# On stability and state feedback stabilization of singular linear matrix difference equations

## Abstract

In this article, we study the stability of a class of singular linear matrix difference equations whose coefficients are square constant matrices and the leading coefficient matrix is singular. Speciffically we analyze the stability, the asymptotic stability and the Lyapunov stability of the equilibrium states of an homogeneous singular linear discrete time system and we define the set of all equilibrium states. After we prove that if every equilibrium state of the homogeneous system is stable in the Lyapounov's sense, then all solutions of the non homogeneous system are continuously depending on the initial conditions and are bounded provided that the input vector is also bounded. Moreover, we consider the case where the equilibrium states of the system are not stable. For this case we provide necessary and sufficient conditions for stabilization.

## 1 Introduction

Linear discrete time systems (or linear matrix difference equations), are systems in which the variables take their value at instantaneous time points. Discrete time systems differ from continuous time ones in that their signals are in the form of sampled data. With the development of the digital computer, the discrete time system theory plays an important role in control theory. Thus many authors have studied the stability of such systems, see . In most cases these articles are referred to regular discrete time systems. In this article we study singular linear matrix difference equations. Thus we consider the singular discrete time system

$F Y k + 1 =G Y k + V k$
(1)

with known initial conditions

$Y k 0$
(2)

where $F,G∈M ( m × m ; F )$, (i.e. the algebra of square matrices with elements in the field ) with $Y k , V k ∈ M ( m × 1 ; F )$ and F is a singular matrix (detF = 0). For the sake of simplicity we set $M m =M ( m × m ; F )$ and $M n m =M ( n × m ; F )$. With $0 m , n ∈ M m n$ we will denote the zero matrix. For V k = 0m,1we get the homogeneous system of (1)

$F Y k + 1 =G Y k$
(3)

Because of the singularity of the matrix F, in order to solve and to study these type of systems there are in the literature two methods. The first method is by using the theory of the Drazin inverse, see , and the second is by using matrix pencil theory and the Weierstrass canonical form which is a generalization of the Jordan canonical form. The advantage of the second method is that it gives a better understanding of the structure of the system and more deep, elegant results. In this article we will present a theory based on the matrix pencil of the system and we will show how the eigenvalues of the pencil are related with the stability of singular systems.

## 2 Mathematical backround

### 2.1 The matrix pencil

Matrix pencil theory has been used many times in articles for the study of linear discrete time systems with constant matrices, see for instance [9, 14, 21, 2733]. A matrix pencil is a family of matrices sF-G, parametrized by a complex numbers, see [14, 21, 23, 27, 3436]. When G is square and F = I m , where I m is the identity matrix, the zeros of the function det(sF-G) are the eigenvalues of G. Consequently, the problem of finding the nontrivial solutions of the equation

$sFX=GX$
(4)

is called the generalized eigenvalue problem. Although the generalized eigenvalue problem looks like a simple generalization of the usual eigenvalue problem, it exhibits some important differences. In the first place, it is possible for det(sF-G) to be identically zero, independent of s. Second, it is possible for F to be singular, in which case the problem has infinite eigenvalues. To see this, write the generalized eigenvalue problem in the reciprocal form

$FX= s - 1 GX$
(5)

If F is singular with a null vector X, then FX = 0m,1, so that X is an eigenvector of the reciprocal problem corresponding to eigenvalue s-1 = 0; i.e., s = ∞.

Definition 2.1.1. Given $F,G∈ M m n$ and an indeterminate $s ∈F$, the matrix pencil sF-G is called regular when m = n and det(sF - G) ≠ 0. In any other case, the pencil will be called singular.

The class of sF-G is characterized by a uniquely defined element, known as complex Weierstrass canonical form, sF w - Q w , see [14, 21, 27, 3436], specified by the complete set of invariants of sF-G.

This is the set of elementary divisors (e.d.) obtained by factorizing the invariant polynomials into powers of homogeneous polynomials irreducible over field . In the case where sF-G is regular, we have e.d. of the following type:

• e.d. of the type $( s - a j ) p j$, are called finite elementary divisors (f.e.d.), where a j is a finite eigenavalue of algebraic multiplicity p j

• e.d. of the type $ŝ q = 1 s q$ are called infinite elementary divisors (i.e.d.), where q the algebraic multiplicity of the infinite eigenvalues

We assume that $∑ i = 1 ν p j =p$ and p+q = m.

Definition 2.1.2. Let B1, B2, . . . , B n be elements of $M n$. The direct sum of them denoted by B1 B2 . . . B n is the blockdiag [B1 B2 . . . B n ].

From the regularity of sF-G, there exist nonsingular matrices $P,Q∈ M m$ such that

$P F Q = F w = I p ⊕ H q P G Q = G w = J p ⊕ I q$
(6)

Where sF w - Q w is the complex Weierstrass form of the regular pencil sF-G and is defined by

$s F w - Q w :=s I p - J p ⊕s H q - I q$
(7)

where the first normal Jordan type element is uniquely defined by the set of the finite eigenvalues,

$( s - a 1 ) p 1 ,…, ( s - a ν ) p ν$

of sF-G and has the form

$s I p - J p :=s I p 1 - J p 1 ( a 1 ) ⊕⋯⊕s I p ν - J p ν ( a ν )$
(8)

The second uniquely defined block sH q - I q corresponds to the infinite eigenvalues

$ŝ q 1 , . . . , ŝ q σ , ∑ j = 1 σ q j = q$

of sF-G and has the form

$s H q - I q : = s H q 1 - I q 1 ⊕ . . . ⊕ s H q σ - I q σ$
(9)

