Contents lists available atScienceDirect
Journal of Functional Analysis
www.elsevier.com/locate/jfa
Shift operators on harmonic Hilbert function spaces on real balls and von Neumann inequality
Daniel Alpaya, H. Turgay Kaptanoğlub,∗
aSchmidCollegeofScienceandTechnology,ChapmanUniversity,Orange,CA 92866,UnitedStatesofAmerica
bBilkentÜniversitesi,MatematikBölümü,06800Ankara,Turkey
a r t i c l e i n f o a b s t r a c t
Articlehistory:
Received5August2020 Accepted15April2021 Availableonline22April2021 CommunicatedbyStefaanVaes
MSC:
primary47A13,47B32,33C55 secondary31B05,33C45,42B35, 46E20,46E22,47B37
Keywords:
Harmonicshift Harmonictypeoperator VonNeumanninequality Drury-Arvesonspace
On harmonicfunction spaces, wedefine shift operators us- ingzonalharmonicsandpartialderivatives,anddeveloptheir basicproperties.Theseoperatorsturnout tobe multiplica- tionsby thecoordinatevariablesfollowedby projectionson harmonicsubspaces.Thisdualitygivesrisetoanewidentity forzonalharmonics.Weintroducelargefamiliesofreproduc- ingkernel Hilbertspacesof harmonicfunctionson theunit ballof Rn andinvestigate theactionof theshift operators on them. We prove a dilation result for a commuting row contractionwhichisalso whatwecallharmonictype.Asa consequence,weshowthatthenormofoneofourspacesG is˘ maximalamongthosespaceswithcontractivenormsonhar- monicpolynomials.WethenobtainavonNeumanninequality forharmonicpolynomialsofacommutingharmonic-typerow contraction.Thisyieldsthemaximalityoftheoperatornorm ofaharmonicpolynomialoftheshiftonG making˘ thisspace anaturalharmoniccounterpartoftheDrury-Arvesonspace.
©2021ElsevierInc.Allrightsreserved.
* Correspondingauthor.
E-mailaddresses:alpay@chapman.edu(D. Alpay),kaptan@fen.bilkent.edu.tr(H.T. Kaptanoğlu).
URLs:http://www1.chapman.edu/~alpay/(D. Alpay),http://www.fen.bilkent.edu.tr/~kaptan/
(H.T. Kaptanoğlu).
https://doi.org/10.1016/j.jfa.2021.109058
0022-1236/©2021ElsevierInc. Allrightsreserved.
1. Introduction
AfterthepioneeringworkofDrury[10] andArveson[6] onextendingthevonNeumann inequality tocommuting operatortuples, there havebeen several other generalizations to Hilbertspace operatorsinother settings.Wecancite [20],[3],[16],[11],[13],and[8]
to name afew. There is alsothe earlier[19] on noncommutingoperator tuples. Butin allvon Neumanninequalitiesthatweknow of,thepolynomialsactingontheoperators are holomorphic functionsof theirvariables. It isthe aimof this workto obtainavon Neumanninequalityinwhichthepolynomials areharmonicintheusualsense inRn.
MultivariableversionsofvonNeumanninequalityoftendependonshiftoperatorson a specific Hilbertfunction space. This immediately brings outthe first major obstacle in dealingwith harmonicity. Eventhe definitionof a shiftoperator on aspace of har- monic functionshasnotbeen madebefore, becauseharmonicity isnotpreservedunder multiplication, and amultiplication byacoordinatevariablemust be followedbysome form ofaprojectiononharmonicfunctions.Wemakeadefinitionandcheck itbyusing another approach.
Another obstacle is to decide which space among harmonic function spaces plays a role like thatofthe Drury-Arvesonspace among holomorphicspaces.We findoutthat consideringafamilyofreproducingkernelHilbertspacesGqofharmonicfunctionsonthe unitballofRnindexedbyq∈ R ismorefeasiblesinceitexposesthecompositionsofthe spaces better.Thenitiseasierto pickoneofthesespaces astheharmoniccounterpart of theDrury-Arvesonspaceusingitsextremalpropertiesinthefamily.
Onemoreobstacleisthatthemorecomplicatedstructure ofharmonicfunctionsper- sistsat theoperatorleveland weare obligedtorestrictourattentioninvon Neumann inequalitytoaclassofcontractionsthatwecall harmonictype.
We now present our major results; for them it helps to have some familiarity with the classical knowledge on harmonic polynomials summarized in Section 3. For m = 0,1,2,. . .,let Pm and Hm denote thehomogeneouspolynomials ofdegree m and spherical harmonicsofdegreem on Rn,respectively.LetHm:Pm → Hm be thestan- dard projection. The zonal harmonics Zm(x,y) arethe reproducing kernels of theHm
with respecttotheL2 innerproductontheunitsphere.Forj = 1,. . . ,n,wedefine the shiftoperators Sj:Hm→ Hm+1actingonx by
SjZm(x, y) := 1 n + 2m
∂
∂yjZm+1(x, y).
Our first main result shows thatSj is closely relatedto the operatorof multiplication Mxj bythejthcoordinatevariable.
Theorem 1.1.Sj= Hm+1Mxj forallj = 1,. . . ,n andm= 0,1,2,. . ..
