h e d i n g s of the 42nd IEEE

Conlerence on Decision and Control

Mami, Hawaii USA, December 2W3 FrP05-2
Accessibility of Switched Linear Systems

Daizhan Cheng Han-Fu Chen

Institute of Systems Science, Chinese Academy Institute of Systems Science, Chinese Academy
of Sciences of Sciences
Beijing 100080, P.R.China Beijing 100080, P.R.China
dchenq@iss03.iss.ac.cn hfchen@control.iss.ac.cn

Abstracl-The controllahility of switched linear control sys- reachable set RT(xo) contains a non-entpry set of interior
tems is investigated. Two kinds of controllabilities, namely, the points.
aeak controllahility and the controllability are proposed. Using
the generalized Chow’s theorem, the Lie algebra approach
has been implemented to reveal the controllable sub-manifold For linear systems the two Lie algebras are degen-
of switched linear systems. The topological S ~ N C ~ U I Tof the erated to the controllable (reachable) subspace C =
controllahle sub-manifolds is investigated. Span{B,AB,. . . ,An-’B). It is well known that a linear
system is controllable, iff dim(C) = n.
I. INTRODUCTION
Since the switched linear system is essentially a nonlin-
In recent years, the study of switched systems attracted ear system, its controllability bas some flavor of nonlinear
more and more attention [4]. The model is based on many systems. Meanwhile, since each particular switching model
practical engineering systems, such as robot manipulators, is linear, its controllability has also some characteristics of
traffic management, power systems etc. The controllability linear systems.
of the system was investigated by [I], [I I]. Some recent ‘This paper considers a switched control system as
results can be found in [71, [SI, [121.
Since switched linear control systems are essentially non- +
k ( t )= A,(,)z(t) B o ( t ) U ( t ) , z ( t )E u(t) E IR”’.
R”,
linear, we review the controllability of non-linear systems (4)
first. Consider an affine nonlinear system where u ( t ) : [O,m) + t is a measurable mapping t =
{1,2,. . . ,N}, and u(t)is a piecewise constant control.
+
x = f(z) g ( z ) u . z E IR”. (1)
11. A MOTIVATING EXAMPLE
Two important Lie algebras related to the controllability are
defined as To begin with, we revisit the results in [SI. The following
definition is commonly used in some recent literatures [SI,
Definition 1.1 131 For system (1)the accessibility Lie algebra [121.
is defined as
L a = {f(x),g(z)}LA: (2) Definition 2.1 A state z E R” is coritrollable at time to,
if there exist a time instant t f > t o and a switching path
and the strong accessibility Lie algebra is defined as U : [to,t f ] -+ L such that x ( t f ;to, x , U ,U ) = 0. The system
(4) is controllable if every x E R”is controllable.
Lo= Iad$g(x) I k 2 O } L A . (3)
Denote the smallest subspace containing B and
A-invariant by (A I B).
The reachable set of system (I), starting from 20, is The main result for this controllability is the following:
denoted by R(xo),the reachable set from xo at time T > 0 (The following condense form can he easily deduced form
is denoted by RT(xo). The reachability (accessibility) of algorithm (35) of [SI.)
system ( I ) is point-wise depending. For system (1) if the
(strong) accessibility Lie algebra has rank = n at zo it is Theorem 2.2 [SI For system (4) the controllable subspace is
said that the system satisfies the (strong) accessibility rank
condition at xg. The basic result is . , ,BA,).
C = ( A , ... A N 1 Bi,. (5)
Hence, the system is contmllable, iff dim(C) = n.
Theorem 1.2 [9], (101 If system ( I ) satisfies accessibiliry
rank condition at xo, the reachable set R(z0) contains a
The Definition 2.1, which is a mimic of its counterpart of
non-empty set of interior points. If system ( I ) sarisfies strong
linear systems, is the controllability at the origin. Theorem
accessibrliry rank condition az 20. then for any T > 0 the
2.2 reveals the largest controllable subspace of the switched
*Work war supported by CNNSF G60274010 and National 973 Project linear systems. This is a significant contribution of [SI.
GI998020308 of China Meanwhile, we would also like to pointed out that as we