Thus, H q is a nilpotent element of $M n$ with index $q ̃ = max { q j : j = 1 , 2 , . . . , σ }$, where

$H q q ̃ = 0 q , q$

and $I p j , J p j ( a j ) , H q j$ are defined as

$I p j = 1 0 ⋯ 0 0 0 1 ⋯ 0 0 ⋮ ⋮ ⋱ ⋮ ⋮ 0 0 ⋯ 0 1 ∈ M p j , J p j ( a j ) = a j 1 ⋯ 0 0 0 a j ⋯ 0 0 ⋮ ⋮ ⋱ ⋮ ⋮ 0 0 ⋯ a j 1 0 0 ⋯ 0 a j ∈ M p j H q j = 0 1 ⋯ 0 0 0 0 ⋯ 0 0 ⋮ ⋮ ⋱ ⋮ ⋮ 0 0 ⋯ 0 1 0 0 ⋯ 0 0 ∈ M q j .$
(10)

For algorithms about the computations of the Jordan matrices, see [14, 21, 3436].

### 2.2 The solution of an homogeneous singular linear discrete time system

In this subsection, we will obtain formulas of the solutions of homogeneous singular linear matrix difference equations.

Definition 2.2.1. Consider the system (1) with known initial conditions (2). Then the initial conditions are called consistent should they satisfy (3) and non-consistent should they not.

For the regular matrix pencil of system (3), there exist nonsingular matrices $P,Q∈ M m$ as applied in (6), see subsection 2.1. Let

$Q= Q p Q q$
(11)

where $Q p ∈ M m p$ is a matrix with columns the p linear independent (generalized) eigenvectors of the p finite eigenvalues of sF-G and $Q p ∈ M m q$ is a matrix with columns the q linear independent (generalized) eigenvectors of the q infinite eigenvalues of sF-G.

Proposition 2.2.1. The initial conditions (2) of the system (3) are consistent if and only if

$Y k 0 ∈ colspan Q p$
(12)

Proof. See [14, 21, 30, 32]

Proposition 2.2.2. Consider the system (3) with initial conditions (2). Then the solution is unique if and only if the initial conditions are consistent. Moreover the analytic solution is given by

$Y k = Q p J p k - k 0 Z k 0 p$
(13)

where $Z k 0 p$ is the unique solution of the algebraic system $Y k 0 = Q p Z k 0 p$.

Proof. See [14, 21, 2833, 3739]

### 2.3 The solution of a non homogeneous singular linear discrete time system

Consider the singular discrete time system (1) with known initial conditions (2).

For the regular matrix pencil sF-G, there exist nonsingular matrices $P,Q∈ M m$ as applied in (6), see also Section 2.1. Let

$P= P 1 P 2$
(14)

with $P 1 ∈ M p m$, $P 2 ∈ M q m$ and

$Q= Q p Q q$

where $Q p ∈ M m p$ is a matrix with columns the p linear independent (generalized) eigenvectors of the p finite eigenvalues of sF-G and $Q q ∈ M m q$ is a matrix with columns the q linear independent (generalized) eigenvectors of the q infinite eigenvalues of sF-G.

Proposition 2.3.1. Consider the system (1) with initial conditions (2). Then the solution is unique if and only if

$Y k 0 ∈ colspan Q p - Q q ∑ i = 0 q * - 1 H q i P 2 V k 0 + i$
(15)

Moreover the analytic solution is given from the formula

$Y k = Q p J p k - k 0 Z k 0 p + ∑ i = k 0 k - 1 J p k - i - 1 P 1 V i - Q q ∑ i = 0 q * - 1 H q i P 2 V k + i$
(16)

where $Z k 0 p$ is the unique solution of the algebraic system

$Y k 0 = Q p Z k 0 p - Q q ∑ i = 0 q * - 1 H q i P 2 V k 0 + i$

Proof. See [14, 21, 2833, 3739]

## 3 Stability of equilibrium state(s) of homogeneous singular discrete time systems

Definition 3.1. For any system of the form (1), with a constant input vector V k = V, Y* is an equilibrium state if it does not change under the initial condition, i.e.: Y* is an equilibrium state if and only if $Y k 0 = Y *$ implies that Y k = Y * for all k ≥ k0+1.

The set of equilibrium states for a given singular linear system in the form of (3) is given by the following Proposition.

Proposition 3.1. Consider the system (3). Then if 1 is not an eigenvalue of the matrix pencil sF-G then

$Y * = 0 m , 1$
(17)

is the unique equilibrium state of the system (3). If 1 is a finite eigenvalue of the matrix pencil sF-G, then the set E of the equilibrium points of the system (3) is the vector space defined by

$E= N r ( F - G ) ∩ colspan Q p$
(18)

where N r is the right null space of the matrix F-G, Q p is a matrix with columns the p linear independent (generalized) eigenvectors of the p finite eigenvalues of the matrix pencil sF-G or from Proposition 2.2.1 the set of the consistent initial conditions of (3).