Inthecourseofprovingthistheorem,weobtainthefollowingidentitiesfortheGegen- bauer polynomials Cmλ and theChebyshev polynomialsTm which seemnew.There, K
istheKelvintransformwhichtransformsaharmonicfunctionontheunitballtooneon itsexterior.
Theorem1.2. Forξ,η∈ S, wehave K
(η· ∂)mK[1]
(ξ) = (−1)mm! Cmn/2−1(ξ· η) (n≥ 3, m = 0, 1, 2, . . .), K
(η· ∂)mK[ log| · | ]
(ξ) = (−1)m(m− 1)! Tm(ξ· η) (n = 2, m = 1, 2, . . .).
Wedefine thespacethatweclaim tobe theharmonic versionoftheDrury-Arveson spaceasthereproducingkernelHilbertspaceG on˘ theunitballofRn withreproducing kernel
G(x, y) :=˘
∞ m=0
1 Am
Zm(x, y),
whereAm isthecoefficientof(x· y)m intheexpansionofZm(x,y).Wecall commuting operators(T1,. . . ,Tn) arowcontractioniftheyareacontractionasatuple.Acontractive normisoneinwhichthetupleofshiftoperatorsisarowcontraction.Anothermainresult ofoursshowsthatthenormofG is˘ as largeas possible.
Theorem1.3. If · isacontractiveHilbertnormonharmonic polynomialsthatrespects theorthogonalityof L2,then· ≤ · G˘1.
Wecall anoperatortuple (T1,. . . ,Tn) harmonic type ifT1T1+· · · + TnTn = 0. The harmonicshiftS = ( ˘˘ S1,. . . ,S˘n) onG is˘ theprimeexampleofaharmonic-typeoperator.
Ourfinalmain resultisavonNeumanninequality.
Theorem1.4. Let(T1,. . . ,Tn) beaharmonic-typerowcontraction onaHilbertspace. If u isaharmonic polynomial,thenu(T1.. . . ,Tn)≤ u( ˘S1,. . . ,S˘n).
Allterminology is explained indetail in an appropriate section inthe paper. After introducinginSection2thebasicnotation,wemakeareviewofharmonic polynomials inSection3andgiveformulasfortheZm,Hm,andK.InSection4,wedefinetheshift operators on harmonic spaces, explain the meaning of the coefficient 1/(n+ 2m), and thenproveTheorem1.1.InSection5,we introduceanewfamilyofreproducing kernel HilbertspacesofharmonicfunctionsandisolateoneofthemasG by˘ makingitclearwhy we need the coefficients1/Am in G.˘ In Section 6, we find the basicproperties of shift operatorsandtheiradjointsactingontheHilbertspacesjustintroduced.InSection7,we investigatetherowcontractionsonharmonicHilbertspaces,explainthetermharmonic type,and proveanessentialdilation resultforharmonic-typeand self-adjoint operator tuples.InSection8,weproveTheorems1.3and1.4.
Acknowledgments
ThefirstauthorthankstheFosterG.andMaryMcGawProfessorshipinMathematical Sciences of Chapman University for its support. The second author thanks Aurelian GheondeaofBilkentUniversityandSerdarAyofAtılımUniversityforusefuldiscussions.
The authors also thank an anonymous referee for suggesting to consider self-adjoint operatorswhichencouragedustoobtaintheresultsonsuchoperatorsinthenexttolast section.
2. Notation
LetB andS betheopenunitballanditsboundarytheunitsphereinRnwithrespect to the usualinner product x· y = x1y1+· · · + xnyn and thenorm |x|=√
x· x, where alwaysn≥ 2.Wewritex= rξ, y = ρη withr =|x|,ρ=|y|,andξ,η∈ S,andusethese throughout withoutfurther comment.When n = 2,the ball is justthe unitdisc D in the complexplanebounded bytheunitcircleT , and x,y∈ D arecomplexnumbersof moduluslessthan1.
In afewplaces, we alsouse thecomplexspace CN and its Hermitianinner product
z,w = z1w1+· · · + zNwN.Wecontinueto use|· |,B,andS in CN too.
Weletσ andν bethesurfaceandvolumemeasuresonS andB normalizedasσ(S)= 1 and ν(B)= 1. Weabbreviatetheall-importantLebesgue classL2(σ) to simplyL2. An overline(·) denotesclosureforsetsandcomplexconjugationforelements;forpolynomials inx,theconjugationaffectsonlythecoefficientsnaturally.Thegreatestintegerlessthan or equalto areal numberisshownby·.Therightside of:= definesitsleftside.
Harmonic functionsby definitionarethosesufficiently smoothfunctionsannihilated by the usual Laplacian Δ := ∂2/∂x21+· · · + ∂2/∂x2n. We leth(B) denote thespace of complex-valued harmonic functionsonB with the topologyof uniform convergence on compact subsets.