0-7803-7924-1/03/$17.00 02003 IEEE 5759

mentioned before, a switched linear system in essence a set of C30 vectorfields, and C = { X I , .. . ,X k } ~ abe the Lie
nonlinear system, so the controllable subspace is not the algebra generared by L. Assume dim(C(x)) = constant 5
overall controllable set of the system. n, x E AI. Then f o r any xo E AI and x E S(x0) there exisf

Example 2.3 Consider thc following system x t , , . . .>XtSE L

k = A , ( ~ ) x , x E IR2, (6)
.
and t l , . . t , E IR, such that
x.* ' . . e?" (20)
for which L = 1 . 2 and x = e,,

Chow's theorem is essential for the controllability of

According to Theorem 2.2, the controllable subspace is { O } . nonlinear control systems. But sometimes it is difficult to
But it is easy to show that the controllable sub-manifold is apply Chow's Theorem to system (4) directly because the
AI, = IR2\{0}. Because along AI the trajectory can go on non-singularity assumption may not be satisfied. Since each
the direction of (1, I), and along A2 it can go on a circle switching model is an analytic system, we will generalize
path. So for any two points, x1,xZ E At,. X I E R(x2) and Chow's theorem to analytic systems without assuming non-
X? E R(X1). singularity. We need some other results for this. The fol-
The above example shown that even though the control- lowing Theorem (generalized Frobinius's Theorem) itself is
lable subspace is { O } , the system is still controllable almost important.
everywhere. So unlike the linear control systems, controllable
subspace is not enough CO describe the controllability of Theorem 3.3 (Generalized Frobinius' Theory) [3] Let AI
switched linear systems. Therefore, in addition to considerihg be an n dimensional C" (analysis) rnanifold and D be an
the controllable subspace of switched linear systems, it is analytic involurive distribution. Then f o r any x E AI there
necessary to consider the controllability sub-manifolds of the exisrs a largest iriregral manifold of D passing thmirgh x.
systems.
111. GENERALIZED THEOREMS OF FROBINIUS 8~ The following lemma is an immediate consequence of
CHOW Campbell-Baker-Hausdorff formula [61.
As in the study of nonlinear systems [3], [ 5 ] , we first define Lemma 3.4 191 Ler A be an aiialytic iuvolutive distribution,
the weak controllability for the switched linear system (4). X E A be an analytic vecrorfield. Assume 5 1 and x2 are
hyo points cormerred b y an integral curve of X , i.e., there
Definition 3.1 1 . For system (4). 5 2 is reacl~blefrom X I . exits t > 0 such that the inregral curve e; (x) of X satisfies
denoted by x2 E R(xl), i f there exisr a piecewise consfant
control arid a T > 0 such rliat the trajectory of the contmlled eF(x1) = 2 2
system satisfies rhot
Then
~ ( 0= )X I , and x(T) = ~ 2 . dim(A(xl))= dim(A(x2)).
2. Sysrem (4) is said IO be controllable at xo if R(x0) =
IR". Sysrem (4) is said to be cor~trollablei f R(x) = R". Using Theorem 3.3 and Lemma 3.4, similar argument as
vx E R". in the proof of Chow's theorem [2] can prove the following
3. 2 2 is weakly reachable froin xi, denoted by x2 E generalized Chow's theorem.
IVR(xl), if here exist po = xl, p l , . .., p , = x2 such thar
e i t h e r p i E R ( p i - I ) o r p i - I ~ R b i )i =
, l , . . . ,s. Theorem 3.5 (Generalized Chow's Theorem) Let Af be an
4. System (4) is said to be weakly conrrollable at xo if n dimensional C" manifold. L = { X I , . .. ,Xk} c V"(AI)
11'R(xo) = IR". System (4) is said to be weakly corirrollable is a ser of C" vector fields, and C = { X l , . . . ,Xk}~.4be
if WR(x) = IR". vx E R". rhe Lie algebra generated by L. For any xo E Ai, deuote by
S,, the largest integral sub-manifold of C passing rhrough
By definition it is clear that if system (4) is (weakly) 20. Then for any x E S,, rhere exist

controllable at a point x E R",it is (weakly) controllable.

x , , , . . . ,xt,E L
To explore the weak controllability, we need Chow's
theorem. and tl , . . ,t, E IR, such that
x t l .. .
= e.y*l (7)
(20).
Theorem 3.2 (Chow's Theorem) [2] Let A f be an n dimen-
sional C" manifold, = { X I , . . .: X k } c V-(Af) be a

