HOUSTON JOURNAL OF MATHEMATICS Volume 21, No. 1, 1995 UNITARILY EQUIVALENT COMPOSITION RANDALL COMPACT OPERATORS K. CAMPBELL-WRIGHT Communicated by Vern I. Paulsen. ABSTRACT. If •b is an analytic function taking the unit disk D into itself then the compositionoperatorC4 canbe definedonthe Hardy spaceHP(D) for i _•p < co by C• (f) = f o•b.In thisworkit is shownthat if somepower of C• is compactand q•hasa nonzeroderivativesat its uniquefixedpoint insidethedisk,thenC4 is unitarilyequivalent to Cq,onH 2(D) if andonly if the equivalence C4, - U*C•,U can be inducedby a unitary composition operatorU = Ca with a(z) = eiøz. 1. Introduction. If •b is an analytic function taking the unit disk D into itself then the composition operatorC4 can be definedon the Hardy SpaceHP(D) for I _<p < oo by C•(f) = f o •b.An overviewof the propertiesof these operatorscanbe foundin the recentbooksby Cowenand MacCluer[7]or Shapiro[13]. Thiswork explores when twocompact orpower compact (thatis,C•/ compactforsomepositiveintegerN) composition operators C• andCe are unitarilyequivalent(that is, C4 = U*C•U for someunitaryoperatorU)on the HardyHilbertspaceH2(D). In the caseof compactor powercompact compositionoperators,•band • haveuniquefixed pointsin D (denotedby a andb respectively) [5,page128].Shapiro[12]hascompletely characterized 1991 Mathematics Subject Classification. Primary 47B38 Secondary 47B07. Key words and phrases. Composition operator, unitary equivalence. Research supported by a Dana Foundation faculty development grant from the University of Tampa. 189 RANDALL 190 K. CAMPBELL-WRIGHT the compact compositionoperatorsin terms of the Nevanlinna counting function.The resultof this workis that for C• compactor powercompact with•(a) • 0 it turnsoutthat C• isunitarilyequivalent to C• onH2(D) if and only if the equivalence C• - U*C•U can be inducedby a unitary composition operator U - C•. This• is necessarily oftheform•(z) - eiez for somereal number•) [11,chapter2], sothat the operatorequivalence is thesameasthe functional equation •(z)The author[2], [3], [4]haspreviously exploredothercomposition operator equivalences and foundthat the obviousequivalences (inducedby an invertiblecomposition operator)arenot alwaysthe onlyones[3, section 4]. The presenttheoremshouldbe compared with the followingresults,in which the obviousequivalencesare the only ones. Define the iterates of •bby •bl - •b, •bn- •bo •bn-1 for n a positive integer. Thenotation •b© willbeusedforderivatives. Theorem A. [3, corollary3.2] Suppose qb,not a diskautomorphism, has its fixedpointa in D andsuppose qbn (0) is neverequalto a for n a positive integer.ThenC• is unitarilyequivalent to C• onH2(D) if andonlyif q•(z)-- ei•b(e-i• z) for somerealnumber O. TheoremA coversmany of the sameoperatorsasthe theoremin this article, but there are alsomany examplesof compactcompositionoperators for which•bn(0) - a for somen. For instance,if •5(z) = O.5(1]a])z1-•'• -ka with a in /]), then •b(a) = a and •b(O)= a. This C4 is compactbecause I1•11•< 1 [11,page28].