On the output-stabilizable subspace
Citation for published version (APA):
Geerts, A. H. W., & Hautus, M. L. J. (1988). On the output-stabilizable subspace. (Memorandum COSOR; Vol. 8827). Technische Universiteit Eindhoven.
Document status and date: Published: 01/01/1988
Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)
Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please follow below link for the End User Agreement:
www.tue.nl/taverne Take down policy
If you believe that this document breaches copyright please contact us at: openaccess@tue.nl
EINDHOVEN UNIVERSITY OF TECHNOLOGY Department of Mathematics and Computing Science
Memorandum COSOR 88-27
On the output-stabilizable subspace
A.H.W. Geerts and M.L.J. Hautus
Eindhoven University of Technology
Department of Mathematics and Computing Science P.O. Box 513
5600 MB Eindhoven The Netherlands
Eindhoven, October 1988 The Netherlands
ON THE OUTPUT·STABILIZABLE SUBSPACE
ABSTRACT
Properties of a certain subspace are linked to well-known problems in system theory.
Keywords:
Output stabilizability, linear-quadratic problem, singular controls, structure algorithm, dissipation inequality.
ON THE OUTPUT-STABILIZABLE SUBSPACE A.H.W. Geerts & M.L.J. Hautus
O.
IntroductionThe objective of this paper is to investigate properties of a space to be denoted !f-, which plays a role in a number of problems in system theory.
In [WO 79, Section 4.4], the Output Stabilization Problem is considered (for the case D = 0):
PROBLEM (OSP): Given the system
L:
x = Ax + Bu, Y = Cx + Dufind a feedback map F such that with the input u
=
Fx, we have y(t) ~ 0 (t ~ 00) for any initial value xo'
If this problem has a solution,
L
is called output stabilizable. A necessary and sufficient condition for output stabilizability was provided by [WO 79, Theorem 4.4]. A slightly different formulation of this condition was given in [HA 80, Theorem 4.10], where it was shown that OSP has a solution iff !f-=
X. Here X denotes the state space Rn, and!f was defined in terms of (~,w)-representations. More generally, the space !f also plays a role in the output-stabilization problem under disturbances, i. e. , the problem of achieving BIBO stability in the presence of a disturbance input term Eq. Then, it turns out, the condition
is im E ~ !f-. The space is also relevant for LQ problems.
Consider the general positive semidefinite infinite horizon LQ problem:
PROBLEM (LQ): Determine a function u: [0,00) ~ RID such that for the solution x of the differential equation
·
x = Ax + Bu, x(O) = x
o
the following quantity is minimized:
!XI
J(x ,u) :=
S
w(x(t),u(t»dt,o 0
where
w(x,u) := x'Qx + 2u'Sx + u'Ru
is a positive semidefinite quadratic form in the variable
(x, u) E IRn+m•
As in [H&S 83], we introduce matrices C and D using the
factorization
[QS R
S'] [C']
=
D' [C D],which is possible because the matrix in the left-hand side is positive semidefinite. Next, we define the output
y :=
ex
+ Du,and the integral to be minimized can be written as
S:ly
12dt .The LQ problem is called regular if D is injective (left
invertible). For regular problems, the problem is well
established and considered classical. Curiously, the problem
of finding necessary and sufficient conditions for the
existence of a solution of this (infinite-horizon) problem
has found little attention even in the regular case. Usually, one is satisfied with the statement that a solution exists if
the system
L
is stabilizable. Of course, this condition isnot necessary. In [GE 88], a necessary and sufficient
condition was given for the regular case (R>O) in terms of
the stabilizability of a suitably defined quotient system.
For singular problems, a reformulation in the style of
[H&S 83] is needed incorporating distributions as possible
inputs, since otherwise, one can not expect that optimal
controls exist even in the finite horizon case. With this
singular problems. It is expressed in terms of the structure algorithm (See [GE 88, Remark 4]).
