Decoupling of multivariable control systems over unique
factorization domains
Citation for published version (APA):
Datta, K. B., & Hautus, M. L. J. (1984). Decoupling of multivariable control systems over unique factorization domains. SIAM Journal on Control and Optimization, 22(1), 28-39. https://doi.org/10.1137/0322003
DOI:
10.1137/0322003
Document status and date: Published: 01/01/1984 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.
Vol.22,No.1,January1984 0363-0129/84/2201-0003$01.25/0
DECOUPLING
OFMULTIVARIABLE CONTROL
SYSTEMSOVER
UNIQUE FACTORIZATIONDOMAINS*
K. B. DATTAS" AND M. L. J. HAUTUS
Abstract. Necessary and sufficient conditions are established for the existence of a state variable
feedback decouplingof anm-input, m-outputtime invariant linearcontrol system over auniquefactorization domain. An explicit computation is provided forthe feedback and the feedforward gain matrix. Also
necessary andsufficient conditionsforthe existence of astability-preservingstatefeedback decouplingare
given.Theresults areillustratedbysomeexamples.
Keywords, decoupling, delay sygtems,systems overrings,multivariablesystems,statefeedback
1. Introduction. Thedesignandsynthesis of noninteractingcontrol in multivari-able controlsystemsby state-variable feedbackwereinitiated by
Morgan
(1964)and definitive results in this direction by establishing necessary and sufficient conditionsfor the existence of a
decoupling
feedback, as well as an explicit construction, were first given by Falb andWolovich (1967). Their resultswere formulated for systemswithreal coefficientsbuttheyareeasilyseen tobeextendibletosystems overarbitrary
fields. The extension of these resultsto systemsover rings, however,is lessobvious.
Onthe otherhand, systemsoverrings have showntopossessawiderange of potential applications such asdelay systems,2-D systems, parametrized systems,discrete time distributed systems, systems with integer coefficients, etc. We refer to the survey papers E. D.
Sontag
(1976), (1981), E. W.Kamen
(1978),and thereferences therein. Thisabundance ofcontrolsystemswhich canconveniently be modelledassystems over rings is a motivation for a systematic investigation of systems over rings. Thisinvestigationwas startedwiththe thesis Rouchaleau
(1972)
and thepaper Rouchaleau,Wyman
andKalman (1972)
andithas receivedmuch attention recently.The purpose ofthispaperistoformulate necessaryand sufficient conditions for the existence of adecouplingstate feedback for alinear time-invariant system over aunique factorizationdomain.This particular class of ring is wideenoughtoencompass
almost all themodels arising from the applicationsmentionedbefore andontheother hand, it allows a complete solution of the problem. The conditions which will be
obtained reduce to the Falb-Wolovich conditions when applied to systems over a field. The method ofproof, however, is completely different fromthe proof in Falb andWolovich (1967). Itcan beregarded ageneralizationto systemsoverrings of the type of proof given in Hautus and Heymann (1980), (1983) and it is based on a
characterization of feedback transformations given inHautus and
Heymann
(1978). Itis possibletoaxiomatize the concept of stability for systemsoverrings in suchawaythatineach particular specification and application(delaysystems,2-Dsystems)
the notion ofstabilitycustomary in thatfieldaccommodates conveniently in the general
framework.
An
examplewillbe given in 2. The treatmentisbasedonwhatwehave called "denominator set". This concept was introduced for systems over a field in*Receivedby theeditorsMay26, 1981, andinrevisedform March8, 1982.Thisresearchwaspartially supported bythe National Science Foundation undergrant ECS-7908673 andby theNational Science
Council,TaiwanundergrantVE80003.
tDepartmentof ElectricalEngineering,liT,Kharagpur-721302, India.Formerlyof theDepartment
ofApplied Physics,CalcuttaUniversity.
tDepartmentof Mathematics, University ofTechnology, Eindhoven,The Netherlands. Thispaper was completedwhile thisauthor was on leave asTRWVisitingLecturerattheDepartmentofElectrical
EngineeringSystems,UniversityofSouthernCalifornia,LosAngeles,California90007.
and Khargonekar and Sontag
(1981)
(where the name Hurwitz set is used).In
the general framework of stability thusprovidedwegivenecessaryand sufficient conditions for the existence ofastability preserving decouplingstate feedback.The problem formulation and the main results are given in 3.
