A polynomial characterization of (A,B)-invariant and
reachability subspaces
Citation for published version (APA):
Emre, E., & Hautus, M. L. J. (1980). A polynomial characterization of (A,B)-invariant and reachability subspaces. SIAM Journal on Control and Optimization, 18(4), 420-436. https://doi.org/10.1137/0318031
DOI:
10.1137/0318031
Document status and date: Published: 01/01/1980
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
providing details and we will investigate your claim.
SIAMJ. CONTROLANDOPTIMIZATION
Vol.18, No.4, July 1980
980 Society for Industrial andAppliedMathematics 0363-0129/80/1804-0007$01.00/0
A POLYNOMIAL
CHARACTERIZATION OF (/, M)-INVARIANTAND
REACHABILITYSUBSPACES*
E. EMRE-? AND M. L. J. HAUTUS{
Abstract.BasedonthestatespacemodelofP.Fuhrmann,a link is laid between thegeometric approach
tolinear system theory, as developed byW. M.WonhamandA. S. Morse, and the approach basedon polynomial matrices. In particular polynomial characterizations of (A,B)-invariant and reachability subspacesaregiven. Thesecharacterizationsare usedtoprovetheequivalence of thedisturbancedecoupling problemand the exactmodel matching problem and alsoto connect thepolynomialmatrixand the geometric approachtotheconstruction of observers.
Finally,constructiveprocedures andconditions aregiven for computing the supremal(A,B)-invariant subspaceandreachability spaceand forcheckingthesolvability ofthe exact modelmatching problem.
1. Introduction.The geometricapproachtolinear systemtheory hasprovedvery successfulinsolvingavarietyofproblems
(see
17]
foradetailed account of thistheory).The principal conceptsin thistheory,which are instrumental inthe description of many results,are(A, B)-invariantsubspaces and reachability(controllability) subspaces.
An
alternativeapproachtolinearsystemdesignhas beendevelopedin
[13],
[14],
[16].
Thistheory dependsto alargeextent onpolynomial matrix techniques. Itis evidentthat a
method for translating results ofonetheorytoanotherisvery desirable,becausesucha
method would yielda better understanding of the relations between the twodifferent
approaches.Thiswould be veryuseful,inparticular since the geometric methodmaybe
viewed as exponent of the so-called "modern control theory" and the polynomial matrix method may be considered ageneralization of the classicalfrequencydomain
methods.
A
number of papers with the objective of translating the results of geometriccontroltheoryintopolynomialmatrix terms haveappeared (e.g.,
[1], [3],
[10], [12]).
It is the purpose of thispaper to show that a very useful link between the twoapproachescanbebasedonthe work ofP.Fuhrmann
([7], [8], [9]).
Specifically,it willbe shown that usingthe statespace modelassociatedwithasystemmatrix, introduced by Fuhrmann, one can give characterizations of the concepts of (A, B)-invariant subspaces and teachabilitysubspaces intermsof polynomialmatrices.Thiswillbe the subject of 3 and 6. One application of the polynomial characterization of (A,
B)-invariant subspaces will be given in 4,where it will be shown that the disturbance decouplingproblem
(see
[
17,Chap.4])
andthe exactmodel matchingproblem(see
[ 16],
[
14], [5])
areequivalentproblems.Another applicationisgivenin 5,where it is shownthat the equivalence of the polynomial matrix and the geometric formulation of
observers can bederived fromthe results of 3.In 7,theconcept ofrowproperness
defined in
[14],
[16]
is used to formulate anecessary and sufficientcondition for the existence of a solution of the exact model matching problem and, hence, of thedisturbance decouplingproblemin termsof degrees of polynomial matrices. Also in 7
a constructive characterization of the supremal (A,B)-invariant subspace and reachability space contained in kerC is given. Finally, in 8, the results of 3 are
extendedtothe situationwhere the system is describedbyRosenbrock’s systemmatrix.
The preliminary 2contains ashort description of Fuhrmann’sstatespacemodel.
* Receivedbytheeditors October18, 1978,and in revised formOctober5, 1979.
5" Centerfor MathematicalSystemTheory, UniversityofFlorida, Gainesville,Florida32611. The work of this authorwassupportedinpart bytheUnitedStatesArmyResearchOfficeunder Grant DAA29-77-G-0225,andbythe UnitedStatesAirForce underGrant AFOSR76-3034Mod.B throughtheCenterfor MathematicalSystemsTheory, Universityof Florida, Gainesville, Florida.
Departmentof Mathematics,EindhovenUniversityofTechnology, Eindhoven,theNetherlands. 420
2. The state spacemodelassociatedwith a matrix fractionrepresentation. Let
K
be a field. Wedenote by
K[s]
the set ofpolynomials, andbyK(s)
thesetof rational functionsoverK.
If 6isanysetand p,qN,
wedenote by6epthe setof p-vectorswithcomponentsin6e and by5epqthe setof pxq matriceswith entriesin5
.
IfA
isapxq matrix, we denote by{A}
the K-linear space generated by the columns ofA.
IfU(s)eKqr[s]
and:Kq[s]K"[s]
is a linearmap, thenU(s)
denotes the resultobtained by applying toeachof the columns of
U(s).
Let
x(s)6K’(s). We
denote by(x(s))_,
the strictly proper part ofx(s)
and by(x(s))_l,the coefficient ofs-1in theexpansionof
x(s)
inpowers ofs-1.
DEFINITION 2.1. Let
T(s)
Keq[s].
ThenXr
denotes the setof
x(s)
K[s]
for
whichthereexists astrictly proper
u(s)
Ka(s)
suchthatT(s)u(s)
x(s).
In
whatfollows,XT-
playsafundamentalrole(compare
theclosely related conceptofrightrational annihilator
[4]).
In
particular,if p=q andT(s)
isnonsingular, thenXT-
{x(s)
K"[s]
T-X(s)x(s)
is strictlyproper}.
In
this particular situationwedefine the mapzrT-:KP[s]X7
x(s)
--
T(s)(T-l(s)x(s))_.
(Compare
[7]
and[8]
wherefurtherpropertiesof this map aregiven.)In
the followingweconsider
XT-
aK-linearspace.Weconsideralinear system whose transfermatrixis givenbythe leftmatrix fraction representation(2.2)
G(s)
T-(s)U(s).
Weassume that
G(s)
isstrictlyproper,T(s)
K[s],
U(s)
Kr[s].
Definethelinearmaps
:X-
---,X-
x(s)
(sx(s)),
(2.3)
YdK
--
XT-
uU(s)u,
:XT-
K
q:x(s)--(T-(s)x(s))-.