In the multi-indexnotation, α = (α1,. . . ,αn) isan n-tuple of nonnegative integers,
|α|= α1+· · · + αn, α!= α1!· · · αn!,00= 1,andxα= xα11· · · xαnn. Letting∂j := ∂/∂xj and∂ := (∂1,. . . ,∂n),wealsohave∂α= ∂1α1· · · ∂nαn,∂αxα= α! andp(∂)=
α
aα∂αfor apolynomialp(x)=
α
aαxα.Soforp(x)=|x|2,
|x|2(∂) = Δ. (1)
ThePochhammersymbol (a)b is definedby
(a)b=Γ(a + b) Γ(a)
when a and a+ b are off the pole set −N ofthe gamma function Γ.This is a shifted risingfactorialsince(a)k = a(a+ 1)· · · (a+ k− 1) forpositive integerk.Inparticular, (1)k = k! and(a)0= 1.Stirlingformulagives
Γ(c + a)
Γ(c + b) ∼ ca−b, (a)c
(b)c ∼ ca−b, (c)a
(c)b ∼ ca−b (Re c→ ∞), (2) whereA∼ B meansthat|A/B| isboundedaboveandbelowbytwopositiveconstants, thatis,A=O(B) andB =O(A),forallA,B ofinterest.Soforexample,1−|x|∼ 1−|x|2 forallx∈ B.Such constantsthatareindependent oftheparametersandthefunctions in the equation are all denoted by the generic unadorned upper case C. We also use A B to meanA=O(B).
WedenoteaninnerproductonafunctionspaceH by[·,·]Handtheassociatednorm by· H.
Definition 2.1.A functionk(x,y) is called thereproducing kernel of aHilbertspace H offunctionsdefinedonB ifk(x,·)∈ H foreachx∈ B and
u(x) = [ u(·), k(x, ·) ]H (u∈ H, x ∈ B).
Thereisaone-to-onecorrespondencebetweenreproducingkernelHilbertspacesand positivedefinitekernels.Weusewordslike positiveandincreasingtomeannonnegative andnondecreasing.
Thealgebraofallboundedlinearoperators onacomplexHilbertspaceH isdenoted B(H).AnoperatorT onH iscalled positive andwewrite T ≥ 0 if[T v,v]H ≥ 0 forall v∈ H.Fora,b∈ H,a⊗ b denotestherank-1 operatordefinedby(a⊗ b)(v)=
v,b
Ha forv∈ H.
3. Harmonicpolynomials
Wereview theessentialsof zonalharmonics andtheKelvin transformfor complete- ness,becausewerefertothese factsmanytimesinthepaper.These resultsaremostly well-knownandcanbe consultedin[7,Chapters4&5].
Form= 0,1,2,. . .,letPmdenotethecomplexvectorspaceofallpolynomialshomo- geneous(withrespecttoreal scalars)ofdegreem onRn. Itisimmediatethat
(x· ∂)pm= m pm (pm∈ Pm). (3)
Let Hm be the subspace of Pm consisting of all harmonic homogeneous polynomials of degree m. By homogeneity, a pm ∈ Pm is determined by its restriction to S, and we freely identifypm with its restriction. The restrictionsof those um ∈ Hm to S are calledsphericalharmonics.WealsoletP andH denoteallpolynomialsandallharmonic polynomialsonRn.
WeregardPmandHmassubspacesofL2inissuesrequiringanormorinnerproduct.
Forexample,ifm= l,thenHmisorthogonal toHlwithrespect to[·,·]L2. Proposition 3.1. [7,Exercise5.12] If um,vm∈ Hm,then
um(∂)(vm) = n(n + 2)· · · (n + 2m − 2)
S
umvmdσ
= 2m(n/2)m
um, vm]L2.
Proof. Letum(x)=
|α|=m
bαxαandvm(x)=
|α|=m
dαxα.By[7,Theorem5.14],theright sidesare
|α|=m
α!bαdα. Itiseasytocheck thatthisis equaltoum(∂)(vm).
Ifpm∈ Pm,thenthereareuniqueul∈ Hl suchthat
pm(x) = um(x) +|x|2um−2(x) +· · · + |x|2kum−2λ(x) (x∈ Rn),
where λ=m/2.This decompositionis simplypm = um+ um−2+· · · + um−2λ when restrictedto S.Let
Hm:Pm→ Hm, pm→ um
bethemapthatprojectspmtoum.TheexplicitformulaforHmrequiressomeconstants cm, m= 1,2,. . .,definedby
cm:=
(−1)m2m(n/2− 1)m, if n≥ 3, (−1)m−12m−1(m− 1)!, if n = 2.
It alsomakesuseoftheKelvin transform K definedonafunctionf by K[f ](x) :=|x|2−nf (x∗) (x= 0), where x∗:= x
|x|2,
whichreducestoK[f ](x):= f (x∗) whenn= 2.NotethatK[f ]= f onS foranyn≥ 2.
TheKelvintransform islinearandinvertiblewith K−1= K.Afunctionis harmonicif andonlyifitsKelvintransformisharmonic.Sou(x)=|x|2−nisharmonicespeciallyfor n≥ 3 whereveritisdefinedsinceitistheKelvintransformoftheconstant1.Forn= 2, itis replacedinformulasbytheharmonicfunctionu(x)= log|x|.
Theorem 3.2.[7,Theorem5.18] Letm≥ 1 and pm∈ Pm.
(a) Hm(pm)(x)= 1 cm
K
pm(∂)|x|2−n
, if n≥ 3, K
pm(∂) log|x|
, if n = 2.