ExampIes,
illustrating theresults,aregiven in 4,and 5isdevotedtotheproofof our main result. 2. Stability ofsystems overrings. Throughout the paper
g
willdenoteaunique factorizationdomain (= UFD) orfactorialring(see
Samuel (1963),Barshay(1969)).
We
usethe notationsg/[z and(z)
todenote the rings ofpolynomialsand rationalfunctions over
,
respectively.A
polynomialq iscalledmonic ifitsleading coefficient equals 1.A
rational functioniscall.ed
causal (or proper) if .it has arepresentation
of the formp/q,where q is a monicpolynomial anddegp_-<deg q.A
denominator set is asubset @ of[z
satisfying the following conditions’(i) ismultiplicative,i.e. 1 andifp,q
(ii) Eachpolynomialp @ is monic (inparticular 0
:).
(iii) 5 issaturated, i.e.ifp and q is monic anddividesp then q
(iv) There existsa g such thatz-a
.
Since a denominator setis multiplicative, itis possibletoassociate with it a ring
of
fractions
tobe denotedby[z] (see
Barshay (1969,Chap.3)).
Specifically[z]
isthe set ofrational functionshavingarepresentation ofthe form
p/q,
where p and q are polynomials and .q.
It is well known and easily seen that[z]
is a ring,even a
UFD
(see Samuel (1963, Thm.4, p. 29)).In
addition,weintroducethesetof causal fractions in[z],
i.e. elements of[z]
that are causal rationalfunctions.
This set isdenoted by
[z
],
or, if thedenominator setdoesnothave tobe specified, by.
LEMMA2.1. isa
UFD.
For a proof, see 5. The set of all monic polynomials, which is denoted 0,
is an example of a denominator set. The corresponding set of causal fractions is denoted
0.
A
(free) linearsystem is identified by aquadruple (A,B, C, D) ofmatricesover ofsuch dimensions thatthefollowing
equationsarewell defined.(2.2) x,+
Ax
+
But,
y,Cxt
+
Du,,
where
x,:=",
utq/:=,
Equations (2.2) give a discrete time interpretation of the system
,v.,
:= (A,B,
C, D).
The system is calledreachableifthe columns of the matrix
[B, AB,
,
A"-IB
spanthe total state space
".
(See Sontag(1976)
fordetails.)To
the systemX
atransfer
function
(2.3) W(z) :=
Wx(z):=C(zI-A)-IB
+Disassociated. This is a matrixwhoseentries are causalrationalfunctions.
For
agiven transferfunction W(z),E
(A,B, C, D)iscalledarealization if(2.3) holds.Other interpretations of
E
canbe given.Systems
over ringscanbeused
tomodel systemswithparameters,systemswithdelays, 2-D systems, neutralsystems(see
Eising(1980),
Hautus
andSontag (1981), E. W.Kamen
(1978), Rouchaleau (1972),Sontag
(1976),(1981)). We
will give an example below.By
a suitable choice of (andsometimes of
g,
see Eising (1980,4.3))
one can accommodate various stabilityDATTA AND
denominator set have been chosen, we call a rational function stable if it is in
Y[z]. A
(single variable) stable transferfunction is an element of[z].
An
nxnmatrix
A
is called astability matrix if det(z!-A)
.
Obviously,W(z)
isstable ifA
isastabilitymatrix.Theconverse isnotalwaystrue,however,foranystable ti’ansfer function matrix, there exists a(free)
realizationX
which is stable, i.e., for whichA
isastabilitymatrix
(see
Sontag (1976)).
Letus give someexamplesof interpretations of systems over rings and particular choicesofdenominator sets.
Example 2.4.
In
the case thatY
(the fieldof realnumbers) stabilityoftenisformulated in terms ofpole location. Specifically, a set C-__.Cis given and amonic denominator q(z) is in ittithas nozeros outside C-.
It
is easilyseenthat,
thus defined,is adenominatorsetprovidedC-f’].
[3Example 2.5.
One
canmodel adelay system withdelays allmultiple of agiven positive real numberz byasystem over the ringg
[cr]
of polynomials inor, whereo-stands for thedelay operator
crx(t)
x(t-r).