(Anevensimpler example is •(z) -- 0.5z.) Theorem B. [3, Theorem3.5]. C4 is isometrically isomorphic to C• on HP(D)for 1 •_p • c•, p • 2 if andonlyif qb(z) - ei•b(e-•ez) for some real number O. TheoremB makesno hypotheses on C4 and C•, but there are very few isometricisomorphisms on HP(D) with p • 2 [9]. The Hardy Hilbert spaceH2(D) hasmanyisometric isomorphisms (unitaryoperators). UNITARILY EQUIVALENT COMPACT COMPOSITION OPERATORS 191 TheoremC. [•, Theorem 5]. Suppose C• is compact forsome positive integerN andqf(a) • O. ThenCq•is similarto C• onHP(D) for I •_p < c• if andonlyif q)-- a o•boa -1 ,forsomediskautomorphism a. The proof of Theorem C is a direct calculationin the spirit of the proof below,but is muchmoreinvolved because similarity(C4-- S-1C•S forS invertible)is a moreflexiblepropertythan unitary equivalence.Whereas unitary equivalencepreservesanglesand normsexactly,similarity only preservesthem approximately. 2. The main theorem. TheoremD. Suppose C• is compact for some positive integer N and qf(a)• O. ThenCq•is unitarilyequivalent to C• onH2(D) if andonlyif qJ(z)- eiø•b(e -iøz) for somerealnumber O. Guyker[10]hasdemonstrated the effectiveness of usingthe orthonormal basis b•-1 -1a12)1/2 (n--O, 1,2,...) - (1 az iz-a - az for studyingthe teducingsubspaces of compositionoperators. One reason for the naturalness of thisbasisisthat (b•} is an orthonormalization of the derivatives kernels K? ) = (•_uz)•+• '•!z• ,so named because If,K?))- f('•)(a). It followsfromGuyker'sdescription [10,page369]of b,•that k--O Onecancalculate C•K• © byusing FakdiBruno's formula [8],[1, RANDALL K. CAMPBELL-WRIGHT 192 pages 823]forthederivative ofacomposition offunctions. ,,-,* u'(,•) <f,•o• >-<cof'K? >-- (f o½5)(") m=O • j a• =n • aj •m ={f, • K• •) • n](•(a))•(•"(a))•2'"(•( •) m=O • j a• =n • aj=m (using thefactthat•(a) = a) Thus, n!(½5,(a))•(½5"(a)) •' . . . (½5('•) (a)) m=O • j a• =n (1!)•al!(2!)•'a2 !.-- Note thatforn>_1,C;K• © isalinear combination ofderivative kernels K•© withm > 1.Thisfactwillcome uplater. Thematrix forC0withrespect tothebasis {b,}(orthederivative dernels {Ka(n)})is lower triangular with diagonal [1,½5'(a),½5'(a)2, ...] [10, page 369] sotheadjoint C;isupper triangular. The adjoint C;iscompact andhasone-dimensional eigenspaces forA: ½5'(a) • [6,page 93],sothe .k- ½5'(a) • eigenvectors ofC•arelinear combinations ofbo, b•,b2, ..., b,, (orofK•,K'•,..., K•('0) ß Proofoftheorem D. <=. Obvious. • . Several restrictions canbemade on½5 and•bwithout loss of generality. IfCoisunitarily equivalent toC•then C•isalso compact and ½5'(a) =•b'(b) [4,Lemma 1].Since Cff=C0t• one can assume that N= 1. One canalso assume thata = b,since [a[= lb[[3,corollary 3.3]andone could conjugate byarotation tomake thefixed points of½5 and •bthesame. Thecases a = 0anda •: 0require substantially different approaches and will be dealt with separately. UNITARILY EQUIVALENT COMPACT COMPOSITION OPERATORS 193 Case 1. a: b: 0, ½5'(0): 0'(0) •= 0, both C0 and C• are compact. The short proof is that C0 unitarily equivalentto C• implies that •5= a o0 oa-• where• is a diskautomorphism taking0 to 0 [4, Theorem 5 - quotedasTheoremC in thispaper].The