Uniformly Bounded Representations and Harmonic Analysis of the 2 x 2 Real Unimodular Group Author(s): R. A. Kunze and E. M. Stein Source: American Journal of Mathematics, Vol. 82, No. 1 (Jan., 1960), pp. 1-62 Published by: The Johns Hopkins University Press Stable URL: http://www.jstor.org/stable/2372876 . Accessed: 03/02/2011 04:25 Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at . http://www.jstor.org/action/showPublisher?publisherCode=jhup. . Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. The Johns Hopkins University Press is collaborating with JSTOR to digitize, preserve and extend access to American Journal of Mathematics. http://www.jstor.org UJNIFORMLY BOUNDED REPRESENTATIONS AND HARMONIC ANALYSIS OF THE 2 X 2 REAL UNIMODULAR GROUP.*1 By R. A. KUNZE and E. M. STEIN. TABLE OF CONTENTS INTRODUCTION CHAPTER1. OPERATOR VALUEDFUNCTIONS ? 2. Lp spaces of operatorvalued functions. ? 3. Interpolationin the generalcase. ? 4. The main interpolationtheorem. C(HAPTERII. UNIFORMLY BOUNDED REPRESENTATIONS ? 5. Uniformlyboundedrepresentations. ? 6. Some lemmasfromFourier analysis. ? 7. Proofsof Theorem5 and Theorem6. CHAPTER III. THE FOURIER-LAPLACE TRANSFORM ON THE GROUP ? 8. Hausdorff-Young theoremfor the group and certainof its implications. ? 9. The discreteseries. CHAPTER IV. APPLICATIONS ? 10. Boundednessof convolutionoperator, ?11. Characterization of unitaryrepresentations of G. * Received November 13, 1958. s This research was supported by the United States Air Force under contract No. AF-49 (638) -42, monitored by the Air Force Officeof Scientific Research of the Air Research and Development Command. 1 2 R. A. KUNZE AND E. M. STEIN. Introduction. This paper deals with a study of the real 2 X 2 unimodular group. Our study of this particulargroup is motivatedby two factors. First, this group has an intrinsicinterest,especiallyin view of its connectionwith severalbranchesof Analysis. Secondly,the 2 X 2 real unimodulargroupaffordsan illuminatingexampleforthe studyof othergroups. We constructa familyof uniformly boundedrepresentations of the group, and considerits implicationwithregardto the Fourieranalysisof the group. These representations withthe followingproperties. They all are constructed act on a fixedHilbertspace S9; theyare determinedby a complexparameter S, 0 < R (s) < 1, and dependanalyticallyon the parameters; finally,when R(s) - 1, theserepresentations are, up to unitaryequivalence,the continuous principal series. The above properties, in particularthe analyticity, togetherwith certain convexityargumentsapplied to operatorvalued functionsyieldthe following: 1) The "Fourier-Laplace" transformof a functionf in L1 (G) exists as an operator-valued function5, whosevalues 5 (s) act on , and whichis analyticin s, 0 < R(s) < 1. 2) When f E Lp (G), 1 ? p < 2, the Fourier-Laplacetransforma can still be defined,and is an operatorvalued functionanalyticin the strip, 1 -/p < R(s) < i/p. 3) A detailedanalysisof the proofsof the above revealsthe remarkable 5 of f is unifact: If fE Lp(G), 1 ? p < 2, the Fourier-Laplacetransform 1' formlyboundedin the operatornormalong the line R (s) In conjunctionwithan analysisof the discreteseriesof representations, 3) impliesthe followingsignificant fact concerningharmonicanalysison the group: Let icbe a functionin Lp(G), 1 ? p < 2. In contrastwiththe (noncompact) abelian situation,the transformation f -> f *k, of convolutionby 7c,is a boundedoperatoron L2(G). We shall now discusscertainof thesefactsin greaterdetail. The representationswe considerarise as follows. Let =[c d]caddbc=1 be an elementof the group. We then consider,for each complexs, the multiplierrepresentations,2 These representations may be put in the form originally obtained by Bargmanu [11 bv means of the transformationa - tan (0/2). (1.1) (1. 2) ~ ~ f(x) t(x) + d 128-2 ->Ibx sgn(bx+ d)c)/(bx I bx+ 3 GROUP. REAL UNIMODULAR c)/ (bx + d)) f((ax+ d 128-2 f ((ax + + d)). The two continuousprincipal series are obtained from these by setting the functionsf to lie in L2 (- oo, oo). s j + it and restricting boundedrepresentations of the uniformly We are led to the construction describedaboveby the followingconsiderations.In the groupwe distinguish a particularsubgroup,namely,the subgroupof lowertriangularmatricesof the form 9 ? ], a 0+ . of either of the principal It may be shown that when the representations seriesare restrictedto this subgroupthen they are all unitarilyequivalent. This raises a natural problem. Can one finda Hilbert space X and representationsU+( , - + it), U-( ,1 + it) unitarilyequivalentto (1. 1), (1. 2) + it) such that U+( ,1 + it) when restrictedto the lower tri(for s angular subgroupare independentof t? The answeris in the affirmative; U+( , s), 0 <R (s) <1, boundedrepresentations the uniformly furthermore, which we construct,are characterizedas the analytic continuationsof the U+(*, 1 + it). It should be added that the representations representations series. U+( ,), 0 < a < 1, are unitarilyequivalentto the complementary In solvingthe above problemand in the actual constructionof these it is natural to considerthe inducedaction of (1. 1), (1. 2) representations, here are F of the functionsf. The considerations on the Fouriertransforms ratherinvolvedbut have an intrinsicinterest. For the analysisrevealsconnectionswith both the so called "iHilbert transform"and the notion of "fractionalintegration." (1. 1) or (1. 2) afford(at It is clear that the multiplierrepresentations -+ it. least formally)an analyticcontinuationof (1. 1) or (1. 2) when s The problem,however,is howto makethis precise,i. e., the problemof finding act and depend an underlyingHilbert space on whichthese representations analyticallyon s. Ehrenpreisand Mautnerhave also dealt withthe problem (1. 1) to values of s 7& I + it. Their result of extendingthe representations concerningthe analyticityof the Fourier transformof an L1 spherical functionon G was one of the motivatingfactsin our work. In addition,it was broughtto our attentionby Ehrenpreisthat a similarresultmighthold betweentheirresultsand for Lp. However,thereare significantdifferences arisingfrom representations bounded uniformly ours. In [5] theyconstruct act on different these representations (1. 1) when s 7, i + it; nevertheless, = 4 R. A. KUNZE AND E. -M. STEIN. Hilbertspaces dependingon s. In [6], and [7] theyconsider,at least implicitly,representations whichact on a fixedHilbert space; but in this case the representations are not uniformlyboundedwhen s 7/=1?