The system then will be of the form
(2.6)
A(cr)x+B(cr)u, yC(cr)x
+D(r)u
where
A, B, C, D
arepolynomial matrices. The systemic significance of the transfer function(2.7)
W(s,r)=D(cr)+C(cr)(sI-A(cr))-XB(cr)
isdescribedin
Sontag (1976). In
particular,applyingaLaplacetransformto(2.6)
yields(2.8)
(s)
W(s,
e-)a(s).
It
is well known (see Hale (1977,7.4))
that X=(A,B, C,D) is (externally) stable iff W(s,e-)
has nopoleinRe
s ->0.Thus,herewedefine(2.9)
:=(p
[s,
r]lp
is monic with respect to s and p(s,e-)
0 forRe
s=>
0}.
When saying p ismonic withrespectto s we mean that p is of the form
p(s,
r)
s"
+p(r)s"-
+" +p,(r)where p,..., p,
R[o,]. It
is easily seen that is a denominator set.In
order thatthesystem beinternally stableone mustrequire that det
(sI-A(cr))
be in.
[3Furtherexamplesdemonstrating thegeneralityof the stability concept described here can be given.
Compare Datta
andHautus
(1981), Eising (1980),Hautus
andSontag
(1981),Kamen (1980).
3. Problem formulation and statement of themainresults. First,wegiveageneral
formulation ofthe decouplingproblem.
We
introducethe i/s-map correspondingtosystem (i.e., system
(2.2))
by(see
Fig.3.1)
(3.1)
W(z):=(zI-A)-B,
so that W
CW
+
D. LetF
and G be dynamical systems withdimensionssuch thattheformula
(3.2)
u-F(z
)x+
G(z
)y+ u
Ws
xFIG. 3.1
into asystem
E,6
withtransfermatrix(3.3)
Wl,O(z)
W(z)L,o(z),
where
(3.4)
Lv,(z)
(I
+F(z)Ws(z))-tG(z).
We
noticethat thesametransfermatrixis obtained if onereplaces(F, G)by(0,Lv,).
A
compensatorin whichF
0 iscalledaprecompensator. IfG isstatic(no
dynamicsin the precompensatorpart) thenwe saythat (F,
G)
is pure (dynamic)feedback
and if, in addition,F
is staticthen (F,G)
is called a static statefeedback.
Our
objectiveistofind acompensator ofaspecified class(precompensator,pure dynamicfeedback,
staticfeedback)such that the resultingtransfermatrix
W,
is diagonal, in which casewe call the resulting system decoupled.
In
order to guarantee that each output can effectively be controlled, werequire in addition that the diagonal elements ofWF,
be nonzero.This is equivalenttorequiring
G
tobenonsingular.Sometimes onewants toimpose stronger conditions, such as theinvertibility (over)
ofG
(compareDatta
and
Hautus (1981)).
Finally, assuming that theoriginalsystem is internallystable,wetrytofind(F, G)such that the resulting system is internally stable. Such acompensator
will be called stabilitypreserving. Notice that we do not attempt to stabilize and to
decouple the system simultaneously. Rather, wetrytodecouple itwhilemaintaining
itsstability. If the system isnot stable atthe outset, it has to be stabilized first and afterwards one has to design the decoupling compensator.
It
will follow from the results of this paper, that one cannot destroy the existence of such a decouplingcompensator whenapplying thestabilizing feedback.
We
assume that q/=,
i.e., the numberofinput and output variables areequal.It
turns out that the problem of decoupling by precompensation or combined compensation (i.e., no restrictions onF, G)
is very simple even ifY
is an arbitrary integral domain.TI-IEOREM 3.5.
In
the situation described above, the following statements areequivalent.
(i) Thereexists a (stability preserving) decoupling combined compensator (F, G).
(ii) Thereexists a (stability preserving)decouplingprecompensator(0, G).
(iii) Wisnonsingular, i.e.,det Wis notidentieallyzero.
Proof.
(ii):ff
(i) is trivial.(i) :::), (iii). Accordingto(3.3)wehave
det
W
detLv,6 detWF,
a 0,(iii)
=:>
(ii). Let adj Wdenote the adjointof W (occurringinCram6r’srule).