Proof. If Y* is an equilibrium state of system (3), then this implies that for

$Y k 0 = Y *$

we have

$Y * = Y k = Y k + 1$

If 1 is not an eigenvalue of the matrix pencil sF-G then det(F-G)≠0 and from the system (3) we have

$F Y * =G Y *$

or

$( F - G ) Y * = 0 m , 1$

Then the above algebraic system has the unique solution

$Y * = 0 m , 1$

which is the unique equilibrium state of the system (3). If 1 is a finite eigenvalue of the matrix pencil sF-G then det(F-G) = 0. If Y* is an equilibrium state of system (3), then this implies that for

$Y k 0 = Y *$

we have

$Y * = Y k = Y k + 1$

This requires that Y * must be a consistent initial condition which from Proposition 2.2.1 is equal to

$Y * ∈ colspan Q p$

Moreover from system (3) we have

$F Y * =G Y *$

or

$( F - G ) Y * = 0 m , 1$

or

$Y * ∈ N r ( F - G )$

So

$Y * ⊆ N r ( F - G ) ∩ colspan Q p$

or

$E⊆ N r ( F - G ) ∩ colspan Q p$
(19)

Let now Y* N r (F - G) ∩ colspanQ p then we can consider

$Y k 0 = Y *$

as a consistent initial condition and

$( F - G ) Y * = 0 m 1$

or

$F Y * =G Y *$

where Y* is solution of system (3) and combined with $Y k 0 = Y *$ we have Y* E or

$N r ( F - G ) ∩ colspan Q p ⊆E$

From (19) and (20) we arrive at (18).

Definition 3.2 . Let $X∈ M r 1$ and $||X||∈F$ be norm of vector X. Then if $A∈ M r r$

$| | A | | = max | | A X | | | | X | | : X ∈ M r 1 , X ≠ 0 r , r$
(21)

Definition 3.3 . Let A = [a ij ]i, j = 1,2,..., rbe a square matrix with $A∈ M r r$ and let X = [x i ]i = 1,2,..., rbe a vector with $X∈ M r 1$, then

$| | A | | 1 = max 1 ≤ j ≤ r ∑ i = 1 r | a i j | | | X | | 1 = ∑ i = 1 r | x i |$
(22)

is by definition the 1-norm for matrices and the 1-morm for vectors respectively, which is simply the maximum absolute column sum of the matrix and the absolute column sum of the vector respectively. Furthermore

$| | A | | ∞ = max 1 ≤ i ≤ r ∑ j = 1 r | a i j | | | X | | ∞ = max 1 ≤ i ≤ r | x i |$
(23)

is by definition the ∞-norm for matrices and the ∞-norm for vectors respectively, which is simply the maximum absolute row sum of the matrix and the maximum absolute row of the vector.

Definition 3.4 . An equilibrium state Y* E of the system (3) is stable in the sense of Lyapounov if, for every δ > 0, there exists ϵ > 0, such that the trajectories starting in

$|| Y k 0 - Y * ||≤δ$
(24)

do not leave

$||Y- Y * ||≤ϵ$
(25)

as k increases indefinitevely.

Definition 3.5 . An equilibrium state Y* E of the system (3) is asymptotically stable if it is stable in the sense of Lyapounov and if every solution starting within (24) converges without leaving (25) to Y* as k increases indefinitely, i.e.

$lim k → ∞ Y k = Y *$
(26)

Definition 3.6. An equilibrium state Y* E of the system (3) is asymptotically stable in the large if, asymptotic stability holds for every equilibrium state of the system. In this case the system (3) is asymptotically stable in the large. Consequently, a necessary condition for asymptotic stability in the large is that there exists a unique equilibrium state since the limit limk→∞Y k is unique. For system (3) this unique equilibrium state is Y* = 0m,1.

Theorem 3.1. Consider the system (3) and its solution (13). Then an equilibrium state Y* E of the singular discrete time system (3) is stable in the sense of Lyapounov if there exist a constant c (0, +∞), such that $J p k - k 0 ≤ c <+∞$, for all k ≥ k0.

Proof. The solution of the system (3) is given from (13),

$Y k = Q p J p k - k 0 Z k 0 p$

where $Z k 0 p$ is the solution of the algebraic system

$Y k 0 = Q p Z k 0 p$

Since the columns of the matrix Q p are linear independent, the matrix is left invertible. Thus we can define its left inverse matrix $Q p - 1 ∈ M p m$ such that

$Q p - 1 Q p = I p$

Then

$Y k = Q p J p k - k 0 Q p - 1 Y k 0$

We assume that there exist a constant c (0, +∞) such that $J p k - k 0 ≤ c <+∞$, for all k > k0. Furthermore let an equilibrium state Y* E. Then

$Y * = Q p J p k - k 0 Q p - 1 Y *$

and easy we obtain

$Y k - Y * = Q p J p k - k 0 Q p - 1 ( Y k 0 - Y * )$

or

$Y k - Y * = Q p 0 m , q J p k - k 0 0 p , q 0 q , p 0 q , q Q p - 1 0 q , m ( Y k 0 - Y * )$
(27)

If we set $Q p = Q p 0 m , q$ and $Q p - 1 = Q p - 1 0 q , m$. Then by taking norms for every k≥k0 in (27) we have

$| | Y k - Y * | | ≤ | | Q p | | J p k - k 0 Q p - 1 | | Y k 0 - Y * | |$
(28)

Hence for any ϵ > 0, if we chose $δ ( ϵ ) = ϵ Q p Q p - 1 c ,$ then for

$| | Y k 0 - Y * | | ≤ δ ( ϵ )$

implies that for every ϵ > 0

$|| Y k - Y * ||≤|| Q p || J p k - k 0 Q p - 1 || Y k 0 - Y * ||≤|| Q p ||c Q p - 1 ϵ | | Q p | | Q p - 1 c ≤ϵ$

or

$| | Y k - Y * | | ≤ ϵ$
(29)

Theorem 3.2. The system (3) is asymptotically stable in the large, if and only if, all the finite eigenvalues of the matrix pencil sF-G lie within the open disc,

$|s|<1$

Proof. The solution of the system (3) is

$Y k = Q p J p k - k 0 Z k 0 p$

Let a j be a finite eigenavalue of the matrix pencil sF-G with algebraic multiplicity p j . Then the Jordan matrix $J p k - k 0$ can be written as

with $J p k - k 0 ∈ M p j$ be a Jordan block. Every element of this matrix has the specific form