onlysuchdiskautomorphism is a rotation•(z) = e•øz. However, it should not be necessaryto quote a similarity theorem (with a long combinatorialproof!) in orderto provean analogous result about unitary equivalence.Unitary equivalenceshouldbe easy enoughto calculate with directly. The followingdirect unitary equivalencecalculation is givenin hopesthat it will be easierto followthan the proofof [4], thus makingthe presentpaperindependent of [4]andmaking[4]moreaccessible. The followinglemma providesa meansfor linking •band 0. It is easily provedby equating Maclaurin series. Lemma A. 65(z)= eiøO(e-'øz ) for somerealnumber0 if andonlyif O(•)(o)= •-(•-•)iøO(•)(o)/orall,• _>o. For a = 0, Guyker's basis reducesto the standard orthonormal basis b,•= z'• (n = 0,1,2,... ) andthederivative kernels areK?) = n!z'•. It followsfrom Formula(2) that (O,(o))•(O"(o))a'---(0(•)(o))• m-'-O • j aj -'-n NowC; andC• areuppertriangular with respect to the standard orthonormal basis and both have one-dimensionaleigenspacescorrespond- ing to the diagonalentriesA = •b'(a)'• [6, page93]. Thus the subspace ofHa(D)spanned by{1,z,z2,... ,z'•} isinvariant under C; orC•. Since C•,= U*C•U andC• = UC•U*, it follows that U andU* arebothupper triangular- and hencediagonal.Thus U(z'•) = c,•z'• wherethe diagonal entries c,• of U have absolute value 1. RANDALL 194 K. CAMPBELL-WRIGHT ApplyUC; andC•U to z• to getUC;(z'•) = C•U(z'•),yielding • m!Cm zm (3) m=0 • j aj =n • ay--m (0'(o))"'(0"(o))",..- (0(•)(o) (l])"xal 1(2!)"2a21.-- (½,(o))-,(½,,(o))-•-... (½(•)(o))-• m=0 • j aj =n (l!)"xal !(2!)"2a2!--- Equatingthe coefficients of z"• in Formula(3) for m = n - 1 yields (n-1)!c,•_• [q•'(O)]'•-•' •"(0) _(n-1)!c,• [•'(0)]'•-•' •"(0) (,•-•)! • (,•-•)! • Thisimplies •b"(0)= c•C---•_• %b"(0). Equating thecoefficients ofzminFormula (3) for m = n- 2 yields '(O)]n--3C•m(O) [(•'(0)] n--4 [•)t/(0)]2 } (n-3)!3! + [¾f (n-4)! 8 =(n _2) ,c,• {[•'(0)]'•-3•'"(0) (0)] n--4 ['•Dt' (0)] 2}' (n-2)!c•-2 (n-3)!3!+ (n-4)! 8 This reduces to •b"'(0)= C•--c,%b"'(0) (usingthe factthat •b"(0)= Cr•c• 2 -- 1 for all n). Proceedingthrough all the coefficients of zm in Formula (3) yields •b(•)(0)= c,•p(•)(0) forall k,n withk < n. Cr•--k + l Case 1A. a- b- O, •'(0) - •b'(O)• O, both • •b"(0)• 0 and %b"(0) • 0. and •p are compact, Onecantakeeiø- •"(o) •"(o)_ c•_• c• foranyn. Thenforallk, k-1 On-k-l-1 UNITARILY EQUIVALENT COMPACT COMPOSITION OPERATORS 195 sothatd(•)(0): e-(•-•)iø!b(•)(0) andd(z): eiø!b(e-•øz ) byLemma A. Case lB. a = b = 0, 9•(0) = !b•(0)• 0, bothC• andC½are compact, = ½"(0)=0. If 4"(0) = 0 thenthis gap(andany subsequent ones)cancausedif- cultyinproving that•(k)(0)= e-(k-•)•øtb(k) (0) (sothatonecanuseLemma A). Tothat end,consider œ= {j- 1' •(•)(0) • 0 andj- 1 isnota multiple of somesmallerelementof œ}.