Stability of Arbitrarily Switched Discrete-time Linear Singular Systems of Index-1

(1)

77

Stability of Arbitrarily Switched Discrete-time Linear Singular

Systems of Index-1

Pham Thi Linh

*

VNU University of Science, 334 Nguyen Trai, Thanh Xuan, Hanoi, Vietnam

Received 22 December 2018

Revised 27 December 2018; Accepted 28 December 2018

Abstract: In this paper, the index-1 notion for arbitrarily switched discrete-time linear singular systems (SDLS) has been introduced. Based on the Bohl exponents of SDLS as well as properties of associated positive switched systems, some necessary and sufficient conditions have been established for exponential stability.

Keywords: Switched system, linear discrete-time singular system, positive system, index-1 system.

1. Introduction

Recently there has been a great interest in arbitrarily switched discrete-time linear singular systems due to their importance in both theoretical and practical aspects, see [1- 4], and the references therein. Consider a switched system consisting of a set of subsystems and a rule that describes switching among them. It is well known that, even if all linear descriptor subsystems are stable but inappropriate switching may make the whole system unstable. On the other hand, since abrupt changes in system dynamics may be caused by unpredictable environmental factors or component failures, it is important to require the stability for some real-life switched systems under arbitrary switching. It should be noted that although there are a few works devoted to stability analysis of SDLS, see [1, 3-5], to our best of knowledge, the problem of investigating the stability for such switched systems via their Bohl exponents or properties of associated positive switched systems has not yet been studied before. Thus, this work was intended as an attempt to fill this gap.

2. Switched discrete-time linear singular systems of index-1

Consider the following autonomous SDLS of the form:

________

_{Tel.: 84-336112702.} Email: linhpt1803@gmail.com

(2)

(k 1) ( 1) ( )k ( )

E_  x k A_ x k (1)

where : N {0} I_N : {1, 2,.., },N N N,is a switching signal taking values in the finite set I_N;

, Rn n

i i

E A  are given matrices, and x k( )Rnare unknown vector for al kN. Suppose that the

matrices Eiare singular for all i1,2,..,N.

We remark that in some works on SDLS [4, 6], instead of (1), a simpler system of the form

( )k ( 1) ( )k ( )

E_ x k A_ x k ,

can be considered. Moreover, all the techniques developed in this paper can easily be applied to the above mentioned SDLS.

Definition 1 System (1) is called an arbitrarily switched singular system of 1 (shortly, index-1 SDLS) if it satisfies the following conditions

(i) rankEi  r n;

(ii) S_ij kerE_i {0}i j, , where S_ij A_i1(ImE_j) { :



A_i



Im }E_j . From condition (ii) in Definition 1 we show that

ker Rn , {1,2,.., }.

ij i

S  E  i j N

Indeed, put W_ij ImA_iImE_j. Then consider linear operators T_ij:S_ij W_ij, defined by T_ijxA_ijx,

we can easily show that kerTij kerAi. According to [7] we have

dimS_ij dimW_ij dim kerT_ij dimW_ij dim kerA_i.

On the other hand

dim dim(Im Im )

dim Im dim Im dim(Im Im ).

ij i j

i j i j

W A E

A E A E



  

 

From last the relation we get

dim dim Im dim Im dim(Im Im ) dim(ker )

dim(Im Im ).

ij i j i j i

i j

S A E A E A

n r A E

   

   



This relation shows that dimSij r.Moreover, from condition (ii) in Definition 1 we have

dimSij r. Hence dimSij r, i.e., ker R

n

ij i

S  E  .

Define the matrix { ,..., ,1 r r 1,..., n}

ij ij ij i i

V  s s h h , whose columns form bases of Sijand kerEi,

respectively, and Qdiag(O I_r, _{n r}_ ), PI_n-Q. Here Oris the r r zero matrix and Imstands for the

m m identity matrix.

Then the matrix Q_ij V QV_ij _ij1 defines a projection onto kerE_ialong S_ijand P_ij I_nQ_ijis the projection onto Sijalong kerEi.

Using similar arguments as in [8-11] we can prove the following results.

Theorem 1 For index-1 SDLS (1), the following assertions hold. (i) Gijk Ej AV QVi ij jk1



(3)

(ii) E Pj jk Ej; (iii) P_jk G E_ijk1 _j;

(iv)V G AV Q_jk1 _ijk1 _i _ij Q.

Proof.

(i) Assume that xkerGijk, we have 0G xijk 

1

(E_j AV QV_{i ij} _jk)x 1

j i ij jk

E xAV QV x .Then 1

j i ij jk

E x AV QV x , thus V QV x_ij _jk1 S_{i j}_, . Furthermore, V QV x_ij _jk1 V QV V V x_ij _ij1 _ij _jk1  Q V V x_ij _ij _jk1 kerE_i.

Since S_ijkerE_i{0}we get V QV x_ij _jk1 0, thus E x_j  AV QV x_{i ij} _jk1 0, hence xkerE_jImQ_jk, i.e.,

jk

xQ x. On the other hand, from the relation Q x V V V QV x_jk  _jk _ij1 _ij _jk1 0, we have xQ x_jk 0. It

means that kerGijk{0}, i.e., the matrix Gijkis non-singular.

(ii) Since Qjkis the projection onto kerEjthen we have E Qj jk 0, i.e.,

( )

j j jk jk j jk

E E P Q E P

.

(iii) From relation G P_ijk _jk  (E_j AV QV_i _ij _jk1)V PV_jk _jk1  E P_j _jk AV QPV_i _ij _jk1 E_j, we get 1

.

jk ijk j

P G E

(iv) From formula of G_ijk E_jAV QV_{i ij} _jk1we have G_ijkV_jk E_jV_jk AV Q_i _ij , thus

i ij ijkVjk j jk

AV QG E V .

The last assertion follows from relations:

1 1 1 1

1 1 1

1

( )

.

jk ijk i ij jk ijk ijk jk j jk

jk jk jk ijk j jk

n jk jk jk

V G AV Q V G G V E V V V V G E V

I V P V Q

   

  



 

   



Theorem 1 is proved.

Using items (iii), and (iv) of Theorem 1, we get

1 1 1

: ijk ;

ijk jk ijk i ij

n r

A O A V G AV

O I

 



 

 _{ } 

 

(2)

1 1

: r

ijk jk ijk j jk

n r

I O

E V G E V

O O

 



 

 _{ } _

 .

Theorem 2 The index-1 SDLS (1) has a unique solution with x(0) x0 Rif and only if x0S(0) (1)

, i.e., the initial condition x0is consistent. In this case, the following solution formula holds.

1 ( ) ( 1) ( 1) ( ) ( 1) (0) (1) (2) (0) (1)

( ) _k _k _k _k _k ... (0).

x k V_ _ _ A_ _ _ _ _ A_ _{ } V_ _ x

Proof.

Multiplying both sides of system (1) by V(1k 1) (k 2)G1( ) (k k 1) (k 2)

 

    , and using the transformation

1 ( ) ( 1)

( ) _k _k ( )

(4)

E__{( ) (}_k_ _k__{1) (}__k_₂₎x k(  1) A__{( ) (}_k_ _k__{1) (}_ _k_₂₎x k( ). (3)

Putting _{x k}( ) : ( ( ) , ( ) ) _{v k} T _{w k} T T

, where _{v k}( )Rr

, _{w k}( )Rn r

, we can reduce system (3) to the following systems

1

( ) ( 1) ( 2)

( 1) ( ),

( ) 0.

k k k

v k A v k

w k

    

  

 

 (4)

System (4) has the solution

1 1

( 1) ( ) ( 1) (0) (1) (2)

( ) ... (0),

( ) 0,

k k k

v k A A v

w k

       

 

hence the solution of system (1) can be written as

( ) ( 1)

( ) ( 1) ( 1) ( ) ( 1) (0) (1) (2)

1 ( ) ( 1) ( 1) ( ) ( 1) (0) (1) (2) (0) (1)

( ) ( )

( )

(0) ...

0

... (0).

k k

k k k k k

x k V x k

v k V

w k

v

V A A

V A A V x

 

       

         



  



  



 

 _ _

 

 

 _ _

 



3. Stability of linear switched singular systems of index-1

Suppose that system (1) is of index-1 and the initial condition x₀ is consistent.

Definition 2 System (1) is called exponentially stable if there exist a positive

constant  and a constant 0  1such that such that for all switching signals and all solutions x of (1)

the following inequality holds

0

( ) k 0.

x k  x  k

3.1. Bohl exponents and exponential stability

To define Bohl exponent for system (1), we first construct the so-called one-step solution operator ( ,k k 1)



  from x k



1



to x(k).

( ) ( 1)

( ) ( 1) ( 1) ( ) ( 1) 1 ( ) ( 1) ( 1) ( ) ( 1) ( 1) ( )

( ) ( )

( )

( 1)

( 1).

k k

k k k k k

k k k k k k k

x k V x k

v k V

w k

V A x k

V A V x k

 

    

      



  



   



 

 _ _

 

 

Then put ( ,k k1) : (k1) ( ) ( k k1)

1 ( ) (k k 1) (k 1) ( ) (k k 1) (k1) ( )k

V_ _ _ A_ _ _ _ _ V_ _ _ we get the following

(5)

( ) ( , 1) ( 1).

x k  _ k k x k

Hence we can define the state transition matrix as

( 1) ( ) ( 1) ( ) ( 1) ( 2)

( , ) :i j _i _i _i ... _j _j _j , i j 0.

          

      

Definition 3 Assume that system (1) is of index-1 and _( , )i j is the state transition matrix. Then Bohl exponent for system (1) is defined as follows:

inf{ R : : ( , ) . i j, , 0}.

B w Mw  i j M ww i j

 _ _ _ _‖ _ _‖ _  _ _{ }

To show the existence of Bohl exponent _Bfor system (1) we will prove that the

set

{ R : _w: ( , ) _w. i j, , 0},

S w M ‖ _ i j‖ M w  i j

is non-empty and bounded from below.

Indeed, from the formula  _ijk V A V_jk _ijk _ij1, , ,i j k{1,2,.., },N we see that the set of matrices ijkis

finite, then there exists a positive constant  0such that

, ,max{1,2,.., } ijk .

i j k N





 ‖  ‖

Thus we obtain that

( , )i j i j, , i j 0,

   

    

‖ ‖

hence S. Besides, for all wSwe have w0. It follows that the set S is non-empty and

bounded from below.

Lemma 1 Assume that system (1) is of index-1 and _( , )i j is the state transition matrix. Then 1

lim max ( ,0) i.

B _i _  i





 ‖  ‖ (5)

Proof.

We carry the proof of Lemma 1 in 3 steps. Step 1.We show the existence of the limit in (5)

Put aimax_ ‖ ( , 0)i ‖ . Then we have aij a ai jfor all ,i j0.According to Polya-Szego [12]

we obtain that 1

lim i

i

ia exists. It means that the limit in (5) exists.

Step 2.Put

1

1 lim max_i _ ( ,0)i i .





 ‖  ‖ We prove  1 B.

Since _B infSthen for all 0there exists w Ssuch that w B , i.e., there exists Mw

such that

( ,0)i M_w( _B )i j, ,i 0.









    

‖ ‖

It follows

1

lim max ( ,0) i .

B

i  ‖  i ‖  

(6)

Step 3. We prove B 1.

From the definition of ₁, for all 0there exists T >0 such that

1

| i | , ,

i

a    i T

i.e.,

1

( ,0)i ( ) ,i i T, .

  

     

‖ ‖ (6)

We will show that there exists M0such that

1

( , )_{i j} _M( )i j, _i _T,

   

     

‖ ‖ . (7)

Indeed, when i j T, for every _{we always have switching signal} *

such that

*

( , )i j (i j, 0)

 _

    . Hence we have

* ₁

( , )i j (i j,0) ( )i j, i j T, .

 _







         

‖ ‖ ‖ ‖

When i j T, we have the following estimate

1 1

( , ) ( ) .

i j

i j i j

i j

  _ 



   

  _ _ 



 

‖ ‖

Choosing

1

max{1, }

T

M 



 

 _ _



  , we get the inequality (7). It means that

1 B   .

Thus we obtain B 1.

Lemma 1 is proved.

Theorem 2 An index-1 SDLS (1) is exponentially stable if and only if _B 1.

Proof.

Necessity.Assume that system (1) is exponentially stable. It follows that there exist a positive

constant M0and 0  1such that

( , )i j M i j ,i j 0.

  

    

Thus, _B1.

Sufficiency. Assume that _B1. Then there exist 0and M0such that

1 B

    and _( , )i j  Mi j ,i j 0.It shows that system (1) is exponentially stable.

Theorem 2 is proved.

3.2. Stability of positive linear switched singular systems of index-1

In this Subsection, we investigate the stability of index-1 SDLS satisfying some positivity condition. Let : { x( ,x x₁ ₂,...,x_r) ,T x_i0}be a positive octant in Rr_,_Int_{( )}_{be the interior of} _.

Consider an order unit norm . _u, defined in [13], [14], and the corresponding order unit space

(7)

Theorem 3 Assume that the matrices A_ijk1 , determined by (2), are positive definite, and there exists

a vector ˆvInt( )such that vˆA v_ijk1 ˆInt P( )for all , ,i j k. Then system (4) is exponentially stable, hence system (1) is also exponentially stable.

Proof.

Since vˆA v_ijk1 ˆInt P( )then there exists a



_ijk(0,‖ ‖uˆ _u)such that the closed ball

1

ˆ ˆ

[ ijk , ijk]

B vA v  . Since _ˆ 1 _ˆ _ˆ _ˆ 1 _ˆ

[ , ]

ˆ

ijk

ijk ijk ijk

u

v A v v B v A v v



   

‖ ‖ we get

1

ˆ ˆ ˆ 0

ˆ

ijk ijk

u

v A v v v



  

‖ ‖ . Let

(0,1) ˆ ijk ijk u v   

‖ ‖ , then

1 _ˆ ₍₁ ₎_ˆ

ijk ij

A v  v.

Put inf{ijk, , ,i j k{1, 2,...,N}}, we obtain

1 _ˆ _ˆ

(1 )

ijk

A v  vfor all , ,i j k. Using the positive

definiteness of matrices A_ijk1 and the monotonicity of ˆ‖ ‖v _uwe get

ˆ

1 1

( 1) ( ) ( 1) (0) (1) (2) (R , . )

1 1

ˆ ( 1) ( ) ( 1) (0) (1) (2)

1 1

ˆ ( 1) ( ) ( 1) (1) (2) (3)

1 1

ˆ ( 1) ( ) ( 1) (1) (2) (3)

ˆ

...

ˆ ...

ˆ

... (1 )

ˆ (1 ) ... ... ˆ (1 ) r v

k k k

k k k v

k v

A A

A A v

v                                       ‖ ‖ ‖ ‖ =‖ ‖ ‖ ‖ ‖ ‖

‖ ‖  (1 ) .k

According to [15], system (4) is exponentially stable. It follows that there exist finite positive

constants 0  1and  0such that

( ) k (0) .

v k   v

‖ ‖ ‖ ‖

Furthermore since the corresponding solution of system (1) is x k( )V__{( ) (}_k_ _k_₁₎( ( ) ,0)v k T T,we have

( ) ( 1)

( 1) ( ) ( 1) ( ) ( 1) 1

( ) ( 1) (0) (1) 1 ( ) ( 1) (0) (1)

( ) ( ( ) ,0)

( (0) ,0)

(0)

(0) .

T T

k k

k T T

k k k k k

k

k k k

k k

x k V v k

V D V v

V V x

                              ‖ ‖ ‖ ‖ ‖ ‖ ‖ ‖ ‖ ‖ ‖ ‖ ‖ ‖ ‖ ‖

Putting 1

,max1,2,.., ij ij

i j N V V

  



 ‖ ‖ ‖ ‖ , we have

The last relation shows that the solution of system (1) is exponentially stable. Theorem 3 is proved.

Example 1 Put

1 2

2 3 0 3 2 0

0 2 0 , 1 6 0 ,

0 0 0 0 0 0

(8)

1 2

1 1 0 2 1 0

0 1 0 ,

1 1 2 1 ,

0 0 1 0 0

A A



   

   

_ _ _ _

   

   

11 12

1 0 0

0 0

0 0 1

1 3

V V

 

 

 _{ } _

 

 

,

21 22

2 0 0 0 2 0 0 0 1

V V

 

 

 _{ } _

 

 

.

We calculate the matrices A_ijk, , ,i j k{1, 2}as

111 112

0

0 0

0 0 1

1 2 1 12 1 2

A A

 

 

 _{ } _

 

 

,

121 122

3 16 1 12 0 0

0 0

1 6 1 1 4

A A

 

 

 

 _{ } _

 

 

,

211 212

7 8 1 3 1 4

0 0

0 0 1

1 6

A A

 

 

 _{ } _

 

 

,

221 222

5 8 1 8 1 16

0 0

0 0 1

5 16

A A

 

 

 _{ } _

 

 

.

Clearly all the matrices A_ijkare positive definite. We choose ˆ_v(9,3)T_Int( )_{and find that}

1

ˆ _ijkˆ

vA vare also inside Int( ). It means that this system satisfies all the condition of Theorem 3, thus

it is exponentially stable.

Acknowledgments

The author is grateful to Professors Pham Ky Anh, Do Duc Thuan and Stefan Trenn for useful discussions leading to the formulation of the problem as well as the results obtained in this paper.

References

[1] D. Liberzon, S. Trenn, On stability of linear switched differential algebraic equations, Proc. IEEE 48th Conf. Decision Control, December (2009) 2156.

[2] Z. Sun, S.S. Ge, Stability Theory of Switched Dynamical Systems, Springer, London, 2011.

[3] S. Trenn and F. Wirth, Linear switched DAEs: Lyapunov exponents, a converse Lyapunov theorem, and Barabanov norms, In Decision and Control (CDC), IEEE 51st_{Annual Conference, (2012) 2666.}

[4] G. Zhai, X. Xu, A unified approach to stability analysis of switched linear descriptor systems under arbitrary switching, Internat. J. Appl. Math. Comput. Sci., 20(2) (2010) 249.

[5] M. Darouach and M. Chadli, Admissibility and control of switched discrete-time singular systems, Syst. Sci. Control Engrg.: An Open Access J., 1:1 (2013) 43.

[6] G. Zhai, X. Xu, D. W.C. Ho, Stability of switched linear discrete-time descriptor systems: a new commutation condition, Internat. J. Contr., 85 (2012) 1779.

[7] S.H. Friedberg, A.J. Insel and L.E. Spence, Linear Algebra, Prentice Hall, New Jersey, 1989.

[8] P.K. Anh, P.T. Linh, Stability of periodically switched discrete-time linear singular systems, J. Difference Equ. Appl., 23(10) (2017) 1680.

[9] P.K. Anh, H.T.N. Yen, Floquet theorem for linear implicit nonautonomous difference systems, J. Math. Anal. Appl., 321 (2006) 921.

[10] P.K. Anh, N.H. Du, L.C. Loi, Singular difference equations: an overview, Vietnam J.Math., 35 (2007) 339. [11] L.C. Loi, N.H. Du, and P.K. Anh, On linear implicit non-autonomous systems of difference equations, J. Difference

Equ. Appl. 8 (2002) 1085.

[12] G. Polya and G. Szego, Problems and Theorems in Analysis I, Springer, Berlin, Heidelberg, 1998.

[13] B.Z. Vulikh, Introduction to the Theory of Cones in Normed Spaces, Izdat. Kalinin Univ., Kaliningrad, 1977. [14] T.S. Doan , A. Kalauch , M. Klose , S. Siegmund, Stability of positive linear switched systems on ordered Banach

spaces, Syst.and Contr. Lett.75 (2015) 14.