TRUNCATED TOEPLITZ OPERATORS AND BOUNDARY VALUES IN NEARLY INVARIANT SUBSPACES ANDREASHARTMANN&WILLIAMT.ROSS 1 1 0 ABSTRACT. WeconsidertruncatedToeplitzoperatoronnearlyinvariantsubspacesoftheHardy 2 spaceH2. Ofsomeimportanceinthiscontextistheboundarybehaviorofthefunctionsinthese n spaceswhichwewilldiscussinsomedetail. a J 9 1 1. INTRODUCTION ] V C If H2 is the classical Hardy space [Dur70], we say a (closed) subspace M ⊂ H2 is nearly . invariant when h f at f ∈M,f(0) =0⇒ ∈M. z m Thesesubspaceshavebeencompletelycharacterizedin[Hit88,Sar88]andcontinuedtobestud- [ ied in [AR96, HSS04, MP05, KN06, AK08, CCP10]. In this paper, we will examine certain 1 propertiesofnearly invariantsubspaces. Weaimto accomplishthreethings. v 1 First,inTheorem3.1weobservethatmuchofwhatwealreadyknowabouttruncatedToeplitz 7 7 operators on model spaces [Sar07] can be transferred, via a unitary operator to be introduced 3 below,mutatismutandistotruncated operators onnearly invariantsubspaces. . 1 Secondly, in Theorem 4.5 we will show that every function in a nearly invariant subspace M 0 1 has a finite non-tangential limit at ∣ζ∣ = 1 if and only if (i) g has a finite non-tangential limit at 1 ζ and (ii) the reproducing kernel functions for M are uniformly norm bounded in Stolz regions : v withvertexat ζ. Thisparallelsaresult byAhern and Clark inmodelspaces. i X Third, in order to better understand the self-adjoint rank-one truncated Toeplitz operators on r a nearlyinvariantsubspaces,wewilldiscussthenon-tangentiallimitsoffunctionsinthesespaces. It turns out (Theorem 5.2) there is a kind of dichotomy: If every function in a nearly invariant subspace M has a boundary limit at a fixed point ∣ζ∣ = 1 then either ζ is a point where every 1 function in M (where g is the extremal function for M - see definition below) has a finite g non-tangentiallimitoreveryfunctioninM hasnon-tangentiallimit0 at ζ. Date:January20,2011. 1991MathematicsSubjectClassification. 30B30,47B32,47B35. Key wordsandphrases. truncatedToeplitzoperator,kernelfunction,nearlyinvariantsubspace, non-tangential limits. ThisworkhasbeendonewhilethefirstnamedauthorwasstayingatUniversityofRichmondastheGaineschair inmathematics.Hewouldliketothankthatinstitutionforthehospitalityandsupportduringhisstay.Thisauthoris alsopartiallysupportedbyANRFRAB. 1 2 ANDREASHARTMANN&WILLIAMT.ROSS A word concerning numbering in this paper: in each section, we have numbered theorems, propositions,lemmas,corollaries andequationsconsecutively. 