$( k - k 0 ) p j a j k - k 0$

The sequence

$( k - k 0 ) p j | a j k - k 0 |$

can be written as

$( k - k 0 ) p j e ( k - k 0 ) l n | a j |$

The system (3) has the unique equilibrium state Y* = 0m,1when for every j, a j ≠ 1, (see Proposition 3.1) and then the system is asymptotically stable in the large, when

$lim k → ∞ Y k = Y *$

Thus this holds if and only if

$ln | a j |<0$

or

$| a j |<1$

Then for k →+∞

$( k - k 0 ) p j e ( k - k 0 ) | ln a j | →0$

or

$( k - k 0 ) p j | a j | ( k - k 0 ) →0$

or for every k≥k0

$J p k - k 0 → 0 p , p$

Then for every initial condition $Y k 0$

$lim k → ∞ Y k = 0 m , 1$

Corollary 3.1. Let r(sF-G) = max1≤j≤ ν|a j | be the spectral radius of the finite eigenvalues of the matrix pencil sF-G. Then the system (3) is asymptotically stable in the large, if and only if

$r ( s F - G ) <1$

### 3.1 Lyapounov theorem on uniform stability

Definition 3.1.1 . The singular linear discrete time system (3) is called uniformly stable if there exists a finite positive constant c such that for any k0 and $Y k 0$ the corresponding solution satisfies

$|| Y k | | 2 ≤c|| Y k 0 | | 2$

Where ||·||2 is the Euclidean norm.

It can be shown for regular linear discrete time systems, see [7, 23], that if a positive scalar function W(Y k ) can be found such that its forward difference ΔW (Y k ), where

$ΔW ( Y k ) =W ( Y k + 1 ) -W ( Y k )$

taken along the trajectory is always negative, then as time increases, W(Y k ) takes smaller and smaller values and finally shrinks to zero, and therefore Y k also shrinks to zero. This implies the asymptotic stability of the origin of the state space. Lyapounov's main stability Theorem, provides a sufficient condition for asymptotic stability. This Theorem states that if there exists a scalar function W(Y k ) satisfying the conditions, W(Y k ) is posistive definite and ΔW(Y k ) is negative definite, then the equilibrium state at the origin is uniformly asymptotically stable.

We consider the singular discrete time system described by (3). We shall investigate the stability of this state by using this method of Lyapounov. Let us choose as a possible Lyapounov function

$W ( Y k ) = Y k * F * TF Y k$

where ()* is the Hermitian tensor and T is a positive Hermitian (or a positive definite real symmetric) matrix. Then

$ΔW ( Y k ) = Y k + 1 * F * TF Y k + 1 - Y k * F * TF Y k$

or

$ΔW ( Y k ) = ( F Y k + 1 ) * TF Y k + 1 - Y k * F * TF Y k$

or

$ΔW ( Y k ) = ( G Y k ) * TG Y k - Y k * F * TF Y k$

or

$ΔW ( Y k ) = Y k * G * TG Y k - Y k * F * TF Y k$

or

$ΔW ( Y k ) = Y k * ( G * T G - F * T F ) Y k$

Since W(Y k ) is chosen to be positive definite, we require ΔW(Y k ) be negative definite. Therefore,

$ΔW ( Y k ) =- Y k * S Y k$

where

$S=- ( G * T G - F * T F )$

must be positive definite. Note that a positive definite T is a necessary and sufficient condition.

Theorem 3.1.1. Consider the singular linear discrete time system (3) with

$n 1 I m ≤F≤ n 2 I m$

A necessary and sufficient condition for the system (3) to be uniformly stable is that, given any positive definite Hermitian (or any positive definite real symmetric) matrix S, there exists a positive definite Hermitian (or any positive definite real symmetric) matrix T with

$m 1 I m ≤T≤ m 2 I m$

such that the matrix

$S=- ( G * T G - F * T F )$

is positive definite. Where n1, n2, m1 and m2 are finite positive constants.

Proof. Suppose T satisfies the stated requirements. Given a consistent initial condition (2) and the corresponding unique solution of the system (3) from (13), we have

$( Y k ) * F * TF Y k - ( Y k 0 ) * F * TF Y k 0 = ∑ j = k 0 k - 1 ( Y j + 1 ) * F * T F Y j + 1 - ( Y j ) * F * T F Y j$

or

$( Y k ) * F * TF Y k - ( Y k 0 ) * F * TF Y k 0 = ∑ j = k 0 k - 1 ( Y j ) * G * T G Y j - ( Y j ) * F * T F Y j$

or

$( Y k ) * F * TF Y k - ( Y k 0 ) * F * TF Y k 0 = ∑ j = k 0 k - 1 ( Y j ) * G * T G - F * T F Y j$

or

$( Y k ) * F * TF Y k - ( Y k 0 ) * F * TF Y k 0 =- ∑ j = k 0 k - 1 ( Y j ) * S Y j$

where S is positive definite and thus we have

$( Y k ) * F * TF Y k - ( Y k 0 ) * F * TF Y k 0 ≤0$

Furthermore

$( Y k ) * F * TF Y k ≤ ( Y k 0 ) * F * TF Y k 0$

or

$( n 1 ) 2 m 1 || Y k | | 2 ≤ ( n 2 ) 2 m 2 || Y k 0 | | 2$

Therefore

$|| Y k | | 2 ≤ ( n 2 ) 2 m 2 ( n 1 ) 2 m 1 || Y k 0 | | 2$

And thus the system (3) is uniformly stable by Definition 3.1.1.