(b) When pm is restricted to S, then Hm (without any need for K) is an orthogonal projection with respectto[·,·]L2
ThespacesHm arefinitedimensional,henceclosed subspacesofL2,and
δm:= dimHm= (n + 2m− 2)(n− 1)m−1
m! (m≥ 1), (4)
which gives δm = 2 for m ≥ 1 when n = 2. When m = 0, H0 = C and δ0 = 1.
The next few are δ1 = n, δ2 = (n− 1)(n+ 2)/2, and δ3 = (n− 1)n(n+ 4)/6. Thus evaluation functionals at points η ∈ S arebounded on Hm, and Hm is a reproducing kernelHilbertspace.ItsreproducingkernelZm(ξ,η) withrespectto[·,·]L2 iscalledthe zonalharmonic ofdegreem;soZmisapositivedefinitefunction.EachZmisrealvalued andsymmetricinitsvariables,henceitisaharmonichomogeneouspolynomialineach ifitstwo variables.ThehomogeneityoftheZmgivesZm(x,y):= rmρmZm(ξ,η);so
Zm(0, y) = Zm(x, 0) = 0 (m≥ 1). (5) Theirreproducingpropertywrittenexplicitlyis
um(x) =
S
um(η) Zm(x, η) dσ(η) =
um(·), Zm(x,·)
L2 (x∈ B, um∈ Hm). (6) ThePoissonkernelis
P (x, η) := 1− |x|2
|x − η|n =
∞ m=0
Zm(x, η) (x∈ B, η ∈ S);
theseriesconvergesuniformly forx inacompactsubsetofB.
ThereisanexplicitformulafortheZmwhichisof majorinteresttous; itis
Zm(x, y) = (n+2m−2)
m/2
l=0
n(n+2)· · · (n+2m−2l−4)
(−1)l2ll! (m− 2l)! |x|2l(x·y)m−2l|y|2l
= Am0(x· y)m+Am1|x|2(x·y)m−2|y|2+Am2|x|4(x·y)m−4|y|4+· · · , (7)
whereAm:= Am0 istheleadingcoefficientobtainedforl = 0.Then
Am:= n(n + 2)· · · (n + 2m − 2)
m! = 2m(n/2)m
m! , (8)
where thenumeratoris thecoefficientinthe equationinProposition3.1. Thefirst few areA0= 1,A1= n,A2= n(n+ 2)/2,andA3= n(n+ 2)(n+ 4)/6.NotethatAm= 2m forallm= 0,1,2,. . . whenn= 2.
Interesting andusefulrelationsinclude
|Zm(ξ, η)| ≤ Zm(ξ, ξ) =
m/2
l=0
Aml= δm (ξ, η∈ S). (9)
If{Ym1,. . . ,Ymδm} isanorthonormalbasisforHm⊂ L2,then
Zm(ξ, η) =
δm
k=1
Ymk(ξ) Ymk(η). (10)
In particular, Z0 ≡ 1 and we takeY01 ≡ 1;also Z1(x,y)= n(x· y) and we canchoose Y1k(x)=√
n xk fork = 1,. . . ,δ1= n. Itisalways possibleto choosetheYmk withreal coefficients.
Theorem 3.3. Every u ∈ h(B) has the homogeneous expansion u = ∞
m=0
um in which um∈ HmandwhichconvergesabsolutelyanduniformlyoncompactsubsetsofB.Letting umk=
um,Ymk
L2∈ C, wealso have
u(x) =
∞ m=0
δm
k=1
umkYmk(x) (x∈ B)
with thesametype ofconvergence.
Whenn= 2,we canusecomplexanalysis andFourieranalysis toconnecttheabove theory to better knownobjects. Then for all m ≥ 1, an orthonormal basis for Hm is {Ym1(x) = xm,Ym2(x) = xm : x ∈ C }. The expansion of a u ∈ h(B) with the Ymk
inTheorem 3.3 withsuitableboundarybehavior is itsFourierseries ontheunitcircle.
Lettingξ = eiφ andη = eiψ,wecanwrite
Zm(x, y) = xmym+ xmym= 2rmρmcos(φ− ψ) (m≥ 1).
It isamazing that this simpleform is equal to thesum in(7) for n= 2, whichis still complicated.
A comparison with thecomplex case donein [15, Section3] when n iseven is very instructive.WethinkofRn asCN byequatingn= 2N .Sphericalharmonicscorrespond to thespace of holomorphic polynomials homogeneousof degree m whichsimply have the form
|α|=m
bαzα. Thedimensionofthis space is (N )m
m! , whichequals 1 foreverym when N = 1.Thecounterpartsofzonalharmonicsarethesesquiholomorphickernels
Mm(z, w) = (N )m
m! z, w m.
Thus the complex version of the sum in (7) has only the leading term Mm, and its coefficientisexactlythedimensionofthespaceofwhichMm isthereproducingkernel.
Note thatAm= δminthe harmoniccaseeven when n= 2. However,writing(7) with x= y = ξ andusing(9),wesee that
m/2
l=0
Aml= δm.By(2),wehave
δm∼ mn−2 and Am∼ 2mmn/2−1 (m→ ∞). (11) Yetthereproducingkernel oftheholomorphic Drury-Arvesonspace is ∞
m=0z,w m and not ∞
m=0
Mm(z,w).