5760
 IV. ACCESSIBILITY VS ITS LIE ALGEBRA
For system (4), the following definition is very natural. In Proof It is easy to see that A i z E La,Bi E La,i =
fact, these concepts are adopted from their counterpart for +
1 , . . . , N. Since A,x Biu is a linear combination of Aix
nonlinear systems. and B,, we have that
f , , = { A i ~ : B ij i = l , . . . , N ) L A .
Definition 4.1 For system ( 4 ) the accessibi/ir).Lie algebra is
defined as Consider a bracket-product of A i z and B i , which means a
term of brackets in any order without &. For instance,
f, := { A , z + BIul i = 1,. .N: U =
,,

and the srivng accessibility Lie algebrn is defined as AI^: 1.432, Bzll, or P I ,[Bz.[ A ~ xA~x111~
, etc.

Csa := ( A I , . . ,. A N ~ B I , . B
. .N ) First of all, if a bracket-product contains more than one Bi,
it equals zero. This is from the fact that a product contains
one Bi is a constant vector field.
We first investigate the strocture of the accessibility Lie Then we have only to consider the terms containing one
algebra e,. Bi and the terns containing no Bi.
Denote the set of { A i } by A, that is A = For terms containing one Bi, using Jacobi identity, it is
{ A I :A2.. . . , A n , ) .A is said to be symmetric if -Ai E A, easy to see that such a term can be expressed as the sum of
Vi. terms of the form
W e consider A i , i = 1,.. . ,N as elements of the Lie A,, . . . Ai,Bi, (10)
algebra gl(n.IR), and denote by A the Lie sub-algebra
generated by {A} . That is, A = { A I , . . ., A N } L AC which is obviously contained in Lsa.
gl(n,lFi). Then we define the set of linear homogeneous For terms containing no Bi, Lemma 4.2 says that
vector fields, denoted by
=A.
{ A l z , . . ., A N Z ) L A (11)
L H ( R " ) = {Ex 1 E E AI,},
From (IO) and (II), equation (9) follows. U
where Af, is the set of n x n matrices. A straightforward We now are ready to use the Lie algebra approach to
computation shows that L H ( R " ) C V"(R") is a Lie sub- investigate the accessibility of switched linear systems. First
algebra of V"'(R"), which is the Lie algebra consists of of all, we can derive a result, which is the counterpart of
the analytic vector fields on El". Now we define a mapping, Theorem 1.2 for nonlinear systems.
@ : L H ( R " ) + gl(n,El) by Q,(Ez):= -E. It follows that
Proposition 4.4 If system ( 4 ) satisfies accessibiliry rank
Lemma 4.2 I. Q, : L H ( R " ) -t gl(n,El) is a Lie algebra condition at xo, then the reachable set R ( x o ) confains a
isomorphism. non-empry set of interior points. If systern ( 4 ) sntisfies strong
2. If f c gl(n,lFi) is a Lie srrb-algebra, then its inverse accessibiliry rank condition at xo, then for any T > 0 the
intage Q,-'(f) is a Lie srrb-algebra of V"(R"), which can reachable set RT(xO)contuins a non-empry set of interior
be expressed as points.
Q,-l(f) = cz. (8) Proof Note that each switching model is an analytic system.
+
Denote L = {Aiz Biu 1 U = constant E R"'}.Assume
system (4) satisfies accessibility rank condition at xo. Then
Proof I . A straightforward computation shows that
there exists a neighborhood U of xo such that dim(f,,(z)) =
Q , ( [ A zB, z ] )= @(BA2- ABx) = - B A + AB, n, x E U . Choose any non-zero vector field X I ( X E L
J
and denote its integral curve as x l ( t l , z ( O ) ) = e;,'(so).
while If all X E L are linearly dependent with XI in U , then
[ @ ( A z ) , Q , ( B x=) ][ - A , - B ] = A B - BA. d i m ( f a ( x ) ) = 1, z E U , which is a contradiction. So
there exist a ty, a vector field XZ E L, and a mapping
Hence @ is a Lie algebra homomorphism. It is obvious that constructed as ? ~ ~ ( t ~ , t ~ ,=x eTe:l(xo).
(0)) Note that its
Q, is a bijective mapping, and so it is an isomorphism.
Jacobian matrix Jn,(ty.O) has full rank. So it is a local
2. The first half statement is obvious. The expression (8) is diffeomorphism. Same argument shows that we can find
directly from the definition of Q, by noting that a Lie algebra (ty,tS) and X, E L (9 may not the same as before),
is a vector space, so the negative sign can be omitted. 0 such that x 3 ( t l , t ~ , t 3 , ~ ( 0=) ) e ~ e ~ e ~ l ( isx ao local
)
diffeomorphism at (ty,t:, 0). Proceeding in this way, finally
Theorem 4.3 For system (4) the accessibilir). Lie algebra can we can construct a mapping
be expressed as
e,
= Ax i f s a . (9) xn(tl,t2,... ,tn,z(0))=e? ~ . . e ~ e ~ ( z o ) ,