Example 3.1.1. Consider the system (3) and let

$F= 2 0 0 0$

and

$G= 1 2 0 - 3$

Let us choose

$- ( G * T G - F * T F ) = I 2$

If the matrix T is found to be positive definite, then the system is uniformly stable. Let $T= a b b d$. Then

$1 0 2 - 3 a b b d 1 2 0 - 3 - 2 0 0 0 a b b d 2 0 0 0 = - 1 0 0 1$

Consequently,

$T= 1 3 2 9 2 9 1 9$

By applying Sylvester's criterion for the positive definiteness of matrix T, we find T is positive definite. Hence, the system is uniformly stable.

## 4 Stability of non homogeneous singular matrix difference equations

The definition of an equilibrium state of a non homogeneous system in the form of (1) is given from definition 3.1.

Proposition 4.1. Consider the system (1) with a constant input vector V k = V. Then if 1 is not an eigenvalue of the matrix pencil sF-G

$Y * = ( F - G ) - 1 V$

is the unique equilibrium state of the system (1). If 1 is a finite eigenvalue of the matrix pencil sF-G, then the set $E ^$ of the equilibrium points of the system (1) is the vector space defined by

$E ^ = N r ( F - G ) ∩ ( colspan Q p - Q q ∑ i = 0 q * - 1 H q i P 2 V )$

where N r is the right null space of the matrix F-G and Q p , Q q , P2 are matrices defined in (11), (14).

Proof. The proof is similar to the proof of Proposition 3.1. Note that from Proposition 2.3.1, the system (1) has a unique solution if and only if the given initial conidtion (2) lies inside the set

$Y k 0 ∈ colspan Q p - Q q ∑ i = 0 q * - 1 H q i P 2 V$

Thus from Proposition 4.1, the unique equilibrium state for a system in the form of (1) with a constant input vector V k = V is Y * = (F - G)-1V, when det(F-G) ≠ 0.

Theorem 4.1. Consider the system (1) with a constant input vector V k = V and a consistent initial condition (2). Then the equilibrium state is asymptotically stable in the large, if and only if, all the finite eigenvalues of the matrix pencil sF-G lie within the open disc,

$|s|<1$

Proof. The general solution of the linear system

$F Y k + 1 =G Y k +V$

is the sum of a partial solution and the solution of the homogeneous system (3). Since an equilibrium state Y* can be assumed as a partial solution, the general solution of this system is

$Y k = Q p J p k - k 0 Z k 0 p + Y *$

Given a consistent initial condition $Y k 0$ and since the columns of the matrix Q p are linear independent, the matrix is left invertible. Thus we can define its left inverse matrix $Q p - 1 ∈ M p m$ and the general solution can be written in the form of

$Y k = Q p J p k - k 0 Q p - 1 ( Y k 0 - Y * ) + Y *$

Let a j be a finite eigenavalue of the matrix pencil sF-G with algebraic multiplicity p j . Then the Jordan matrix $J p k - k 0$ can be written as

with $J p k - k 0 ∈ M p j$ be a Jordan block. Every element of this matrix has the specific form

$( k - k 0 ) p j a j k - k 0$

The sequence

$( k - k 0 ) p j | a j k - k 0 |$

can be written as

$( k - k 0 ) p j e ( k - k 0 ) l n | a j |$

The system has a unique equilibrium state when for every j, a j ≠ 1, because then det(F-G) ≠ 0 and the unique equilibrium state is Y* = (F - G)-1V. In this case the system is asymptotically stable in the large, when

$lim k → ∞ Y k = Y *$

Thus this holds if and only if

$ln | a j |<0$

or

$| a j |<1$

Then for k →+∞

$( k - k 0 ) p j e ( k - k 0 ) | ln a j | →0$

or

$( k - k 0 ) p j | a j | ( k - k 0 ) →0$

or for every k≥k0

$J p k - k 0 → 0 p , p$

From (30)

$Y k - Y * = Q p 0 m , q J p k - k 0 0 p , q 0 q , p 0 q , q Q p - 1 0 q , m ( Y k 0 - Y * )$
(31)

If we set $Q p = Q p 0 m , q$ and $Q p - 1 = Q p - 1 0 q , m$, then by taking norms for every k≥k0 in (31), for every consistent initial condition $Y k 0$ we have

$|| Y k - Y * ||= Q p J p k - k 0 Q p - 1 ( Y k 0 - Y * ) ||≤|| Q p || J p k - k 0 Q p - 1 || Y k 0 - Y * ||→0$

and thus

$lim k → ∞ Y k = Y *$

or

$lim k → ∞ Y k = ( F - G ) - 1 V$

Knowing from Proposition 2.3.1 the closed formula that provides the unique solution of the singular system (1), we can prove the following Theorem.

Theorem 4.2. The unique solution of the non homogeneous finite (k0 ≤ k ≤ k N ) discrete time system with a bounded input vector V k (1) is bounded if every equilibrium state of the homogeneous system (3) is stable in the sense of Lyapounov.