Thereforewe must find theharmonic counterparts of thez,w m and we are ledto the
Xm(x, y) := 1
AmZm(x, y) = (x· y)m− · · · , (12) whichwecallthexonalharmonics.ThefirstfewareX0= 1,X1(x,y)= x· y,
X2(x, y) = (x· y)2−|x|2|y|2
n , X3(x, y) = (x· y)3− 3
n + 2|x|2(x· y)|y|2. Byhomogeneity, (9),(4),and (8),weseethatforallx,y∈ B,
|Xm(x, y)| = rmρm|Xm(ξ, η)| = (rρ)m|Zm(ξ, η)|
Am ≤ (rρ)mδm
Am ≤ (rρ)m< 1 (13) incomplete analogywith|z,w m|≤ |z|m|w|m< 1 for z,w intheunitballofCN. 4. Shiftoperators
Wedefinetheshiftoperatorsonharmonicfunctionsfirstinanunusualway,butlater showthattheyareequivalent essentiallyto multiplicationsbythecoordinatevariables.
Themotivationforourdefinitionliesintheobservation 1
m + 1
∂
∂wj
z, w m+1
= zjz, w m
forz,w∈ CN andtherealizationthatXm(x,y) replacesz,w m.
Definition4.1. For1≤ j ≤ n,wedefine thejth shiftoperator Sj:Hm→ Hm+1 acting onthevariablex byfirstletting
SjXm(x, y) := 1 m + 1
∂
∂yjXm+1(x, y)
and thenextendingtoallofHmbylinearityandthedensityoftheXm(·,y) inHm. Note that all this make sense; a partial derivative of a harmonic function is again harmonic,andfinitelinearcombinationsofthereproducingkernelsZmandhenceofthe XmaredenseinHmin· L2.Alsonote thattheshiftactsonthefirstvariablex,but the partial derivativeis with respect to the second variable y. So occasionally we also use notation like Sx or ∂y to indicatethe variables onwhich theyact. In termsof the morefamiliar zonalharmonics,
SjxZm(x, y) = Am
Am+1 1 m + 1
∂
∂yjZm+1(x, y) = 1 n + 2m
∂
∂yjZm+1(x, y).
Let’s denoteby Sj∗:Hm→ Hm−1 the adjointofSj with respectto [·,·]L2 inwhich the reproducingkernel ofHmis Zm= AmXm. Thisis ofcoursethe jthbackward shift operator.FirstwesetSj∗(X0)= Sj∗(1)= 0.Nextform≥ 1,using(6),symmetryofXm
initsvariables,anditsreal-valuedness, weobtain (Sjx)∗Xm(x, y) =
(Sjt)∗Xm(t, y), Am−1Xm−1(x, t)
L2
= Am−1
Xm(t, y), SjtXm−1(x, t)
L2
= Am−1
Xm(t, y), 1 m
∂
∂xjXm(x, t)
L2
= 1 m
Am−1
Am
AmXm(t, y), ∂
∂xjXm(x, t)
L2
= 1 m
Am−1 Am
∂
∂xj
Xm(x, y),
wherey actsjustlikeaparameter.ThislastformulaforSj∗isindependentofthepartic- ular form ofthefunction onwhichit acts,so worksequallywell forum∈ Hm inplace of Xmbylinearityand densityagain. Thus
Sj∗um= 1 m
Am−1 Am
∂jum= 1
n + 2m− 2∂jum (um∈ Hm, m≥ 1). (14) It is clearthat theSj∗ commute with eachother. Sothe shift S = (S1,. . . ,Sn) is a commutingtuple.
Shiftoperatorsonholomorphicfunctionspacesareoperatorsofmultiplicationbythe coordinate variables. Here we make a distinction between the two, becausethe latter doesnotingeneralcarryharmonicfunctionstoharmonicfunctionsunliketheformer.If f,g arefunctionsonthesamedomain,welet
Mgf = gf be theoperatorofmultiplication byg.
OneofourresultsconcernsalimitedversionoftheobviousfactthatifS isthetuple of shift operators ona space of holomorphic functionson a domain in CN and p is a holomorphicpolynomialinN complexvariables,then p(S)= Mp.It isafirstresulton theconnectionbetweenshiftsand multiplicationoperators.
Proposition 4.2.If u isa harmonic polynomial,then u(S)(1)= u.In other words,1 is acyclic vectorforh(B).
Proof. Itsufficesto consideru= um∈ Hm. By(6),repeated useof(14),and Proposi- tion3.1,
um(Sx)(1)(x) =
um(Sη)(Z0)(η, y), Zm(x, η)
L2
=
Z0(η, y), um((Sη)∗)(Zm)(x, η)
L2
=
Z0(η, y), 1
n(n + 2)· · · (n + 2m − 2)um(∂η)(Zm)(x, η)
L2
= 1
2m(n/2)m
um(∂η)(Zm)(x, η), Z0(y, η)
L2
= 1
2m(n/2)mum(∂y)(Zm)(x, y)
=
S
um(·) Zm(x,·) dσ(·) = um(x).