By
definition, forx(s)XT,
we have4x(s)= sx(s)-T(s)c(s)
for somec(s)6Kq[s].
Since
T-l(s)x(s)
andT-l4,x(s)
arestrictlyproperitfollowsthatc(s)
mustbe constant.Hence
(2.4)
x(s)
sx(s)- T(s)c
for some c
K
q,
dependingonx(s).
Thefollowingresult isprovedin
[7].
THEOrEM 2.5. The system
,
:= ((,.,
)
with state spaceXT-
is an observablerealization
of
G(s).
The realization isreachableiff
T(s)
andU(s)
areleft
coprime.We
will call this realizationZ
the T-realization ofG(s).
Conversely, ifwearegivenanobservable system
Z
(C,A,
B),thenwe construct aleftmatrixfractionrepresentation of the transfer matrix of
Z
in thefollowing way.Let(2.6)
C(sI-a)
-=
T-(s)S(s),
whereS and T areleft coprime. Define
(2.7)
U(s):=S(s)B.
422 E. EMRE AND M. L. J. HAUTUS
THEOREM2.8. The T-realization
of
G(s)
T-I(s)U(s),
whereTandUare
defined
by(2.6)
and(2.7)
isisomorphictothesystem,.
Proof.
Using the dual of[11,
Cor.4.11],
we see that $(s) is a basis matrix ofXT.
Hence
thelinearmap:K
XT
:x--S(s)x
is anisomorphism. Usingtheequation
T(s)C=S(s)(sI-A),
whichfollows from
(2.6),
onederives easily therelationsg6e 6eA, 6eB, 6e C.In
particular, g6ex=g(S(s)x)=zrT.(S(s)(sI-A)x)+zrT-(S(s)Ax)= zrT-(T(s)Cx)+
zrT-(S(s)Ax)
$(s)AxYAx.
It
follows that(C,A, B)
and (c,g,
)
areisomorphic. [3Using Theorem2.8,wemaytransform results obtained for the particular realiza-tion
(’,
,
)
toanyobservable system.3. (,)-invariant
sulslaees.
Wegiveacharacterization ofthe(’,
)-invariantsubspaces of the T-realization ofatransfer matrix
G(s)
T-l(s)U(s),
asdefinedin the previous section.For
the definition of(4, )-invariantsubspaceswereferto17].
THEOREM 3.1. Let
(s)
be a qm
polynomical matrix. Then{(s)}
is an(, )-invariantsubspace
ofXT-iff
thereexistCa
gqm,
F1 Kr’and
A
K""such
that
(3.2)
T(s)C1
+
U(s)F1
(s)(sI-A1).
Proof.
Suppose
that{(s)}
is an(’, )-invariant subspace,i.e.,(3.3)
{(s)}
c_{(s)}
+
Im
.
Applying
(2.4)
toeachcolumn of(s),
wefine that4(s)=
l(S), where(3.4)
a(s)
:=s(s)-T(s)C1
for some
C1
K
q". Onthe otherhand,(3.3)
implies(3.5)
l(S)
(s)A1
+
U(s)F1
for some
A
K
""
andF1
K
r".
Combining(3.4)
and(3.5)
yields(3.2).
Conversely,ifweassume
(3.2),
then(3.6)
T-l(s)(s)
(C1
+
T-I(s)U(s)F1)(sI-A1)
-1isstrictlyproperand hence
{(s)}
Xr.
Furthermore,if wedefinel(s)
by(3.4),then(3.5)
followsfrom(3.2)
and, hence,{l(S)}
XT-.
Itfollowsthat’(xI)’(S))
"rt’T(SXIt(s))
7rT(XI(S)
+
T(s)C1)
xItl(s).
Thus,
(3.5)
implies(3.3).
El
If the matrix
(s)
occurring in Theorem 3.1 has full columnrank, itispossibletogive an interpretationtothematrices
A
1,F1,C1.
Forin that case there existsa K-linearmap
:XT
K
satisfying(3.7)
(s)
F.
Then
(3.2)
impliesIt
followsthat{(s)}
is an(M-
-)-invariantsubspace,andthatA
isthe matrixoftherestriction of s4- to
{(s)}
withrespecttothe basismatrix(s). In
addition,F1
isthe matrix(withrespecttothebasismatrix
(s)
of{(s)}
andthe naturalbasisinK r)
of.
Finally,wehave(3.9)
(s)--
C1
so that
C1
isthematrixof the restriction of to{(s)}
withrespecttothe basismatrix(s)
of{(s)}
andthenaturalbasisofKq.
The last result givesacharacterizationof(d, )-invariantsubspacescontained in
ker.
COROLLARY 3.10. Let
(s)Kq"[s].
Then{(s)}
is an (M,)-invariantsubspacecontained in ker
iff
thereexist matricesF, A
such that(3.11)
U(s)F
(s)(sI-Ax).
Proof.
Accordingto(3.9), wemusthaveC1
0 informula(3.2).
[3COROLLARY3.12.
Xu
isthe largest (M, )-invariantsubspaceo]XTcontainedinker
’.
Proof.
According to (3.10), we have for an arbitrary(M,
)-invariant subspace{(s)}
contained in ker:
W(s)= U(s)F(sI-A)
-.
Hence
{(s)}c__Xu (see
Definition2.1).
It remains to be shown thatXu
itself is an(s4, )-invariant subspace. If
(s)
is a basis matrix ofXu
then there exists astrictly propermatrixQ(s)such that U(s)Q(s)(s).
Let (F1,A1,B1)
be a reachable realiza-tionof Q(s),so that(I)(s)
U(s)FI(sI-A1)-IB1.
Itfollows fromLemma3.13 thatW(s)
:=U(s)F(sI-AI)
-1isapolynomialmatrix.
By
Corollary 3.10,{(s)}
isan(s4, )-invariant subspace. Hence{(s)}_ {(s)}.
Onthe other hand,since(s)
(s)B,
itfollows that{(s)}
c__{(s)}.
Consequently,Xu
{(s)}
{(s)}
is an(M, )-invariantsubspace.I-I
LEMMA3.13.
Let O(s)
gln[s],
A
K
nn,
BK
nr,
(A,B)
reachable.If
O(s)
(sI-A)-aB
is apolynomialmatrix thenO(s)(sI-A)
-
isapolynomialmatrix.Proof.