Proof. The solution of system (1) is given from the formula (16),

$Y k = Q p J p k - k 0 Z k 0 p + ∑ i = k 0 k - 1 J p k - i - 1 P 1 V i - Q q ∑ i = 0 q * - 1 H q i P 2 V k + i .$

or

$Y k = Q p 0 m , q ( J p k - k 0 0 p , q 0 q , p 0 q , q Z k 0 p 0 q , m + ∑ i = k 0 k - 1 J p k - i - 1 0 p , q 0 q , p 0 q , q P 1 0 q , m V i ) - - Q q 0 m , p ∑ i = 0 q * - 1 H q i 0 q , p 0 p , q 0 p , p P 2 0 p , m V k + i .$
(32)

If we set $Q p = Q p 0 m , q$, $Q p - 1 = Q p - 1 0 q , m$, $P 1 = P 1 0 q , m$ and $P 2 = P 2 0 q , m$, then by taking norms in (32)

$| | Y k | | ≤ | | Q p | | J p k - k 0 Z k 0 p + | | Q p | | ∑ i = k 0 k - 1 J p k - i - 1 | | P 1 | | | | V i | | - | | Q q | | ∑ i = 0 q * - 1 H q i | | P 2 | | | | V k + i | |$
(33)

If every equilibrium state of the system (3) is stable in the sense of Lyapounov, the matrix $J p k - k 0$ is bounded,

$|| J p k - k 0 ||≤ c 1$
(34)

and moreover for a finite discrete time system, k0 ≤ k ≤ k N , we have

$∑ i = k 0 k - 1 | | J p k - i - 1 | | ≤ ( k N - 1 - k 0 ) c 1$
(35)

If the input vector V k is bounded, we have

$|| V k ||≤ c 2$
(36)

By applying (34), (35) and (36) into (33) we have

$|| Y k ||≤|| Q p || c 1 Z k 0 p +|| Q p || k N c 1 || P 1 || c 2 -|| Q q || c 3 || P 2 || c 2$

where

$c 3 = ∑ i = 0 q * - 1 || H q i ||$

and thus we have proved that every solution of a finite discrete time system in the form of (1) is bounded.

Theorem 4.3. Let the system (3) be asymptotically stable in the large. Then after a δ perturbation in the set of consistent initial conditions of the non homogeneous discrete time system (1), the unique solution changes by an amount depending on δ.

Proof. The solution of system (1) is given from the formula (16),

$Y k = Q p J p k - k 0 Z k 0 p + ∑ i = k 0 k - 1 J p k - i - 1 P 1 V i - Q q ∑ i = 0 q * - 1 H q i P 2 V k + i .$

where $Z k 0 p$ is the solution of the algebraic system

$Y k 0 = Q p Z k 0 p - Q q ∑ i = 0 q * - 1 H q i P 2 V k + i$

Since the columns of the matrix Q p are linear independent, the matrix is left invertible. Thus we can define its left inverse matrix $Q p - 1 ∈ M p m$ such that

$Q p - 1 Q p = I p$

Then

$Z k 0 p = Q p - 1 Y k 0 + Q p - 1 Q q ∑ i = 0 q * - 1 H q i P 2 V k + i$

and the solution can be written as

$Y k = Q p J p k - k 0 Q p - 1 Y k + Q p - 1 Q q ∑ i = 0 q * - 1 H q i P 2 V k + i + ∑ i = k 0 k - 1 J p k - i - 1 P 1 V i - Q q ∑ i = 0 q * - 1 H q i P 2 V k + i$

If we perturb the initial conditions of the system accordingly

$|| Y k 0 - Ỹ k 0 ||≤δ$

then the solution of the system with initial conditions $Ỹ k 0$ changes to

$Ỹ k = Q p J p k - k 0 Q p - 1 Ỹ k + Q p - 1 Q q ∑ i = 0 q * - 1 H q i P 2 V k + i + ∑ i = k 0 k - 1 J p k - i - 1 P 1 V i - Q q ∑ i = 0 q * - 1 H q i P 2 V k + i$

and substracting $Ỹ k$ from Y k , we obtain

$Y k - Ỹ k = Q p J p k - k 0 Q p - 1 ( Y k 0 - Ỹ k 0 )$

or

$Y k - Ỹ k = Q p 0 m , q J p k - k 0 0 p , q 0 q , p 0 q , q Q p - 1 0 q , m ( Y k 0 - Ỹ k 0 )$
(37)

If we set $Q p = Q p 0 m , q$ and $Q p - 1 = Q p - 1 0 q , m$. Then by taking norms for every k≥k0 in (37) we have

$Y k - Ỹ k ≤|| Q p || J p k - k 0 Q p - 1 Y k 0 - Ỹ k 0$
(38)

If every equilibrium state of the system (3) is stable in the sense of Lyapounov, the matrix $J p k - k 0$ is bounded,

$J p k - k 0 ≤ c 1$

Then from (38) we have

$Y k - Ỹ k ≤ c 1 c 2 c 3 δ$

where $c 1 = Q p , c 2 = Q p - 1$. Therefore if we chose ϵ = ϵ(δ) = c1c2c3δ, we obtain,

$Y k - Ỹ k ≤ ϵ$

## 5 State feedback stabilization

In this section we will study the case where the system (3) has non stable equilibrium states. We state the following question. Under what conditions is it possible to apply to the initial unstable system (3) an external input vector

$V k =B U k$
(39)

where $B ∈ M m r$, $U k ∈ M r 1$ and a state feedback law

$U k = G ̃ 1 Y k$
(40)

such that the new system has stable equilibrium states. If it is possible, the system (3) will be called stabilizable. So, the initial system (3) takes the form, first

$F Y k + 1 =G Y k +B U k$
(41)

and after applying the feedback law (40)

$F Y k + 1 = G ̃ Y k$
(42)

where $G ̃ =G+B G ̃ 1$ and $G ̃ 1 ∈ M r m$ is the feedback gain.