œis a finitesetwhichgenerates (by taking integerlinear combinations of elementsof œ) the samealgebraicideal 27 that is generated by {j - 1 ß9½•)(0) • 0}. Let G bethegreatest common factor ofœ(also theGCFof27).Forj - 1inœ' •pO') •(j)(ø) _ c•_j+• c• foranyn. (0) Since G is some integer linear combinationof elementsof œ it followsthat c, is constantfor any n. Thus Cr•--G c"can be expressed as ½Giofor some Cr•--G •. Thenforanyk,•(•)(0) • 0 implies that Cn-k+l = ½(k-1)iO•(k)(O), sothat•(k)(0)= e-(k-•)iø!b(k)(0). ByLemma A, •(z)= eiø!b(e-iøz ). Note that in Case 1A, 0 is uniquely determined,while in Case lB several(G to be precise)valuesof 0 sufficeto make½GiO constantand •(z) = e•ø!b(e-•øz). If oneconstructed the setœin Case1A, it wouldbe œ= {1}, whichgenerates the ideal27= g. Case 2. a = b • 0, 9•(a): !b•(a)• 0, bothC• andC½arecompact. Letf• = b•+ •=o z•,•b•denote a generator oftheone-dimensional eigenspace of C• corresponding to ,X= 9•(a)'• [•, page[}3]andlet #,• = b•+ Y•=o!/•,•b•denote a generator ofthe,X= 9•(a)• eigenspace ofC}. One can use Formula (1)forb•interms ofK?) and Formula (2)for to find that fo = go = boand f• = #• = b• + abo.It is not necessary to completelycalculatethe higherordereigenvectors in orderto completethis proof. Someof thesecalculationswill be broughtup as needed. C4 unitarilyequivalent to C½implies that G : U*C•U forsome unitary operator U and Uf,, = c,,#,, with c• constant. Sincef0 = go = bo, Icol= 1. In fact,onecanassume withoutlossof generality that co= 1 (otherwisedivideU by co). 196 RANDALL K. CAMPBELL-WRIGHT One can now prove by induction that c• = 1 and the' derivatives •b(•)(a)= •p(•)(a)forall n, sothatU = I and•b= •p.Thiswillcomplete the proof.It, is alreadyknownthat •b(ø)(a) = a = •b(ø)(a), andco= 1 by assumption.For n = 1, it hasbeenobserved that •b•(a)= •p•(a).Also, a = (b• +abo,bo) = (f•,bo) = (Ufo, Ubo)= (c•b• + c•abo,bo)= cxa.Thus, cl=l. Assume thatcj = I and&(J)(a)= •(J)(a)forallj < n. An explicit calculationof the coefficient x.,o in the eigenvector f. will helpshowthat c• = 1 and&(•)(a)= •(•)(a). (Note:othercalculations, likex•,•_•, lead to the sameconclusion.) n--1 (4) • =• + • •,• j=0 n--1 =(-a)•(1-i•i•)•/•K•+ • x•,j(-a)J( 1-i•i•)•/•K• j=0 +• (terms involving K?)) k=l (by Formula(1)). UsingFormul• (2) and (4) andequatingthe coe•cients ofK• intheequality C•f• = •(a)•f• yields n--1 (-a)•(1- Ia•)l/•+ •xmj(-a)J(1- al•)•/• j=O n--1 =•,(a)•(-a)•(1 -lal•)l/•+ &'(a) • • /•,•(-a)Y( 1-lal•)X/•, j=O which yields n-1 (-•)• + • •,•(-•/= 0, j=0 n--1 (•) xmo = -[(-a) • + • xmj(-a)J] ß j=l UNITARILY EQUIVALENT COMPACT COMPOSITION OPERATORS 197 One can showthat c• - 1. First observethat for j < n, Then x•,0 = c•,y•,,oand Formula(5) gives n--1 n--1 (-a)•+ • x•,j(-a)5-- c•(-a)•+ • x•,j(-a)5. j=l j=l so that c• = 1. Thus, x•,,5 = y•,,j for j < n and f• = g•. It follows byequating thecoefficients ofK• inC;K• © = C•K• © (see Formula (2)) that •b(•)(a)= •h(•)(a).By induction onn, c• = 1 and•b(•)(a)= •h(•)(a)forall n, so•b= •h. [] 3. Concluding comments. The similaritycalculationin [4] waswritten usingeigenfunctions of C•, but it wouldreadthe sameif it hadbeenwritten usingeigenfunctions of C; andpointevaluation kernels. Thustheunitaryequivalence calculation in Case I of the proof of Theorem D in this paper can be consideredas a specialcaseof the similaritycalculation in [4]. Forthe similarityC• = S-xC•S the calculations revealthat S is lowertriangular,but onemust