In the present paper, this issue is also addressed and it
will be shown in particular that output stabilizability (and
hence !f- = X) is necessary and sufficient for the existence
of a solution of the LQ problem. This intuitively rather
obvious condition turns out to be relatively difficult to
prove. In the case that D is left invertible (the regular
case), the condition will be shown to be equivalent to the
existence of a positive semidefinite solution of the
Algebraic Riccati Equation associated with the LQ problem
mentioned above. Finally, the condition !f-
=
X will bebrought into connection with the dissipation inequality, to
be discussed in section 3.
1. The output-stabilizable subspace
Consider the system
r:
x=
Ax + Bu, Y=
Cx + Du,where A e IRnxn, B e IRnxm, C e IRPxn and D e IRPxm•
In connection with the LQ problem, we have to admit
distributions as inputs. These distributions are chosen from
the class ~ of impulsive-smooth distributions on IR with
1mp
support on [O,~) introduced in [H&S 83]. Such a distribution
can be written as a sum of a function which is smooth on
[O,~) and an impulsive distribution with support in {O}.
However, in the LQ problem, only distributions such that the
output is regular are admitted. The class of such inputs is
denoted 'IL.--~ Using the notation y (~) for limt ..~ y (t), we
introduce the following spaces:
r:J
'!J2 := {X 0 E XI 3u e
~: IOJ
Iy (t) 12
dt < OJ }
o
and for every F E R~n, we define the space
F
'!J
I
:=
{xo E XI If u=
Fx then y(OJ)=
O}'!J~
:= {xo E XI If u = Fx then IOJjy (t) 12dt < OJ}.o
Finally, the space ~- is defined as
~-:= <AlimB>
+
X- (A)+
V Lwhere <AI imB> is the reachable subspace of
L ,
X- (A) thestable subspace of A, and V
L denotes the weakly unobservable
space of L, i. e., the space of initial states X
o such that
there exists a regular control (hence without impulses) such
that the output is identically zero.
Our main result is:
THEOREM For every
L,
weF. In addition, there
i = 1,2.
have '!J
i
exists
= ~- and '!JF s;; ~- for every
i
F such that '!JF = ~- for
i
Because of the relation '!JF
=
~- for some F, we will refer toI
~- as the output-stabilizable subspace. Some of the
statements of the Theorem are rather obvious. In particular,
if we use the feedback control u
=
Fx, the resulting outputy(t) will have the property that it tends to zero
exponentially fast when either X
o e '!J~ or Xo E '!J~.
Consequently, '!JF
=
'!JF. Furthermore, it is trivial that1 2
'!JF s;; '!J for all F. Also the inclusion ~-s;; '!J is easily shown.
i i 1
In fact, any element of x E!f- can be decomposed as
x
=
Xl + x2' where XlE <AI im B> + X-(A) , the stabilizable
subspace and x
2 E VL' It is known (See [WO 79, Ch.3]) that
there exists a feedback F such that with the input u = Fx and
initial state x, the state x (t) tends to zero. For the
initial value x, there exists a smooth input such that y is
2
identically zero. A similar reasoning can be used for proving
the inclusion ~-~ ~ . Finally, the fact that there exists an
2
F such that ~F
=
~- is known (compare [HA 80]). The difficult1
part of the Theorem is the proof of the inclusions ~. ~
~-1
(i
=
1,2). This proof is given in [G&H 88].An interesting special case is ~-
=
X.
This condition can bereformulated as: For each X
o e
X
there exists an input u e ~such that y(oo) = O. According to the Theorem, an equivalent
formulation is: There exists a feedback u
=
Fx such that thehas the
stable, i.e.,
solution to
resulting closed loop system is output
y(oo)
=
a
for every xo e X. This is theproblem OSP mentioned in section 1. The equality ~2 =
X
saysthat in the optimal control problem, there exists for any X
o
an input u e ~ such that the performance J (xo' u) < 00. This
will be seen to be a necessary and sufficient condition for
the existence of an optimal control in ~.