Above,Sη and∂η mustbeinterpretedasSy
y=η and∂y
y=η,respectively.Notethatwe usetheharmonicityofumonlyinpassingtothelast line.
Ifthefunctionactedonismorecomplicatedthan1,thenweofferthefollowingpartial result.
Proposition4.3. If pm∈ Pm,then
pm(Sx)(X)(x, y) = 1 (1 + )m
pm(∂y)(Xm+)(x, y).
Proof. FollowingthesameideaandnotationintheproofofProposition4.2, pm(Sx)(Z)(x, y) =
pm(Sη)(Z)(η, y), Zm+(x, η)
L2
=
Z(η, y), pm((Sη)∗)(Zm+)(x, η)
L2
=
Z(η, y), pm(∂η)(Zm+)(x, η)
(n + 2 )(n + 2 + 2)· · · (n + 2 + 2m − 2)
L2
= 1
2m(n/2 + )m
pm(∂η)(Zm+(x, η), Z(y, η)
L2
= 1 2m(n/2 + )m
pm(∂y)(Zm+)(x, y).
Lastly, we pass to the xonal harmonics X = Z/A and simplify the resulting coeffi- cient.
Our goal now is to express the shift operators Sj : Hm → Hm+1 in terms of the operators of multiplication by the coordinate variables Mxj : Hm → Pm+1, where for both j = 1,. . . ,n.ItisTheorem 1.1andwerestateit.
Theorem 4.4. For all j = 1,. . . ,n and m = 0,1,2. . ., if Sj : Hm → Hm+1, then Sj= Hm+1Mxj.
So the harmonic shiftsare really Toeplitz operators. Wefacilitate the longproof of this theorem withsomecomputationallemmas inwhichj = 1,. . . ,n,y is aparameter, and allpartial derivativesandKelvintransformsarewithrespecttox.
Lemma 4.5.Fora,b∈ R,easy computations give
∂j|x|a= a|x|a−2xj and ∂j(x· y)b= b(x· y)b−1yj, (y· ∂)|x|a= a|x|a−2(x· y) and (y· ∂)(x · y)b= b(x· y)b−1|y|2.
Lemma 4.6. If a polynomial p has |x|2 as a factor, then p(∂)|x|2−n = 0 and also p(∂)log|x|= 0 whenn= 2.Consequently
Xm(∂, y)|x|2−n= (y· ∂)m|x|2−n and Xm(∂, y) log|x| = (y · ∂)mlog|x| (n = 2).
Proof. The firststatementfollows immediately from (1) and theharmonicity of|x|2−n andoflog|x| whenn= 2.Thesecondstatementfollowsfromtheexplicitformsofzonal harmonics in (7), because all the terms in Xm(x,y) except the first have a factor of
|x|2.
This lemma isveryuseful, becauseitletsustreatXm asthesingle term(x· y)m in thepresenceof K orHm likeitsholomorphiccounterpartz,w m.
Lemma 4.7.Form≥ 1,
(y· ∂)m|x|2−n= cm Xm(x, y)
|x|n+2m−2 (n≥ 3), (y· ∂)mlog|x| = cm
Xm(x, y)
|x|2m (n = 2);
and hence
K
(y· ∂)m|x|2−n
= cmXm(x, y) (n≥ 3), K
(y· ∂)mlog|x|
= cmXm(x, y) (n = 2).
The identitiesform≥ 3 aretrue also form= 0 ifweset c0= 1.
FromLemma4.7,Theorem1.2 followseasilywhichwerestate.
Theorem4.8. Forξ,η∈ S, wehave K
(η· ∂)mK[1]
(ξ) = (−1)mm! Cmn/2−1(ξ· η) (n≥ 3, m = 0, 1, 2, . . .), K
(η· ∂)mK[ log| · | ]
(ξ) = (−1)m(m− 1)! Tm(ξ· η) (n = 2, m = 1, 2, . . .), where Cmλ is the Gegenbauer (ultraspherical) polynomial of degree m and indexλ, and TmistheChebyshev polynomialof thefirstkindof degree m.
Proof. Let’s first note that K[1] = |x|2−n for n = 3 and K[ log|x|] = −log|x| for n= 2. Thefirstidentity isaconsequenceof thewell-knownrelationbetweenthe zonal harmonicsandGegenbauerpolynomials;see[12,(14.8)] forexample.Thesecondidentity holdsbecauseofthecloseconnectionbetweenGegenbauerpolynomials withparameter 0 andChebyshevpolynomials;see[18,18.1.1] forexample.
The identities in Lemma 4.7 and Corollary 4.8 seem new; we are unable to locate theminstandardreferencessuchas[18].Theyalsogivefurtherindicationthatthexonal harmonicsXmareimportantintheirownrightsincethecoefficientsintheidentitiesin Lemma4.7 withtheZmarenotsimpleknownones.
Proof of Lemma4.7. Thesecondset ofidentitiesfollowsimmediately fromthefirstset, andfortheseweproceedbyinductiononm.Wegivetheproofonlyforn≥ 3;theproof for n = 2 is obtained by replacing |x|2−n by log|x| and setting n = 2 in appropriate places.Form= 1,byLemma4.5,
(y· ∂)|x|2−n= (2− n)|x|−n(x· y) = c1
X1(x, y)
|x|n .
