Schur Class Operator Functions and Automorphisms of Hardy Algebras
Paul S. Muhly
1and Baruch Solel
2Received: June 12, 2007
Communicated by Joachim Cuntz
Abstract.
LetE
beaW ∗
-orrespondeneoveravonNeumannalge- braM
andletH ∞ (E)
betheassoiatedHardyalgebra.Ifσ
isafaith-ful normalrepresentationof
M
onaHilbert spaeH
, thenonemayform thedual orrespondene
E σ
andrepresentelementsinH ∞ (E)
as
B(H )
-valuedfuntionsontheunitballD (E σ ) ∗
. ThefuntionsthatoneobtainsarealledShurlassfuntionsandmaybeharaterized
in terms of ertain Pik-like kernels. We study these funtions and
relate them to systemmatries and transfer funtions from systems
theory. Weusetheinformationgainedtodesribetheautomorphism
groupof
H ∞ (E)
intermsofspeialMöbiustransformationsonD (E σ )
.Partiularattentionisdevotedtothe
H ∞
-algebrasthatareassoiatedtographs.
2000 Mathematis Subjet Classiation: 46E22, 46E50, 46G20,
46H15,46H25,46K50,46L08,46L89,
Keywords and Phrases: Hardy Algebras, Tensor Algebras, Shur
lassfuntions,
W ∗
-orrespondene,nonommutativerealizationthe- ory,Möbiustransformations,freesemigroupalgebras,graphalgebras,Nevanlinna-Pikinterpolation
1
SupportedinpartbygrantsfromtheNationalSieneFoundationand fromthe U.S.-
IsraelBinationalSieneFoundation.
2
SupportedinpartbytheU.S.-IsraelBinationalSieneFoundationandbytheFundfor
thePromotionofResearhattheTehnion.
1 Introduction
Let
M
beaW ∗
-algebraandletE
beaW ∗
-orrespondeneoverM
. In[31℄webuiltanoperatoralgebrafromthisdatathatwealledtheHardyalgebraof
E
and whih wedenoted
H ∞ (E)
. IfM = E = C
-the omplexnumbers,thenH ∞ (E)
isthelassialHardyalgebraonsistingof allbounded analytifun-tionsontheopen unit dis,
D
(seeExample 2.4below.) IfM = C
again,butE = C n
, thenH ∞ (E)
is thefree semigroupalgebraL n
studied by DavidsonandPitts[17℄,Popesu[32℄andothers(seeExample2.5.) Oneoftheprinipal
disoveriesmadein [31℄,andthesoureofinspirationforthepresentpaper,is
thatattahedtoeahfaithfulnormalrepresentation
σ
ofM
thereisadualor-respondene
E σ
, whihisaW ∗
-orrespondeneovertheommutantofσ(M )
,σ(M ) ′
, and theelementsofH ∞ (E)
denefuntions onthe openunit ballofE σ
,D (E σ )
. Further,thevaluedistributiontheoryofthesefuntionsturnsout to be linked through our generalization of the Nevanlinna-Pik interpolationtheorem [31, Theorem 5.3℄ with the positivity properties of ertain Pik-like
kernelsofmappings betweenoperator spaes.
Inthesettingwhere
M = E = C
andσ
isthe1
-dimensional representationofC
onitself,thenE σ
isC
again. TherepresentationofH ∞ (E)
intermsoffun-tionson
D (E σ ) = D
isjusttheusualwaywethink ofH ∞ (E)
. Inthissetting,our Nevanlinna-Pik theorem is exatlythe lassialtheorem. If, however,
σ
is arepresentationof
C
on aHilbert spaeH
,dim(H ) > 1
, thenE σ
maybeidentiedwith
B(H)
andthenD (E σ )
beomesthespaeofstritontrationson
H
, i.e., all those operators of norm stritly less than1
. In this ase, thevalueof an
f ∈ H ∞ (E)
at aT ∈ D (E σ )
is simplyf (T )
,dened throughtheusual holomorphi funtional alulus. OurNevanlinna-Piktheorem givesa
solutiontoproblems suh asthis: given
k
operatorsT 1 , T 2 , . . . , T k
allofnormlessthan
1
andk
operators,A 1 , A 2 , . . . , A k
,determinetheirumstanesunderwhihoneanndaboundedanalytifuntion
f
ontheopenunitdisofsupnorm at most
1
suh thatf (T i ) = A i
,i = 1, 2, . . . , k
(See [31, Theorem6.1℄.)On the other hand, when
M = C
,E = C n
, andσ
is one dimensional, the spaeE σ
isC n
andD (E σ )
is the unit ballB n
. Elements inH ∞ (E) = L n
arerealizedasholomorphifuntionson
B n
thatlieinamultiplierspaestud- iedin detailbyArveson[5℄. Moreaurately,thefuntionalrepresentationofH ∞ (E) = L n
in termsofthesefuntions expressesthisspaeasaquotient ofH ∞ (E) = L n
. The Nevanlinna-Pik theoremof [31℄ontainsthoseof David- son andPitts [18℄, Popesu[34℄, and Ariasand Popesu [4℄, whih dealwithinterpolationproblemsforthesespaesoffuntions(possiblytensoredwiththe
boundedoperatorsonanauxiliaryHilbertspae). Italsoontainssomeofthe
results of Constaninesu and Johnson in [16℄ whih treatselementsof
L n
asfuntionsontheballofstritrowontrationswithvaluesintheoperatorson
a Hilbert spae. (See their Theorem 3.4 in partiular.) This situation arises
when onetakes
M = C
andE = C n
, but takesσ
to besalar multipliation onanauxiliaryHilbert spae.Ourobjetiveinthepresentnoteisbasiallytwofold. First,wewishtoidentify
those funtionson
D (E σ )
that arisefromevaluatingelementsofH ∞ (E)
. Forthis purpose, weintrodueafamilyof funtions on
D (E σ )
that weallShurlassoperatorfuntions(seeDenition3.1). Roughlyspeaking,thesefuntions