+it. The precedingconsiderations, in particular3) above, lead to characterizationsof the representations of the group. This may best be understood in thefollowingcontext. As a resultof theworkof Bargmann[1], Godement [8], and Harish-Chandra[14], attentionhas been focusedon a particular class of representations, thosewhichare "square integrable"in the following sense; a representation g -->U0 on a Hilbert space X is of this type if the function = cp=cp(g) (Ug4 is in L2(G) for everyX, in 59. The square integrableirreducibleunitary representations of the 2 X 2 real unimodulargroup are essentiallythe representationsofthe discreteseries. We are able to give a similarcharacterization of the representations of the continuousprincipal series. An irreducible unitaryrepresentation g -> U0 is equivalentto one of the latterif and onlyif ) E Lq(G) (Ugq$,? holdsforall E,V C 9l and all q > 2, but not for q - 2. An analogouscharacterizationholds for the representations of the complementary series. (See Theorem10 and its corollary,in ? 11.) One of the main ideas motivatingthispaper was the desireto extendthe classical Hausdorff-Young theoremto the group. We recall the formof the Hausdorff-Young theoremfor Fourier transforms, as given by Titchmarsh. Let f C Lp, 1 ? p ? 2, and let F(x) =- (2r)-f e$ixY f(y) dy. Then F J< 11 (27r)1-/P11 f tIIp wherel/p+1/q=1. The mostconvenientmethodforprovingthis theoremis by using a convexityprinciplefor linear transformationsintroducedby M. Riesz. This principleallows one to "interpolate" betweenvariousboundsof linear transformations. For a general discussionof this methodof proofwe referthe readerto [3]. An extensionof the above theoremto locallycompactabelian groups,via the Riesz convexityprinciple,is given in Weil's book [23]. An abstractgeneralization of thistheoremto arbitrary locallycompactunimodular groupshas been givenby one of us, [16]. This generaltheoremwas proved REAL UNIMODULAR GROUP. 5 by what amountsto an extensionof the convexityprincipleto linear transbetweenoperatorvalued functions. formations boundedrepreDue to the analyticstructureof the familyof uniformly seintations of G, it is possibleto provea versionoftheHausdorf-Youngtheorem whichis much strongerthan its classical analogue (see Theorem7 in ? 8). The proof of Theorem 7 necessitatesyet another extensionof the Riesz convexityprinciple-fromthe case of a single fixedlinear transformation to a familyof transformations dependinganalyticallyon a parameter.3 It seems quite likelythat many of the resultsdescribedabove hold not only for this group but for certainothergroups as well (e. g. the comiplex classicalgroups). We hope to returnto this matterat a later time. We now proceedto describethe organizationof this paper. In Chapter I, which consistsof ?? 2, 3, and 4, we consideroperator valued functionsand we provethe basic convexity(interpolation)theorems. ?? 2 and 3 are quite generalin nature. However,in ? 4 the subject matter is tailoredto fitthe situationwhicharisesin the 2 X 2 real unimodulargroup. of our familyof ChapterII concernsitselfwiththe actual construction representations.In ? 5 thegeneralbackgroundand theoremsare stated. Their proofs,however,requiresomeextensiveFourieranalysis. This is done in ? 6. rn ? 7 we returnto the proofsof the statedtheorems. Combiningthe resultsof ChaptersI and II, we study the "Fourierforthegroupin ChapterIII. This leads to our extension Laplace " transform of the Hausdorff-Young theoremwhichis containedin ? 8 . In ? 9, we complete the Fourier analysisof a functionon the group by a considerationof the discreteseries. ChapterIV containssome applicationsof the above. In ? 10 we are mainly concernedwith the theoremthat convolutionby a functionin LI, 1 ? p < 2, is a boundedoperatoron L2. Some implicationsof this result are also deduced. Finally,in ? 11, we deal with characterizations of various of the group and with a relatednotion--the"extendability" representations of a representation to Lp. We shouldlike to observethat,exceptfor some notation,the contentsof ChaptersI and II are independentof each other. Since ChapterI is of a more technicalnature,the readermightwell begin with ChapterII which deals with uniformlyboundedrepresentation of the group. 8 In the case of numerical valued functions this extension was obtained by one of us in [20]. 6 R. A. KUNZE CHAPTER AND E. M. STEIN. I. OPERATOR VALUED FUNCTIONS. 2. Lp spaces of operatorvalued functions. In this part we provetwo purelytechnicaltheorems. In theseresultswe have ignoredvariouspossible and have restrictedour attentionto a rathersimple situation generalizations whichappears to be adequate for our purpose. We beginby introducingenoughterminology to state the theorems. a will denote Throughoutthe paper X4 complexseparableHilbertspace. The ring of all boundedoperatorson & will be denotedby S. If A is any is any orthonormal basis of 59, non-negativeoperatorin 13and * then 1, (2. 1) tr(A) >: (A&i n=l is non-negative and independentof the choice of basis. The bounidof ain A in B will be denotedby 11A jl, and we shall put i A (A*A)1. The p-th normof A is thengivenby = (2.2) IA II (tr(j A jP))1/1, where 1 _ p <oo. The letter M will always stand for a regular measure space4 overa locallycompactspace witha countablebasis for open sets; the underlyingtopologicalspace will also be denotedby 311. We shall consider functionson M whose values are boundedoperatorson St. If F is such a function,we say that F is measurableprovided (2. 3) t >(F (t) d, ), tC E M. is a measurablenumericalfunctionon 11M for each pair of vectors@,q in S1. If F, G are measurable,our assumptionsimplythe measurabilityof F H-G. FG, and F*, these being definedin the obviousway; thus for example F* is givenby F*(t) = [F(t)]*. An operatoron 69 is said to be of finiterank if it is reducedby a finite limensionalsubspace. The set of all such operatorsis a two sided * ideal in B and will be denotedby S. (2. 4) By a simple functionwe shall mean a functionF on M to E having only a finitenumberof distinctvalues, each non-zerovalue being assumedon a set of finitemeasure. 4 For a general discussion of measure theory on locally compact spaces see Halmos [10, Chapter 101. REAL UNIMODULAR THEOREM F(t) norms11 1. If F is a measurableoperatorvalued functionon M, the 1? IIP, p ? oo, are measurableas functionsof t, and therelations F 11- (2.5) IIF fP (2.6) , GROUP. (f iF(t) esssup 11F(t) 11X IIPdm(t) )1/P, 1?p <oo, definenormsrelativeto whichthe collectionLp(M, 3) of all measurableF with 11 F IIP <co is a complexBanach space (one identifiestwo functionsif theydifferonlyon null sets). Moreover,the formula fF= (2. 