In addition to V
L, the strongly reachable subspace WL plays
an important role in the (singular) optimal control problem.
It is defined as the space of points x e X for which there
1
exists an impulsive input u e U
L such that x(O+)
=
Xl' It caneasily be seen that W
L =
a
iff D is injective, whichcorresponds to the regular case for the LQ problem.
Furthermore, it can be seen (see [H&S 83, section 3]) that
strongly reachable states are also strongly (null-)
controllable, i.e., there exists for any such state an
impulsive control in ~ for which the state jumps to zero
instantaneously, so that x(O+) = 0, while y remains zero.
Obviously, for such initial states, the minimal cost of the
LQ problem equals zero.
2. Optimal control
It is clear that the existence of a solution of LQ for a
certain X
(see [GE 88, property is semidefinite minimizing J(xo'u). It follows
for the regular problem, this the existence of a positive Algebraic Riccati Equation
of such a solution for all x implies~·
=
X,
hence thatL
iso 2
output stabilizable. Conversely, it is shown in [GE 88] for the regular case that the existence of u e CUI: such that J (x ,u) < 00 , is also sufficient for the existence of an
o input u e ~ Corr.2] that equivalent to solution of the ARE:~(P) := C/C + A'P + PA - (PB + C/D) (D/D)-l(B/P + D/C) = O.
The smallest positive semidefinite solution of to be denoted by P-, yields the minimal integral:
this equation, value of the
min {J (xo' u)
I
u e ~}For singular problems, the ARE is not defined, because D'D is
not invertible. However, as is shown in [H&S 83] for the left-invertible case and in [GE 88+] for the general case, a sequence of transformations can be constructed that transform
L
into a system of the form.
x=Ax+Bu +Bu
o 0 1 1
Y
=
Cx
+5
u ,
o 0
where the following properties are satisfied:
i) D is left invertible and contains the independent o
columns of D.
ii) rank
5 =
rank T (s), where T (s) ois the transfer function of
L.
: =
D + C(sI - A) -lB is the derivative of is an invertible by the structure iii) im B1 ~ WI:. iv) C(WI:) ~ im50.
v) u=
1/(p)ii, where ii :=
[~:],
po
and hence acting as distributions. Finally, ~polynomial matrix, computed
algorithm.
The transformed system
f
can be reduced further by means of a feedback transformation, i.e., a transformation of the form u = Fx + u, where F :=-0 (0'0)
-lO/C. This yields the systemo 0
(A ,:S,c
,0),
whereo 0
-:=C+DF.oThe result is the following system:
x=Ax+:Suo +:su,
0 0 1 1
y = C x + o u .
0 0 0
The following properties can be proved:
Ao(WI:) S;; W
t ,
VI: + WI: = <ker Co lAo> = V~,
O/C
= O.o 0
It follows from these properties together with the properties iii and iv mentioned before that the output y is independent of u. Hence u represents the nonuniqueness of the optimal
1 1
control. Selecting in particular u
1 = 0 leaves us with a regular control problem, since D is left invertible.
o
Consequently, the optimal control can be expressed in terms of the Riccati equation corresponding to system
r
:= (A,:So'C,Oo)· Let us denote this equation by i(p) =O.
It is shown in [GE 88+] that the smallest positive semidefinite solution of this equation determines the minimal cost of the problem. It follows from these results that the existence for each x of an input u such that J (x , u) < 00, i. e., theo 0
condition ~2 =
X,
implies the existence of an optimal control. Consequently, output stabilizability ofL
is equivalent to the existence of a positive semidefinite solution of the equation ~ (P) = O. Finally we note that ker P = VI: + WI:' in particular, WI: S;; ker P where P~(P) = O. This property is an easy consequence of the equality V~ + W~ = Vf .
3. The dissipation inequality
Another way of characterizing the optimal controls can be expressed in terms of the dissipation inequality ([WI 71]).