Proposition 5.1. Consider the singular system (3) and the corresponding singular system of the form (41). We also suppose that some of the finite eigenvalues of the matrix pencil sF-G does not lie within the open disc |s| < 1. Then this system becomes stable by a feedback law of the form (40) if

$rank [ s F - G , B ] =m$
(43)

Proof. Since the system (3) has finite eigenvalues of the matrix pencil sF-G that don't lie within the open disc |s| < 1, there exists non stable equilibrium state(s). We consider the system (41) and we want to apply an appropriate feedback law in the form (40), such that all finite eigenvalues of the matrix pencil of the system (41) lie within the open disc |s| < 1. Thus

$det ( s F - G ̃ ) ≠ 0$

and

$rank ( s F - G ̃ ) = m$
(44)

Thus we require

$rank [ s F - G , B ] = rank [ s F - G , B ] I m - G ̃ 1 = rank( s F - G - B G ̃ 1 )= rank ( s F - G ̃ ) =m$

or

$rank [ s F - G , B ] = m$

Example 5.1. Consider the system (3) with

$F = 1 0 - 1 0 1 0 0 0 0 , G = - 2 0 2 - 1 1 3 - 1 0 0 1$
(45)

Then

$det ( s F - G ) = ( s + 2 ) s + 2 3$

the equilibrium state is not stable. We consider now the system (41) with

$B = 1 0 0$
(46)

Since for $s =2, 2 3$

$rank[ s F - G , B ]=3$

the system is stabilizable. We assume the feedback law

$U k = 3 2 0 0 Y k$

and by applying it in (41) we get the system (42), with

$G ̃ = - 1 2 0 2 - 1 1 3 - 1 0 0 1$
(47)

with

$det ( s F - G ̃ ) = s + 1 2 s - 2 3$

and since the matrix pencil doesn't have the finite eigenvalue 1 the system has the unique equilibrium state 0m,1and moreover since all the finite eigenvalues are inside the unite circle, the unique equilibrium state is asymptotically stable in the large and

$lim k → ∞ Y k = 0 m , 1$

## 6 Conclusions

In this article, we studied the stability of a class of linear singular discrete time systems whose coefficients are square constant matrices and the leading matrix coefficient is singular. We presented a theory based on the matrix pencil of the system and we showed how the eigenvalues of the pencil are related with the stability of singular systems. We studied the stability of systems in the form of (3) and the behavior of the solution Y k as k increases from k0 to ∞. Furthermore we reformulated the Lyapounov Theorem for uniform stability to be applied in singular systems. After we considered the system (1) and proved that all solutions of the non homogeneous system are bounded provided that the homogeneous system (3) is asymptotically stable in the large. Moreover we provided properties to avoid a "blow up" in the system when having small perturbations in the initial conditions. Finally for the case of not stable equillibrium states we gave necessary and sufficient conditions for state feedback stabilization. As a further extension of this article, we can discuss possible applications based on the presented approach, as is the very famous Leondief model, see , or the very important Leslie population growth model and backward population projection, see also , the Host-parasitoid Models in physics, see  or the distribution of heat through a long rod or bar as suggested in . Furthermore another interesting case for further studies is the case of systems with a singular pencil. This is the case where the constant matrices of the system are not square or they are square with an identically zero matrix pencil. For all these, there is some research in progress.

## References

1. 1.

Arvanitis KG, Pasgianos GD, Kalogeropoulos G: Design of adaptive three-term controllers for unstable bioreactors: a Lyapunov based approach. (English) WSEAS Trans Syst 2006, 5(9):2126-2134.

2. 2.

Arvanitis KG, Kalogeropoulos G, Giotopoulos S: Guaranteed stability margins and singular value properties of the discrete-time linear quadratic optimal regulator. (English) IMA J Math Control Inf 2001, 18(3):299-324. 10.1093/imamci/18.3.299

3. 3.

Bistritz Y: Stability testing of two-dimensional discrete linear system polynomials by a two-dimensional tabular form. (English). IEEE Trans. Circ Syst I Fund Theory Appl 1999, 46(6):666-676. 10.1109/81.768823

4. 4.

Campbell SL: Singular Systems of Differential Equations. Volume 1. Pitman, San Francisco; 1980.

5. 5.

Cao J, Zhong S, Hu Y: A descriptor system approach to robust stability of uncertain degenerate systems with discrete and distribute delays. (English) J Control Theory Appl 2007, 5(4):357-364. 10.1007/s11768-006-6188-7

6. 6.

Chen Y: Permanence and global stability of a discrete cooperation system. (English) Ann Diff Equ 2008, 24(2):127-132.

7. 7.

Ogata K: Discrete Time Control Systems. Prentice Hall; 1987.

8. 8.

Dai L: Singular Control Systems. In Lecture Notes in Control and information Sciences Edited by: Thoma M, Wyner A. 1988.

9. 9.

Dassios IK: Perturbation and robust stability of autonomous singular linear matrix difference equations. Appl Math Comput 2012, 218: 6912-6920. 10.1016/j.amc.2011.12.067

10. 10.

Debeljkovi DLj Stojanovi SB, Vinji NS, Milinkovi SA: A quite new approach to the asymptotic stability theory: discrete descriptive time delayed system. (English) Dyn Contin Discr Impuls Syst Ser A Math Anal 2008, 15(4):469-480.

11. 11.

Chen G, Liu ST: Linearization, stability, and oscillation of the discrete delayed logistic system. IEEE Trans Circ Syst 2003, 50: 822-826. 10.1109/TCSI.2003.812618

12. 12.

Han QL: A descriptor system approach to robust stability of uncertain neutral systems with discrete and distributed delays. (English) Automatica 2004, 40(10):1791-1796. 10.1016/j.automatica.2004.05.002

13. 13.