NextweassumethefirstidentityinLemma4.7holdsform,andshowthatitalsoholds form+ 1.Byapplying theinductionhypothesis,differentiatingwiththequotient rule, andusingLemma4.5,weobtain
(y· ∂)m+1|x|2−n= (y· ∂)
cm Xm(x, y)
|x|n+2m−2
= cm|x|n+2m−2(y· ∂)Xm− Xm(n + 2m− 2)|x|n+2m−4(x· y)
|x|2n+4m−4
= cm|x|2(y· ∂)Xm− (n + 2m − 2)(x · y)Xm
|x|n+2m .
Writing outthecoefficients,wemustshow
(x· y)Xm− |x|2
n + 2m− 2(y· ∂)Xm= Xm+1. By(7) andLemma4.5,theleftsideequals
n + 2m− 2 Am
m/2
l=0
n(n + 2)· · · (n + 2m − 2l − 4)
(−1)l2ll! (m− 2l)! |x|2l(x· y)m−2l+1|y|2l
− 1 Am
m/2
l=0
n(n + 2)· · · (n + 2m − 2l − 4)
(−1)l2ll! (m− 2l)! 2l|x|2l(x· y)m−2l+1|y|2l
− 1 Am
m/2
l=0
n(n + 2)· · · (n + 2m − 2l − 4)
(−1)l2ll! (m− 2l)! (m− 2l)|x|2l+2(x· y)m−2l−1|y|2l+2
= 1 Am
m/2
l=0
n(n + 2)· · · (n + 2m − 2l − 2)
(−1)l2ll! (m− 2l)! |x|2l(x· y)m−2l+1|y|2l
+ 1 Am
m/2+1
l=1
n(n + 2)· · · (n + 2m − 2l − 4)
(−1)l2l−1(l− 1)! (m − 2l + 1)!|x|2l(x· y)m−2l+1|y|2l
= m + 1 Am
m/2
l=1
n(n + 2)· · · (n + 2m − 2l − 2)
(−1)l2ll! (m + 1− 2l)! |x|2l(x· y)m+1−2l|y|2l+ extra term + (x· y)m+1,
where theextra termis due to l =m/2+ 1 in the second sumon theprevious line.
Theright sideequals
n + 2m Am+1
(m+1)/2
l=1
n(n + 2)· · · (n + 2m − 2l − 2)
(−1)l2ll! (m + 1− 2l)! |x|2l(x· y)m+1−2l|y|2l+ (x· y)m+1. Since (m+ 1)/Am = (n+ 2m)/Am+1, the two sides are equal except for the extra termandthatthesumontherightsideendsperhapsatalargervalue.Whenm iseven, thel givingrisetotheextratermequalsm/2+ 1,andthenm− 2l + 1=−1< 0 inthat term, sothere isreallynoextraterm.Also(m+ 1)/2= m/2=m/2,andthesums onbothsidesendatthesamevalue.Whenm isodd,thel givingrisetotheextraterm equals (m+ 1)/2.Also(m+ 1)/2= (m+ 1)/2.Sotheextratermandthelasttermin thesumontherightarethesame.
Proof of Theorem4.4. Againwewritetheproofonlyforn≥ 3.Itsufficestodotheproof onlyforum(·)= Xm(·,y) asinDefinition4.1.InviewofLemma4.6andbyTheorem3.2, allweneedtoshowis
1 m + 1
∂
∂yj
Xm+1(x, y) = 1 cm+1
K ∂
∂xj
(y· ∂)m|x|2−n
forallm= 0,1,2,. . .,whereK and∂ arewithrespecttox.ApplyingLemma4.7onthe rightandcombiningtheconstants,thisequationtakes theform
∂
∂yj
Xm+1(x, y) =− m + 1 n + 2m− 2K
∂
∂xj
Xm(x, y)
|x|n+2m−2
(15)
By(7) andLemma4.5, theleft sideequals n + 2m
Am+1
(m+1)/2
l=0
n(n + 2)· · · (n + 2m − 2l − 2)
(−1)l2ll! (m− 2l)! xj|x|2l(x· y)m−2l|y|2l
+
(m+1)/2
l=1
n(n + 2)· · · (n + 2m − 2l − 2)
(−1)l2l−1(l− 1)! (m + 1 − 2l)!yj|x|2l(x· y)m+1−2l|y|2l−2
.
ByLemma4.5,
∂
∂xj
Xm(x, y)
|x|n+2m−2
= |x|n+2m−2∂jXm− Xm(n + 2m− 2)|x|n+2m−4xj
|x|2n+4m−4
= |x|2∂jXm− (n + 2m − 2)xjXm
|x|n+2m .