ThenWadj W (det W)L
Choose a
Y
such that z-a @. Then, for sufficiently high k,(z-a)-k
adj W Gis causaland stable since theentriesof
W
arestable.Hence,
byW. G (z
a)-k
(det W)./,G is a stable decoupling compensator, which has an internally stable realization. If
G isinternallystable, then thetotal realization isinternallystable. [3
The condition for the existence of pure feedback decoupling compensators is moreinvolved. Toformulateit weneedsome notation.
We
write(3.6)
W(z)where wi(z) denotesthe ith rowof
W.
Let di(z) denoteaGCD over of the entries of wi(z). Such a GCD existsbecause is aUFD.
An
explicit construction of suchaGCD isgiven in Lemma 3.11. We can write
w
dw*
for suitablew*
withentries in.
Hence(3.7) W(z A(z)
W*(z
),where A(z)=diag(dl,...,d,,) and
,
W*
is the matrix consisting of the rowsW ,’’’,Wm.
Now..
we areinthe positiontoformulatethemain result ofthispaper.THEOREM 3.8. Let
,
be a reachable, internally stablesystem with respectto the denominatorset@. Then the followingstatementsareequivalent"(i) can be decoupled by a stability preserving static state
feedback
with Ginvertible
over.
(ii) Y_, can be decoupled by a stability preserving, stable dynamic state
feedback
withGinvertibleover (andF(z) stable).(iii) Y.,
can
be decoupledby astableprecompensatorL which is invertible over.
(iv)
W*,
as given in (3.7), isinvertibleover.
Theproofofthisresultwillbe given in 5.
in
the theorem it is assumedthat the gain matrix G isinvertible, although thisis not necessary in the original problem formulation: G nonsingular would do.
It
ispossible to generalize the theorem to this more general case, but the formulation
becomes more involved. There are two remarkable consequences of Theorem 3.8,
already noted for systems over fields in Hautus and
Heymann
(1980), (1983). Inthe firstplace, if decouplingis possible by dynamicstate feedback, itis also possible bystaticfeedback.
In
thesecondplace,the conditionfor theexistenceofadecouplingstatefeedbackdoesnotdependonthe realization, provided the realizationisreachable
(forsystems over fieldsthislatter restriction isnotnecessary).
The GCD’sused indefining
W*
are not unique andconsequentlyso is notW*
itself.
However,
the invertibility ofW*
isindependent of the particularchoice of theGCD’s,
as easily can be seen. The condition onW*
can be checked by computingw(z)
=detW*
and checking whether(w(z))-l
.
Whether this condition can beverified effectively depends on
.
In
the particular case that @ 0, the set of allextent,expandw(z) inpowersofz
-x,
W
(Z)
WO q-WIZ-1-[-W2Z-2’’[-"
and
(w(z))
-1o
iffw0is invertible overY.
Itisevennotnecessarytocomputew(z).
COROLLARY3.9.If
0
in Theorem 3.8 andif
W*
isexpandedasW*(z)
W*o
+
W’z
-
+...then the condition (iv)
of
Theorem 3.8 may bereplaced by"W*o
isinvertibleovergt.When restricted to the case that
Y
is a field, thisis exactly the condition given inFalb and Wolovich(1967).
Althoughthe particular choice of the GCD’s used in the definition of
W*
is ofno relevance for condition (iv) of Theorem 3.8, it will turn out that for the actual
construction of a decoupling feedback it is imperative that the
GCD’s
satisfy anadditional condition. To express this condition we use the notation (p,q
l[z])
for theGCD’s
of elements inY[z].
DEFINITION 3.10. Let wl,’.’,w,,. A GCD d of Wl,"’,w,, is called
admissible ifthere exist polynomials q and p such thatpd andqwi/pd, 1,...,m
arepolynomials, and(p,
qlY[z])
1.The existence and the construction of an admissible GCD is guaranteed by the following result.
LEMMA 3.11. Let wx,’’’,Wm and letq be a leastcommon denominator
of
Wl,..’,w,,.
Define
fii:=qwi and let v:=(,61,.."
,/,,lY[z]).
Finally, write pi:=ffi/V.Choose a
e
such thatz-ae
anddefine
lz:=min{degq-degpili=l,...,m}.