2. PRELIMINARIES If H2 is the Hardy space of the open unit disk D (with the usual norm ∥ ∥) and I is inner, let K = H2 IH2 be the well-studied model space [Nik86]. Note that H2,⋅as well as K , are I I regarded, via⊖non-tangential boundary values on T = ∂D, as subspaces of L2 = L2(T,dθ/2π) [Dur70]. Itis wellknownthatH2 isa reproducingk∶ernel Hilbertspacewithker∶nel 1 k (z) = , λ ∶ 1 λz − as isK withkernel I 1 I(λ)I(z) kI(z) = − . λ ∶ 1 λz Note that kI is bounded and that finite linear comb−inations of them form a dense subset of K . λ I AlsonotethatifP is theorthogonalprojectionofL2 onto K , then I I (P f)(λ)= f,kI , λ∈D. I λ ⟨ ⟩ Note that for every λ ∈ D this formula extends to f ∈ L1 = L1 T,dθ 2π so that we can define P forL1-functions(which ingeneral doesnotgivean H1-func(tion)./ ) I By a theorem of Ahern and Clark [AC70], every function f ∈ K has a finite non-tangential I limitat aboundarypointζ ∈T, i.e., f ζ = limf λ ( )∶ ∠λ→ζ ( ) exists for every f ∈ K , if and only if ζ ∈ ADC I . Here ζ ∈ ADC I means that I has a finite I ( ) ( ) angularderivativeinthesenseofCarathe´odory,meaning limI λ =η ∈T ∠λ→ζ ( ) and limI′ λ exists. ∠λ→ζ ( ) Moreover,wheneverζ ∈ADC I , thelinearfunctionalf ↦f ζ is continuouson K givingus I ( ) ( ) akernel functionkI forK at theboundarypointζ. That istosay ζ I f ζ = f,kI f ∈K . ζ I ( ) ⟨ ⟩ ∀ Sarason [Sar07] begana study,taken up by others, ofthe truncatedToeplitz operatorson K . I These operators are defined as follows: For ϕ ∈ L2, define the truncated Toeplitz operator A ϕ denselyon theboundedf ∈K by I A f =P ϕf . ϕ I ( ) Let denotetheA whichextendtobeboundedoperatorsonK . Certainlywhenϕisbounded, I ϕ I T then A ∈ , but there are unbounded ϕ which yield A ∈ . Moreover, there are bounded ϕ I ϕ I T T truncatedToeplitzoperatorswhichcannotberepresentedbyaboundedsymbol[BCF+10]. Much TRUNCATEDTOEPLITZOPERATORSANDBOUNDARYVALUESINNEARLYINVARIANTSUBSPACES 3 is known about these operators (see the Sarason paper [Sar07] for a detailed discussion)but we listafew interestingfacts below: (1) is aweakly closedlinearsubspaceofoperators onK . I I T (2) A ≡0ifand onlyifϕ∈IH2 IH2. ϕ + (3) The operator Cf = zfI (considered as boundary functions on T) defines an isometric, ∶ anti-linear, involution on K for which CA C = A∗ = A . The operator C is called a I ϕ ϕ ϕ conjugation. Thismakes acollectionofcomplexsymmetricoperators[GP06, GP07]. I T (4) A boundedoperatorAon K belongsto ifand onlyiftherearefunctionsϕ ,ϕ ∈K I I 1 2 I T so that A A AA∗ = ϕ kI kI ϕ . z z 1 0 0 2 − ( ⊗ )+( ⊗ ) In theabove,weusethenotationf g fortherank-oneoperator ⊗ f g h = h,g f. ( ⊗ )( )∶ ⟨ ⟩ (5) An operatorAon K isarank-onetruncatedToeplitzoperatoron ifand onlyifitcan be I writtenas constantmultipleofoneofthefollowingthreetypes kI CkI, CkI kI, kI kI λ λ λ λ ζ ζ ⊗ ⊗ ⊗ where λ ∈ D, ζ ∈ ADC I . The two first ones are the non-selfadjoint and the last ones ( ) are theselfadjointtruncatedToeplitzoperators. (6) Sedlock[Sed](seealso[GRW])showedthateverymaximalalgebrain canbewritten I T as thecommutantofageneralizationoftheClark unitaryoperator. (7) In [CGRW10] they show, for two inner functions I and I , that is spatially