examineeachdiagonalin the limit as n goesto infinity in order to reach equalitiesbetween•b and •h. This requiresa good deal of combinatorial bookkeeping.By contrast,in the unitaryequivalence C4 = U*C•U it turns out that U is diagonalwith diagonalentriesof absolutevalueone,so that many terms drop out of the analysis. Thus the calculation in Case 1 of this papercanbe viewedasthe coreequalitythat [4]hasto approachin the limit. One of the obstaclesin doing explicit calculationswith composition operatorsis missingderivatives.This paperconsiderspowercompactcom- positionoperators with •b•(a)• 0. If •b•(a)= 0 thenthe spectrum of C• reduces to {0, 1} andmucheigenfunction structureis lost[5],[6, Theorem 4.1]. It is clearthat differentmethodswouldbe requiredto considerequivalences of thesecomposition operators.Someroughprogress is madeusing normestimates in [3, Lemma2.5, Theorem2.6],but nothingapproaching the precisionof the present paper. 198 RANDALL K. CAMPBELL-WRIGHT As another example of the impact of missingderivatives, in Case 1 of TheoremD, where•b(O)= 0 and•b•(0) • O,onecancomplete the calculation in onestepif •b•(0) • 0 but morework(andan algebraicideal)is required if •b"(0)= 0. Future investigationsinvolving closecontact with the finer structure of compositionoperatorswill surelyface more of these "missingderivative" issues. My regardsto the refereeof this paper, who made severalsuggestions to improve the presentation. And my thanks to the referee of an earlier version of this paper, who suggestedthat I do the calculation in Case 1 rather than quotingthe (muchmoredimcult)Theorem5 of [4]. REFERENCES 1. Abramowitz, M. and Stegun, I. A., eds.,, Handbook of mathematical functions, with formulas, graphs, and mathematical tables,Dover, New York, 1965. 2. Campbell-Wright, R.K., On the equivalenceof composition operators, Thesis, Purdue University, 1989. 3. 4. __, , Equivalent compositionoperators,Integ. Eq. Op. Th. 14 (1991), 775-786. Similar compactcomposition operators,Acta Sci.Math. (Szeged)58 (1993), 473-495. 5. Caughran, J.G. and Schwartz, H.J., Spectraof compactcompositionoperators,Proc. Amer. Math. Soc. 51 (1975), 127-130. 6. Cowen,C.C., Composition operators on H 2, J. OperatorTheory9 (1983),77-106. 7. Cowen, C.C. and MacCluer, B.D., Composition operatorson spacesof analytic functions, CRC press, Boca Raton, 1995. 8. Fak di Bruno, F., Note sur une nouvelle formule de calcul diffdrentiel, Quart. J. Pure Appl. Math. i (1857), 359-360. 9. Forelli, F., The isometriesof Z p, Canad. J. Math. 16 (1964), 721-728. 10. Guyker, J., On reducingsubspaces of compositionoperators,Acta Sci.Math. (Szeged) 5a 0989), a69-a76. 11. Schwartz, H.J., Composition operators on H v, Thesis, University of Toledo, 1969. 12. Shapiro,J.H., The essentialnorm of a compositionoperator,Ann. of Math. (2) 125 (1987), 375-404. 13. •, Composition operators and classicalfunction theory, Springer Verlag, New York, 1993. Received November 10, 1993 Revised version received August 1, 1994 DEPARTMENT OF MATHEMATICS, UNIVERSITY OF TAMPA, TAMPA, FLORIDA 33606 E-mail address: rcampwri@cfrvm.bitnet