Then
d:=v
.(z-a)-*"
is anadmissible GCD
of
Wx, ",w,,.For a proof see 5. The lemma provides us with an actual construction of an
admissible
GCD,
atleast ifwehaveanalgorithmfor computingGCD’s
inY[z].
This is for instancethe case whenY
Rio,I,
seeBose(1976).
Now
wecanspecify Theorem 3.8"PROPOSITXON3.12.
If
theelementsdi
usedintheconstructionof
W*
areadmissibleGCD’s
of
withen thereexistsa staticfeedback
(F,G)
suchthatLv.
W*-I.
Foraproofsee 5.
Onceithasbeenproved thatagiven precompensatorL canbe implemented by
a(static) feedback (F, G), the actual computation ofF and G isstraightforward. Let
usstart from
L
-1,
ratherthanL
(recallthatL-1W*).
The relationL
-1Lv,-l
readsG-I(I
+FWs)=L
-1=Mo+MlZ
-1+...,
wherewehave expandedL-1
intopowers of z
-1.
Invertibility ofL-1over implies thatM0
isinvertible. SinceWs
is strictly causalwe musthave(3.13) G
=M1.
Then it follows that
GL-I=I=:
V(z)
is strictly causal and the mapF
has to be computed fromFW, (z V(z ).
Onsubstitution of
W
(zI-A)-IB
into this equation and expanding both sides asaseriesin z
-x,
one obtainsDATTA AND
wherewehave used theexpansion
V(z)
Vlz
-1+
V2z
-2+...
Because
ofreachabil-ity, the map
[B,
AB,...,A"-IB]
has a right inverse and henceF
can be solveduniquelyfrom
(3.14).
Notice thatthemaps
F
andG
are uniquelydeterminedbyL. In
particular,thestability-preservationproperty of
(F, G)
isautomatically guaranteed bytheinvertibilityof
W*
over.
4. Examples.
Example 4.1. Consider the system
’=
(A,B, C, D)
over[cr],
whereB=
C-or
2+cr+1
cr D=0.Suppose
thato.
It
is easily seen thatE
isreachable. The transfermatrixhas the following expansionW(z
CBz
-1+
CABz
-2 +.--[ltr
tr+ll
]
_1[
lz
+
2tr]
-2 cr+l20.2+0._
1 z 2o"+
2o.2+o._1
2o’2-o
"+1]
-3 20"3+2
z +" ’.Thesystemcanbe decoupled bystatefeedback since
W0*
[
+1]
isinvertible. Accordingto(3.13), wehaveFurthermore,
F
hastobe computed from(3.14),
whereV
isthe coefficient of z in theexpansionofW*o
-
W(z).
(NoticethatW*(z)=
z-1W(z).)
It
follows thatF[0
1 1tr-1]
[01
tr+l tr+lo’2-1]
1 tr tr 0
"2+
1 o’-1 o’-1 0"2-O"+2
-i-1
Equating thefirst twocolumns we obtain
The transfer function of theresultingsystem is
z-lL
Example 4.2. The purpose of this example is to show that the reachability
condition inTheorem 3.8 is essential.Let
N[o-],
N
No
The systemis notreachable,since
2
0
and hence allreachablevectorshavetobedivisiblebytr.Thetransfer functionis
Following theproceduredescribedatthe beginning ofthissectionweobtain
W*
toO" Z--Wo*
+...0 1 where
W’
=Iis invertible. We show that, nevertheless, decoupling by dynamic state feedback is impossible. Suppose that (F,G) is a dynamic state feedback decoupling the system. Then
W(I+FW)-IG=A
for some diagonal matrix A (recall thatW=
Ws).Equivalently,
(4.3) W
AG-I(I
+FW).Expanding thematrices
D, W, F
inpowersof z-1wehaveW
Wlz
-1+ W:zz
-z+..,
A
=Do+DlZ-l+D:z-2+...,
F
Fo
+Flz
-1+",
where2
Substitution into
(4.3)
yieldsWlz
-
+
W2z-2
+
(Do+Dlz-
+"")G-I(I
+
(Fo+’’
")(Wlz-
+"")).
The coefficients of Zo yields: D0=0, and of z
-"
W1
=D:G
-
hencetrG
=D.
Inparticular,
G
isdiagonal. Finally, equating the coefficients of z-
weobtainWz
DG-1Fo
W
+ D2G
-
o-2Fo
+ D2G
-1.
Thematrix
D2G
-1 isdiagonal and hence r2F0,2
tr. whichhasno solution in[r].
Furtherexamplescanbefound inDattaand Hautus
(1981).
5. Proofs.Proof of
Lemma 2.1. Itiswell known that[z]
isaUFD
if is amultiplicativeset
(see
Samuel (1969, Thm. 4, p.29)). We
define an isomorphism which mapsonto
l[z]
for some@1.
Theisomorphismisq"
r(z)->(z):=r
a+
definedon (z),where a is chosensuch that z-a
,
and1
:={z(z)lp
e,
ndegp}.
@1
is easily seen to be a multiplicative set in9[z].
The map q is invertible, andq r(z)
r((z- a)-
The homomorphism propertiesarereadilyverified.It
remains-1
36 K. 13. DATTA AND M. L. J. HAUTUS
Letr
p/q
.
Thenz"(z)
where n degq. Notice that because of the causality of r, the numerator
z"ff(z)
isapolynomial. Conversely,letr
l[z],
say p(z)r(z)
z 4(z)
for some q withdeg q n.Then
_,
(z._-__a.)"p(
1)
o
r(z)=
q(z) \z-a
since(z -a)-1 and hencep((z
-a)-),
and q and hence(z-a)"/q(z).
Since
l[z
is aUFD
and isomorphicto,
itfollows thatRemark. need not be adenominator set, in particular, the elements of neednotbemonic.
Remark. The isomorphism q is a standard device for transferring properties
known about quotient ringstorings of causalquotients (Eising(1980, 4.2),
Hautus
and
Sontag (1980)).
Proof
of
Lemma 3.11. Becauseof the assumption thatg
isaunique factorization domain, the ringg[z]
is alsoaunique factorizationdomain(see
Barshay(1969,
Thm.4.7)). Let
usdenote by theset of polynomials v of the form v uw, where u is aunit and w
.
If v is any polynomial in[z],
we can factorize it into primes v p Pro.As
aconsequence,wecandecomposev into vv/v
where v istheproductof theprimefactorsof v whichare in and v/
consists of theother factors.
Except
for unit factorsthisdecomposition is unique. Ifweinsiston aunique decomposi-tion we can achieve this by requiring thatv-
be monic.We
callv-
the-part
and+
v the
non-
part ofv.The following results follow easily from this definition.PROPOSITION5.1.
(i) (ab
)/
a/b
/,
(ab)-
a-b-(ii)p[q
(in[z ])
iff
p/[q/
andp-[q-.
+
(iii) (GCD (v,..., v,))
/=GCD
(v,...,
v,). (iv)If
p,
qlP
thenq.
In
view of thesaturation conditionimposedon,
property (iv)follows from the fact that divisors of elements of are monic upto aunit factor.Afterthispreparation we are inthe positiontoprove that d
:= v(z-a)-
is an admissibleGCD
ofw1,..,
w,, over.
In
the first place d is a divisor of w,..., w,because(recallthat wi=piv/q)
Wi
pit)(Z
a)U
pi(z-a)",
d qv q
since q
and/z
+degpi=<degq.Now
letdt
be also adivisor of wt,...,Win.We
havetoshowthat
d/d1
,
Let
dl
a/, wi/d =:
a
b/ci,withfl,c
We
noticethatbi
Wi
aidl
,
andhence
qob
i,Bci.
+b?
Taking the
non-
partweobtaina/?.
Thisshows thata/l/
ing[z].Observing that v is a GCD ofthe/i’s
and hence v+ is a GCD ofthe/’s,
we conclude thata
+Iv
+,
sayv+ ya+with yg [z ]. It
follows thatd
flv
/33,ve[z]
=a(z-a)
ot-(z-a)
since the denominator is in
.
It
remainsto beshown thatd/d1
is causal. Choosesuch thatdeg q degpi
+
Ix (recall the definition ofIx).
Thenwehave thatd d wi
dl
Widl
v q
(z-a)
iscausal,since ai is.
It
follows that dis aGCDofwl,., ’, w,,. Finally,weshow thatd is admissible.
We
choose p(z- a)
",
and q as already defined. Thenpd v is apolynomial and pd and q are coprime.
In
addition, qwi/pd-pi is alsoapolynomial.Thiscompletestheproof.
Proof of
Theorem 3.8 (andProposition 3.12).(i) =),(ii)is evident.
(ii)
:=>
(iii).If(F, G)
is astability preserving,stable dynamicstatefeedback and ifG is invertible, then, according to(3.3), W isdecoupledalsobytheprecompensator
LF,
defined by(3.4). It
remains to be shown that the entries ofLF.
are in and thatLF,
is invertible over.
It
is easily seen thatLv.
is invertible as a rational--1
matrix and that
Lr,
G-I(I
+FWs)
has entries in.
So,
only the stability of itself has tobe shown.By
assumption, the resulting system is internally stable. This implies that thematrixV
WsLF,
is stable.Sinceitfollows that
(I
+
FWs)-I
I
-FWs
(I
+
FWs)-1LF.C
G-FV
isstable.
(iii)::> (iv).
Suppose
that for some nonsingular diagonal matrixE
diag(el,’’’,e,,) we have W
EL,
where L is amatrix invertible over.
Thefirst row ofthismatrixequation reads[w11,
wlm]=el[111,
11,,,],
which implies thatel is a divisor ofwl
=[w11,
",wl,,].
Sincedi
isaGCD
of wl itfollows that el divides dl, i.e.,
dl
hie1
for somehi
s.
Similar results hold for theotherrows,sothat we canwrite
A
EH
whereH
diag(h1, ",h,).It
follows thatEL
WA
W*
EHW*
andhenceL
HW*.
ThereforeW
*-1HL
-1iscausal and stable.(The nonsingularity
ofthe matricesinvolvedisobvious).
(iv)
::>
(i).IfW*
isinvertiblewedefineL
W*-1,
andformula(3.7)
readsAND M. L. J. HAUTUS
Hence,
the system is decoupled bythe precompensatorL which is amatrix over,
hencecausal and stable.We
havetoshow thatF
andG exist such thatL
LF,
(see(3.4)). To
this extent, we formulate a generalization to systems over of a result given for systemsover a fieldinHautusandHeymann (1978).
The result inquestionis:
THEOREM 5.2.
Let
(A,B, C, D)
denote a reachable system andL be abicausal isomorphism (i.e., causal and with a causal inverse). Then there exists a static statefeedback
compensator (F,G), with Ginvertible, andLLI,
iff
for
each polynomialu
"[z]
wehave:If
Wsu
ispolynomialthenL-lu
ispolynomial.Theproof of thisresultis completely analogoustotheproofin the fieldcase,so it will not be repeated here.
Contrary
tothe field case, however, the reachability of the system is essential. (A counterexample in the nonreachable case canbe deduced from Example4.2.) In
order to apply the theorem to ourL,
wehave toprove thatL-lu
W*u
is polynomialwhenever u andWsu
arepolynomial.We
show that the following stronger statement: u andWu
are polynomial impliesW*u
ispolynomial holds,providedW*
isconstructed via admissibleGCD’s.
(Thiswill alsoproveProposi-tion
3.12.)
Letus assume thatWu
ispolynomial.Thefirst entry of thisvector iswlu,whichis apolynomial. Let
dl
bean admissibleGCD of w, and letw’
d-(lw.
We
have to show that
W*lU
is polynomial. According to Definition 3.10, there existpolynomialsp and q such thata := pdand v
/a,
where v :=qw arepolynomials and(a,
ql[z])
1. Since wlu=q-Xvxu
ispolynomialwe haveqlv u.
Also,alvxu,
sinceu and v
x/a
are polynomial.Hence
qalvu,
q and a being coprime. Butthis means thatw*u=d-Xwxu
(qa)-Xvxup
is polynomial. The same argument applies to eachrow. This shows thatLcanbe realized bystatefeedback.
It
remainstobe shown that the resulting systemisinternallystable. SinceW,v.,
the i/s-mapwhich results after feedback is applied, isequaltoWLF,
oand hencestable,the desired resultfollowsfrom:LEMMA
5.3. LetW,
(zI-A)-IB
beareachable i/s-map. ThenW
isstableiff
det
(zI-A).
Proof.
Since(zI-A)-l=adj
(zI-A)/det
(zI-A), the "if" part is obvious. Toprove the "only-if" partwenotethat because of the reachability of(A, B),there exist polynomial matrices
P(z)
andQ(z)such that(zI
-A )P(z
+
BO(z
I
(see
KhargonekarandSontag
(1981, Lemma3.2)). It
follows that(zI-A)-=P(z)+
W(z)Q(z)is stable whenever
Ws(z)
is.But
then also det(zI-A)-a=
1/det
(zI-A)
is stable,and hence det
(zI-A)
(since issaturated).
Acknowledgment.
K.
B.Datta
isgratefultotheEindhovenUniversity of Tech-nology,Department
of Mathematics for offeringhim a Research Fellowship duringthe tenure of which he workedontheproblemreported here.
REFERENCES
J.BARSHAY(1969),TopicsinRingTheory,W. A.Benjamin,NewYork.
N. K. BOSE (1976),An algorithmfor GCFextractionfrom two multivariable polynomials,Proc. IEEE, pp.185-186.
K. I. DATTA ANDM. L. J. HAUTUS (1981),Decoupllngofsystemsoverauniquefactorizationdomain,
inProc.International Symposiumon MathematicalTheory otNetworks andSystems, N. Levan, ed., Santa Monica,CA,pp. 35-39.
F.EISING(1980),2-D systems, analgebraicapproach, Ph.D. Dissertation,Dept.ofMathematics,Univ. ofTechnology,Eindhoven.
P. L.FALBANDW.A. WOLOVICH (1967),Decouplinginthe design and synthesisofmultivariable control
systems,IEEETrans. Aut. Contr., AC-12,pp. 651-659.
J.HALE (1977),TheoryofFunctionalDifferentialEquations, Springer,NewYork.
M. L. J.HAUTUSANDM. HEYMANN (1978),Linearfeedback--analgebraic approach,thisJournal,.16, pp.83-105.
(1980), New results in linear feedback decoupling, in Analysis and Optimization of Systems, A. Bensoussan andJ. L. Lions, eds., Lecture Notes inControl and Information Sciences 28, Springer,NewYork, pp. 562-577.
(1983),Linearfeedbackdecoupling-transferfunctionanalysis,IEEE Trans. Aut.Control,toappear. M. L. J. HAUTUSANDE. D.SONTAG (1981),Anapproach todetectability and observers, in Algebraic and Geometric Methods i.n Linear System Theory, C. Byrnes and C. Martin, eds., Reidel, Dordrecht.
E. W. KAMEN(1978),LecturesonAlgebraicSystemTheory:Linear systems over rings,NASAContractor Report3016.
(1980), Onthe relationship betweenzero criteriafortwo-variable polynomials and asymptotic stability ofdelaydifferentialequations,IEEETrans. Aut.Control,AC-25, pp. 983-984.
P.P.KHARGONEKARANDE. D.SONTAG(1981),On therelationbetween stablematrixfraction factoriz-ationsand regulable realizationsoflinear systems over rings,IEEE Trans. Aut.Control,AC-27, pp. 627-638.
B. S. MORGAN,JR. (1964), The synthesisoflinearmultivariable systemsbystate-variablefeedback,IEEE Trans. Aut.Control, AC-9, pp.405-411.
A.S. MORSE(1976), Systeminvariantsunderfeedbackandcascadecontrol,inMathematicalSystemsTheory, G.MarchesiniandS.K.Mitter,eds.,Lecture Notesin Economics and MathematicalSystems,131, Springer,Berlin,pp. 61-74.
Y. ROUCHALEAU(1972),Linear,discrete-time,finite-dimensionaldynamicalsystems over someclassesof
commutative rings,Ph.D. Dissertation, Stanford Univ., Stanford,CA.
Y.ROUCHALEAU, B. F. WYMANANDR. E. KALMAN(1972), Algebraicstructureoflineardynamical
systems,III.Realization theoryover a commutative ring,Proc. Nat.Acad.Sci.,69, pp. 3404-3406. P. SAMUEL (1963), Anneaux factoriels, Redaction de ArtibanoMicali,SociedadedeMatemiticadeSo
Paulo.
E. D.SONTAG (1976),Linear systems over commutative rings: a survey, Ricerche diAutomatica,7, pp. 1-34. (1981),Linear systems over commutativerings:a(partial)updated survey,Proc.8th WorldCongress