isomor- 1 2 I1 T phic to if and only if either I = ψ I φ or I = ψ I z φ for some disk I2 1 2 1 2 T ○ ○ ○ ( ) ○ automorphismsφ,ψ. As we have already mentioned, a nearly invariant subspace M is a closed subspace of H2 such that f f ∈M,f 0 =0⇒ ∈M. ( ) z A result ofHitt [Hit88] says that if g (called the extremalfunction for M) is theuniquesolution totheextremalproblem (2.1) sup Reg 0 g ∈M, g =1 , { ( )∶ ∥ ∥ } thenthere isan innerfunctionI so that M =gK I and moreover,themap U K M, U f =gf, g I g ∶ → isisometric. Note that necessarily we have the condition I 0 = 0 since g ∈ M = gK and so 1 ∈ K . If I I one were to choose any g and any inner I, then g(K) is, in general, not even a subspace of H2. I 4 ANDREASHARTMANN&WILLIAMT.ROSS Sarason showsin[Sar88]thatwhen I 0 =0,everyisometricmultiplieron K takes theform I ( ) a (2.2) g = , 1 Ib − where a and b are in the unit ball of H∞ with a2 b2 = 1 a.e. on T. As a consequence, gK is I a(closed)nearly invariantsubspaceofH2 with∣ e∣x+tre∣m∣ alfunctiong as in(2.1). Remark 2.3. From now on, whenever we speak of a nearly invariant subspace M = gK , we I will always assume that I 0 = 0 and that g is extremal for M (as in (2.1)) and of the form in ( ) (2.2). We will say that gK is a nearly invariant subspace with extremal function g and suitable I innerfunctionI withI 0 =0. ( ) Our first step towards defining a truncated Toeplitz operator on the nearly invariant subspace M =gK , as Sarason did forK ,is tounderstandP , theorthogonalprojectionofL2 ontoM. I I M Lemma 2.4. Let M = gK be a nearly invariant subspace with extremal function g and associ- I atedinner functionI, I 0 =0. Then ( ) P f =gP gf , f ∈M. M I ( ) Proof. To show that the map f gP gf is indeed P , we need to show that this map is the I M identityon M andvanishesonL↦2 M(. ) ⊖ Letf =gh∈gK withh∈K . Thenforevery λ∈D wehave I I P ggh λ = ggh,kI = gh,gkI . I λ λ ( )( ) ⟨ ⟩ ⟨ ⟩ But,fromourpreviousdiscussion,multiplicationbytheextremalfunctiong isanisometryfrom K ontoM andso I gh,gkI = h,kI =h λ . λ λ ⟨ ⟩ ⟨ ⟩ ( ) Thus g λ P gf λ =f λ , I ( )( )( ) ( ) inotherwords,f gP gf is theidentityonM. I ↦ ( ) Tofinishtheproof,weneed toshowthat themapf gP gf vanisheson L2 M. Clearly, I when f ∈H2 = h h∈H2,h 0 =0 then P gf =0.↦When(f ∈)H2 M we hav⊖e 0 I { ∶ ( ) } ( ) ⊖ P gf λ = gf,kI = f,gkI =0 I λ λ ( )( ) ⟨ ⟩ ⟨ ⟩ sincegkI ∈M and f M. λ ⊥ ∎ Corollary2.5. Thereproducingkernel forM =gK isgiven by I 1 I λ I z kM z =g λ g z − ( ) ( ). λ ( ) ( ) ( ) 1 λz − Proof. ObservethatkM z = P k z =g z P gk z and λ ( ) ( M λ)( ) ( )( I λ)( ) P gk z = gk ,kI = k ,gkI = gkI λ =g λ kI z . I λ λ z λ z z λ ( )( ) ⟨ ⟩ ⟨ ⟩ ( )( ) ( ) ( ) ∎ TRUNCATEDTOEPLITZOPERATORSANDBOUNDARYVALUESINNEARLYINVARIANTSUBSPACES 5 3. TRUNCATED TOEPLITZ OPERATORS ON NEARLY INVARIANT SUBSPACES Weare nowabletointroducetruncatedToeplitzoperatorsonnearly