5761
which is a local diffeomorphism at (ty,... t:) E IR". Then
~ Proof: Case 1. Assume d x o c Csa.By Lemma 3.4, at any
.
there exists an open neighborhood V of (ty:... t:), such point x E IVR(xo)
that r;'(V) c R(x0) is an open set. The conclusion follows.
Next, we assume (4) satisfies strong accessibility rank
diin(CC,(x))= dim(Cso(x)). x 6 W R ( x 0 ) .
condition at 20. Similar to the discussion for nonlinear The the weakly reachable set of x o is the reachable set of
systems [lo], an additional variable x,+1 = t can be added + +
x O ,which is zo V . Since xo 6V , ( x o V )n V = 4.
to the original system to gel an extended system as Case 2. Assume d x ~ Csa.Then

I x = Ax
X"+, =
+Bu
1
It is easy to show that the extended system satisfies the
If IVR(xo)n V
the origin. But
diln(c&o))
# 4, then
> dim(C&o))
x o is path-wise connected with

accessibility rank condition at ( x 0 , T )for any T > 0. The

dim(CC,(0))= dim(Csa(0)).
conclusion follows from the fact that for the original system
RT(x0) has a non-empty set of interiors in IR", iff for the This contradicts to the assumption. 0
extended system R ( ( x 0 , T ) )has non-empty set of interiors
in R"+'[91. 0 Remark 5.3 From Proposition 5.2 ir is clear that ifthe initial
We have some further results for switched linear systems point x(0) = xo V then there is no piecewise constaiit
as follows. control leading it to iero.
From the decomposed form of (12) we denote
Proposition 4.5 For system ( 4 ) the weakly reachable set of
any x E IR", denoted by IVR(x), is the largest integral sub-
A0 := { AT'I i = 1,... ,N } L A . (13)
manifold of Cmpassing through x. We define a projection r : A + A0 in a natural way by
Proof First, still use the fact that each vector field in letting r ( A i ) = A:', i = 1,.. . ,N.Then the following is
C, is analytic. Hence the generalized Chow's theorem is an immediate consequence of the definition.
applicable. Replacing negative time interval ti in (7) by -ti
and exchanging two ends of the segment of the integral curve, Proposition 5.4 The mpping R is a Lie algebra honionior-
x i with x i + l . the conclusion is an immediate consequence phism. Moreovez
of the generalized Chow's theorem. 0
C, = Aox2 @ C,,, (14)
V. CONTROLLABILITY AND WEAK
CONTROLLABILITY where @ is the direct sum (of TWO subspaces).
It is obvious that Ao is uniquely determined by
In this section we will first reveal a topological structure
of (weak) controllable sub-manifolds for a switched linear A0 : = { A i 1 1 5 i 5 N ; A , $ ! k e r ( a ) } .
system with dim(C,) < n.
Let V be the controllable subspace of system (4). The From Proposition 5.2 and the structure of C, the topolog-
following proposition is from the definition and Theorem 2.2. ical structure of the (weak) controllability sub-manifolds is

Proposition 5.1 For osrem ( 4 ) the coritrollable subspace

is V = I ( C s a ) ( 0 ) ,the iriregral sub-manifold of the strong
.
characterized as follows:
System (4) is globally controllable iff dim(&) = n.
As dim(&) < n, system (4) can not even be globally
accessibility Lie algebra passing through 0. weakly controllable. In this case the state space is split
Now the state space can be split as x = ( d ,z2), where into two disjoint parts: subspace V and its complement
x 1 is the coordinates of V . Since V contains B, and is Ai-
invariant for all i, system (4) can be expressed as . v c = IR=\v.
Recall (12). V = { x E IR" I x2 = 0}, so V c = { x E
IR" I x2 # 0). Assume dim(V)= n - k . If k > 1, then
V e is a path-wise connected open set, and if k = 1,
then V" is composed of two path-wise connected open
sets: {x I x2 < 0) and { x I x 2 > 0).
The following proposition gives a partition for possible
(weakly) controllable sub-manifolds. Based on the above argument,'it is clear that the (weak)
controllability of system (4) is determined by its (weak)
Proposition 5.2 Consider system (4). For any x o E R" i f controllability on V". In the later discussion we assume V c
is a path-wise connected open set. As for k = 1, it will be
x V , then
replaced by two path-wise connected open sets: { x 1 z2 < 0)
IVR(zo) n V = 4.
{z 1x2 > 0).
For further discussion we need some auxiliary definitions.

5762
Definition 5.5 For a subspace V c IR" a transformation P Proot (i) (Necessity) If there exists a zo E V c with
is called V-invariant, if PV c V . dim(dox0) = s < k, then W R ( x o ) c .7(La)(xo),and
Note that in system (12) the controllahle subspace according to the generalized Chow's theorem, the right hand
side is a sub-manifold of dimension n - k s < n, and V" +
can not he (weak) controllable.
(Sufficiency) Let 5 0 ~ x 1E V". We have to find a T > 0,
Then P is V-invariant, iff a switching law u(t)with switching moments
0 =to < ti < ... < t, =T,

and a switched vector field

Definition 5.6 System (12) is V-invariant, feedback block X ( t )= &(t) + Bv(t)uo(t): 0 C t < T ,
diagona1i:able. if there exist a V-invariant rransfonnation such rhat
+
P. feedback controls U, = K,x z',. i = 1,. . . ,N such rlwr XI = x ( T ) =:e . . . ehY,'( x o ) ,
the feedback models have block-diagonal fonn. 77wt is, by
lertirig I = Px, (12) becomes where X , = X ( t ) , ti-l 5 t < ti. d; = t; - t i - 1
Denote the starting point and the destination by

where the same A, B are used for notational ease, birr r l i q respectively. We design the control in three steps. First,
are different from those in (12). since V is controllable, we can find switching law and
controls such that at a moment TO we have x ~ ( T o =) 0.
Remark 5.7 For a h e a r Vstem, let V be the controllable Correspondingly, sz(T0)is well known.
subspace. Then it is always V-invariant, feedback block Now by the controllability, there are Xi = Af;x+Bi,ui,,
diagonalizable. 111fact, Brunovsky canonical form is of this k = l,.. . ,s, such that
rye. x : = x ' ( T ) = e , :X' ...e ~ ( D ) .
In the following we assume system (12) is V-invariant, (17)
feedback block dia.gonalizable. Then we have only to con- Let (SI,... C (1,2;.. . s ) besuchthdtthecorrespond-
sider the controllability of (15). It is obvious that the Lie ing
algebras do of (12) and (15) are the same (precisely, they A,, E Aol t = 1 , . .. 1.
are isomorphic by do U P;'dOP22).
The following lemma is necessary for proving the main
result.
where Xz,= A:2x2.
Lemma 5.8 For system ( I ) let U be a potli-wise connected Then in step 2, we can formally move .'(To) to zf (in
open set, and dim(LC,(x))= n, x E U. Then for any z E U weak sense). This can he done because of Lemma 4.8. Note
the largest integral sirb-manifold of die Lie algebra is that in step 2 the first part of states x 1 remains in the origin
Z ( L ) ( X )3 U. by setting all controls 'U, = 0. Now in step 3, we use the
controls in (17) to move x1 to the destination x i . At the
same time, according to (IS), x 2 moves from xf to x:.
Proot If it were not true, then there would exist a point (ii) Same argument for weak controllability can be used
p E U, which is also on the boundary of Z, i.e., p E z\Z. for controllability. The only difference is that since Ao is
Denote by Z ( p ) the largest integral sub-manifold of La symmetric, by Chow's theorem, x i is physically reachable
passing through p . Then Z ( x )U Z ( p ) is a connected integral
from x2(To). 0
manifold of L= passing through x , and it is larger than Z ( x ) .
VI. AN ILLUSTRATING EXAMPLE
which is a contradiction. U
The following is the main result on controllability.
Example 6.1 (i) Consider the following system with n = 4,
Theorem 5.9 For system (15) assume dim(V)= n - k. ( i ) m = 1 and N = 2. Two models ( A t , B , ) ,i = 1 . 2 are as
V e is weakly controllable, iff follows

dim(A0x) = k. x E Vc. (16)

(ii) Assume A0 is symmetric. Then V c is conrrollable, if
(16) holds. 0 0 0 1

5763
We skip some routine computation, and show the results VIII. ACKNOWLEDGMENTS
directly. The controllable subspace is The authors gratefully acknowledge the anonymous re-
viewers' comments on paper presentation and on the exam-
ples.

Let z = (21 z2 z3 x4) E IR4. Then IX. REFERENCES

[ I ] EzzineJ., A.H. Haddad, Controllability and observahil-
ity of hybrid systems, Int. J. Cona, Vol. 49, No. 6, pp.
2045-2055, 1989.
where T,(Af) stands for the tangent space of a manifold A.I [2] Hermann, R.. Differenrial Geoinetq and the Calculus
at z. Define a subspace of Variations, Academic Press, 1968.
[3] Isidori,A., Non-linear Control Systems, 3nd ed.,
v = {x E IF? I x3 = X? = O}. Springer-Verlag. 1995.
Then we know that the weakly reachable set of z is [41 Liberzon, D., and A S . Morse, Basic problems on
stability and design of switched systems, IEEE Control

I
W R ( 2 )= 1(13,,)(0) = v,
IR4\v. x g v ,
zE v
where 1 ( D )is used for the integral manifold of a distribution
Systems Magazine, Vo1.19, No. 5 , pp. 59-70, 1999.
[5] Mohler, R.R., Nodinera Systems, Volume 11, Appli-
cations to Bilinear Control, Prentice Hall, New Jersey,
1991.
D. [6] Olver,P.J., Applications of Lie Groups to Differenrial
(ii) Now assume we add two following switching models Equations, 2nd ed., Springer-Verlag, 1993.

i'': : :).[;): (" H ;;l).(i)

( A , :Bi),i = 3 , 4 to (i) [7] Sun, Z., D. Zheng, On Reachability and stabilization of
switched linear systems, I€€€ Trans. Aur. Contr, Vol.
46, No. 2, pp. 291-295, 2001.
0 0 - 1 0 . [SI Sun, Z., S.S. Ge, T.H. Lee, Controllability and reacha-
0 0 0 - 1 0 0 1 bility criteria for switched linear systems, Astomatica,
Vol. 38, pp. 775-786, 2002.
Then One that A. is the weak 191 Susmann, H.J., Prbits of families of vectorfields and
controllability becomes controllability. Precisely, integrability of distributions, Trans. American Math.
Soc. 180, 171-188, 1973.
R ( x ) = Z(CC,,)(O) = v, 2: E v [IO] Susmann, H.J., A general theorem on local controlla-
IR4\V, xev. bility, SIAM J. Conrr O p t , Vol. 25, 158-194, 1987.
Note that we need pre-feedback controls to block diagonalize [ I I ] Szigeti, E , A differential-algebraic condition for con-
trollability and observability of time varying linear
A3 and A4 first. 0
systems,
VII. CONCLUSION [I21 Xie, G., D. Zheng, and L. Wang, Controllability of
In this paper the controllability of switched linear system is switched Linear systems, IEEE Trans. Auroin. Contr,
considered. First, we showed that the controllability defined 140 I - 1405, 2002.
in [SI and [I21 is the largest controllable subspace, which is
equivalent to the point controllability of the origin. In general,
when the controllability subspace is a proper subspace, the
(weak) controllable sub-manifolds are investigated. The main
contribution of the paper consists in that we have
1. generalized the Chow's theorem for analytic case,
which was used thoroughly for investigating reachable set
of switched linear systems;
2. defined the (strong) accessibility Lie algebra and proved
the relationship between (strong) accessibility of the system
with the rank of the (strong) accessibility Lie algebra;
3. provided a clear topological structure for all possible
controllable sub-manifolds;
4. given necessary and sufficient conditions for the con-
trollability, weak controllability and their local counterparts
of a large class of switched linear systems.

5764