To this extent we define the dissipation matrix
_ [C' C+A' P+PA
F (P) :
=
B'P+D'C
PB+C'
OJ .
0'0The dissipation inequality or the linear matrix inequality
then reads:
F(P) ?; 0
For every solution P of the dissipation inequality the following holds:
rank F(P) ?; rank T(s),
where the right-hand side denotes the global rank of the rational matrix T (s) (see [SCH 83]). Solutions of the dissipation inequality for which equality holds in the above inequality are called rank-minimizing solutions of the dissipation inequality.
We consider the set:
r := {PI P=P', F(P) ?;
a}.
Then r ¢
t
because 0 e r. We are more particularly interestedin the rank-minimizing solutions of the dissipation inequality. Therefore we also introduce
rmin:= {P e r
I
rank F (P) = rank T (s) } .The following results hold: (See [GE 88+]) i) F(P) ?; 0 implies W~ ~ ker P.
ii) P e r . iff P e r , ~ (P) = 0 and W~ s;; ker P.
m~n w
iii) P e r . (Recall that W~ ~ ker P ).
m~n w
The following theorem relates the dissipation inequality with the rest of this paper:
THEOREM
L
is output stabilizable i f and only i f there existsP e r such that P ?;
o.
In addition, i f such a P exists,min
the smallest such P equals the matrix P-, representing the minimal cost of the optimal-control problem.
The dissipation inequality enables us to formulate the
conditions for the existence and the actual computation of
the cost of the optimal-control problem directly in terms of
the original data of the problem. The characterization of P
in terms of the dissipation inequality is more general than
the characterization using the ARE, because it includes the
singular case. Note that the ARE is not defined in the
singular case, because the matrix D'D is not invertible.
Hence one has to use the ARE of the transformed system in
order to express P-.
REMARK It is shown in
y (t) ~ 0 for t ~ 00
equivalentlyI (x/ker
does not hold. It
[G&H 88] that if u is such that either
or S;lyl2dt < 00 then P-x(t) ~ 0, or
P-) (t) ~
O.
Remarkably, the converseis possible that P-x(t) ~ 0 (t ~ 00),
whereas y(t) does not converge to zero. _
Finally, we note that a condition for output stability can
gi ven in the spirit of the results of [HA 70]. In fact, a
more general formulation is:
PROPOSITION Let ~ be an (A,B)-invariant subspace of X. Then
<Alim B> + X- (A) + ~ = X i f and only i f
VA
e
C, Re A ?;OV~I
e
cn[~[A
-
AI,B]=
0 A~~
=
0~ ~
=
0].The proof is straightforward. The condition for output
stabilizability is obtained by taking ~
=
VL• We obtain an
equivalent formulation by choosing a basis matrix V for V
L•
REFERENCES
[GE 88] Ton Geerts, A necessary and sufficient condition for
solvability of the linear-quadratic control problem without stability, Syst. & Contr. Lett., vol. 11, pp. 47 - 51, 1988
[GE 88+] A.H.W.Geerts, All optimal controls for the singular
linear-quadratic problem without stability; a new interpretation of the optimal cost, Memorandum CeSOR
87 - 14, Eindhoven University of Technology, 1987,
to appear in Linear ~g. &Appl.
[G&H 88] A.H.W.Geerts & M.L.J.Hautus, The output-stabilizable
subspace and linear optimal control, in preparation.
[HA 70] M.L.J.Hautus, Stabilization, controllability and
observability of linear autonomous systems, Nederl.
Akad. Wetensch. Proc. Ser. A, 73, pp. 448 - 455,
1970.
[HA 80] M.L.J.Hautus, (A,B)-invariant and stabilizability
subspaces, a frequency domain description,
Automatica, vol. 16, pp. 703 - 707, 1980.
[H&S 83] M.L.J.Hautus & L.M.Silverman, System structure and
singular control, Lin. ~g. & Appl., vol. 50, pp.
369 - 402, 1983.