AftertakingtheKelvintransform,therightsideequals
(m + 1)
xjXm− 1
n + 2m− 2|x|2∂Xm
∂xj
= m + 1 Am
(n + 2m− 2)
m/2
l=0
n(n + 2)· · · (n + 2m − 2l − 4)
(−1)l2ll! (m− 2l)! xj|x|2l(x· y)m−2l|y|2l
−
m/2
l=0
n(n + 2)· · · (n + 2m − 2l − 4)
(−1)l2ll! (m− 2l)! 2l xj|x|2l(x· y)m−2l|y|2l
−
m/2
l=0
n(n + 2)· · · (n + 2m − 2l − 4)
(−1)l2ll! (m− 2l − 1)! yj|x|2l+2(x· y)m−2l−1|y|2l
= m + 1 Am
m/2
l=0
n(n + 2)· · · (n + 2m − 2l − 2)
(−1)l2ll! (m− 2l)! xj|x|2l(x· y)m−2l|y|2l
+
m/2+1
l=1
n(n + 2)· · · (n + 2m − 2l − 2)
(−1)l2l−1(l− 1)! (m + 1 − 2l)!yj|x|2l(x· y)m+1−2l|y|2l−2
.
Since thecoefficients multiplying thesums areequal,the twosides areequalexcept perhaps inthe upperlimitsof thesums.Let m beeven. Theupper limitonthe leftis l =(m+ 1)/2= m/2=m/2.Inthesumwithxj,thisisalsotheupperlimitonthe right.Thesumwithyj ontherightendswithl = m/2+ 1,butthenm+ 1−2l = −1< 0, so this term is not really there. Next let m be odd. The upper limit on the left is l = (m+ 1)/2= (m+ 1)/2 =m/2+ 1. Inthe sumwith yj,this is also theupper limit onthe right.Thesum withxj onthe right endswith l = (m− 1)/2,so it seems as ifthetermontheleftwithl = (m+ 1)/2 isextra,butthenm− 2l = −1< 0,sothis termisnotreallythere,either.
Example 4.9. Let’s compute the action of shifts on a very simple harmonic function.
Let’s find S1u andS2u for u(x)= x1. We apply Theorem 4.4 and follow therecipe in Theorem 3.2separately for n≥ 3 and n= 2.Straightforward computations yield that S1x1 = x21− |x|2/n for any n ≥ 2. On theother hand,simply S2x1 = x2x1 since this product isalreadyharmonic.
5. HarmonicHilbertfunctionspaces
Weareinspiredbyafewearlierworksindefiningnewreproducingkernelswithdesired properties.In[16],familiesofweightedsymmetricFockspacesofholomorphicfunctions thatincludetheDrury-Arvesonspacearestudiedfollowing [6]. In[12],Bergman-Besov kernels are definedas weightedinfinite sumsofzonal harmonics muchlike the Poisson kernel.Andwehavealreadynotedthattherighttoolisthexonalharmonicsratherthan thezonal harmonics.
Definition 5.1.Letβ :={βm> 0: β0= 1, m= 0,1,2,. . .} beasequencesatisfying
lim sup
m→∞
δm Am
βm
1/m
≤ 1. (16)
Wedefine positivedefinite kernelsby
Gβ(x, y) :=
∞ m=0
βmXm(x, y) (x, y∈ B)
and spacesGβ asthereproducingkernelHilbertspacesgeneratedbythese kernels.
Wecanalsowrite
Gβ(x, y) =
∞ m=0
δm
k=1
βm
Am
Ymk(x) Ymk(y) (x, y∈ B)
by(10).Nothingabouttheboundedness,summability,ormonotonicityofβ isassumed atthispoint.Butthecondition(16),via(13),ensuresthattheseriesdefiningthekernels Gβ converge absolutely and uniformly on compact subsets of B× B and hencedefine harmonicfunctionsof x,y∈ B.Forany β,
Gβ(0, y) = Gβ(x, 0) = 1 (17)
by(5) since β0 = 1, Gβ(x,y)= Gβ(y,x), and Gβ is real-valued. The Gβ depend onx andy viax· y sincetheZm areconstantmultiplesof Gegenbauerpolynomials ofx· y;
see[12,(14.8)]
Theorem5.2. Theelements of Gβ are harmonicfunctions onB.
Proof. This is by [5, p. 43]; the result there is stated for sesquiholomorphic kernels, butworks equallywell for harmonic kernels since both function classeshave thesame topology,thetopologyofuniformconvergenceoncompactsubsets.Thehypothesesthere aresatisfied,becauseeachGβ(x,y) islocallybounded by(13) andaharmonicfunction ineachvariableonB.
AlsoeveryGβ− βmXm= ∞
l=mβlXl ispositive definite.Then by[5,Theorem II.1.2], everyHm iscontinuouslyimbeddedineachGβ.
Theorem5.3. Thespace Gβ coincideswith thespaceofharmonic functionsu on B with expansionsasin Theorem3.3forwhich
u2Gβ :=
∞ m=0
um2Gβ:=
∞ m=0
Am
βm um2L2 <∞ (18) equippedwith theinnerproduct
u, v
Gβ :=
∞ m=0
um, vm
Gβ :=
∞ m=0
Am
βm
um, vm
L2. (19)
Moreover,
Wmkβ :=
βm
AmYmk: k = 1, . . . , δm, m = 0, 1, 2, . . .
isanorthonormal basisforGβ.
Proof. Weadapttheproofof[5,Theorem III.3.1] tooursituationmainlytoshowhow theL2 normcomesin.Considerthespaceofu∈ h(B) satisfyingthefinitenesscondition