invariantsubspaces. Cer- tainlywheneverϕ isaboundedfunctionwecan useLemma2.4tosee thattheoperator AMf =P ϕf =gP gϕf , f ∈M, ϕ M I ∶ ( ) iswell-defined and bounded. The most general situation is when the symbol ϕ is a Lebesgue measurable function on T and g 2ϕ ∈ L2. Then, for every bounded h ∈ K , the function g 2ϕh belongs to L2 and so I P g∣ 2∣ϕh ∈K , and,by theisometricmultiplierproperty ofg inK∣ ∣, weget I I I (∣ ∣ ) 2 P ϕh =gP g ϕh ∈gK =M. M I I ( ) (∣ ∣ ) Notethatbytheisometricpropertyofg,thesetgK∞,whereK∞ isthesetofboundedfunctions I I inK , is densein gK . Hence, inthissituation,AM is denselydefined. I I ϕ Let M denote the densely defined AM, g 2ϕ ∈ L2, which extend to be bounded on M and ϕ T ∣ ∣ recall that U K M,U f =gf,isunitary. g I g ∶ → Theorem3.1. LetM =gK beanearlyinvariantsubspacewithextremalfunctiongandsuitable I inner function I, I 0 = 0. Then for any Lebesgue measurable function ϕ on T with g 2ϕ ∈ L2 ( ) ∣ ∣ we have Ug∗AMϕ Ug =A∣g∣2ϕ. Proof. Let h∈K ,then I AMϕ Ugh=AMϕ gh=gPI g 2ϕh=UgA∣g∣2ϕh. ∣ ∣ ∎ Theaboveshowsthatthemap AM A ϕ ∣g∣2ϕ ↦ establishesaspatialisomorphismbetween M (theboundedtruncatedToeplitzoperatorsonM) T and (theboundedtruncatedToeplitzoperatorsonK )induced,viaconjugation,bytheunitary I I T operatorU K M =gK ,i.e., M =U U∗. g I I g I g ∶ → T T InviewoftheresultsinSarason’spaper[Sar07]onecanusetheabovespatialisomorphismto provethefollowingfacts: (1) isaweakly closedlinearsubspaceofoperators onM. M T (2) AM ≡0ifand onlyif g 2ϕ∈IH2 IH2. ϕ ∣ ∣ + (3) Recalling the conjugation C mentioned earlier, define C = U CU∗. This defines a g g g ∶ conjugation on M and C AC = A∗ for every A ∈ M, and since AM ∗ = AM (as g g T ( ϕ ) ϕ can be verified by direct inspection), we see that M is also a collection of complex T symmetricoperators(withrespect toC ). g 6 ANDREASHARTMANN&WILLIAMT.ROSS (4) IfS =U A U∗,thenaboundedoperatorAonM belongsto M ifandonlyifthereare g g z g ∶ T functionsϕ ,ϕ ∈M so that 1 2 A S AS∗ = ϕ kM kM ϕ . g g 1 0 0 2 − ( ⊗ )+( ⊗ ) (5) Recall from our earlier discussion of [Sar07] that A is a rank-one truncated Toeplitz operatoronK ifandonlyifitcanbewrittenasconstantmultipleofoneofthefollowing I threetypes(two ofwhichare non-selfadjoint,and oneofwhich isselfadjoint) kI CkI, CkI kI, kI kI λ λ λ λ ζ ζ ⊗ ⊗ ⊗ whereλ∈D,ζ ∈ADC I . It followsimmediatelythat forλ∈D, ( ) U kI CkI U∗ =U kI U CkI =gkI gCkI, g λ λ g g λ g λ λ λ ( ⊗ ) ⊗ ⊗ U CkI kI U∗ =U CkI U kI =gCkI gkI, g λ λ g g λ g λ λ λ ( ⊗ ) ⊗ ⊗ are rank-one truncated Toeplitz operators on M = gK . What are the self-adjoint trun- I cated Toeplitz operators on M? The result here is the following: Suppose that ζ ∈ ADC I , then ( ) gkI gkI ζ ζ ⊗ is a self-adjoint rank-one truncated Toeplitz operator on M. Conversely, if A is a non- trivial rank-one self adjoint truncated Toeplitz operator on M, then there exists a ζ ∈ ADC I sothatA=gkI gkI. ( ) ζ ⊗ ζ (6) Recall from ourearlierdiscussionthateverymaximalalgebrain can bewrittenasthe I T commutant of a generalization of the Clark unitary operator. Conjugating by U we get g an analogousresultforthemaximalalgebrasof M. T (7) Again from our earlier discussion, for two inner functions I and I , that is spatially 1 2 I1 T isomorphicto if and only if either I = ψ I φ or I = ψ I z φ for some disk I2 1 2 1 2 T ○ ○ ○ ( )○ automorphisms φ,ψ. Using Theorem 3.1 we get the exact same result for M1, M2. T T Noticehowthisresult isindependentofthecorrespondingextremalfunctionsg and g . 1 2 4. EXISTENCE OF NON-TANGENTIAL LIMITS With trivial examples one can show that there is no point ζ ∈ T such that every f ∈ H2 has a non-tangentiallimitatζ. However,therearemodelspacesK andζ ∈Tsothateveryf ∈K has I I a non-tangentiallimitat ζ. Thissituationwas thoroughlydiscussedby Ahern and Clark [AC70] withthefollowingtheorem. Forζ ∈Tand α >1 let z ζ Γ ζ = z ∈D ∣ − ∣ <α α ( )∶ { ∶ 1 z } −∣ ∣ bethestandardStolzdomainswithvertexat ζ. Theorem 4.1 (Ahern-Clark). Let I be an inner function. Every function f ∈ K has a non- I tangential boundary limit at ζ ∈ T if and only if for every fixed Stolz domain Γ ζ at ζ, the α ( ) family F = kI λ∈Γ ζ α,ζ λ α ∶ { ∶ ( )} TRUNCATEDTOEPLITZOPERATORSANDBOUNDARYVALUESINNEARLYINVARIANTSUBSPACES 7 isuniformlynormbounded,i.e., 1 I λ 2 sup{∥kλI∥2 = −1 ∣ (λ2)∣ ∶λ∈Γα(ζ)}<∞. −∣ ∣ Theoriginal proof ofthis theorem involveda technical lemmausing operatortheory. We will givea new proofwhich avoids thistechnical lemma. This willallowus to considerthesituation ofnearly invariantsubspaceswheretheoperatortheorylemmaofAhern-Clark nolongerworks. Also note that the uniform boundedness of the families F is equivalent to the condition ζ ∈ α,ζ ADC I . ( ) Proofof Theorem 4.1. From the uniform boundedness principle it is clear that if every function in K has a non-tangential limit at ζ ∈ T, then the kernels kI are uniformly bounded in Stolz I λ domainsΓ ζ . α ( ) Letusconsiderthesufficiencypart. If,forfixedαandζ,thefamilyF isuniformlybounded, α,ζ then there exists a sequence Λ = λ ⊂ Γ ζ such that kI converges weakly to some kI,Λ ∈K , i.e.,foreveryfunction∶f {∈Kn},n≥1 α( ) λn ζ I I (4.2) f λ = f,kI f,kI,Λ = fΛ as n . ( n) ⟨ λn⟩→⟨ ζ ⟩ ∶ ζ →∞ Observethat f λ f 0 Azf λ = ( )− ( ) λ ( )( ) and f λ N−1f j λj ANz f λ = ( )−∑λjN=0 ̂( ) , ( )( ) wheref j isthej-thFouriercoefficient off. As aresultwehave ̂ ( ) ∞ ANf 2 = f j 2 0 asN . ∥ z ∥ j∑=N∣̂( )∣ → →∞ Apply(4.2)to thefunctionsANf to get z f λ N−1f j λj AzNf,kλI = AzNf λ = ( )−∑λjN=0 ̂( ) . ⟨ ⟩ ( )( ) Hence f λ N−1f j λj ⟨AzNf,kζI,Λ⟩ = nl→im∞⟨AzNf,kλIn⟩=nli→m∞ ( n)−∑λjNn=0 ̂( ) n fΛ N−1f j ζj = ζ −∑j=0 ̂( ) . ζN 8 ANDREASHARTMANN&WILLIAMT.ROSS Nowusingthefact that ANf 0 