7) f trF(t) dm(t) is meaningfulforF in L1(M, 3) and definesan integralsatisfying fFjI 11F111. (2.8) 2. If F is any measurableoperatorvalued functionon M and vanishingoutsidecompact 1 ? p ? oo, thereexistsimplefunctionsSi, S2, sets such thatFSn is integrable, THEOREM (2.9) lir fFSnr n and 1iSn lq _ 1, i/p + l/q that lr (2. 10) = l/p IF Ilp 1. Furthermore, if p and q are indices such + l/qC? 1 and F, G are measurable,then F IlP11 G lq FG Iir? 11 11 Finally, the simple functionsvanishingoutside compact sets are dense inl Lp(IM,3) for all p such that 1? p < oo. Except for the minor complicationthat we are dealing with operator valued functions,the proofsof theseresultsinvolvenothingnew and proceed includedmost of the details for along staildardlines. We have nevertheless thebenefitof thereader. We mentionthatone mightobtainsimilaralthough less explicittheoremsas consequencesof knownresultsfromdirectintegral integration;however,it seems theoryand the theoryof non-commutative situation inappropriateto complicatean essentiallysimplermeasure-theoretic by such considerations. Furthermore,in our applicationwe require these resultsin the ratherexplicitand concreteformgiven above. We begin by recallingsome of the facts about the trace and the p-th normsmentionedearlier. As a generalreferenceto this part, we referto a paper [4] of Dixmier. gE 5 Let t, D be fixedvectorsin &l and defineE,7 by En,,?(g) = 8 R. A. KUNZE AND M. E. STEIN. An operatorE is of finiterank if and only if it is a finitesum of operators of the formEe n and moreover (2. 11) tr(E4,-q)=- Let S., 1 C p :0oo denote the collection of all bounded operators A on A IIP< 00. A positive operator A is in B., 1 ? p < oo, if and Sl such that 11 only if its spectrum Ak,A2, is discrete and 00 E XnP < 00. n=l In this event,IfA 11 p= (2. 12) whichimplies, vP)1/p, (' 2 n=1 1A IlPis a non-decreasing functionof p. (2.13) If A E 3p, B E ,3q,and 1/r=1/p+1/q? A11p 11 B 11q,where1 _ p, q C 00. 1, then j]AB llr (2. 14) If A E B and 1 ? p coo, thereexist operatorsE1,E2, finite'ranksuch that 11 En IIq ? 1 and limtr(AE,) n wherel/p +A1/q - of A 1v, 1. In case A is of finiterank,thereexistsanothersuch - 11 A IIP. Iq ? 1 and tr(AE) operator E with 11E is a Banach space under the normgiven by (2. 2), and the (2. 15) f3p collectionE of operatorsof finiterankis densein Sp for 1 ? p < oo. (2.16) If 1 p<oo and B E3q, where1/p+ l/q= 1, then A--tr(AB), A E 3p is a boundedlinearfunctional,kB on Op, and the map B withthe conjugatespace of S3p. B identifies3q Given an orthonormal basis 4, of 9 we formthe set 0 of all finiterationallinear combinationsof the operatorsEij and (- 1)Ei , where (2. 17) Eij = Ets @J. ?Z is denumerable,and one easily verifiesthat the productof two members of i is again in ?Z. LEMMA 1. ?) is dense in R,,, 1 < p < oo. 9 GROUP. REAL UNIMODULAR Suppose E is an operatorof finiterank and that P is a projectionof E. Then for 1 ? p < oo and A, B in 9D we have finiteranksuch that EP E-AB 11 P-B 112 11B 112 +11 A 2 11 111 C11E-A EP-AB v 11 11 112. The inequalitiesfollowfrom (2.12) and (2.13). Now because .Z is dense in thep-th in B2, we see that any operatorof finiterank can be approximated normby elementsof i). An applicationof (2. 15) finishesthe proof. As a corollarywe obtainthe fact that S , 1 _ p < so, is separable. The collectionof all measurableoperatorvalued functionson M will be denotedby 3(M). LEMMA 2. If F C 3 (M) and E is an operatorof finiterank, t -* tr(F (t) E) is a measurablefunctionon M. There exist vectors , *, ,* E in X such that n = Eti.d n=1 n E EF(t), 7 and by (2. 11), Thus F(t)E= n=l tr(F(t) E) whichimpliestr(F(t)E) LEMMA -, r ) E(F(t) n=1 is measurableas a functionof t. 3. If 1? p?oo functionon M. n = and FCE3(M), t- 11F(t)I p is a measurable Let A belongto BZ. From (2. 13), (2. 14), and Lemma 1 we see that (2.18) A IIp= Sup{ tr(AE) I: E C 0, 11E l1qC 1}, 11 wherei/p -- 1/q 1. ReplacingA by F(t) and applyingLemma 2 we see that 1F (t) 1p is measurable;this followsfromthe fact that the least upper boundof a countablecollectionof measurablefunctionsis again measurable. 4. A functionF on M to p,,1 ? p < oo, is measurableif and t -tr onlyif (F (t) B) is measurableon M forall B in 3q (l/p + 1/q= 1). LEMMA For everypair of vectors4, -qin *, B=Ee,, is in Sq; henceFCE3(M), providedtr(F(t)B) is measurableas a functionof t for all B in S3q. Conversely,suppose FCE 3(M). Let BCESq. By (2.13) tr(F(t)B) existsand is finitefor each t in M. If p 1, = 10 R. A. KUNZE AND E. M. STEIN. n-*o = iSany orthonormalbasis of 9, and is thereforemeasurable. where is, 4 If 1 <p <oo, thereexist, by (2. 15), operatorsE1,EF, , of finiterank O 11 such that B-E 0. Thus B q , Itr(F(t)B) -tr(F(t)E,) | 11 F(t) IIP11 B- Nn 1q > 0. By Lemma 2, tr(F(t)E,) is measurablein t foreach n, and hencetr(F(t)B) is also. The resultjust establishedtogetherwith (2. 16) showsthat a measurable functionF on ,Alto B. is weaklymeasurableas a functionon If to the separableBanach space B.; thus F is also stronglymeasurable.5 The proofof Theorem1 now followsfromthe precedinglemmasand the wellknowntheory5 of theLebesgueintegralextendedto functionswithvalues in a Banach space. II1 provingTheorem2 it is convenientto establish LEMMA 5. Suppose F n = i=l fiA1, where fi is a measurable numerical function on M, fifj= 0, i=/=j, and Ai E S. F llP ( 11 (2. 19) n Then for 1? p <oo, fi IPP 11 1A1 j)I'/P, and in case 11 F 11,< ??, fF (2. 20) n E (f fi(t)dm(t) (trAi). i=l 0 (i7=j), Since f (t)fj(t) f F(t) IIPP r1 dm(t) IIIP F n Ai IIPP) dnm (t) f ( j=1 Ifi(t) P11 n ~E(fr fi(t)|Pdm (t) ) 11AjJJIPP If 1FI1< 00,then fJF ftr(F (t) ) dm(t)- 6 tr( E f5(t)Ai) dm(t) i=1 For a discussion of these points see Hille and Phillips [15, Chapter 3]. 11 GROUP. REAL UNIMODULAR zf(t)trAt)dm(t) f( j=1 n =Y. (rff(t) dm (t) ) tr(At). i-l sets are LEMMA6. The simple functionsvanishingoutside comnpact densein Lp(M, J) for 1?5 p < oo. Let f,g be measurablenumericalvaluedfunctionson MI,and let A, B E S. By simpleestimatesand the precedinglemmawe have (2.291) A -B A IIP+ 119 gIP 11 f - gIP 11 ~1fA -gB IIP' 11 IIP, funcwhere1 ?< p < oo. Thus if e > 0, A E B., and if f is the characteristic tionof a set of finitemeasure,we can choosea compactset withcharacteristic g aindan operatorB of finiterank such that fuinction fA-gB 11p< e. 11 (2. 