Wedecompose therational matrixO(s)(sI-A)
-
into itspolynomialand itsstrictly proper part"
O(s)(sI -A)
-
P(s)
+
R
(s);then
Ro
:=R(s)(sI-A)= O(s)-P(s)(sI-A)
is apolynomial of degreezeroandhenceconstant. Itfollowsthat
Ro(sI-A)-XB
O(s)(sI-A)-B-P(s)B
is astrictlyproper polynomial and hencezero. Since (A,
B)
is reachable,thisimpliesRo
0 andhence424 .E.EMRE AND M.L.J. HAUTUS
The foregoing implies that the set of
(M,
)-invariant subspaces in ker is uniquely determined by the numerator polynomial matrix of the matrix fractionrepresentation of the transferfunction matrix:
COROLLARY3.14.Let
U(s)
gqXr[s],
Ti(s)gqXq[s],
1,2, such thatG,(s) :=
T
.1(s)U(s)
isstrictly proper
for
1,2.Let %,Mi, i)betheTi-realization
of
Gi(s)fori
1,2. ThenM
Xcr
is an (M1, 31)-invariant subspaceo]
Xrl
contained in ker1
iff
M is an(2,
32)-invariantsubspaceo]’Xr2
containedin ker2.
REMARK
3.15. Theorem 3.1 may be specializedtothecaseU(s)
0,that is, 0.In
thiscasewehavea realizationofG(s)
0 with the samestatespaceXr
andthe samemap as before.
An
(M, )-invariant subspace ofXr
then is just an M-invariant subspace. Thusweobtainthe following characterization of M-invariant subspaces.PROPOSITION. Let
(s)
be a qxm polynomial matrix. Then,{(s)}
is an sC-invariantsubspaceof Xr
iff
thereexistC
K
q", A1
Km"
suchthatT(s)C1
(s)(sI
A 1).
4. Exact modelmatchinganddisturbance decoupling. Ifwe have anobservable system
(C,
A, B)
with statespaceX
thenwemayconsidertheproblem of characterizingthe (A, B)-invariant subspaces contained in kerC. Using the isomorphism given in
Theorem2.8,wetransform theproblemtothecase ofa suitable T-realization.
For
thiscase wemayappealtoCorollary3.10, by whichacomplete characterization is given.It
is important that,asalready notedinCorollary 3.14,thischaracterizationdepends only
onthe numerator
poly_nomial
U(s). Consequently,wehavethefollowingresult.THEOREM. Let
,
(C,A,
B) be a realization with state space Xof
atransfer
matrixG(s)
T-l(s)U(s),
andlet (q,sg,
3)
bethe T-realizationof
G(s).
If
,
andZ
are isomorphic by the isomorphism
."
X-,Xr,
thenM
_
X
is an(A,
B)-invariant subspacecontained in kerc
iff
thereexist constantmatricesF1,A
satisfyingU(s)F
W(s)(sI-A
), where q(s) isa basismatrixof
(M).
Thuswe seehowcharacterizationsfor
(,
)-invariant subspacesof the particularstate spacemodelY., can be generalizedto arbitrary
(observable)
state space models.In
this section we usethe theorydevelopedthusfartoshow the equivalenceoftheexactmodel matching problem and the disturbance decoupling problem.
PROBLEM 4.1. (Disturbancedecoupling problem
(DDP)).
Given thesystem(4.2)
2(t)
Ax(t)
+
Bu(t)
+
Eq(t), y(t) Cx(t),where
(C, A)
isobservable,determine a constant matrixF
such thatif
u(t)
Fx(t),
>-_O,
the outputy(t) doesnotdependonq(t),
>=
0.The following result has been given in
[17,
Thm.4.2]
in aslightly different but equivalent formulation"THEOREM4.3. Problem 4.1 hasasolution
iff
thereexists asubspaceM
of
thestatespacesuchthat
AM
_
M
+ {B},
{E}
c_M
c_kerC. l-1PROBLEM 4.4.(Modifieddisturbance decouplingproblem
(MDDP)).
Givensystem(4.2),
determine constant matricesF
andDsuch thatif
u(t)
=Fx(t)+Dq(t),theoutputdoesnotdependonq(t).
In
the modified problem, one assumes that not only the state but also the disturbance is directly available for measurement. Similarly to(4.3)
we have the following result.THEOREM4.5.Problem 4.4hasasolution
iff
thereexists asubspaceM
suchthatAMcM+{B},
{E}cM+{B},
MckerC.The exactmodel matchingproblemis defined as follows.
PROBLEM 4.6. Given
transfer
function
matricesGx(s)
andG2(s)
determine a (i)strictly properor (ii)properrational matrix
O(s)
such thatGl(s)O(s)=G2(s).
Problem 4.6(i) will be called the exact model matching problem
(EMMP),
andProblem4.6(ii)willbecalledthe
modified
exactmodel matchingproblem(MEMMP).
Itis thepurpose of thissection to show that theexistence ofasolution of Problem 4.1 is
equivalenttotheexistenceofasolution of Problem4.6(i).Similarly" Problem 4.4 hasa
solution itt Problem 4.6(ii) has a solution.
We
will concentrate on the modifiedproblems.The originalproblemscanbe dealtwith similarly.
Firstwehavetoindicate which
MEMMP
correspondstoagivenMDDP
andvice versa. Letusstart withsystem(4.2).
The dataGx(s)
andG2(s)
ofMEMMP
arethen defined byGI(S)
:-C(sI-A)-IB,
G2(s)
:=C(sI-A)-XE.
Conversely,if we are given
Gx(s)
andG2(s)
inMEMMP,
we construct anobservable realization(C, A,
[B, El)
of thetransfer matrix[Gx(s), G2(s)].
ThenC,
A,
B,E
arethe data forMDDP.
Thus,wehavea one to onecorrespondencebetweenMEMMP’s
andMDDP’s.
Following Theorem 2.8,weassume that
C(sI-A)-X= T-l(s)S(s)
with
T(s)
and $(s) relatively prime, andU(s)=S(s)B;
and we consider theT-realization
(,
,
N)
of Gl(S)=T-l(s)U(s).
According to Theorem 2.8, the map5:x
S(s)x :K
Xr
is anisomorphism. Consequently,weintroduce thepolynomialmatrix
R (s)
:= $(s)E asrepresentativeofEinXr.
ThenwehaveG2(s)=
T-I(s)R
(s)
andwe canstatethe following result.
THEOREM4.7. Let
{(s)}
bean (sg, )-invariantsubspacein ker,
sothatthereexistconstant matrices
Fx
andA
satisfying(4.8)
U(s)F1
(s)(sI
A 1).
In
addition,assumethat{R (s)}_
{(s)}+{U(s)},
sothatthereexist matricesB1
andD1
such that
(4.9)
R (S)= xIt(s)B1
+
U(s)D1.