aredened sothat aPik-likekernelthatonemayattahtoeahone isom-
pletely positive denite in the sense of Barreto, Bhat, Liebsher and Skeide
[14℄. In Theorem 3.3 weuse their Theorem 3.2.3 to give aKolmogorov-type
representation of the kernel, from whih we derive an analogueof a unitary
systemmatrix
A B C D
whosetransferfuntion
A + B(I − L ∗ η D) −1 L ∗ η C
turns out to be the given Shur lass operator funtion. We then prove in
Theorem 3.6that eahsuhtransferfuntion arises byevaluatinganelement
in
H ∞ (E)
at pointsofD (E σ )
andonversely, eah funtion inH ∞ (E)
hasarepresentationintermsofatransferfuntion. Themeaningofthenotationwill
bemadepreisebelow,but weuseitheretohighlighttheonnetionbetween
our analysis and realization theory as it omes from mathematial systems
theory. Thepointto keepinmindis thatfuntionson
D (E σ )
thatomefromelementsof
H ∞ (E)
arenot, apriori, analytiin anyordinarysenseand itisnot at alllear what analyti features theyhave. Our Theorems3.1 and3.6
together with[31, Theorem 5.3℄ showthat theShur lassoperator funtions
are preisely the funtions one obtainswhen evaluating funtions in
H ∞ (E)
(ofnormatmost
1
)atpointsofD (E σ )
. Thefatthateahsuhfuntionmayberealizedasatransferfuntion exhibitsasurprisinglevelofanalytiitythat
isnotevidentin thedenition of
H ∞ (E)
.Ourseondobjetiveisto onnettheusualholomorphipropertiesof
D (E σ )
with the automorphisms of
H ∞ (E)
. As aspae,D (E σ )
is theunit ball of aJ ∗
-triplesystem. Consequently,everyholomorphiautomorphismofD (E σ )
istheomposition ofaMöbiustransformation andalinearisometry [20℄. Eah
ofthese implementsanautomorphismofthealgebraofallbounded, omplex-
valued analyti funtions on
D (E σ )
, but in our setting only ertain of themimplement automorphisms of
H ∞ (E)
- those for whih the Möbius part isdeterminedbyaentralelementof
E σ
(seeTheorem4.21).Ourproofrequiresthefatthattheevaluationoffuntionsin
H ∞ (E)
(ofnormatmost1
)atpointsof
D (E σ )
arepreiselytheShurlassoperatorfuntionsonD (E σ )
. Indeed,thewhole analysisisanintriate point-ounterpoint interplayamongelements
of
H ∞ (E)
, Shur lass funtions, transfer funtions and lassial funtiontheory on
D (E σ )
. In the last setion, we apply our general analysis of theautomorphisms of
H ∞ (E)
to the speial ase ofH ∞
-algebras oming fromdireted graphs.
Inonludingthisintrodution,wewanttonotethatapreprintofthepresent
paperwasposted on thearXiv on June 27, 2006. Reently, inspired in part
by ourpreprint, Ball,Biswas,Fang andter Horst [8℄ wereable to realizethe
Fokspaethat wedesribehereintermsofthetheoryofompletelypositive
denitekernelsadvaned byBarreto,Bhat,LiebsherandSkeide[14℄thatwe
also use (See Setion 3 and, in partiular, the proof of Theorem 3.3.) The
analysis of Ball et al. makes additional ties between the theory of abstrat
Hardyalgebrasthat wedevelophereandlassialfuntiontheoryontheunit
dis.
2 Preliminaries
Westartbyintroduingthebasidenitionsandonstrutions. Weshallfollow
Lane[24℄forthegeneraltheoryofHilbert
C ∗
-modulesthatweshalluse. LetA
beaC ∗
-algebraandE
bearightmoduleoverA
endowedwithabi-additive maph·, ·i : E × E → A
(referred to asanA
-valuedinner produt) suh that,for
ξ, η ∈ E
anda ∈ A
,hξ, ηai = hξ, ηia
,hξ, ηi ∗ = hη, ξi
, andhξ, ξi ≥ 0
,withhξ, ξi = 0
only whenξ = 0
. Also,E
is assumed to be ompletein thenormkξk := khξ, ξik 1/2
. We writeL(E)
for the spae of ontinuous, adjointable,A
-modulemapsonE
. ItisknowntobeaC ∗
-algebra. IfM
isavonNeumannalgebraandif
E
isaHilbertC ∗
-moduleoverM
,thenE
issaidtobeself-dualinaseeveryontinuous
M
-modulemapfromE
toM
isgivenbyaninnerprodutwithanelementof
E
. LetA
andB
beC ∗
-algebras. AC ∗
-orrespondenefromA
toB
is aHilbertC ∗
-moduleE
overB
endowed witha struture of aleftmoduleover
A
viaanondegenerate∗
-homomorphismϕ : A → L(E)
.Whendealingwithaspei
C ∗
-orrespondene,E
,fromaC ∗
-algebraA
toaC ∗
-algebraB
, itwill be onvenientsometimes to suppress theϕ
in formulasinvolvingtheleft ation and simplywrite
aξ
ora · ξ
forϕ(a)ξ
. This shouldausenoonfusioninontext.
If
E
is aC ∗
-orrespondene fromA
toB
and ifF
is a orrespondene fromB
toC
, then thebalanedtensor produt,E ⊗ B F
is anA, C
-bimodule thatarriestheinner produtdenedbytheformula
hξ 1 ⊗ η 1 , ξ 2 ⊗ η 2 i E⊗ B F := hη 1 , ϕ(hξ 1 , ξ 2 i E )η 2 i F
TheHausdorompletion ofthis bimodule isagaindenotedby
E ⊗ B F
.InthispaperwedealmostlywithorrespondenesovervonNeumannalgebras
that satisfy some natural additional properties as indiated in the following
denition. (Forexamplesand moredetailssee[31℄).