22) The conclusionof the lemma followsfromthe fact5 that finitelinear combinationsof functionsof the formfA are dense in Lp(M1,i). LEMIA 7. If F is a simplefunctionwithcompactsupportthereexists a simplefunctionS with compactsupportsuch that S 1l _ 1, and F IIp, fFS (2. 23) 1/p+1/q==1. where1?poo, , A, of finiterank and mutuallydisjoint There exist operatorsA, f5such that functionsf measurablesubsetswith characteristic n F ===EfAi. i=1 By (2. 14) thereexistsan operatorEi of finiteranksuch that Ei A11 p. Let tr(A5E) ci (2. 24) 11F KllplP 11Ai jq ? 1 aild JpP-1 and put S (2. 25) Then fF5 =X f n cifiE. j=1 c,f,(AgE ) n~~~~~~~~ ==ci(ffj(t)dm(t)tr(AiEi) = i=1 = F II. c 11 fi IIPP1Al IIP= 11 12 R. A. KUNZE AND E. M. STEIN. Also n qo 11sJS jjgqq - ECjqi j=1 ii q 11 Iff,llq~1E E* lgqq n <_ E 11F !(p1-p'q 11 Ai jp(p-1_)q 11 f j=1 n z f, IIPP Ai IIP 11 11 -11 F1 P_P lipp 1. j=1 Finally,sinceF has compactsupport,so does S. Proof of Theorem 2. SupposeF, G are measurable. To establish(2. 10) we use (2. 13) whichimplies 11FG lr?_f (1F(t) IIP' G(t) fF( (t 11 F)IPP IIq)r dm(t). dm (t) ) "P (f 11G (t) 11 qqdmn(t)) I-q providedp 74oo and q 74oo. The other two cases arse treated by similar arguments. As the case p oo is somewhatexceptionaland requiresseparatetreatment,we shall prove (2. 9) onlyfor p such that 1 ? p < oo.8 Suppose first of all that 11 F,1 P< oo. By Lemma 6, thereexistsimplefunctionsF1, F, with compact supportssuch that 11 F -F IIP-> 0. Choose S,, for Fn in accordancewithLemma 7. By (2. 10) FS, is integrableand I fFSnf- FnSn I 11 F-Fn IIp?. If 11 F liP=- , let F. be the productof F and the characteristic functionof a set of finitemeasure containedin {t: F(t) lpc n} and chosen so that Fn 1p- oo. Then iiF,, J1p Thus we can choose a simple function Sn <oo. with compact support contained in the support of F, such that, 1iS,,lq ? 1 and II, < 1/n. fFnSn F11F Then FS,=== FnSn and f FSn 1F lIP. 3. Interpolationin the general case. In this sectionwe provea rather general interpolationtheoremfor operatorvalued functions. Let 4 be a complexvalued functionwhose domain contains a strip, X< Rz ?< /. We shall say that 1 is admissible on the strip if (D is analytic in c < Rz </, continuousin a ?< Rz _/, and satisfiesthe growthcondition 6 We do not need the exceptional case p - oo in our application. REAL UNIMODULAR (3.1) 13 GROUP. Sup log Ij> (x + iy) I CekIvI, whereC and /iare constantsdependingon d1; we requirealso that ,usatisfies the additionalcondition (3.2) If <7r/ 12 dii, are admissible on a given strip and if v1, v2 are complex numbers it is easilyverifiedthatthe combinations v11 + v2d12, )142 are also admissible. A complexvalued functionon a measurespace will be called a simple functionif it can be expressedas a finitelinear combinationof characteristic functionsof measurablesets of finitemeasure. Now let M1, M2 be measurespaces, and let D be a strip, -? Rz?<,l8. Suppose B,, z E D, is a complexvalued bilinearformdefinedfor all simple functionsfl, f2 on M1, M2. We shall say that the collection{B,} is an admissiblefamilyof bilinearformson D if (3. 3) 4)(z) Bz (flyf2) is admissible onD foreachpairofsimplefunctions, fl,f2 on M1,M2. We now introducesome notationand terminologywhich will remain fixedthroughout this part. The stripa < Rz ? ,Bwill be denotedby D and we shall put (3. 4) y-==(1 - ) (X+ Tfl 0< T<1. We supposepo,piL, qo, q1 are givenindicessuchthat1 ? or q1# o. The indicesp, q are then determinedby (3. 5) 1/p=- (1-r) (3. 6) 1/q pi, qi C oo, and qo so l/po + 71/pi, (1-r) 1/qo + 71/q1. The conjugateindicesof q0,q1,q will be denotedbyqo', q1',q'. Finally,Ao,Al will denotenon-negativefunctionssuch that (3. 8) logAi(y) A Ae6ll, 8 < 7r/ a). Withminorchanges,the proofof Theorem1 [20] yields the following convexityprinciple. LEMMA 8. and suppose (3. 9) Let {Bz} be an admissible family of bilinear forms on D, IBa+iy(fl,f2)I Ao(y)11 11 f,IIPo f2 liqo R. A. KUNZE 14 AND E. M. STEIN. f2 IjqL' 11 Bp+i8(f1if2)I ?A1(y) 1If. IIPO (3. 10) for all simple functionsf,,f2 on MA1, M2. Then for simple functionsf. f, we also have (3.11) |jY(fl f2) IAT- jj q' 11 fll P 11f2 The constantA, is given explicitly,in termsof the Poisson kernelfor the strip,by logA, (3.12) - logAo [(-)y] w(1 r + f logAl[( , y) dy a)y]w(r, y)dy, where w(1, y) - sin7r/ (cosh7ry+ cos7r). By a boundedsubsetof a regularmeasurespace we mean any measurable subsetof a compactset. Now let N be an arbitrarymeasurespace. Suppose fromsimplefunctionsf on N to measurTZ,z C D, is a lineartransformation able operatorvaluedfunctionsT, (f) = F, on AM. We shall say that { T,} is an admissiblefamilyon D if (F, (t)$,y) is locallyintegrableon M1and (3.13) ?)(Z) =--j (Fz(t) , n)dm(t) is admissibleon D for each choiceof vectors$, 7 in 94, simplefunctionf on N, and boundedsubsetK of M. THEOREM 3. Let N be a measurespace, and suppose {Tj}, z C D, is an admissiblefamilyof linear transformation fromsimplefunctionsf on N to measurableoperatorvalued functionsTz(f) F-F on M. Suppose further that the followingtwo conditionsare satisfiedfor each simple functionf. (3. 14) (3.15) T+iy(f) l oC Ao(y) 11 IIP. 11 T (+iY() 11 IIf i C A1(y)11 f IY1. Then it is also true that (3-16) q AT11 11Tny(f)11 f fIP In provingthe theoremit is convenientto establishthe followinglemma. LEMMA 9. If {Tz}j, z ED, is an admissiblefamilyand S is a simple operatorvalued functionon M whichvanishesoutsidea compactset in M. Then trP,(t)S(t) is integrableand 15 REAL UNIMODULAR GROUP. (3. 17) 1(z) = f trFz(t)S(t) dm(t) is admissibleon D for each simplef on N. functionof a Suppose firstthat S == kE , wherek is the characteristic boundedset. Then t-rFz(t) S(t) -= kc(t)(Fz(t)t,t) and the resultfollowsby assumption. The generalcase followsby linearity. Proof of the theorem. Our assumptionsimply q #oo.7 that Thus to show Ty(f)11q 11 C A7 11fIIP in view of Theorem2, to showthat it suffices, I ftrlFy(t)S(t) dm(t) C? A, 11 S 11k f IIP 11 (3. 18) for each simple functionS vanishingoutsidea compactset. The idea of the proofis to reduce this problemto one concerningan admissiblefamily of bilinear forms. We shall then apply Lemma 8 to completethe argument. Supposethenthat S is a simplefunctionwithcompactsupport. We can n expressS as E k1Ej,wherekI,1c2, j=1 functionsof , cknare the characteristic mutuallydisjoint boundedsubsetsK1 and each EI is an operatorof finite rank. Now let El U= Ei I be the canonical polar decompositionof Ei, and let (3.19) j A,P>O X,ij> ?, be the spectraldecompositionof I Ei The pairs of indices i, j will then rangeovera finitesetwhichwe shall call M2. To each complexvaluedfunction g = {gij) definedon M2 we associate an operatorvalued functionG on M whichis given by (3. 