ThenQ(s) :=
Fl(sI-Ax)-XB1 +Dx
is asolutionof
MEMMP.
Conversely, let Q(s) beasolution
of
MEMMP
and let(F1, A
x,Bx, Dx)
bea reachablerealizationof
Q(s). Then426 E.EMRE ANDM. L. J.HAUTUS
Proof.
If(s)
satisfies(4.8)
and(4.9)
thenU(s)Q(s)-
(s)B1
+
U(s)D1
R (s),
which implies Gl(s)Q(s)
G2(s).
Conversely thelatterequation impliesU(s)Q(s)-R (s).
Hence,(4.10)
U(s)FI(sI-A1)-IB1
R(s)-
U(s)D1.
Since
(A1, B1)
isreachable itfollowsfromLemma
3.13 that(4.11)
(s)
:=U(s)F(sI-A1)
-1is apolynomial.Now
(4.10)
and(4.11)
imply(4.9)
and(4.8).
COROLLARY 4.12.
MEMMP
has a solutioniff
the correspondingMMDP
has asolution.
Similarly oneproves
PROPOSITION4.13.
EMMP
hasasolutioniff
thecorrespondingDDP
hasasolution. Thus, ifwe want to solve(M)EMMP,
we may construct the dataA, B, C, E
of(M)DDP
andsolve the latter problem. Thenwedo notonlyobtain asolution Q(s)of(M)EMMP
but alsoarealizationOf this solution.In
thisrespect,it is importantto notethat thesolutionof
(M)EMMP
onlydependsonthenumeratorpolynomialsU(s)
andR (s).
Consequently, byasuitable choice ofT(s) (not
necessarily equaltotheoriginaldenominatorpolynomial)wemaytrytoobtainasimple
(M)DDP;
compare[3]. We
willformulate thisideamoreexplicitlyin 6.Alsoin 6,wewillgive existence conditions
for asolution of
(M)EMMP
and,hence, of(M)DDP
in termsofU(s)
andR (s).
The following result states that if disturbance decoupling is at all possible by a
(dynamic)controldependingcausally uponq(t),thenitis possible byafeedbackcontrol
of the form u
Fx
+
D
lq.COROLLARY 4.14. Let there existaproper rational matrix
H(s)
such that,if
the controluH(s)q
isusedin(4.2),
theoutput doesnotdependonq. ThenMDDP
hasasolution.
If
thereexists a strictly propermatrixH(s)
with this property, thenDDP
has asolution.
Proof.
If thecontrol uH(s)q
isused in(4.2),
then the transferfunction matrixfrom q to y is
Gl(s)H(s)+ G(s).
If y does notdependon q, then thistransfermatrix mustbezero,henceGI(s)H(s)
-a2(s),
that is,
-H(s)
is a solutionofMEMMP.
Consequently, byCorollary 4.12,MDDP
hasa solution.5. Observers.
We
considerseveral formulationsof the observerproblem,which isawell-knownprobleminlinear system theory. Further referencesonthe subjectcanbe found in
[14],
[15],
[16],
[19], [20],
[2]
and[6].
Thus fartwotypes of formulation of this problem have appeared in the literature: the geometric formulation
(see
[19], [20],
and[2])
and the polynomialmatrixformula-tion
(see
[14],
[15],
and16]).
Here,
our purpose is(based
on the results on the connections of the geometric theory oflinearsystemsandpolynomialmatrixapproaches developedin3)
to show explicitly the algebraic equivalence of the geometric and the polynomial matrixformulations of this problem, including the case where some of the inputs may be
Let 5; (C,
A, B)
be a given system over R. Let C- be a subset of C satisfyingC-f’lR#
.
We
call a rationalfunctionu(s)
stable (withrespecttoC-)
ifu(s)
hasnopoles in C\C-.
In
the continuous time interpretation ofE,
one might choose C-={s
eCIRe
s< 0}
and in discrete time C-{s
eCI
Isl
<
1},
but also different choices ofC- arepossible.We
assume that CNqn,
BeNnr
and that in addition to,
we are given afeedthroughmatrix
D
x Nqr.In
continuous time, the interpretation(Y_,,D)
reads:(5.1)
2Ax
+
Bu,
yCx
+
Du,
and the transfer function of
(, D)
isG(s)
G.D(S)
C(sI-A)-IB
+
D.DEFINITION 5.2.LetL
".
A
system(,, )
(,
:,
,
:)
is anL-observerof
(,
D)
iffor
every initial valueXoof
E,
Yo
of
X
and every controlfunction
u, theoutputof
(5.3)
=+iy,
=+Dy
satisfies"
-
Lx
isstable (inparticularrational).The observerusesonlythe output of(Y_,,
D).
If onewants toconsider the situation in whichpartialor totalknowledge of theinputof5;isavailable, onecan incorporatethis intheproblembyasuitable choice ofD.
In
particular,if the input iscompletely known,oneintroduces newmatrices,/
and a newoutput37
ofE
accordingtosothat
7
x
+/u
represents thetotal data availablefortheestimation ofLx.
Letususe thefollowingnotation"G(s)
:=G.,(s)=
C(sI-A)-B
+D,
(S)
:=O,,l(S)
(sI-
/)-IB
-1-/,
GL(S)
:=L(sI-A)-IB.
Thenwehavethe following result.
THEOREM 5.4.
Let
the system be reachable and let be observable. Then the followingstatements areequivalent:(i) (Y.,,
D)
isanL-observerof
(:E,
D).
(ii)GL(s)= G(s)G(s)
andtr(a)_ C-. (iii) Thereexists arealmatrixM
suchthatDD
O,MA
AM +BC,
LCM+DC,
MB
BD
and
o’(A
c_C-.Proof.
(i)=), (ii). IfXo 0,o
0, thenwehave-L
=H(s)a,
428 E.EMREAND M.L.J. HAUTUS
particular, u
e^’uo
withAC,
wefind:9-L=
y.(s-A)-"-H(s)Uo.
Since
)3-L
hastobe stable for all nN,
h C,Uo
,
itfollowsthatH(s)
O. Furthermore,ifXoO,
uO,
then:-L,f
(sI-)-$o.
Since
(C,
A)
is observable,thestabilityof:
-L
implies thatr(A)
_
C-. (ii)=>
(iii).We
use the matrix fraction representationO’o(S)
:=B’(sI-A’)-IC
’=
T-I(s)U(s)
defined
by B’(sI-A’)
-=
T-(s)S(s),
U(s)= S(s)C’,
andweconsidertheT-realiza-tion
(,
M,
)
ofG’o(S) (see
Theorem
2.8).
