Symp. on Sing. Syst.,
Atlanta, GA (1987) 20-24.

[17] Luenberger, D. G., ”Time-Invariant Descriptor Systems”, Automatica, 14 (1978)

473 - 480.

[18] Muller, P. C., ”Stability of Linear Mechanical Systems with Holonomic Constraints”,
Appl. Mech. Rev. (11), part 2, November (1993), 160 - 164.

[19] Owens, D. H., D. Lj. Debeljkovic, ”Consistency and Liapunov Stability of Linear
Decsriptor Systems: a Geometric Approach”, I MA Journal of Math. Control and
Information, (1985), No.2, 139 - 151.

[20] Owens, D. H. D. Lj. Debeljkovic , ”On Non-Lyapunov Stability of Discrete De-

scriptor Systems”, Proc. 25th Conference on Decision and Control, Athens, Greece
(1986) 2138 - 2139.

[21] Pandolfi, L., ”Controllability and Stabilization for Linear Systems of Algebraic and
Differential Equations”, JOTA, 30 (4) (1980) 601-620.

[22] Takaba, K., N. Morihira, T. Katayama, ”A Generalized Lyapunov Theorem for

Descriptor System”, Systems & Control Letters, (24) (1995) 49 - 51.

[23] Tseng, H. C., P. V. Kokotovic, ”Optimal Control in Singularly Perturbed Systems:

The Integral Manifold Approach”, IEEE Proc. on CDC, Austin, TX (1988) 1177-
1181.

[24] Wu, H., K. Muzukami, ”Lyapunov Stability Theory and Robust Control of Uncer-
tain Descriptor Systems”, Int. J. Science, Vol.26, (10), (1995), 1981-1991.

[25] Zhang, Q., G. Dai, J. Lam, L .Q. Zhang, M. De La Sen, ”Asymptotical Stability and
Stabilization of Descriptor Systems”, Acta Automatica Sinica, Vol. 24 (2), (1998a),
208 - 211.

Email: ddebeijikovic@alfa.mas.ac.yu