20) n G (t)- _k.7i(t) Y: g{jUxP+j. f j=1 Then G is a simple functionwith compactsupport,and by an elementary computationwe get (3.21) 7 G*(t)G(t) itv 7 ls(t)IgHI'2Pi,. The case q-=oo couldbe dealt withby a moreinvolvedargument. R. A. KUNZE AND E. M. STEIN. 16 Now for 1 Cp 11G IPP==ftr[(G*(t)G(t))I/2]dm(t) <oo, which implies kijjpP)"P. G IIP ( X I gij I11 11 (3.22) i,j is independentof p being,in fact,equal to the measureof KE, Since 11 kIIpPP G IIP. We we can introducea measure in M2 relativeto which1 9 g IIP 11 observethat this relationis also valid for p - 0o. Because the maps f-F, g -* G are linear it followsthat the equation Bz(f, g) (3. 23) ftr(Fz(t) G(t) ) dm(t) = definesa bilinearformfor each z in D. In this formulaf is an arbitrary simplefunctionon M=-N, and g is any complexvalued, obviouslysimple, functionon M2. By Lemma 9, in particularby (3.17), {B4 is an admissible familyon D. Now Fa+iyG1i1 C Ba+iy(f,g) : I 11 Fa+1Yl o 11G lIqo', 11 and using (3.14) we get I Ba+iy(f,g) _Ao(y) 1If|IPO 11 g 1Iqo'. (3. 24) By similarestimateswe obtain g 1a I B:+V(f,g) ? A1(y) || f ||P111 (3. 25) Thus by Lemma 8, I B7(f, g) I ? AT11 g IIq'. As this holds for all g, f1IP11 we may take g Then G-=S {Aij}. f trFy (t) S (t) dm(t), By(f,g) (3. 26) and whichimplies (3. 18). 4. The main interpolationtheorem. In orderto proveour resultsfor the 2 X 2 real unimodulargroup,we use, in additionto factsaboutthe group certain convexityarguments. The basic and most and its representations, importantfact along theselines is establishedin this section;withthe intent of clarifyingthe situation,we have presentedit in a slightlymore general formthan our applicationrequires. An operatorvaluedfunction5 definedon an open strip,aO < Rs </3o, in the complexs-planeis said to be analyticif (J (s) d,-) is analyticfor all 6,s in N9. We shall say that a is of admissiblegrowthin the stripif (4.1) sup log1 THEOREM (+ i) :-SCell ,u<.ii/(go - a). 4. Let N be a measurespace and T be a linear map fronm REAL UNIMODULAR 17 GROUP. oniN to analyticoperatorvaluedfunctionssuch that5 = Tf simplefunictions is of admissiblegrowthon the stripz, < Rs < ,P3 for each simplefunctiont f. Suppose that for c, < a <,B <,38 we have (4.2) sup -oo <t<oo (4. 3) (}I Ja (I + 11a(A + oo it) 1o (1 + it) 1122It j24( t l)ec -t I)2b AO1 f 11, dt) Al?A1If 112 forall simplef, wherea, b, c are real and a ? 0. Then we may contclude ( (4.4) 00 00 || a (y+it) gq(1 + t )qddt)1/ *?< AT||f IIP, where 1 <p<2, 1, y=-+T(/31/p+1/q and theparameterr is determinedby i/p = 1 a), d =c+r(a+br/2. c), Remarks. Beforewe provethe theorem,we noticethat the result (4. 4) is intermediate-inthe sense of Riesz-Thorinconvexity-betweenthe hypotheses (4. 2) and (4. 3). It shouldbe notedthat the singularityat t 0 of the measure tI12a( + I t |)2b dt does not persistin the conclusion;only the influenceof I t |2a for t near infinityremains. The proofgivenbelowcouldbe generalizedin severaldirections. We may begin with a generalpair of indices (po, q0), (pi, q1) instead of (1,oo) and (2, 2). We might also considermore general measuresthan those of the form tI (12a( + I t I) 2b dtgivenabove. Weshallnotconsider thesegeneralizations here. It should be pointed out that the proof given below would be much simplerif a, b, and c were zero. In that case the left-handsides of (4. 2) and (4.3) would be translationinvariantin t. Since the basic methodof the proofconsistsof translationalong verticalliiies of the strip,we are forced to overcomethe lack of translationinvariance by somewhatcomplicated devices. At severalpointsin the proofit will be convenientto referto the easily verifiedresultgivenbelow: LEMMA10. If v is real and 8 > 0, thereexistsa constantA > 0 such that (4. 5) for- (+ o<y,t I y + tl) < A (1 + I y l)v(1 + I t ) <o. Proof of the theorem. We shall obtain the result as a consequenceof Theorem 3. To do this we set M1 (-oo, oo) and put 2 18 R. A. KUNZE dm (4.5) = AND E. M. STEIN. (1 + I t 1)2(a+b-c) dt, wheredt is Lebesgue measure. Given a simple functionf on N we form 5= Tf and set Fz(t) =5(z (4.6) + it) (1 + I t | )c-a(z + it)a for c-?Rz ?< /. Since a> 0 we may choose a single valued branchof the factor (z -,l + it) a which is analytic in ac< Rz < / and continuouson is analytic in z for each t and is jointly -<?Rz?_,8. Thus (Fz(t)>-q) continuousin z, t forall vectors4, - in 54. Furthermore, the transformation T0 definedby Tx(f) F, is linearand maps simplefunctionson N to measurable operatorvalued functionson M. We shall now estimate Fz (t) 1.o By (4. 6) and the condition (4. 1) that 5 is of admissiblegrowthin oc? Rz ?< /, we find, = (x+i(y+ + I t)c-a I - ,+i (y + t)Ia )II.(1 t) IIFz(t)II1.=11 ll <A (x+i(y+ t))1 0(1 + t )c(1 + IYI)a (4.7) Hence, log 11Fz (t) 11.o CeAIY+t I + log (1 + I t l c + log (1 + I y | a + log A. This estimatetogetherwiththe aboveimpliesthe condition(3. 13) that {T.} be an admissiblefamily. Now, for z a + iy we find,using (4. 2), (4. 7), that = + i(y + t))110(1+ I t c)(1 + I Y I)a A IIflI1(1+ I y+ t )-C(1 + It C)(1 + y 5( (a < A 11 ||Fa+iy(t) lIoo ? )a Thus by Lemma 10, we obtain (4. 8). A (1 11Ta+iy(f) 11.o0 (4.8) f K1 + Iy la-cI 11 Next we shall estimate11T+iy (f) 112. jjFp+y112' 00 5 (/3+ i(y + 11 We have, Iy + t 12a(1+ I t I)2c-2adm. t)) 1122 Now makinguse of (4. 3), we obtain jj .+y 1122 ?(A1 If112)2- x0sup <t<x0 [(1 + ? (A1 f/112)2 sup [(1 + -co<t<9o y + t I)-2b(1+ I t 1)2c-2a(1+ y + t I)--2b(1+ I t 1)2b]. I t 1)2(a+b-c)] 19 REAL UNIMODULAR GROUP'. Thus by Lemma 10, (4.9) 11l2< A(1 + ||F+iy I yj)II71 If 112. Having (4.8) and (4.9) we can apply Theorem3 and concludethat 1ITy(f) IIqC AT1If II (4.10) Now Tey(f)Iq 11 f 115 (y+it)l1qq1( + I t I),c-qaIy00 CAT11f IJll. Hence IL _00 + I t I)qc-qa(j + 5?|(y + it)q(1 Since Tq Thus 3+it Igaddm = t I),qa(l + It 1)2(a+b-c) dt ? A 1ff 1q 2, =qc+2(a+b-c). qd=gc+qT(a+b-c) oo1 qq( + 1t I))qd dtC< A 11f JjVq, (-y+ it) 11 whichprovesthe theorem. 5. Uniformlybounded representations. We now considerthe group G of 2 X 2 real unimodularmatrices,and we firstrecall some of the known of G. factsconcerningthe representations We representan elementg C G, by g=[c da ] ad -bc 1, and denoteby g (x) the fractionallinear transformation Then (5.1) g(x) (ax + c)/(bx + d), -oo < x < oo. (9192) (x) =92(g1x) (bx + d)-2, bx+ d=/=0. and dg(x)/dx We now introducetwo "multipliers" 4+ and 4v. These are definedby (5.2) (5.3) p+(g,x,s) = -(g, x, s) bx+d I28-2 sgn (bx + d)p+ (g, x, s), wheres is an arbitrarycomplexnumber. 