Then the equationG(s)G(s)= GL(s)
mayberewritten as
Hence,
ifwewrite(5.5)
then(5.6)
U(s)
+
T(s)D’)’(s)
S(s)L’.
W(s)
:=(U(s)+
T(s)D’):’(sI-’)
-,
W(s)C’
SL’
U
+
TD’)D’.
Since
(C, A)
isobservable,itfollows fromLemma3.13 thatW(s)
isapolynomial matrix.Equation(5.5)
impliesU(s)B’
+
T(s)D’B’=
W(s)(si-A’).Hence, {W(s)}
is an(M,
)-invariantsubspace, and(5.7)
M(s)
sg(s)- T(s)D’B’ = (s)A’
+
U(s)B’
(see
(3.4)). We
consider again themap
defined in the proof of Theorem 2.8"x
S(s)x.
ThenwedefineM’
b-lW(s),
so thatS(s)M’
W(s).
Itfollows from(5.7)
that
A’M’
MSfM’dW(s)
W(s)A’
+
U(s)B’
M’A’
xC’B’.
Hence,
A’M’
M’A’ +
C’B’.
Furthermore,(5.6)
implies"T-l(s)W(s)
’-
T-(s)S(s)L’
+ T
-1UD’=
Since the left-hand Side is strictly
proper,
itfollowsthatD’D
0 and5M’
C’ L’ +
C’D
O,
Hence,
M’C’-L’+C’D’
=0.Finally,itfollows from
(5.7)
thatT-l(s)U(s)’
+
T-(s)(s)fix’
sT-XW(s)-D’:
’.
Hence,
since
T-l(s)U(s)
andT-l(s)(s)
arestrictlyproper. (iii)::>
(i).A
short calculationyields-Lx
C(-Mx),d
d--
Mx
A
Mx
),and the result follows fromr(A)_ C- [3
The equivalence (ii):> (i) isgiven in
15],
andthe equivalence (i):> (iii), withthe a priori assumption(intheproofof(i)::), (ii))that-
Mx
isstable, isgivenin[2].
Notice that here the stability of-
Mx
is aconsequence,ratherthan
anassumption(see
the proof of (iii)::),(i)).For
the situation of availability of the whole input, thiswas also shown in[6].
REMARK
5.8. The results of thissectioncaneasilybeextendedtosystems overan arbitraryfieldK,
providedanappropriate definitionofstable rational function has been defined. Suchadefinitioncanbe given as follows:Let
:t/be amultiplicativesubsetofK[s]
(i.e.,p(s)
l,g(s)
ell::>p(s)g(s);
1).
Then we say that a rational functionr(s)e K(s)
is stable ifr(s)
has the representationr(s)=
p(s)/q(s) withp(s)
K[s], q(s)
l. Then the stable functions formaring.In
the situation describedabovewehave
{p(s)
K[s]lp(s)
0=>
sC-}.
In
the general situationTheorem5.4 remains validifone replacesthe conditionr(A)
C-with"(sI-
fi)-i
isstable".A
particularexample,which is relevantfor discrete time systems, over arbitrary fields,is/:=
{s"
In
=0,1,...}.
An
observer constructed according this multiplicative set is called a deadbeatobserver.
{i. Reaehalility
snlslaees.
Let(s)
be a full column rank basis matrix of an(,
)-invariantsubspace.Recall
theinterpretation of the matricesA
1,F1,C1
givenin(3.7),
(3.8)
and(3.9).
LetB1
beanyconstantmxp matrixsuchthat{(s)B1}
{U(s)},
say
(s)B1
U(s)L1.
Then
B1
isthematrix ofthe (codomain) restrictionofY3L1
to{(s)}. It
followsthat (g-3)’Ll
v(s)A
B
vfor everyv e
K p.
Consequently,(6.1)
(4-IL)={(s)[B,...,
An-IB1]}.
Thisformula immediatelyimpliesthe following result.
THEOREM6.2.
Let
(s)
bea (full
columnrank)
basis matrixof
an
(4, )-invariantsubspace. Then
(i)
{(s)}
isa teachabilitysubspaceiff
thereexistsa constantmatrix
B
such that{T(s)B1}
_
{U(s)},
and(A1,
Ul) isreachable(here
A1
isgivenby(3.2)).
(ii)If
B
isa constantmatrix such that430 E.EMRE AND M. L. J. HAUTUS
then
{q(s)[B1,’’’,
A’-IB1]}
isthe supremalreachabilitysubspace containedin
{(s)}.
Letus nowconsiderreachabilitysubspacescontainedinker
c.
Let(s)
be a basis matrixofsuch aspace. AccordingtoCorollary3.1 0, thereexist matricesF1
andA such that(6.4)
(s)
U(s)FI(sI-A1)
-1.
Itfollows from Theorem6.2 that there exists
B1
suchthat (A1,B1)
is reachable and{(s)B1}
_
{U(s)},
say(s)B1
U(s)L1. Hence
(6.5)
U(s)Q(s) U(S)tl,where Q(s):=
FI(sl-A1)-IB1.
Also, since(s)
has full column rank, (F1,A1)
isobservable,asfollows from
(6.4).
Hence(F1, A
1,B1)
is a minimalrealization ofQ(s).COROLLARY6.6. Thereexists anontrivial reachabilitysubspacecontained in ker
iff
{ U(s)}
0xt,
{o}.
Proof.
If(s)
is a basis matrix of the(,
)-invariant subspaceXtr
and(s)=
U(s)FI(sI-A1)
-1,
then the supremal reachability subspace contained inXt
(or,
equivalently,inker
)
is nontrivial iffB1
0,whereB1is a matrixsatisfying(6.3).
[3According to
(6.5), O(s)-L1
is a nontrivial right zero matrix ofU(s).
Consequently, if the supremal reachability subspace contained in is nonzero, then
U(s)
is not left invertible. The converse, however, is not true.For
example, ifU(s)
Ul(S),0]
whereUl(s)
isleft invertible,then it iseasilyseen thatU(s)
is notleft invertible, and{U(s)}
71Xtr
{0}.
In
orderto giveanecessary and sufficient condition forthe existenceofamaximalreachability subspace containedinkerc,
weconsider theK[s]-module
(6.7)
A:={v(s)
Kr[s]l
U(s)v(s)
0}.
Thismoduleisgenerated by the columns ofamatrix
M(s)
(see 15,
Thm.3.1]).