Hinrichsen D, Son NK, Ngoc Pham HA: Stability radii of higher order positive difference systems. Syst Control Lett 2003, 49(5):377-388. 10.1016/S0167-6911(03)00116-6

14. 14.

Kalogeropoulos GI: Matrix Pencils and Linear Systems, Ph.D Thesis. City University, London; 1985.

15. 15.

Kipnis MM, Malygina VV: The stability cone for a matrix delay difference equation. (English) Int J Math Math Sci 2011, 2011: 15. (Article ID 860326)

16. 16.

Klamka J: Controllability of dynamical systems. Matematyka Stosowana 2008, 50(9):57-75.

17. 17.

Klamka J: Controllability of nonlinear discrete systems. Int J Appl Math Comput Sci 2002, 12(2):173-180.

18. 18.

Klamka J: Controllability and Minimum Energy Control Problem of Fractional Discrete-Time Systems. In Chapter in Monograph New Trends in Nanotechnology and Fractional Calculus. Edited by: Baleanu D, Guvenc ZB, Tenreiro Machado JA. Springer-Verlag, New York; 2010:503-509.

19. 19.

Lewis FL: A survey of linear singular systems. Circ Syst Signal Process 1986, 5: 3-36. 10.1007/BF01600184

20. 20.

Ostalczyk P: Stability analysis of a discrete-time system with a variable-, fractional-order controller. (English) Bull Pol Acad Sci Tech Sci 2010, 58(4):613-619.

21. 21.

Papachristopoulos DP: Analysis and applications of linear control systems, Ph.D. Thesis. University of Athens, Greece; 2008.

22. 22.

Rachik M, Lhous M, Tridane A: On the improvement of linear discrete system stability: the maximal set of the F -admissible initial states. (English) Rocky Mt J Math 2004, 34(3):1103-1120. 10.1216/rmjm/1181069845

23. 23.

Rugh WJ: Linear System Theory. Prentice Hall International, London; 1996.

24. 24.

Sandefur JT: Discrete Dynamical Systems. Academic Press, San Diego; 1990.

25. 25.

Steward GW, Sun JG: Matrix Perturbation Theory. Oxford University Press, Oxford; 1990.

26. 26.

Martynyuk AA, Slyn'ko VI: Stability results for large-scale difference systems via matrix-valued Liapunov functions. (English) Nonlinear Dyn Syst Theory 2007, 7(2):217-224.

27. 27.

Karcanias N, Kalogeropoulos G: Geometric theory and feedback invariants of generalized linear systems: a matrix pencil approach. Circ Syst Signal Process 1989, 8(3):375-397. 10.1007/BF01598421

28. 28.

Dassios IK: Homogeneous linear matrix difference equations of higher order: regular case. Bull Greek Math Soc 2009, 56: 57-64.

29. 29.

Dassios IK: On a boundary value problem of a class of generalized linear discrete-time systems. Adv Diff Equ 2011, 2011: 51. 10.1186/1687-1847-2011-51

30. 30.

Dassios I: On non-homogeneous generalized linear discrete time systems. In Circ Syst Signal Process. Springer; 2012. 10.1007/s00034-012-9400-7

31. 31.

Grispos E, Giotopoulos S, Kalogeropoulos G: On generalised linear discrete-time regular delay systems. J Inst Math Comput Sci Math Ser 2009, 13(2):179-187.

32. 32.

Kalogeropoulos G, Arvanitis KG: A matrix-pencil-based interpretation of inconsistent initial conditions and system properties of generalized state-space systems. IMA J Math Control Inf 1998, 15(1):73-91. 10.1093/imamci/15.1.73

33. 33.

Kalogeropoulos G, Stratis IG: On generalized linear regular delay systems. J Math Anal Appl 1999, 237(2):505-514. 10.1006/jmaa.1999.6458

34. 34.

Gantmacher RF: The Theory of Matrices I, II. AMS Chelsea Publishing, Chelsea, New York; 1959.

35. 35.

Mitrouli M, Kalogeropoulos G: A compound matrix algorithm for the computation of the Smith form of a polynomial matrix. Numer Algor 1994, 7(2-4):145-159.

36. 36.

Kalogeropoulos GI, Psarrakos P, Karcanias N: On the computation of the Jordan canonical form of regular matrix polynomials. Linear Algebra Appl 2004, 385: 117-130.

37. 37.

Martin HM, Tovbis A: On formal solutions of linear matrix differential-difference equations. Linear Algebra Appl 2002, 345: 29-42. 10.1016/S0024-3795(01)00449-9

38. 38.

Hassel MP: The Spatial and Temporal Dynamics of Host-Parasitoid Interactions. Oxford University Press, New York; 2000.

39. 39.

Cheng H-W, Yau SS-T: More explicit formulas for the matrix exponential. Linear Algebra Appl 1997, 262: 131-163.

## Acknowledgements

I would like to express my sincere gratitude to my supervisor Professor G. I. Kalogeropoulos for making personal notes and notations available to me that I used in the article as well as for his helpful and fruitful discussions that led to necessary changes and modifications in the proof of the Theorems. Moreover, I am very grateful to the anonymous referees for their valuable suggestions that clearly improved this article.

## Author information

Authors

### Corresponding author

Correspondence to Ioannis K Dassios.

Competing interests

The author declares that they have no competing interests.

## Rights and permissions

Reprints and Permissions

Dassios, I.K. On stability and state feedback stabilization of singular linear matrix difference equations. Adv Differ Equ 2012, 75 (2012). https://doi.org/10.1186/1687-1847-2012-75

• Accepted:

• Published:

### Keywords

• matrix difference equations
• linear
• discrete time system
• stability
• equilibrium state
• pencil
• singular 