20 R. A. KUNZE AND E. M. STEIN. Next we considerthe "multiplierrepresentations" g -v (g, s) givenforfunctionsf on the real axis by v+(g, s) : (5. 4) f(x) -- q+(g,x,s)f(g(x) ) From these, one may obtain the irreducible unitary representationsof G. They fall into threeclasses.8 a) The two continuous principal series oo00<t <oo g -> v+~(g,1/2+it), wherethe Hilbert space is the space L2 of square integrablefunctionson K< x<po, with the usual measure. b) The complementaryseries g ->v+ (g,v) , O < of< 12 The Hilbertspace,in this case, is definedby the innerproduct (5. 5) wherea, c) au (f,h)u= 00 f 00 00 f(x)h(y) 00 I x -y2 dxdy, P(2a)cos(o-r)/7r. The two discrete series, 0,1,2, g --> +~(g, k) k=-O, We shall not need the exact formof theserepresentations. The Plancherelformulafor G was derivedby Harish-Chandra[13]. It involvesrepresentations of type a) and c) and not of typeb). To state it we make the usual definition U(f) f(g)U(g)dg for uniformlybounded representations g -- U (g) and f in L1 (G). Using thisnotationthePlancherelformulaassertsthat,wheneverf C L1 (G) nL 2(G), f 11 1122 1/2 .i:i v+(f, 1/2+ + 1/2 00o t tanhrtdt it) 1122 V-(f, 1/2 + it) 2' tcoth7rtdt 8Except for notation, these representations are those of Bargmann [1]; the differenceof notation is discussed more fully in the proof of Theorem 10 in ? 11. 21 REAL UNIAMODULAR GROUP. + + oo E 11 D+(f, k) 1122(k k=O E k=O + 1/2) D-(f,k)l22(k+ 1). 11 Here 11112means the usual Hilbert-Schmidt norm for operators as used in ?2 above. One of our main results is contained in the following theorem. THEOREM 5. tations There exists a separable Hilbert space X9 and represeng- U+(g,s) of G on t with the followingproperties: 1) 9-- U+(g,s) is a continutous representation of G on &9 for each complexs in the strip0 < R(s) < 1. g -- 2) g-- U+(g, 1/2 + it) is unitarilyequivalent to the representation v+ (g, 1/2 + it) of the continuousprincipalseries definedabove. 3) g-- U+(g. u). 0 <a< 1/2 is unitarily equivalent to the representation g -> v+(g, ) of the complementary series. 4) If e and - are two vectorsin 91, thenthe functions s (U?(g,s),), g fixed, are analyticin 0 < Rs < 1. 5) sup 11U+(g,s) ? Au(I + I t )2, g s-= + it, 0 <Ko- 1.< Furthernmore, the constantA(y is bounded on any intervalof theform0 < a ?r ? /3K 1. It is known that for each t, the representations v-( 1/2 + it) and v+( 1/2 - it) are unitarily equivalent. Hence the same fact holds for the representations U+ (, 1/2 + it) and U+(, 1/2 -it). As the next theorem shows, these equivalences are to some extent already inherent in the " analytic structure" of the representations g -* U? (g, s); the theorem also describes some adidtional, and rather interesting, relations among the representations U? ( *,s). THEOREM 6. The following symmetries exist: 9 As in ? 2, the ordinary bound of an operator A is denoted by 112K11X. 22 R. A. KUNZE AND E. M. STEIN. 1) The representations U+( , s) and U+( *,1s- ) are contragredient. Similarly,U-( , s) and U(*, 1-s) are contragredient. 2) U+( , s) contragredient. U+(, 1- s). Hence U+( ,s) and U+(.,s) are also 3) There existsa fixednon-scalarunitaryoperatorS such that for all s in 0 <Rs< 1, SU-(, s)S-1l U-( , 1-s). Thus U( ,s) is unitarilyequivalentto the contragredient of U-( ,s). Remarks. (i) It should be observed that the known result [1] concerning the reducibility of the representation U- (, 1/2) is implied by 3). (ii) The representations U+ ( , s) for s / 1/2 + it and s 7/4 are unitarily equivalent to representations introduced by Mautner and Ehrenpreis [5]. These they show are not equivalent to unitary ones. They also assert that the representations are uniformlybounded. However, the more definite statement contained in 5) of Theorem 5 is crucial for our purposes. The proof of Theorem 5 is lengthy and requires some vigorous (iii) classical Fourier analysis. This is contained in ? 6, which is, for the most part, somewhat technical. At first reading the reader may prefer to pass on to ? 7. 6. Some lemmas from Fourier analysis. We shall begin by introducing a class of Hilbert spaces, which will be seen 10 to be related to the Lp spaces via the Fourier transform. These spaces San are indexed by a parameter O< f < 1, and are given by the norm 00 The spaces &9(r, 9,ru corresponding to any pair of indices Orl,U2, such that < fT1, U2 < 1, are naturally related by a family of unitaries which we shall now exhibit. Let s, = ol + it, and S2 = (2 + it2 where -c < tl, t, <oo. Now o let V (s1, S2) (6. 2) be the mappingwiththe domain6V0, givenby F (x)- F(x) Ix 812, FE U1. 10Although many of the results of this section are probably known, they do not seem to be accessible in the literature in the manner in which we need them. REAL UNIMODULAR 23 GROUP. Then 00 W (sl, S2)F I 11qf22 12 1 X F(X) 2(T121-2) 1X 12LT2-1 dx - 11F 11012. This facttogetherwith (6. 2) showsthat W(S2, sl) is the inverseof W(sl, S2). In what follows,we shall be mainly concernedwith the pair of spaces 9u and & For the sake of convenience we shall set TV,= W (s, 1 - s). The mapping W( is of particularinterestbecause it implementsa duality between 9u and 9(1u. In orderto make this,statementprecise,we shall introducesome additional notation. Throughoutthis section and the one that follows,it will frequentlybe convenientto put (6. 3) 00 F(x) G(x) dx. (F, G) This notationwill be used with the understanding that F, G are measurable complexvaluedfunctionsdefinedon -so < x < oo such thatFG is integrable. The innerproductin 9( will be denotedby (, ) and we shall sometimes set 1-o a=o'. LEMMA (6.4) 11. If F E94a and GE <, O <i,1 then (F7,G) = (F,WuG)an (WuF,G)0u= FIfu 1G II0'. (F, G) I _ 11 (6. 5) Furthermore, if A is a boundedoperatorontSa, the operator (6. 6) A'- WuA*Wu-l, whereA* is the (Hilbert space) adjoint of A is characterizedas the unique boundedoperatoron 59u' such that (6. 7) G) (A(F), (F A'( G)) forall F in 9u and all G in 59ga. To prove (6. 4) we firstobserve that 2oa'- (WuF, G)ua= 00 F (x) 2X x 1G2(1 1 1- X) 21. x 11-2c dx (F, G) JF (x) G(x) (F, Wa G)a. X 11-2o 1 X 12a-1 dx Thus 24 R. A. KUNZE AND E. M. STEIN. Now, by Schwartz'sinequality, lu'11G I(F G) C 11WuF]? t', and (6. 5) followsfromthe fact that Wa is an isometry. Suppose A is a boundedoperatoron 9(u, and that F E San G E 9 . By (6. 4), the fact that preservesinnerproducts,and a secondapplicationof (6. 4) we findthat WoJ (A (F), G) = (A (F), Wa G) a -(F, A *Wu,G)a (WuF, WuA*Wa/G)a (F, WuA*Wa G). Thus (6. 