COROLLARY6.8. Thereexists anontrivial reachability subspace containedin kerc
iff
themodule Adefined
in(6.7)
is notgeneratedbya constant matrix.Proof.
LetM(s)
be a generator matrix of minimal degree, sayM(s)=
Mos
k+"
"+Mk.
Thens-kM(s)=
O(s)-L1,
whereO(s)=Mls
-1+.+Mks
-kand
L1
=-M0.
WehaveU(s)O(s)=U(s)L1
and
U(s)L1
O,
since, otherwise,[M(s)-skMo,
M0]
would be agenerator matrixoflowerdegree than k.Itfollows that
{U(s)L1} {U(s)}f’]Xu,
sothat{U(s)}fqXcr#{0}.
Conversely, suppose that A is generated by a constant matrix, sayD,
and thatv{U(s)}f’lXtz,
say vU(s)c
U(s)r(s),
wherec is a constant vector andr(s)
is astrictlyproperrational vector.
It
follows that there existsa rationalvectorq(s)
such thatc-r(s)=Dq(s).
Decomposing q(s) into a polynomial and a strictly proper partq(s)
=ql(s)+q2(s), we conclude that c=Dql(s), so that vU(s)c
0. Hence,{ U(s)}
o
x
{0}.
Nowwe have a procedure for constructing reachability subspaces contained in
ker
.
Choosing anymatrixL1
suchthat{U(s)L1}_Xu,
wehaveU(s)O(s)= U(s)L1
for some strictly proper
O(s).
If(F1,A1,B1)
is aminimalrealizationofO(s),
itfollowsthatq(x) :=
U(s)FI(sI-A 1)
-1isabasis matrix ofareachabilitysubspace, provided the columns ofq(s)areindependent.In
general,it seems difficult toformulateconditionsupon
L1
andQ(s)thatguarantee that(s)
hasfullcolumnrank.A
sufficientconditionfor this isthat Q(s)be astrictlyproperrational matrix with minimal McMillandegree
satisfying the equation U(s)Q(s)=
U(s)LI.
Indeed, ifin thiscase(s)
doesnothave full columnrank, thereexists(s)
withlesscolumns than such that{(s)}
{(s)}.
Since
{(s)}
is an (sg, )-invariant subspace, there exist F2,A2
such that(s)=
U(s)F2(sI-A2)
-a.
Also,there existsD1
suchthat(s)= (s)D1. Hence,
U(s)O(s)--
xIt(s)B1
((s)D1B1
U(s)Q2(s)U(s)L1,
where Q2(s):=
F2(sI-A2)-aDIB1
has lower McMillandegreethanO(s).
THEOREM6.9.
Let
L1
bea constant matrixsuchthat {U(s)LI}={U(s)}fqXu.Let
Q(s) be a strictly proper rational matrixof
minimalMcMillan degree, satisfying theequation U(s)Q(s)
U(s)LI.
Let(F1, AI, B1)
bea minimal realizationof
Q(s). Thenxlz(s)
:=U(s)FI(sI-AI)
-1isabasis matrix
of
thesupremalreachability spacecontained in ker.
Proof.
The supremal reachability subspace contained in kerc
is the (unique)minimal (, )-invariantsubspace satisfying
(Im )
(3 o/_ 7/"_
o/g, where o/# isthesupremal (s
g,
)-invariant subspacegontained in ker.
To see this, observe that an(4, )-invariantsubspace7/"satisfying
(Im )
f3/d/"_
7/’_
is(
--)-invariantfor every such that is(1
-’)-invariant. Indeed,(
")
7/"_
(4
-)
//VW
and
(4- Y3’)
7/’_
7/’+
Im
imply(1-
)c_ Wf3(+Im )=
o//.+o/g.f-)im_
.
Since
{U(s)}f]Xu={U(s)L1}={xIt(s)B}_{xIt(s)}Xu,
and because of the minimalMcMillandegree ofQ(s), the resultfollows. 71
In
thenextsection,it willbe shownhow Theorem 6.9canbe usedfor the explicit construction of the supremal reachability subspace.7. Constructive characterizations. Conditions for solvability and the charac-terization of solutions of variousproblemscanbe made explicit by theuseofrowand column propermatrices
(see
[16]).
Thiswill bethe subject ofthissection.If
R
KPq[s]
hasrowsrx(s),
",rp(s),
thendegri(s)iscalledthe ithrowdegreeof
R (s).
The coefficientvectorofs’
inri(s),whereui degri(s)iscalled theithleadingcoefficient
rowvector, and is denoted by[ri]r. We
denote by[R]r
thematrixof leading coefficient row vectors, that is, theconstantmatrix with rows[rx]r,
",[rp]r.
Similarly,[R ]c
denotes the matrix ofleadingcoefficient column vectors,thatis,[R
]c
([R’])’. A
matrix iscalledrow
(column)
proper if[R
]([R
]c)
isnonsingular.A
rowpropermatrixiseasilyseen tobe right invertible.Conversely, wehave
(see [16,
Thm,2.5.7])
LEMMA 7.1.
If
L(s)
KPq[s]
is right invertible thereexists a unimodular matrixM(s)
KPP[s]
such thatM(s)L(s)
is rowproper withrowdegreesu,,
upsatisfying,
<-...<=
up.If
L(s) KPq[s]
is not right invertible, them exists a unimodularmatrixM(s)
such thatM(s)L(s)=[
Ll(s)]
0where
L(s)
is rowproper withrowdegrees,
<-.<=
’l. Thenumberof
rowsof
Ll(S)
equals the rankof
L(s).
The row degrees u are independent of
M(s)
(which is not unique) andwill becalledthe row indices of
L(s).
The following result
(see
[14, Property
2.2])
states a simple criterion for thepropernessofarational matrix
T-(s)
U(s)
if the denominator polynomialmatrixisrow432 E. EMRE AND M. L.J.HAUTUS
LEMMA
7.2.LetT(s)
berowproper withrowdegrees1)1, 1)q.If
therowdegreesof
U(s)
arehi,",
hq thenT-(s)U(s)
isproperiff
h <-_vi(i 1,.,
q)and strictly properiff
Ai<
1}i (i i, ",q).Observethat if
T
is not rowproper,there exists a unimodular matrixM(s) such thatT(s):=
M(s)T(s)
is row proper. If we define Ua(s):= M(s)U(s), we haveT-(s)U(s)
T-
(s)U(s),
andwemayapply Lemma7.2.Letus now consider
(M)EMMP
asdefinedinProblem 4.6.Assume
thatwehaveamatrixfraction representation
T-(s)[
U(s), R
(s)]
of[G(s), G_(s)].
Thenthe equation forO(s)
reads(7.3)
U(s)Q(s)=R(s).In
order that this equation has a(not
necessarily proper) rational solution, it isnecessary and sufficient that rank U(s)=rank
[U(s),R(s)].
For the existence of aproper solution additional conditionshave to beimposed. Writingdown the ith rowof
(7.3)
Ui(S)((S)- ri(s),
wenotethat anecessaryconditionfor the existence ofapropersolutionisdeg ui(s)>=
degri(s).Thefollowingresult shows that this is also sufficient provided that
U(s)
hasthe formwith
U(s)
row proper. According to Lemma 7.1, this can always be obtained by premultiplying(7.3)
with a suitableunimodular matrixM(s).
THEOREM7.4.Let
M(s)
bea unimodularmatrixsuch thatU(s)
M(s)R(s)
I_R(s)J
M(s)U(s)
0
where
U(s)
isrowproper Let the row degreesof
U(s)
be v, 1)l and let the rowdegrees
of
R(s)
beA,.,
At.
Then(7.3)
hasaproper solutioniff
R(s)=0
andAi<-_(i 1,...,
l).
Equation(7.3)
has a strictly proper solutioniff
R(s)=0
and Ai<
(i=1,...,l).Proof.
The conditions are necessary accordingto theforegoing discussions. Nowassumethatthe conditionshold. Then there existsL grl
such that
U(s)L
isa rowproper matrix with rowdegrees u,
,
Ul. DefineO(s)
:=L(UI(s)L)-IRI(S).
Then Q(s) satisfies
(7.3). It
follows from(7.2)
that Q(s)is proper. Theproofforthestrictlypropersolution is similar.
We
can express the result of Theorem 7.4 in a way not involving explicitlythematrix
M(s):
COROLLARY 7.5. Equation
(7.3)
hasaproper solutioniff
U(s)
and[U(s), R(s)]
have thesamerankandthesame row indices.In [14],
noexplicitconditionfor thesolvabilityisgiven.In
[5],
acondition is given intermsof the kernel ofthematrixU(s), R
(s)].
The conditions given in Theorem 7.4 andCorollary7.5 aredirectlyexpressed in termsof the matricesU(s)
andR (s).
The set
Xu
is the largest(M,
)-invariant subspace contained in kerc.
By
definition
x(s)
Xu
iff the equationhas a strictly proper solution
v(s).
Therefore, using Theorem 7.4, we can give aconstructive characterizationof
Su.
COROLLARY 7.6. Let
M(s)
be as in Theorem 7.4. Thenx(s)Xu
iff
y(s) :=M(s)x(s)
satisfies
the conditionsdegyi(s)
<
vi (i i, l),yi(s) 0 (i=/+1,...
,q).
Here
yi(s)denotes the ithcomponento] y(s). In
particular,if
weintroducetherow vectorw(s)
:=[s
-,
,1],
thenM-(s)
W(s)
is abasismatrixof
Xu,
whereW(s)
:=[
W(s)
with
Wx(s)
:= diag(wl-a(S),
Wl-X(s)).
Oneway of solving(7.3),alreadymentioned in 4,isthereformulation of
(7.3)
as a(M)DDP. In
doing so,itisnotnecessaryto use theoriginal denominatormatrixT(s).
Onemight tryto find a newdenominatormatrixTx(s)
suchthatT-
(s)U(s)
isstrictlyproperand
Tl(s)
isassimpleaspossible.IfwechooseTl(s)
rowproper,thenaccordingtoLemma 7.2,it sufficesfor the strictpropernessof
T
-1U,
thattherowdegreesofTx
arelargerthan therowdegreesof
U.
Ifwedenote the latterbyAx,,
A1,the simplestchoiceof
Tx(s)
isTl(s)=
diag(sX,+x,
.,
sXl+l).
For
this computation,itisnotnecessarythatU(s)
be inrowproperform.Butifwetransform
U(s)
such thatithastheform given in Theorem7.4,then the dimension of the statespacewill be minimal.Theseideas areworkedoutin more detail in[3].
We
conclude thissection with a constructionof the supremal reachability subspacecontained in ker
.
Tothisend,weconsiderthespaceA
:-{v(s)K(s)l
U(s)v(s)-O},
and wechoose aminimalbasis for
A (see [5]),
thatis, abasis forA(see
(6.7))
which iscolumn proper. We define
La
:-[M]c.
Furthermore we choose anyD(s)Kl[s]
whichhas thesamecolumndegrees
M(s)
and suchthat[D]c
L
Thenweobserve(by
Lemma
7.2)
that,ifN(s):=LaD(s)-M(s),
then
O(s)
:=N(s)D-X(s)
is strictlyproper.NowwehaveTHEOREM7.7.(i)
{ U(s)L1}
Xu
({ U(s)},
(ii)
O(s)
is astrictlyproperrational matrixof
minimalMcMillandegreesatisfying(7.8)
U(s)O(s)=U(s)La.
Hence,
if
(F1,
A1,B1)
is aminimal realizationof
O(s),
then(s)
:=U(s)Fx(sI-Ax)
-1is a basis
of
thesupremalreachability subspace containedin ker.
Proof.
(i)SinceU(s)M(s)
0,itiseasily seen that(7.8)
issatisfied. Thisimpliesthat{ U(s)L 1}
c__Xu
71
{ U(s)}. Suppose
thatthere existsa matrixL
Offull column rank suchthat
{U(s)L1}c_
{U(s)/Sx},
andU(s)/S1
U(s)O(s)
for some strictlyproperO(s).
Let),/
beright coprimepolynomialmatricessuch that((s)- l(s)l-a(s), and/(s)
iscolumnproperwith
[D]
I.ThenU(s)(N(s)- LID(S))
O.434 E. EMRE AND M.L.J. HAUTUS
over
K(s).
ButthenL1
cannothave more columns thanL1.
Consequently,{U(s)L}
{U(s)L}.
(ii)
Suppose
thatO(s)-
lql’(s)-(s)
hasalower McMillandegree thanO(s)
andthat
N(s)
andD(s)
arerelatively primeandthatD(s)
iscolumnproperwith[D
(s)]
I.
Thenwehave
U(s)(N(s)-LD(s))
O,
andhence,
N(s)-LID(S)= M(s)R(s). By
the "predictable degreeproperty" (see [5,
3,Remark]),
this implies that the sum of the column degreesofD(s),
and hence deg detD(s)
isnotlessthandegdetD(s),
which contradictsourassumption.8. Generalization tosystemsrepresented by Rosenbrock’ssystemmatrix.
In
this section,weindicatehow the result of 3canbe generalizedtothe casewhere the system isrepresented byasystem matrixT(s)
U(s)]
(8.1)
P(s)
V(s)
W(s)J’
where
T(s)
K’’[s]
is nonsingular andP(s) K
(q+l)(q+r)Is].
We
assume that the transfer function matrixG(s)
:=V(s)T-I(s)U(s)+
W(s)
and thematrix
T-l(s)U(s)
are strictlyproper. Ifthe latter conditionisnotsatisfied,wecan obtain this by strict system equivalence
(see
[13,
3.1]).
Indeed, ifwedefineU(s)
:=rr(U(s)),
then
O(s)
:=T-l(s)(U(s)
Ul(S))
isapolynomial matrix.Therefore,
[
T(s)
P(s)
:=I.-V(s)
U(s)
W(s)
+
V(s)O(s)
isapolynomial system matrixwiththe same transfermatrix
G(s).
In
[9],
it isshown that the mapssC
Xr
--,Xr
x(s
zrr(sx (s )),
3
K
-->XT
UU
s u,"XT
">Kl’x(s)
’’>(V(s)
T-l(s)x(s))_l
yield a realization
(,
M,
)
ofG(s)
which is reachable iffT(s)
andU(s)
are left coprime, and observable iffT(s)
andV(s)
areright coprime.It
is easily seen that Theorem 3.1 is equally valid in this situation. Instead ofCorollary3.10weget
THEOREM8.2.
Let W(s)
beaqx
mpolynomialmatrix. Then{(s)}
isan(M,
)-invariantsubspace in ker
iff
there existsCI
K
",
F
K
"*,
A
K
""
and anxmpolynomialmatrix
(s)
such that(8.3)
P(s)
IV1]
[*(s)](si_Ax)"
F
!_(s).l
Cx,
El,A1
we have(3.2)
and hence(3.6).
Butthenxlt(s)
V(s)
T-(s)W(s))_l
=((V(s)C +(G(s)-
W(s)F))(sI-A)-)-((V(s)C-
W(s)F)(sI-A)-)_
since
G(s)
and(sI-A)
-
arebothstrictlyproper.Now
itfollows fromLemma(8.5)
that
(8.4)
(s)
:=(-V(s)CI
+
W(s)Fx)(sI-Ax)
-is apolynomialiff
xlt(s)=
0. Combining(3.2)
and(8.4)
yieldsthe desiredresult.LEMMA
8.5.Let
Q(s)Ktn[s],
A
K
.
If
(O(s)(sI-A-1))_l
=0, thenO(s)(sI-A)
-x is apolynomialmatrix.TheproofisanalogoustotheproofofLemma3.13 and will beomitted. ThegeneralizationofCorollary3.12 can beexpressedintermsof the map
Kq[s]"
r
/x(s)/x(s).
Kq+l[s].->
y(s)J
L
COROLLARY
8.6. The largest (sg,Yd)-invariant subspaceof
XT-contained
in keris(Xp).
Theproofis similarto theproofofCorollary3.12 andwillbe omitted.
Acknowledgment. Oneof the authors
(E.
Emre)
wouldlike tothank theDepart-mentof Mathematics of theEindhovenUniversity ofTechnology forfinancialsupport and hospitalitywhilethisresearch wasbeing done.
REFERENCES
[1] G. BENGTSSON,Outputregulation andinternalmodelsmafrequencydomainapproach,
Automatica-J.IFAC, 13(1977),pp. 333-345.
[2] P. BHATTACHARYYA,Observer designforlinear systems withunknowninputs,IEEE Trans.Automatic Control, AC-23(1978),pp. 483-484.
[3] E.EMRE,Nonsingularfactors ofpolynomialmatricesand (A,B)-invariant subspaces, Memorandum
COSOR78-12,Dept.of Math.,Eindhoven Univ.of Technology;thisJournal,toappear. [4] E. EMREAND L. M. SILVERMAN,Relativelyprimepolynomialmatrices: Algorithms, Proc. IEEE
Conferenceon Decision andControl,Houston, TX,1975.
[5] G. D. FORNEY,Minimal basesofrationalvectorspaces,withapplicationtomultivariablelinearsystems, thisJournal,13(1977),pp.,493-520.
[6] T.E.FORTMANNANDD. WlLLIAMSON,Designforlow-orderobserversforlinearfeedbackcontrollaws,
IEEE Trans.AutomaticControl, AC-17(1972), pp. 301-308.
[7] P.FUHRMANN,Algebraicsystemtheory, ananalyst’spointofview,J.FranklinInst.,301 (1976),pp. 521-540.
[8]
,
Linearalgebra andfinitedimensional linear systems,Math.Rep. 143,BenGurionUniv. oftheNegev,Beersheva, Israel.
19]
,
Onstrict systemequivalence and similarity,Internat. J.Control, 25(1977),pp. 5-10. [10] A. G. J. MACFARLANE AND N. KARCANIAS, Relationships between state-space andfrequency-responseconcepts,Preprints of7th.WorldCongress IFAC, Pergamon Press, NewYork,1978, pp. 1771-1779.
11 M.L.J.HAUTUSANDH.HEYMANN,Linearfeedback--analgebraicapproach,thisJournal, 16(1978),
pp. 83-105.
[12] A. S. MORSE,Minimal solutionstotransfermatrix equations,IEEETrans.AutomaticControl, AC-21
436 E.EMRE AND M.L.J. HAUTUS
[13] H. H.ROSEN3OCK,Statespace and multivariabletheory, Wiley,NewYork, 1970.
[14] S. H. WANGAND E.J. DAVISON, A minimizationalgorithmforthedesignoflinearmultivariable systems,IEEE Trans.AutomaticControl, AC-18(1973),pp. 220-225.
[15]
,
Observing partialstatesforsystems withunmeasurable disturbances, Ibid., AC-23 (1978),pp. 481-483.[16] W.A. WOLOVICH,LinearMultivariableSystems,Springer-Verlag,NewYork, 1974.
[17] W.M. WONHAM,LinearMultivariable Control:AGeometricApproach,LectureNotesin Mathematical
Systemsno.,101, Springer-Verlag,NewYork, 1974.
[18]
,
GeometricMethodsin the structural synthesisol linear muttivariable controls. Proc. ofJoint AutomaticControlConference, SanFrancisco,June1977.[19]
,
Dynamicobservers: Geometrictheory, IEEE-AC-15,pp. 258-259.[20] W. M.WONHAMAND