when N , weobtainfrom ANf,kI,Λ 0 that ∥ z ∥→ →∞ ⟨ z ζ ⟩→ fΛ N−1ζjf j lim ζ −∑j=0 ̂( ) =0, N→∞ ζN and, sincethedenominatorin theabovelimitisharmless,weget ∞ f j ζj =fΛ. ̂ ζ j∑=0 ( ) Thismeansthat theFourierseries forf convergesat ζ. NowtakeanysequenceΛ = λ convergingnon-tangentiallytoζ. Inviewoftheuniform n n≥1 boundedness of the family F∶ {, th}ere exists a weakly convergent subsequence Λ′ = kI . α,ζ ∶ { λnl}l≥1 Repeating theaboveproof, weobtain ∞ ζjf j =fΛ′. ̂ ζ j∑=0 ( ) In other words, for every sequence λ converging non-tangentially to ζ there exists a sub- n n≥1 sequence λ such that f λ { } convergesto thesamelimit nj j≥1 nj j≥1 { } { ( )} ∞ l = f j ζj. ζ ∶ j∑=0 ̂( ) It followsthatf has afinitenon-tangentiallimitat ζ. ∎ We will now prove an analog of Theorem 4.1 for nearly invariant subspaces where the sit- uation is a bit more complicated. Certainly a necessary condition for the existence of finite non-tangential limits for every function in gK at ζ is that g has a finite non-tangential limit at I ζ (since indeed g is extremal and so g ∈ gK ). As we will see with the next example, it is not I possible to deduce the existence of the non-tangential limits for every function in gK merely I from theuniformnormboundednessofthereproducing kernelsin aStolzangles. Example 4.3. Givenany innerfunction I, with I 0 =0, wehavealready cited Sarason’s result [Sar88]statingthat everyisometricmultiplieron K( )isoftheform I a (4.4) g = , 1 Ib − whereaandbarefunctionsintheballofH∞ with a2 b2 =1a.e.onT. LetΛ = 1 4−n 1 n≥1 and Λ = 1 2−n , which are both interpolatin∣g∣se+qu∣e∣nces. Observe that Λ co{nsi−sts of}the 2 n≥1 2 sequence{Λ −to wh}ich we have added in a certain way the (pseudohyperbolic)midpoints of two 1 consecutive points of Λ . Let B be the Blaschke product whose zeros are Λ . In particular, B 1 i i 1 vanishesonΛ andisbigonΛ Λ ,say B λ ≥δ >0forλ∈Λ Λ (thiscomesfromthefact 1 2 1 1 2 1 that B is an interpolating Blas∖chke prod∣uct(, a)n∣d that on Λ Λ ∖we are pseudohyperbolically 1 2 1 ∖ farfrom thezeros ofB ). 1 Next,define 1 1 a= B 1 2 + 4 TRUNCATEDTOEPLITZOPERATORSANDBOUNDARYVALUESINNEARLYINVARIANTSUBSPACES 9 which is a function oscillating on Λ between the values 1 2 (on Λ ) and a λ 1 2 ≥ δ 4, 2 1 λ∈Λ Λ . / ∣ ( ) − / ∣ / 2 1 ∖ By construction we have 1 4 ≤ a ≤ 3 4. In particular there is an outer function b in the ball 0 of H∞ such that a2 b0 2 =/1 a.e∣.∣on T/. Moreover 0 < √7 4 ≤ b0 = √1 a2 ≤ √15 4 < 1 a.e. on T and this∣ex∣ te+n∣ds∣by themaximum/minimumprincip/lesto∣th∣edisk (−b ∣b∣eing oute/r). As 0 a consequence, for every inner function J, the function g = a 1 Jb will be a bounded outer 0 function(actuallyinvertible). Wewillusethisfact inparticul/a(rfo−rJ =)IB : 2 a a g = = . 1 Jb 1 I B b 0 2 0 − − =±bin(4.4) Then 1 1 =0 whenλ∈Λ1 g λ = a λ δ ∣ ( )− 2∣ ∣ ( )− 2∣{ ≥ 4 whenλ∈Λ2 Λ1. ∖ Hence g has no limitat ζ =1. Choosenowan innerfunction I such thatζ =1∈ADC I . Then, the kernels kM = g λ gkI are uniformly bounded in a fixed Stolz region Γ 1 at ζ =(1.)Indeed λ λ α g isbounded,theke(rn)elskI areuniformlyboundedinΓ 1 by1∈ADC I ((s)eeTheorem4.1), λ α and kM = g λ kI . ( ) ( ) λ λ ∥ ∥ ∣ ( )∣∥ ∥ Theorem 4.5. Let M =gK bea non-trivialnearlyinvariantsubspacewithextremal functiong I and corresponding inner function I. Then every function in M has a finite non-tangential limit atζ ∈Tif andonlyif thefollowingtwo conditionsaresatisfied: (1) g has afinitenon-tangentiallimitatζ. (2) ForeveryStolzregionΓ ζ atζ,thefamilyFM = kM λ∈Γ ζ isuniformlynorm bounded,i.e., α( ) α,ζ ∶ { λ ∶ α( )} 1 I λ 2 sup{∥kλM∥2 =∣g(λ)∣2 1− (λ)2∣ ∶λ∈Γα(ζ)}<∞. −∣ ∣ Proof. If every function in M has a finite non-tangential limit at ζ, then (1) holds since g ∈ M. Condition(2)holdsby theuniformboundednessprinciple. Nowsupposethatconditions(1)and (2)hold. Case1: ζ ∈ADC I . In this case, everyfunction in M has a non-tangentialboundarylimitat ζ sinceeveryfunctio(n)inK has as wellas g does. I Case2: ζ ∈ ADC I . From the Ahern-Clark theorem (Theorem 4.1), we know that ζ ∈ ADC I ifan/d onlyi(f ) ( ) 1 I λ 2 kλI 2 = −1 ∣ (λ2)∣ ∥ ∥ −∣ ∣ is uniformly bounded in Stolz regions Γ ζ . Hence, if ζ ∈ ADC I , then there is a sequence α Λ = λ in Γ ζ such that kI 2( ) . Recall t/hat kM(=)g λ gkI so that kM = ∶ { n}n≥1 α( ) ∥ λn∥ → +∞ λ ( ) λ ∥ λn∥ 10 ANDREASHARTMANN&WILLIAMT.ROSS g λ kI . Inorderthatthislastexpressionisuniformlyboundeditisnecessarythatg λ ∣0(whne)n∣∥nλn∥ . ( n)→ →∞ Now, since FM is a uniformly bounded family, there exists a subsequence Λ′ = λ so thatkM convergα,eζs weaklyto afunctionkM,Λ′. ∶ { nj}j≥1 λnj ζ Letus introducethefollowingoperator Q =U ANU∗ N g z g from M =gK toitself. Let f =gh∈gK =M. Recall thatU∗f =h. Then I I g h z N−1h m zm QNf z = UgAzNUg∗ f z = UgANz h z =g z ( )−∑mzN=0̂( ) ( )( ) (( ) )( ) (( )( ) ( ) f z g z N−1 = h m zm. ( ) ( ) zN − zN m∑=0̂( ) NowsinceQ f ∈M, weget, fromtheweak convergenceofkM kM,Λ′ as j , N λnj → ζ →∞ f λ g λ N−1 (4.6) ⟨QNf,kζM,Λ′⟩=jl→im∞(QNf)(λnj)=jl→im∞( (λNnnjj) − (λNnnjj) m∑=0̂h(m)λmnj). Thelastsumappearingaboveisaharmlesspolynomialconvergingtosomenumberwhenj . Alsoλ ζ and g λ 0when j . Hence thelimitin (4.6)isequal to →∞ nj → ( nj)→ →∞ fΛ′ Q f,kM,Λ′ = ζ , N ζ ζN ⟨ ⟩ where fΛ′ = f,kM,Λ′ , ζ ζ ⟨ ⟩ whichexistsbyconstruction. Since Q f = U ANh = gANh = ANh 0 as N , N g z z z ∥ ∥ ∥ ∥ ∥ ∥ ∥ ∥→ →∞ wehave fΛ′ lim ζ = lim Q f,kM,Λ′ =0 N→∞ ζN N→∞⟨ N ζ ⟩ whichis possibleprecisely whenfΛ′ =0. ζ Theaboveconstructionisindependentofthechoiceofthesequence λ sothatforevery n n≥1 such sequence there is a subsequence λ such that f λ 0 {whe}n j . As in the proofofTheorem4.1 weconcludethat{ nljim}j≥1 f λ exist(sannjd)i→s, infact, equ→al t∞o0. λ→ζ ∠ ( ) ∎