Definition 2.1
LetM
andN
bevonNeumannalgebrasandletE
beaHilbertC ∗
-moduleoverN
. ThenE
isalleda HilbertW ∗
-moduleoverN
inaseE
isself-dual. Themodule
E
isalledaW ∗
-orrespondenefromM
toN
inaseE
isaself-dual
C ∗
-orrespondenefromM
toN
suhthatthe∗
-homomorphismϕ : M → L(E)
, giving the left module struture onE
, is normal. IfM = N
weshallsay that
E
isaW ∗
-orrespondene overM
.Wenote thatif
E
isaHilbertW ∗
-moduleoveravonNeumannalgebra,thenL(E)
isnotonlyaC ∗
-algebra,butis alsoaW ∗
-algebra. Thusitmakessenseto talkaboutnormalhomomorphismsinto
L(E)
.Definition 2.2
An isomorphism of aW ∗
-orrespondeneE 1
overM 1
anda
W ∗
-orrespondeneE 2
overM 2
is a pair(σ, Ψ)
whereσ : M 1 → M 2
isan isomorphism of von Neumann algebras,
Ψ : E 1 → E 2
is a vetor spaeisomorphism preserving the
σ
-topology and fore, f ∈ E 1
anda, b ∈ M 1
, wehave
Ψ(aeb) = σ(a)Ψ(e)σ(b)
andhΨ(e), Ψ(f )i = σ(he, f i)
.When onsidering the tensor produt
E ⊗ M F
of twoW ∗
-orrespondenes, one needs to takethe losure of theC ∗
-tensor produt in theσ
-topologyof[6℄ in order to get a
W ∗
-orrespondene. However, we will not distinguish notationallybetweentheC ∗
-tensorprodutandtheW ∗
-tensorprodut. Notealsothatgivena
W ∗
-orrespondeneE
overM
andaHilbertspaeH
equippedwith anormalrepresentation
σ
ofM
,weanform theHilbert spaeE ⊗ σ H
by dening
hξ 1 ⊗ h 1 , ξ 2 ⊗ h 2 i = hh 1 , σ(hξ 1 , ξ 2 i)h 2 i
. Thus,H
is viewed as aorrespondenefrom
M
toC
viaσ
andE ⊗ σ H
isjust thetensor produtofE
andH
asW ∗
-orrespondenes.Note alsothat,given anoperator
X ∈ L(E)
andan operatorS ∈ σ(M ) ′
, themap
ξ ⊗ h 7→ Xξ ⊗ Sh
denes a bounded operator onE ⊗ σ H
denoted byX ⊗ S
. TherepresentationofL(E)
that resultswhenone letsS = I
,isalledthe representation of
L(E)
indued byσ
and is often denoted byσ E
. Theomposition,
σ E ◦ ϕ
isarepresentationofM
whihweshallalsosayisinduedby
σ
,but weshallusually denoteitbyϕ(·) ⊗ I
.Observe that if
E
is aW ∗
-orrespondene over a von Neumann algebraM
,then we may form the tensor powers
E ⊗n
,n ≥ 0
, whereE ⊗0
is simplyM
viewed as the identity orrespondene over
M
, and we may form theW ∗
-diret sum ofthe tensor powers,
F (E) := E ⊗0 ⊕ E ⊗1 ⊕ E ⊗2 ⊕ · · ·
to obtaina
W ∗
-orrespondeneoverM
alledthe(full)Fokspae overE
. Theationsof
M
onthe left and right ofF(E)
are the diagonal ations and, when it isonvenient to do so, we make expliit the left ation by writing
ϕ ∞
for it.That is,for
a ∈ M
,ϕ ∞ (a) := diag{a, ϕ(a), ϕ (2) (a), ϕ (3) (a), · · · }
, wherefor alln
,ϕ (n) (a)(ξ 1 ⊗ ξ 2 ⊗ · · · ξ n ) = (ϕ(a)ξ 1 ) ⊗ ξ 2 ⊗ · · · ξ n
,ξ 1 ⊗ ξ 2 ⊗ · · · ξ n ∈ E ⊗n
.The tensor algebra over
E
, denotedT + (E)
, is dened to be thenorm-losedsubalgebra of
L(F(E))
generated byϕ ∞ (M )
and the reation operatorsT ξ
,ξ ∈ E
,dened bytheformulaT ξ η = ξ ⊗ η
,η ∈ F (E)
. Wereferthereaderto[28℄forthebasifats about
T + (E)
.Definition 2.3
([31 ℄)GivenaW ∗
-orrespondeneE
overthe von Neumannalgebra
M
,theultraweaklosureofthetensoralgebraofE
,T + (E)
,inL(F(E))
,isalledthe HardyAlgebraof
E
,andisdenotedH ∞ (E)
.Example 2.4
IfM = E = C
, thenF(E)
an be identied withℓ 2 ( Z + )
or,through the Fouriertransform,
H 2 ( T )
. Thetensor algebrathen isisomorphito the dis algebra
A( D )
viewed asmultipliation operatorsonH 2 ( T )
andtheHardy algebraisrealizedasthe lassial Hardyalgebra
H ∞ ( T )
.Example 2.5
IfM = C
andE = C n
, thenF(E)
an be identied with thespae
l 2 ( F + n )
, whereF + n
is the free semigroup onn
generators. The tensoralgebrathen iswhat Popesu refersto as the non ommutative dis algebra
A n
andtheHardyalgebraisitsw ∗
-losure. ItwasstudiedbyPopesu[32 ℄andby DavidsonandPittswho denoteditby
L n
[17 ℄.Weneedtoreviewsomebasifatsabouttherepresentationtheoryof
H ∞ (E)
andof
T + (E)
. See[28,31℄formoredetails.Definition 2.6
LetE
beaW ∗
-orrespondeneover avon NeumannalgebraM
. Then:1. Aompletelyontrativeovariantrepresentationof
E
onaHilbertspaeH
isapair(T, σ)
,where(a)
σ
isanormal∗
-representationofM
inB(H)
.(b)
T
is a linear, ompletely ontrative map fromE
toB(H )
that isontinuousinthe
σ
-topologyof [6 ℄onE
andthe ultraweaktopologyon
B(H).
()
T
is a bimodule map in the sense thatT (SξR) = σ(S)T (ξ)σ(R)
,ξ ∈ E
,andS, R ∈ M
.2. Aompletelyontrativeovariantrepresentation
(T, σ)
ofE
inB(H )
isalled isometri inase
T (ξ) ∗ T (η) = σ(hξ, ηi)
(1)forall
ξ, η ∈ E
.Itshould benotedthat theoperatorspaestrutureon
E
towhihDenition2.6 refers is that whih
E
inherits when viewed as a subspae of its linkingalgebra.
As weshowedin [28, Lemmas3.43.6℄andin [31℄, ifaompletely ontrative
ovariant representation,
(T, σ)
, ofE
inB(H )
is given, then it determines aontration
T ˜ : E ⊗ σ H → H
dened by the formulaT(η ˜ ⊗ h) := T (η)h
,η ⊗ h ∈ E ⊗ σ H
. TheoperatorT ˜
intertwines therepresentationσ
onH
andtheinduedrepresentation
σ E ◦ ϕ = ϕ(·) ⊗ I H
onE ⊗ σ H
;i.e.T ˜ (ϕ(·) ⊗ I) = σ(·) ˜ T .
(2)Infatwehavethefollowinglemmafrom[31,Lemma 2.16℄.
Lemma 2.7
Themap(T, σ) → T ˜
isabijetion betweenallompletelyontra-tiveovariantrepresentations
(T, σ)
ofE
ontheHilbertspaeH
andontrativeoperators
T ˜ : E ⊗ σ H → H
thatsatisfyequation(2). GivensuhaT ˜
satisfyingthis equation,
T
,denedby the formulaT(ξ)h := ˜ T (ξ ⊗ h)
, together withσ
isa ompletely ontrative ovariant representation of
E
onH
. Further,(T, σ)
isisometri ifandonly if
T ˜
isan isometry.The importane of theompletely ontrative ovariantrepresentationsof
E
(or, equivalently, theintertwining ontrations
T ˜
asabove)is that theyyieldallompletelyontrativerepresentationsofthetensoralgebra. Morepreisely,
wehavethefollowing.
Theorem 2.8
LetE
beaW ∗
-orrespondeneoveravonNeumannalgebraM
.Toevery ompletely ontrative ovariantrepresentation,
(T, σ)
,ofE
there isa unique ompletely ontrative representation
ρ
of the tensor algebraT + (E)
that satises
ρ(T ξ ) = T (ξ) ξ ∈ E
and
ρ(ϕ ∞ (a)) = σ(a) a ∈ M.
The map
(T, σ) 7→ ρ
isabijetion between theset ofall ompletelyontrativeovariant representations of
E
and all ompletely ontrative (algebra) repre-sentationsof
T + (E)
whose restritions toϕ ∞ (M )
are ontinuouswith respettothe ultraweak topologyon
L(F(E))
.Definition 2.9
If(T, σ)
is aompletely ontrativeovariant representation of aW ∗
-orrespondeneE
overavonNeumannalgebraM
,weall the repre-sentation
ρ
ofT + (E)
desribed in Theorem 2.8 the integrated form of(T, σ)
andwrite
ρ = σ × T
.Remark 2.10
Oneoftheprinipaldiulties onefaesindealingwithT + (E)
and
H ∞ (E)
istodeidewhenthe integratedform,σ × T
,ofaompletelyon-trative ovariant representation
(T, σ)
extends fromT + (E)
toH ∞ (E)
. Thisproblemarises alreadyinthesimplestsituation,vis. when
M = C = E
. Inthissetting,
T
is given by asingle ontration operator on aHilbert spae,T + (E)
is the dis algebra and
H ∞ (E)
is the spae of boundedanalyti funtionsonthe dis. Therepresentation
σ × T
extendsfromthe disalgebratoH ∞ (E)
preisely whenthere isnosingularparttothe spetral measureofthe minimal
unitary dilation of
T
. We arenot aware of aomparableresult inourgeneralontext but we have some suient onditions. One of them is given in the
following lemma. Itisnotaneessaryonditioningeneral.
Lemma 2.11
[31 , Corollary 2.14℄ Ifk Tk ˜ < 1
thenσ × T
extends to a ultra-weakly ontinuousrepresentation of
H ∞ (E)
.In[31℄weintroduedandstudiedtheoneptsofdualityandofpointevaluation
(forelementsof
H ∞ (E)
). These playaentralroleinouranalysishere.Definition 2.12
LetE
beaW ∗
-orrespondeneoveravonNeumannalgebraM
andletσ : M → B(H )
beafaithfulnormalrepresentationofM
onaHilbertspae
H
. Then theσ
-dual ofE
,denotedE σ
,isdenedtobe{η ∈ B(H, E ⊗ σ H ) | ησ(a) = (ϕ(a) ⊗ I)η, a ∈ M }.
An important feature of the dual
E σ
is that it is aW ∗
-orrespondene, but overthe ommutant ofσ(M )
,σ(M ) ′
.Proposition 2.13
With respettothe ationofσ(M ) ′
andtheσ(M ) ′
-valuedinnerprodutdenedasfollows,
E σ
beomesaW ∗
-orrespondeneoverσ(M ) ′
:For
Y
andX
inσ(M ) ′
,andη ∈ E σ
,X ·η·Y := (I⊗X )ηY
,andforη 1 , η 2 ∈ E σ
,hη 1 , η 2 i σ(M) ′ := η 1 ∗ η 2
.Inthefollowingremarkweexplainwhatwemeanbyevaluatinganelementof
H ∞ (E)
atapointintheopenunit ballofthedual.Remark 2.14
The importaneof this dual spae,E σ
,is that itislosely re-latedtotherepresentationsof
E
. Infat, theoperatorsinE σ
whosenormdoesnot exeed
1
arepreisely the adjoints of the operators ofthe formT ˜
for ao-variant pair
(T, σ)
. In partiular, everyη
in the openunitball ofE σ
(writtenD (E σ )
) gives rise to a ovariant pair(T, σ)
(withη = ˜ T ∗
) suh thatσ × T
extendstoarepresentation of
H ∞ (E)
.Given
X ∈ H ∞ (E)
wean applythe representation assoiatedtoη
toit. Theresultingoperator in
B(H)
will bedenotedbyX b (η ∗ )
. ThusX(η b ∗ ) = (σ × η ∗ )(X ).
In this way, we view every element in the Hardy algebra as a
B(H )
-valuedfuntion
X b : D (E σ ) ∗ → B(H)
onthe open unitballof
(E σ ) ∗
. Oneof ourprimaryobjetivesistounderstandthe rangeofthe transform
X → X b
,X ∈ H ∞ (E)
.Example 2.15
SupposeM = E = C
andσ
the representation ofC
on some HilbertspaeH
. ThenitiseasytohekthatE σ
isisomorphitoB(H)
. FixanX ∈ H ∞ (E)
. Aswementionedabove,thisHardyalgebraisthelassialH ∞ ( T )
andweanidentify
X
withafuntionf ∈ H ∞ ( T )
. GivenS ∈ D (E σ ) = B(H )
,it is not hard to hek that
X(S b ∗ )
, as dened above, is the operatorf (S ∗ )
denedthroughthe usualholomorphi funtionalalulus.
Example 2.16
In [17 ℄ Davidson and Pitts assoiate toevery element of thefreesemigroupalgebra
L n
(seeExample2.5)afuntionontheopenunitballofC n
. ThisisaspeialaseofouranalysiswhenM = C
,E = C n
andσ
isaonedimensionalrepresentationof
C
. Inthis aseσ(M ) ′ = C
andE σ = C n
. Note, however, thatourdenitionallowsustotakeσ
tobetherepresentationofC
on an arbitraryHilbert spaeH
. If we doso, thenE σ
isisomorphi toB(H ) (n)
,the ntholumn spae over
B(H )
, andelements ofL n
dene funtionson theopenunit ballof this spaeviewed asaorrespondene over
B(H)
with valuesin
B(H )
. This is the perspetive adopted by Constantinesu and Johnson in [16 ℄. In the analysis of [17℄ it is possible that a non zero element ofL n
willgive rise tothe zero funtion. Weshall show in Lemma 3.8 that, byhoosing
an appropriate
H
wean insure thatthis does nothappen.Example 2.17
Partof the reentwork ofPopesuin [35℄maybe astin ourframework. We will follow his notation. Fix aHilbert spae
K
, and letE
bethe olumn spae
B(K) n
. Take, also, aHilbert spaeH
and letσ : B(K) → B(K ⊗ H)
be the representation whih sendsa ∈ B(K)
toa ⊗ I H
. Then,sinetheommutantof
σ(B(K))
isnaturallyisomorphitoB(H )
,itiseasytosee that
E σ
isthe olumnspae overB(H )
,B(H) n
. It follows thatD (E σ )
isthe openunit ballin
B(H) n
. Afreeformalpowerseries withoeientsfromB(K)
is aformal seriesF = P
α∈ F + n A α ⊗ Z α
whereF +
n
isthe freesemigroupon
n
generators,theA α
areelements ofB(K)
andwhereZ α
is themonomialin nonommuting indeterminates
Z 1
,Z 2
,...,Z n
determined byα
. IfF
hasradiusofonvergeneequalto
1
,thenonemay evaluateF
atpointsofD (E σ ) ∗
togetafuntionon
D (E σ ) ∗
withvaluesinB(K⊗H)
,vis.,F ((S 1 , S 2 , · · · S n )) = P
α∈ F + n A α ⊗ S α
. See[35 , Theorem 1.1℄. Infat, under additional restritions ontheoeientsA α
,F
maybeviewedasafuntionX
inH ∞ (B(K) n )
insuhawaythat
F ((S 1 , S 2 , · · · S n )) = X(S b 1 , S 2 , · · · S n )
inthesensedenedin[31 ,p.384℄ and disussed above in Remark 2.14. The spae that Popesu denotes by
H ∞ (B(X ) n 1 )
arises whenK = C
,and isnaturallyisometrially isomorphi toL n
[35 ,Theorem3.1℄. Wenotedinthe preeding examplethatL n
isH ∞ ( C n )
.The point of [35 ℄, at least in part, is to study
H ∞ (B(X ) n 1 ) ≃ L n = H ∞ ( C n )
through all the representations
σ
ofC
on Hilbert spaesH
, that is, throughevaluating funtions in
H ∞ (B(X ) n 1 )
at points the unit ball ofB(H ) n
for allpossible
H
's. The spaeB(K) n
isMoritaequivalenttoC n
inthesenseof [30 ℄, at leastwhendim(K) < ∞
, and, in that asethe tensor algebrasT + (B(K) n )
and
T + ( C n )
are Morita equivalent in the sense desribed by [15℄. The tensoralgebra
T + ( C n )
, in turn, is naturally isometrially isomorphi to Popesu's nonommutative dis algebraA n
[33 ℄. The analysis in [15 ℄ suggests a sensein whih
C n
andB (K) n
are Morita equivalent even whendim(K) = ∞
, andthattogetherwith[30℄suggeststhat
H ∞ (B(K) n )
shouldbeMoritaequivalenttoH ∞ (B(X ) n 1 ) ≃ H ∞ ( C n )
. ThiswouldsuggestanevenloseronnetionbetweenPopesu's free power series, and all that goes with them, and the perspetive
wehave takeninthis paper, whih,asweshall see, involvesgeneralizedShur
funtionsandtransferfuntions. Theonnetionseemslikeapromisingavenue
toexplore.
In[31℄ weexploited theperspetiveofviewing elementsof theHardyalgebra
as
B(H )
-valued funtions on the open unit ball of the dual orrespondene to prove a Nevanlinna-Pik type interpolation theorem. In order to state itwe introdue some notation: For operators
B 1
andB 2
inB(H )
, we writeAd(B 1 , B 2 )
forthemapfromB(H)
toitselfthatsendsS
toB 1 SB ∗ 2
. Also,givenelements
η 1 , η 2
inD (E σ )
, we letθ η 1 ,η 2
denote the map, fromσ(M ) ′
to itselfthat sends
a
tohη 1 , aη 2 i
. Thatis,θ η 1 ,η 2 (a) := hη 1 , aη 2 i = η 1 ∗ aη 2
,a ∈ σ(M ) ′
.Theorem 2.18
([31 ,Theorem5.3℄)LetE
beaW ∗
-orrespondeneoveravon NeumannalgebraM
andletσ : M → B(H )
beafaithfulnormalrepresentation ofM
onaHilbertspaeH
. Fixk
pointsη 1 , . . . η k
inthediskD (E σ )
andhoose2k
operatorsB 1 , . . . B k , C 1 , . . . C k
inB(H )
. Then thereexistsanX
inH ∞ (E)
suhthat
kX k ≤ 1
andB i X(η b i ∗ ) = C i
for
i = 1, 2, . . . , k,
if and only if the map fromM k (σ(M ) ′ )
intoM k (B(H ))
denedbythe
k × k
matrix(Ad(B i , B j ) − Ad(C i , C j )) ◦ (id − θ η i ,η j ) −1
(3)
isompletely positive.
That is,themap
T
,say,givenbythematrix(3) isomputedbytheformulaT ((a ij )) = (b ij ),
where
b ij = B i ((id − θ η i ,η j ) −1 (a ij )B j ∗ − C i ((id − θ η i ,η j ) −1 (a ij )C j ∗
and
(id − θ η i ,η j ) −1 (a ij ) = a ij + θ η i ,η j (a ij ) + θ η i ,η j (θ η i ,η j (a ij )) + · · ·
We lose this setion with two tehnial lemmas that will be needed in our
analysis. Let
M
andN
beW ∗
-algebrasand letE
be aW ∗
-orrespondene fromM
toN
. Given aσ
-losed suborrespondeneE 0
ofE
we know thatthe orthogonal projetion
P
ofE
ontoE 0
is a right module map. (See [6,Consequenes1.8(ii)℄). Inthefollowinglemmaweshowthat
P
alsopreservestheleftation.
Lemma 2.19
LetE
be aW ∗
-orrespondene from the von Neumann algebraM
to the von Neumann algebraN
, and letE 0
be a subW ∗
-orrespondeneE 0
ofE
that is losed in theσ
-topology of [6 , Consequenes 1.8 (ii)℄. IfP
is the orthogonal projetion from
E
ontoE 0
, thenP
is abimodule map; i.e.,P (aξb) = aP (ξ)b
for alla ∈ M
andb ∈ N
.Proof.
ItsuestohekthatP(eξ) = eP (ξ)
forallξ ∈ E
andprojetionse ∈ M
. Forξ, η ∈ E
andaprojetione ∈ M
,wehavekeξ + f ηk 2 = kheξ, eξi + hf η, f ηik ≤ kheξ, eξik + khf η, f ηik = keξk 2 + kf ηk 2 ,
where
f = 1 − e
. So,foreveryλ ∈ R
wehave(λ + 1) 2 kf P (eξ)k 2 = kf P (eξ + λf P (eξ))k 2 ≤ keξ + λf P (eξ)k 2
≤ keξk 2 + λ 2 kf P (eξ)k 2 .
Hene,forevery
λ ∈ R
,(2λ + 1)kf P (eξ)k 2 ≤ keξk 2
and,thus,
(I − e)P (eξ) = f P (eξ) = 0.
Replaing
e
byf = I − e
wegeteP((I − e)ξ) = 0
and,therefore,P(eξ) = eP (eξ) = eP (ξ).
Sine
M
isspanned byitsprojetions,wearedone.Lemma 2.20
LetE
beaW ∗
-orrespondeneoverM
,letσ
beafaithfulnormalrepresentationof
M
ontheHilbertspaeE
,andletE σ
betheσ
-dualorrespon-deneover
N := σ(M ) ′
. Then(i) The left ation of
N
onE σ
is faithful if and only ifE
is full (i.e. ifand only if the ultraweakly losed ideal generated by the inner produts
hξ 1 , ξ 2 i
,ξ 1 , ξ 2 ∈ E
,isall ofM
).(ii) Theleft ationof
M
onE
isfaithfulifand onlyifE σ
isfull.Proof.
We shall prove (i). Part (ii) then follows by duality (using [31,Theorem 3.6℄). Given
S ∈ N
,Sη = 0
for everyη ∈ E σ
if and only if forall
η ∈ E σ
andg ∈ E
,(I ⊗ S)η(g) = 0
. Sine the losed subspaespannedby the rangesof all
η ∈ E σ
is allofE ⊗ M E
([31℄), this is equivalent to theequation
ξ ⊗ Sg = 0
holdingforallg ∈ E
andξ ∈ E
. Sinehξ ⊗ Sg, ξ ⊗ Sgi = hg, S ∗ hξ, ξiSgi
, we nd thatSE σ = 0
if and only ifσ(hE, Ei)S = 0
, wherehE, Ei
isthe ultraweakly losedidealgenerated byall innerproduts. Ifthis idealisallofM
wendthattheequationSE σ = 0
impliesthatS = 0
. Intheother diretion,ifthisisnotthease, thenthisideal isoftheform
(I − q)M
forsomeentralnonzeroprojetion
q
andthenS = σ(q)
isdierentfrom0
butvanisheson
E σ
.3 Schur class operator functions and realization
Throughout this setion,
E
will be a xedW ∗
-orrespondene over the von NeumannalgebraM
andσ
willbeafaithfulrepresentationofM
onaHilbertspae
E
. We then form theσ
-dual ofE
,E σ
, whih is aorrespondeneoverN := σ(M ) ′
, and we writeD (E σ )
for its open unit ball. Further, we writeD (E σ ) ∗
for{η ∗ | η ∈ D (E σ )}
.The following denition is learly motivated by the ondition appearing in
Theorem2.18andShur'stheorem fromlassialfuntion theory.
Definition 3.1
LetΩ
be asubset ofD (E σ )
and letΩ ∗ = {ω ∗ | ω ∈ Ω}
. Afuntion
Z : Ω ∗ → B(E)
will be alled a Shur lass operator funtion (withvaluesin
B(E)
)if,foreveryk
andeveryhoieofelementsη 1 , η 2 , . . . , η k
inΩ
,the map from
M k (N)
toM k (B(E))
denedbythek × k
matrixof maps,((id − Ad(Z(η i ∗ ), Z(η ∗ j ))) ◦ (id − θ η i ,η j ) −1 ),
isompletely positive.
Note that,when
M = E = B(E)
andσ
isthe identityrepresentationofB(E )
on
E
,σ(M ) ′
isC I E
,E σ
isisomorphitoC
andD (E σ ) ∗
anbeidentiedwiththe open unit dis
D
ofC
. In this ase our denition reovers the lassial Shur lass funtions. More preisely, these funtions are usually dened asanalyti funtions
Z
from anopensubsetΩ
ofD
into the losed unit ballofB(E)
butitisknownthatsuhfuntionsarepreiselythoseforwhihthePikkernel
k Z (z, w) = (I − Z (z)Z(w) ∗ )(1 − z w) ¯ −1
is positivesemi-denite onΩ
.The argument of [31, Remark 5.4℄ shows that the positivity of this kernelis
equivalent, in ourase, to the onditionof Denition 3.1. This ondition, in
turn, isthesameasassertingthatthekernel
k Z (ζ ∗ , ω ∗ ) := (id − Ad(Z(ζ ∗ ), Z(ω ∗ )) ◦ (id − θ ζ,ω ) −1
(4)isaompletelypositivedenitekernelon
Ω ∗
in thesenseofDenition3.2.2of[14℄.
Forthesakeofompleteness,wereordthefatthat everyelementof
H ∞ (E)
ofnormat mostonegivesrisetoaShurlassoperatorfuntion.
Theorem 3.2
LetE
beaW ∗
-orrespondeneoveravonNeumannalgebraM
and let
σ
be afaithful normal representation ofM
inB(H)
for some Hilbertspae
H
. IfX
isanelementofH ∞ (E)
ofnormatmostone,thenthefuntionη ∗ → X b (η ∗ )
dened in Remark 2.14 is a Shur lass operator funtion onD ((E σ )) ∗
with valuesinB(H)
.Proof.
OnesimplytakesB i = I
foralli
andC i = X(η b ∗ i )
in Theorem2.18.Theorem 3.3
LetE
beaW ∗
-orrespondeneoveravonNeumannalgebraM
.Suppose also that
σ
a faithful normal representation ofM
on a Hilbert spaeE
and thatq 1
andq 2
are projetions inσ(M )
. Finally, suppose thatΩ
is asubset of
D ((E σ ))
and thatZ
is a Shur lass operator funtion onΩ ∗
withvalues in
q 2 B(E)q 1
. Then thereisaHilbert spaeH
,anormalrepresentationτ
ofN := σ(M ) ′
onH
and operatorsA, B, C
andD
fullling the followingonditions:
(i) Theoperator
A
lies inq 2 σ(M )q 1
.(ii) Theoperators
C
,B
,andD
,areinthespaesB(E 1 , E σ ⊗ τ H )
,B(H, E 2 )
,and
B(H, E σ ⊗ τ H)
,respetively,andeahintertwinestherepresentations ofN = σ(M ) ′
on the relevant spaes (i.e. , for everyS ∈ N
,CS = (S ⊗ I H )C
,Bτ(S) = SB
andDτ (S) = (S ⊗ I H )D
).(iii) Theoperator matrix
V =
A B C D
,
(5)viewedas anoperator from
E 1 ⊕ H
toE 2 ⊕ (E σ ⊗ τ H )
, isaoisometry,whih isunitaryif
E
isfull.(iv) For every
η ∗
inΩ ∗
,Z(η ∗ ) = A + B(I − L ∗ η D) −1 L ∗ η C
(6)where
L η : H → E σ ⊗ H
is dened by the formulaL η h = η ⊗ h
(soL ∗ η (θ ⊗ h) = τ(hη, θi)h
).Remark 3.4
BeforegivingtheproofofTheorem3.3,wewanttonotethattheresultbears astrong resemblane tostandardresultsin the literature. Weall
speial attention to [1 , 2, 7, 9 , 10 , 11 , 12 , 13℄. Indeed, we reommend [7 ℄,
whih is a survey that explains the general strategy for proving the theorem.
What isnovel inourapproahisthe adaptation of the resultsin the literature
toaommodateompletelypositive denitekernels.
Sinethe matrixin equation(5) andthefuntion inequation (6)arefamiliar
onstrutsinmathematialsystemstheory,morepartiularlyfrom
H ∞
-ontroltheory(see,e.g.,[38℄),weadoptthefollowingterminology.
Definition 3.5
LetE
beaW ∗
-orrespondeneover avon NeumannalgebraM
. Supposethatσ
isafaithfulnormalrepresentation ofM
onaHilbertspaeE
and thatq 1
andq 2
are projetions inσ(M )
. Then an operator matrixV = A B
C D
,where the entries
A
,B
,C
,andD
,satisfyonditions(i)
and(ii)
of Theorem 3.3for somenormal representation
τ
ofσ(M ) ′
on aHilbert spaeH
,isalleda systemmatrixprovidedV
isaoisometry (that isunitary,ifE
is full). If
V
is a system matrix, then the funtionA + B(I − L ∗ η D) −1 L ∗ η C
,η ∗ ∈ D (E σ ) ∗
isalledthe transferfuntion determinedbyV
.Proof.
Aswejustremarked,thehypothesisthatZ
isaShurlassfuntionon
Ω ∗
meansthat thekernelk Z
in equation(4)isompletelypositivedeniteinthesenseof[14℄. Consequently,wemayapplyTheorem3.2.3of[14℄,whihis
alovelyextensionofKolmogorov'srepresentationtheoremforpositivedenite
kernels,to ndan
N
-B(E) W ∗
-orrespondeneF
andafuntionι
fromΩ ∗
toF
suhthatF
isspannedbyN ι(Ω ∗ )B(E)
andsuhthatforeveryη 1
andη 2
inΩ ∗
andeverya ∈ N
,(id − Ad(Z (η ∗ 1 ), Z(η 2 ∗ ))) ◦ (id − θ η 1 ,η 2 ) −1 (a) = hι(η 1 ), aι(η 2 )i.
Itfollowsthatforevery
b ∈ N
and everyη 1 , η 2
inΩ ∗
,b − Z(η ∗ 1 )bZ(η ∗ 2 ) ∗ = hι(η 1 ), bι(η 2 )i − hι(η 1 ), hη 1 , bη 2 iι(η 2 )i
= hι(η 1 ), bι(η 2 )i − hη 1 ⊗ ι(η 1 ), bη 2 ⊗ ι(η 2 )i.
Thus,
b + hη 1 ⊗ ι(η 1 ), bη 2 ⊗ ι(η 2 )i = hι(η 1 ), bι(η 2 )i + Z (η ∗ 1 )bZ(η 2 ∗ ) ∗ .
(7)Set
G 1 := span{bZ(η ∗ ) ∗ q 2 T ⊕ bι(η)q 2 T | b ∈ N, η ∈ Ω ∗ , T ∈ B (E) }
and
G 2 := span{bq 2 T ⊕ (bη ⊗ ι(η)q 2 T ) | b ∈ N, η ∈ Ω ∗ , T ∈ B(E) }.
Then
G 1
is a subN
-B(E) W ∗
-orrespondene ofB(E) ⊕ F
(where we usethe assumption that
q 2 Z(η ∗ ) = q 2 Z (η ∗ )q 1
) andG 2
is a subN
-B(E) W ∗
-orrespondeneof
B(E)⊕(E σ ⊗ N F)
. (ThelosureinthedenitionsofG 1 , G 2
isinthe
σ
-topologyof[6℄. ItthenfollowsthatG 1
andG 2
areW ∗
-orrespondenes [6,Consequenes1.8(i)℄). Denev : G 1 → G 2
bytheequationv(bZ(η ∗ ) ∗ q 2 T ⊕ bι(η)q 2 T ) = bq 2 T ⊕ (bη ⊗ ι(η)q 2 T ).
It followsfrom (7)that
v
is anisometry. Itis alsolearthat itisabimodulemap. Wewrite
P i
fortheorthogonalprojetionontoG i
,i = 1, 2
andV ˜
forthemap
V ˜ := P 2 vP 1 : q 1 B(E ) ⊕ F → q 2 B(E) ⊕ (E σ ⊗ N F).
Then
V ˜
is apartial isometry and, sineP 1 , v
andP 2
are all bimodule maps(seeLemma 2.19),sois
V ˜
. WewriteV ˜
matriially:V ˜ =
α β γ δ
,
where
α : q 1 B(E) → q 2 B(E)
,β : F → q 2 B(E )
,γ : q 1 B(E) → E σ ⊗ F
andδ : F → E σ ⊗ F
and all these maps are bimodule maps. LetH 0
be theHilbert spae
F ⊗ B(E) E
and note thatB(E ) ⊗ B(E) E
is isomorphitoE
(andtheisomorphismpreservestheleft
N
-ation). TensoringontherightbyE
(overB(E)
)weobtainapartialisometryV 0 :=
A 0 B 0
C 0 D 0
:
E 1
H 0
→
E 2
E σ ⊗ H 0
.
Here
A 0 = α ⊗ I E
,B 0 = β ⊗ I E
,C 0 = γ ⊗ I E
andD 0 = δ ⊗ I E
. Thesemapsare well dened beausethe maps
α, β, γ
andδ
arerightB(E)
-module maps.Sinethesemaps arealso left
N
-modulemaps,soareA 0 , B 0 , C 0
andD 0
.Bythedenitionof
V 0
,itsinitialspaeisG 1 ⊗ E
anditsnalspaeisG 2 ⊗ E
.Infat,
V 0
induesanequivalene ofthe representationsofN
onG 1 ⊗ E
andon
G 2 ⊗ E
.Itwillbeonvenienttousethenotation
K 1 N K 2
iftheHilbertspaesK 1
andK 2
arebothleftN
-modulesandtherepresentationofN
onK 1
isequivalenttoasubrepresentationoftherepresentationof
N
onK 2
. Thismeans,ofourse,thatthereisanisometryfrom
K 1
intoK 2
thatintertwinesthetworepresentations.Ifthetworepresentationsareequivalentwewrite