[SCH 83] J .M. Schumacher, The role of the dissipation matrix
in singular optimal control, vol. 2, pp. 262 - 266, 1983
[WI 71] J.C.Willems, Least squares stationary optimal
control and the algebraic Riccati equation, IEEE
Trans. Automat. Contr., vol. AC - 16, pp. 621 - 634,
1971.
[WO 79] W.M.Wonham, Linear Multivariable Control: A
EINDHOVEN UNIVERSITY OF TECHNOLOGY Department of Mathematics and Computing Science
PROBABILITY THEORY, STATISTICS, OPERATIONS RESEARCH AND SYSTEMS THEORY
P.O. Box 513
5600 MB Eindhoven - The Netherlands Secretariate: Dommelbuilding 0.02 Telephone: 040 - 473130
List of COSOR-memoranda - 1988
Number Month Author Title
M 88-01 January F.W. Steutel, Haight's distribution and busy periods. B.G. Hansen
M 88-02 January J. ten Vregelaar On estimating the parameters of a dynamics model from noisy input and output measurement
M 88-03 January B.G. Hansen, The generalized logarithmic series distribution. E. Willekens
M 88-04 January J. van Geldrop, A general equilibrium model of international trade with C. Withagen exhaustible natural resource commodities.
M 88-05 February A.H.W. Geerts A note on "Families oflinear-quadratic problems": continuity properties.
M 88-06 February Siquan. Zhu A continuity property of a parametric projection and an iterative process for solving linear variational inequalities.
M 88-07 February J. Beirlant, Rapid variation with remainder and rates of convergence. E.K.E. Willekens
M 88-08 April Jan v. Doremalen, A recursive aggregation-disaggregation method to approxi-1.Wessels mate large-scale closed queuing networks with multiple job
Number Month Author
-
2-Title
M 88-09 April J. Hoogendoom, The Vax/VMS Analysis and measurement packet (VAMP):
R.c.
Marcelis, a case study.A.P. de Grient Dreux, I.v.d. Wal,
R.I.Wijbrands
M 88-10 April E.Omey, Abelian and Tauberian theorems for the Laplace transform E. Willekens of functions in several variables.
M 88-11 April E. Willekens, Quantifying closeness of distributions of sums and maxima SJ. Resnick when tails are fat.
M 88-12 May E.E.M. v. Berkum Exact paired comparison designs for quadratic models.
M 88-13 May J. ten Vregelaar Parameter estimation from noisy observations of inputs and outputs.
M 88-14 May L.Frijters, Lot-sizing and flow production in an MRP-environment. T. de Kok,
J. Wessels
M 88-15 June J.M. Soethoudt, The regular indefinite linear quadratic problem with linear H.L. Trentelman endpoint constraints.
M 88-16 July lC. Engwerda Stabilizability and detectability of discrete-time time-varying systems.
M 88-17 August A.H.W. Geerts Continuity properties of one-parameter families of linear-quadratic problems without stability.
M 88-18 September W.E.J.M. Bens Design and implementation of a push-pull algorithm for manpower planning.
M 88-19 September A.I.M.Driessens Ontwikkeling van een informatie systeem voor het werken met Markov-modellen.
3
-Number Month Author Title
M 88-21 October A. Dekkers Global optimization and simulated annealing. E. Aarts
M 88-22 October J. Hoogendoom Towards a DSS for performance evaluation of VAX/VMS-c1usters.
M 88-23 October R.de Veth PET, a performance evaluation tool for flexible modeling and analysis of computer systems.
M 88-24 October J. Thiemann Stopping a peat-moor fire.
M 88-25 October H.L. Trentelman Convergence properties of indefinite linear quadratic J.M. Soethoudt problems with receding horizon.
M 88-26 October J. van Geldrop Existence of general equilibria in economies with natural Shou Jilin enhaustible resources and an infinite horizon.
C.Withagen
M 88-27 October A. Geerts On the output-stabilizable subspace. M. Hautus