7) is satisfiedby the operatorA' WuA*Wu-l. That A' is the unique operatorwith this propertyfollowsfromthe easily establishedfact that,GE 59(, and (F, G) =0 forall F E &u impliesG(x) =0 a.e. It should A. also be observedthat (A')' In additionto the tu spaceswe shall considerthe Lp spaces,1 < p < o, of functionsf definedon -o < x < so and normedby = 11 fIIP - 00 "P. I f(x) IPdx) Since the parametera ranges between0 and 1 and 1 ? p < co thereshould be no confusionbetweenthe norms 11 lI, and 11IIP For a functionf definedon- Ko x <oo, the Fourier transformF is definedby (6. 6) F (x) (27)-f eiav f(y)dy and the inverse Fourier transform is given by (6. 7) f(x) (2i)-feixyF(y)dy. Here and throughoutthis sectionwe shall adhere to the followingconvention. Pairs of functionswhichare relatedto each otherby either (6. 6) lowercaseand capital letterssuch or (6. 7) will be denotedby corresponding we take for grantedsuch standardfacts as as f,F or g, G. Furthermore, the Planchereltheorem,and the sense in which these transformsexist for functionsin Lp, 1 < p ? 2, as well as the equivalenceof (6. 6) with (6. 7) on f or F. (See e. g. [21] ). To be morespecific, undersuitablerestrictions a theorem we shall make use of two well knownresults on Lp transforms, of Titchmarsh(the so-called Hausdorff-Young theorem),and the Parseval formulafor L.,Lq. These resultsmay be stated as follows: REAL UNIMODULAR 25 GROUP. LEMMA 12. If f,G E Lp, 1 ? p ? 2, and theirFourier transforms F, g are given by (6. 6), (6. 7), then a) F 11q?_ A 11 f IIP, 11 b) lglq?A IIG 1p, wherel/p + l/q == 1, and c) (f, g) (F, G). Now the relationbetweenthe L. spaces and the SL, spaces mentioned earlieris containedin the followinglemma. LEMMA form (6.6). 13. Let f E Lp, 1 < p 2, and let F denoteits Fourier transI-1l/p. Let a Then O <o1/2, F E9tU. and iF IIa?AuIfil, (6.8) I(; 1l<p_2. The class of FE St9uwhichare Fourier transforms of f E Lp is dense in 0< a < p?2, 1/2. Analogously, let FEW u 1/2 ? (r <1, and let = 1 -1/p. 2 ? p < co, the inversetransform(6. 7) existsin Lp norm,and Then 1/2 --- < 1. u Au11 F lII 11 f IIP-:: Considerfirstthe case 1 <p ? 2. By a theoremof Hardy and Littlewood (see [11], p. 375), (f F(x) |P x Now., Ap 1jf IIP, Ip-2 dx)1/P ? 1 < p2. 00 F I F(x) ulf2 I ' IF(x) n 00 ( 00 F (x) 1Iqdx)/a I IX 12,y-l s dx ~~~~~~~~00 00 | F (x) I I X I f-1P dx) ll/ by Holder's inequality. Furthermore,(2u 1) p = p -2. Thus usilig the inequalitiesof Titchmarsh(Lemma 12) and Hardy and Littlewoodwe obtain 1JF j2?A1 A 11 fJ2, 1 <p 2. This proves (6. 8). If F is the characteristic functionof a finiteinterval,then f given by (6. 7) is in Lp forall p > 1. Hence finitelinearcombinations of characteristic functionsof finiteintervalsare contained among the Fourier transforms (6. 6) of f E Lp. Thereforethe image of Lp, 1 < p 2, under the Fourier 26 R. A. KUNZE AND E. M. STEIN. transform is dense in the corresponding space Nu, 1 1/p. This concludesthe considerationof the case 1 < p ? 2. The second part of the lemma,which deals with the case 1/2 <r < 1, 2 ? p < so, followsfromthe firstpart by duality. We shall brieflyindicate 1 1 - 1lp', where1/p+ 1/p' the argument. Put &' 1- o. Then ' and 1 < p'?`2. By Lemma 11, Nut and No, are dual and it is well known thatLp and Lp are dual. The secondpart of the lemma then followsupon identifying(6. 7) with the adjoint of (6. 6) consideredas a mappingfrom Lp,to Stu (properlyspeaking,thiscan be doneonlyon a densesubsetof N,). LEMMA 14. Let 00 K(F) (x) Then {J K(F) -e I x/y11 I x-y1-kF(y)dy. ,00 ^00 00 (X) 12 dx Aa jI 00 F(x) 12 dx if -1/2 < c < 1/2. This lemmais known. The prooffollowseasily fromTheorem319 of [12]. There,a moregeneraltheoremon integraloperatorswhosekernelsare homogeneousof degree-1 is given. The maill discussionof this sectionis containedin the lemma below. We shall deal with operatorsacting on &1(J. It will be convenient,however, to specifytheactionoftheseoperatorsbyexhibitingtheiractionon the Fourier of the functionsin question. transforms Thus we considerthe multiplicationoperators (6. 9) Mt +: (6. 10) mSt-: f(x) f (X) I X 121t f (X) sgn(x) IX x 2it f (X) Now if F is the Fourier transformof fC L1 n L2, we shall denotethe Fouriertransforms of mt+(f), mt-(f) by M1It+(F),Mt-(F). It will also be convenient to introducethefollowingclass 2Z of functions: F C i) if F is C? and vanishes in a neighborhoodof zero and outside a compact set. Clearly 2Z is dense in each N9(u, 0 < a < 1. Furthermore, 2) is containedin the image of L1 n L2 underthe Fouriertransform,(6. 6). LEMMA 15. If F C O, thenMt' (F) C 59L7 foreachqf such that0 < and the transformations h i n tF Mto(F)l F C9 have unique bounzded extensionsto all of Nu,. 0 < 1. REAL UNIMODULAR 27 GROUP. The extensions, whichwill also be denotedby t+,Alt-are unitaryon &I and, in general,the bound,11 MIt 11aof Mt' consideredas an operatoron satisfies (6. 11) It IIa-A(1 + t ), M1 0 <u < 1. Since the restrictions of mt+, mt-to Lp, 1 ? p < co, map Lp isometrically ontoLp, the Planchereltheoremimpliesthat Mt' extendto unitaryoperators on j. To treat the case a 7?&1/2, one would like to expressMt+ and Mt- as integraloperatorsof convolutiontype. However,the kernelsin questionare not locally integrable,and we must thereforeproceedratherindirectly." We introducethe transformation mre: f (X) O < e< _> X I-e+2itf (X) 1/2, whichmaps L1 n L2 into L1 n L2. PuttingF forthe Fouriertransform(6. 6) of fC L1 n L2, we denotethe Fourier transformof mte(f) by Mte(F). Now let FE 5Z'. Since F is the Fouriertransformof an f C L lnL2, we can formMtI(F), and, as is easily verified,by the Planchereltheorem, (6.12) 11MtE(F) -Mt+ (F) 112 0, as E -> . Next, we claim that (6.13) MlIte(F) = ae,t F(y) I x-y le-1-2it dy forF in Z),where ae,t = Jr(1 -E + 2it)cos[Ir/2(1 -E + 2it)]. This followsfromthe fact [2, p. 43] that the Fouriertransformof e-bixaI X J-e+2it b> 0 is ( 6.14) (2r)-r(1e+2t + iX),e-1-2 it + (_bix)e-1-2it]. This convergesto (6. 15) (27r)-la. t I X 12it "The following observations may help clarify the situation. When t = 0, M,+ reduces to the identity transform. This may be regarded as convolution by the Dirac kernel. When t = 0, Mt- reduces to the so-called " Hilbert transform," which apart from a constant factor may be viewed as a convolution by the function 1/x. In this case our result was proved by Hardy and Littlewood [11], whose argument we extend to the general case. 28 R. A. AND KUNZE E. M. STEIN. as b-> 0+,and is boundedbyA I x j'-1,withA independentof b. Now (6. 13) theoremsand the Planchereltheorem. followsby standardconvergence Togetherwith FC Z, considerG(x) Ix 16-IF(x). Since FC CO and vanishesin a neighborhoodof zero and outside of a compactset, the same may be said of G(x). Thus we may apply formula (6. 10) to G as well. Call - (x) A6(x) =Mte(G) (6. 16) x I f-1MItE(F) (x). - Then by (6. 13), +00 + a,x Ae(x)=ae,t (6.17) ] 2E-1-2it F(y)lxaF(y) [|Y y dy. It is easy to verfy,(by the Lebesgue dominatedconvergencetheorem) that (6.18) If we use (6.12) with G in place of F, (6. 18) anid (6.16), x xIX'-MAIt+ (F) (6.19) (FcC asE --0, lAE(x)-AO(x)12-*0, 112 M1It+ (G) 112+ we obtain || AO (X) 112. As has already been noted || Mt+ (G) 112 = G 112 || while 11 and G 112 11 I X |ff-'F 112 i (F) 11 11I 1A-IMt+ 11F jjgn Mt+(F)1 Substitutingthe above in (6. 19) leads to (6. 20) || MIt (F) || _ || F II u+ || AO(X) 112- to estimate11AO(X) therefore It remains + 00 Ao(x) =ao,t + =ao.t y I-F(y) - oo 1 - X 1/1 y Ix I 112. 1-IF(y)I I X-y x If-"F(x) x 1-1-2it y I-'-2It dy j y I0yF(y)dy. , and with j y IjAF (y) in place We now applyLemma 14, witha-f=u of F(y). We thenhave 11AO(X) 112C Ag I aoIt| Recalling(6. 17) we have 112=A( | ao,t F (x) llu However ao,t=- (1/)r(1 + 2it)cos 17r(1 + 2it). REAL UNIIODPULAR 29 GROLUP. Hence by well-knownestimatesin the theoryof the r function,see [22], p. 151, it followsthat Iaot 6A(l + t). Combiningthis withthe abovewe obtain: || Ao(X) + 112 AA(l t F) 11Fo. Togetherwith (6. 20), this implies _- Ag(1 + I t 1 1F 11Mft+(F) 11ag 11aor This was our desiredresultfor Iit+. The prooffor Mt- is verysimilar. The onlychangethat occursis that we use the fact that the Fouriertransformof sgn(x) I x 1-,+2it is I (27r) b,tsgn(x) where bet (i/r) r(1 -E+ xle-1-2it 2it) sin 27r[1 -,+ 2it] This concludesthe proofof the lemma. LEMMA 16. The estimatesfor Mt+ and Mt- may be strengthenedas follows. Let e> 0, then _ A,Je(L + Mt+ 11,g I t I)(l+)k-l O< f< 1 withAor,,independentof t. Proof. Let us consider[t+, and assumethat 1 _ < 1; the othercases are treatedanalogously. We have alreadynoted that MIt+is unitaryon 6Vi. Thus we have + +00 (6. 21) I Mt+(F) 12 dx) +00 + I|Ff12 dx). By the lemmawe have just proved,we have,if a? (6. 22) (4 Mt+(F) 12 1 X 120To-1 dx)^ + f) _Aao(l + I t I)l( < 1, i()2 1 1 x 12ao-l x Notice that the above inequalitiesare of th-esame nature,exceptfor the weightfunctionswhich determinethe measuresin question. Now it is possibleto "interpolate" betweenthese two inequalities,and obtainintermediateones fromthem. Of coursewe have alreadyused many 30 R. A. KUNZE AND E. M. STEIN. varianltsof this type of argumentin ? 3 and ? 4 above. The particular theoremwe need is containedin [20], (Theorem 2). To apply it we argue as follows: Choose ao, so that ur< 2u- 1= ,o < 1. (1-0) We may then write *0 + 0(2o -1) 0(2uo-1), with 0 < 0 < 1. Notice that in the above, u =4 when 0 =0, when0 1. The resultof applyingTheorem2 of [20] is H-owever, + co ao 0 =- (2,o'-1) /(20r0-1). Thus we chooseco-close enoughto 1 so that.0? (2a - 1) (1 +,E). resultbecomes oo (f M|t+(F) I and a Hence the dx) 12 j x 12,r-I ?Aci (1 @ + I't I) (oIr1) (I( I F(x) 12 j X 12a-I dx). Our lemmais therefore proved. We observethat the above proofyieldsthe inequality Ruwemarkc. Aeue ?eAco6m A simple argumentthen allows us to deduce the followingfact: The constantA,, whichappears in (6. 11) maybe takento be uniformly bounded in everyclosedsubintervalof a lyingin 0 < o < 1. This observation will be of use later. CHIAPTER II. UJNIFORMLY BOUNDED REPRESENTATIONS. 7. Proofs of Theorem5 and Theorem6. Beforepresentingthe details of the argument,we shall brieflydiscussthe main steps involvedin the cong --- U- (g, s). structionof the representations Our representations v. constructedon the space are fromrepresentationsg -whic (g, s) on ,12 s m be+it. The operatorsm(g, s) and s) are relatedby Thi(g, 12 For the definitionof the Hilbert space 8W a see ( 6. 1 ) . REAL (7. 1) U (g,s) UNIMODULAR = W(s,j )V 31 GROUP. (g,s)W(1,s), where W(s, 1) is the unitarytransformation (6. 2) of 9a onto 4 The 1 + are Vobtained the representations it) g (g, by simply transferring representations g v-(g, + it) of the continuousprincipalseries fromL2 to S, by means of the Fourier transform. We also obtain the operators V' (g, s), 0 < Rs < 1, via the Fouriertransform in a similar,but technically more involved,fashionfromthe representations g -* v' (g, s). To definethe " operatorsVI (g, s) for <R (s) < 1 it is convenientto extendthe notation r'= 1-cr to complexs with 0 < Rs < 1 by settings' 1 s; the transformations -- s' is then simplyreflectionabout the line a =. Now the 1 to an s with < Rs < 1 is definedto be the representation corresponding of the representation to s'. Thus we put13 contragredient corresponding (7.2) V+(g,s)=[TJ(g-1,s')]', <Rs<1. It followsthat I [V,(g-,~ s) ]' (7. 3) -V(g, sI), O < Rs < 1. It will be shown in the course of the proof that the apparentlyarbitrary definition(7. 2) is the natural one to make. As a firststep in the proofwe shall establishthe followinglemma. 17. The multipliersp+ given by (5. 2), (5. 3) satisfy LEMMA a) p+(g,x,s) = 4(g,x,s), b) +(9192, )c X,S) -=0?(g1 X,S) +(g2,glx (9, g-lx,s ) dg-I(x) ldx s), ?> (9-l x, 1 -s) The firstrelation,a) is immediate,b) is essentiallya consequenceof the chain rule for derivativesapplied to (5. 1), and c) followsby simple computationsfromb) upon setting92 =- g and g1 g-1. As the followinglemma shows, it is natural to restricts so that = 0<Rx< 1. LEMMA 18. Suppose s = cr+ it, where0 ? a ? 1. Then foreach g C G, the operatorsv (g,s) are isometricon Lp, wherep= (1 )-1. Silnce the case p = oo is easily verified,we shall suppose 1 < p < o0. Making the transformation x -* g (x) we findthat 13If A is an operator on a, A' is the operator on Sa' given by (6.6) and (6.7). 32 R. A. 3 x00 I f(x) _00 Now Iv?(g,s)f(x)P KUNZE AND 00 IPdx I bx+ d E. M. STEIN. I bx+ d 1-2 1 f(g(x) ) IPdx. 00 f(g(x)) P, and since (2- 1(2a-2)p 2)p= it followsthat 1vV (g, s) f IIP= 11 f IP. It is interestingto observethat whenp p' is given by p' = 2, ( o) -1 its conjugateindex Thus the operators ve(g, s) and v (g, s') give (1 -o')-1. riseto a pair of isometricrepresentations of G on Lp, Lp wherep) (l )-1 and s + it. Moveover,as thefollowinglemmashows,theserepresentations are contragredient. = LEMMA 19. Let s =a + it and p any gE G, fE Lp, and hC Lp, (v (g, s) f, h) (7. 4) = (l < r,0 < 1. Thzen for (f, v+(g-1,s') h). To provethis we make the tranisformation x - g-1(x) and findthat ^00 (v+ (g, s) f,h) (g, x, s) f (g (x) )h(x) dx oo = (g, g-lx, s)f(x)h(g-lx) (dg-1(x)/dx)dx. