Sei sulla pagina 1di 4

Systems & Control

North-Holland

Letters

4 (1984)

199-202

June 1984

Eigenvalue placement for generalized linear


systems
Vinicius

A. ARMENTANO

Department
of Electrical
Engineering,
6166, Campinas 13100 - SP, Brazil

*
FEC,

UNICAMP,

C. P.

Received 5 November
1983
Revised 5 March 1984
This paper deals with some aspects of eigenvalue placement
by state feedback for generalized
linear systems described by
Ei = Ax + Bu, where E is a singular map. It is shown that
controllability
of the infinite eigenvalues of the pencil (SE - A)
is equivalent
to the existence of a state feedback map which
assigns those eigenvalues
to pre-specified
complex numbers. A
procedure
for the assignment
of all eigenvalues
to distinct
complex numbers is also discussed.
Keywords:
Generalized
linear systems, Controllability,
value placement,
Infinite eigenvalues,
State feedback.

Eigen-

1. Introduction
In this paper we consider the generalized linear
system
2:

Ei=Ax+Bu,
XEX:=R,
rank B=m,

UE%:=R,

(1)

where E is a singular map and the pencil (SE - A)


is regular [7].
We term the set of eigenvalues of (SE - A) as
the set of complex numbers for which the pencil
(SE-A)
loses rank. It is well known [l] that the
regular pencil (SE - A) has rank E eigenvalues,
counting the finite and infinite ones.
The question of eigenvalue placement for Z can
be posed in the following way: does there exist a
state feedback control law, u = Fx, such that the
closed loop pencil (SE - A - BF) is regular and
has pre-specified (finite) eigenvalues? If so, give
* This work was carried out while the author was with the
Department
of Electrical
Engineering,
Imperial
College,
University
of London,
England.
It was supported
by the
Conselho
National
de Desenvolvimento
Cientifico
e
Tecnolbgico,
CNPq-Brazil.

conditions for existence and a procedure to compute F.


Cobb [2] has been the first author to deal with
the above problem. He has shown that the finite
eigenvalues of (SE - A) must be controllable in
order to be freely assignable by state feedback. He
has also given a condition to bring the infinite
eigenvalues of (SE - A) to finite positions of the
complex plane, by means of a state feedback map,
without pre-specifying
those positions. It has been
pointed out by Verghese [l] and also shown in [3]
that this condition corresponds to the requirement
of controllability
of the infinite eigenvalues.
A stronger result is presented here, namely, it is
shown that controllability
of the infinite eigenvalues is equivalent to the existence of a state feedback map which assigns those eigenvalues to prespecified complex numbers.
After this paper was first submitted, the author
became aware of a work [6] which proves a
similar result by a different method.
We also present an extension of the procedure
given by Moore [5] regarding the computation of a
map F which simultaneously assigns all the eigenvalues of the pencil (SE - A - BF) to distinct
complex numbers and the corresponding
closed
loop eigenvectors. We shall follow throughout the
notation introduced by Wonham [4].
2. Assignment

of the infinite eigenvalues

It has been shown by Gantmacher [7] that there


exist nonsingular maps M and N such that
M(sE-A)N=[~;L
dimL=r,

J-I

],

dimJ=n-r,

where the eigenvalues of L coincide with the finite


eigenvalues of (SE - A) and the map J is nilpotent
(i.e. has zero eigenvalues).
This work
here.

was pointed

out by a reviewer

whom

we thank

199

Volume

4. Number

SYSTEMS

& CONTROL

Let p be the number of Jordan blocks of J of


dimension greater than one. Let n, + 1, i EP, be
the dimension of a Jordan block i, ni > 1. Then the
regular pencil (sE - A) has p infinite eigenvalues
of respective orders ni, i EP [l].
Let i := N-lx and let i be partitioned as 2 =
XI) where x, E R and x, E R-. Accord(4
ingly, let

LETTERS

June 1984

Let y = rank J = rank J, and note that y =


,spnrr
i.e. y is the total number of infinite eigenc
values. The next result has been proved in [2] and
is concerned with the property of converting infinite eigenvalues into finite ones by state feedback.
Theorem 1. There exists F : W+ % such that
degree det( sJ - Z - B,,,F) = y
if and only if 2 is controllable at infinity.

Hence, by using (2), we obtain


decomposition for 2:

the following

k, = Lx, + Bu,

(3)

Jx, = x, + B,u.

(4)

Consider J as map from w to YY, i.e. J: W+


w. Then since J is nilpotent it follows that there
exists a decomposition
w-= 7-r-, @ wz
such that
Mat B, =

4
[1
B

(5)

where Jz has all elementary divisors of J of order


greater than one.
From (3) it is obvious that controllability of the
finite eigenvalues of (sE - A) is equivalent to the
existence of a map F such that the eigenvalues of
L + B,F take on pre-assigned symmetric complex
numbers.
The modal criterion for controllability
of the
infinite eigenvalues of (sE - A) says such eigenvalues are controllable if the pencil [sE - A B]
has no infinite eigenvalues [l].
Proposition 1 below gives equivalent necessary
and sufficient conditions for controllability
of the
infinite eigenvalues [3]. In the following we shall
denote .%?:= Im B.
Proposition 1. The system 2 is controllable at infinity if either of the following conditions hold:
(i)

ImJ,+.G?,=7Kz,

(ii)

(Jt].G9z):=972+J2.S?z+

Im J + ker J + G?,= W,

(iv)

ImE+A

200

kerE+B=Z.

Theorem 2. There exists a map F : W+ @such that


the set of roots of
det(sJ-I-B,F)=O
is A if and only if the system 2 is controllable at
infinity.
Proof. (I) From Proposition 1 (ii) we have that
( J2]9Yz) = 71y; and thus the basis constructed by
Wonham [4] can be used so that J2 and B, are
represented as

0 1

(6)
Mat B, =

where 4, is cyclic with minimal polynomial equal


to (Y,, the i-th invariant polynomial of J. Since J is
nilpotent, it follows that (Y~= s,+l with n, 2 n2 >,
...

>n

It is %o shown in [4] that the pair (Ji, bii) is


controllable. Thus we may assumethat Jjj and b,,
are represented as
0 1
00

a-. +J~-.G@12=Wz,

where q is the index of nilpotency of Jz (and of J),


(iii)

Let Ai be a set of n, symmetric complex numbers and let A := Up,, A,. We then obtain:

0
1

. ..
*.:
.

.& =

0
.
0
1

...

...

. .:

(74

Volume

b,;=

4, Number

SYSTEMS

II
0
0
:

& CONTROL

G'b)

(,7,+1)x1

The submatrix B, plays no role in the present


analysis.
Define F: W-+ ?Lby FIW, = 0 and F/-W; = F,
where F, is given by

(8)
-------0

iEP.
From the definition of F and (5) it follows that
Mat(l+B,,F-sJ)=

,,B:kT-,

From (6), (7) and (8) it follows that (I + B,*F, sJz) is upper triangular with matrices P,(s), i Ep,
in its diagonal given by

LEITERS

June

1984

Remark. The above theorem has shown that each


infinite eigenvalue of order n, can be converted
into n, pre-specified symmetric finite eigenvalues
by meansof a real state feedback map. A different
basis has been used in [6] to prove a similar result.

3. A procedure for eigenvalue and eigenvector


assignment
In this section we discuss the simultaneous assignment of all eigenvalues and their corresponding closed loop eigenvectors by state feedback.
The content of the next proposition is essentially
an extension of a result by Moore [5].
Proposition 2. Let the systemZ be controllable at its
infinite eigenvalues.Let { X, }, i E h, h = rank E, be
a symmetric set of (finite) h distinct complex numbers. Supposethere exists a subspaceY:= span{ vi },
i E h, such that
(i) vi E Z if Xi is real and v, = v: (complex
conjugate) if A, = AT,
(ii) the vectors { v, }, i E h, are linearly independent and v, E %,,, := (X, E - A)-%,
(iii) Ivn ker E = 0,
Then there exists a real map F such that
(A+

BF)v,=hiEv,,

iGh,

and the pencil (SE - A - BF) is regular.


<@=[a.,

,:

0:

m-1 :,+1....(,+,,

It can be readily seen that the expansion by the


last row of P;(s) yields
~,(s):=det
=f

I.1

+ . . . +fr.n,+,r

whence
det(l+

P,=[XE-A

B],

and let

P,(s)
sl +f,.*s+

Proof. As in [5] associate with each complex number X the matrix

B,F-sJ)=,fiIpj(s).

It is clear that zeros of p,(s) can be assignedto


A, by a suitable choice of the vector i.T.
(*) Assume that I: is not controllable at infinity. This implies by Theorem 1 that there is no F
such that
degreedet(sJ-I-B,,.F)=y,
which contradicts the hypothesis.

Q,=

[ h1
;

be a compatible partitioned matrix whose columns


span ker Px. It can be shown that rank B = m
implies that the columns of Nh are linearly independent.
Since v, E X,, = Im N,, it follows that v, = N,,k,
for some kj which is unique and
(X,E - A) N,,k, + B R1,k; = 0.
Define F, : Y+ 6 by
FOv,= -Rx,k,,

iEn.
201

Volume

4, Number

SYSTEMS

& CONTROL

It remains to define a suitable extension F to F,


such that the pencil (sE - A - BF) is regular.
Since dim V= h, dim ker E = n - h and Vii
ker E = 0, it follows that
V&kerE=X.

Comments. Proposition 2 admits a certain converse: Let {hi}, i E k, be the set of finite eigenvalues of the (not necessarily regular) pencil (SE - A
- BF). Then

(9)

E(VG

ker E) = EX=

(A + BF)u,

Im E.

Let x E X be represented in (9) and consider


any extension P: X+ @ to F,. Let Ex and (A +
BF)x be represented in the decomposition
EV-&=X,

(10)

where .@is any subspace of dimension n - h which


complements Im E to X.
In the decompositions (9) and (10) the maps E
and A + BP admit the following representations:

[ 1
I
o

MatE=

0
o,

MatA+Bk=

4,
At,
[ 1
o

22

where dim I = h and the eigenvalues of A,, coincide with {hi}, i E h.


It is clear that the pencil (SE-A
- BP) is
regular if and only if A,, is nonsingular.
Let Q; be the projection on Xalong Im E. Then
A,, nonsingular is equivalent to the nonsingularity
of
Q,(A

+ Bfi)I

ker E.

(11)

Write

with F, := FI ker E and note that there exists a


map F, : ker E + % such that (11) is nonsingular if
and only if the zero eigenvalues of Q, A( ker E are
controllable by Q, B, i.e.
(12)

Since ZZ is controllable at infinity, it follows


from Proposition 1 (iv) that (12) holds. Therefore
an appropriate F, can be chosen so that (11) is
nonsingular; define F: X+ 42 by
Fly=

F,

and

The definition

202

FI ker E = F,.
of a real map F in case complex

=X;Eu,,

for some vectors u,, i E 1. It can be shown that the


vectors u,, i E 1, satisfy the conditions (i)-(iii) of
Proposition 2.
It can also be shown that the subspace F in the
statement of Proposition 2 exists for any symmetric set of distinct complex numbers { Xi }, i E h,
provided that the system 2 is controllable at its
finite and infinite eigenvalues. In this case we can
conclude from the results in [2] that there exists F
such that (sE - A - BF) is regular and has {A,},
i E h, as the set of finite eigenvalues. This implies
that there exists ui E X(X;), i E li, such that
span{ u,} n ker E = 0.
A possible way of defining Y is to choose
linearly independent vectors ui E X( Xi) but u, @
ker E. Then, generically [4] C= span { ui } is such
that V77 ker E = 0.

References
PI

Q,(A]kerE+B)=X.

June 1984

conjugate numbers are assigned can be handled in


the same way as in [5].

Also note that


EY=

LETTERS

G.C. Verghese,
B.C. Levy and T. Kailath,
A generalized
state-space
for singular
systems, IEEE
Trans. Automat.
conrro126
(1981) 811-831.
PI D.J. Cobb. Feedback and pole placement in descriptor
variable systems, Internat.
J. Conrrol6
(1981) 1135-1146.
The pencil (SE - A) and controllabih131 V.A. Armentano,
ty-observability
for generalized
linear systems, Report
83.13, Electrical
Engineering,
Imperial
College, University
of London.
Linear Multivariable
Control: A Geometric
[41 W.M. Wonham,
Approach (Springer,
New York, 1979).
151B.C. Moore, On the flexibility offered by state feedback in
multivariable
systems beyond
closed-loop
eigenvalue
assignment, IEEE rrans. Automat. Control 21(1976)
689-692.
and F.L. Lewis, A result on the placement of
161K. Aqaldiran
infinite
eigenvalues
in descriptor
systems, Electrical
Engineering,
Georgia Institute of Technology,
Atlanta, GA.
The Theoy of Matrices,
Vol. II (Chelsea,
171 F.R. Gantmacher,
New York, 1959).

Potrebbero piacerti anche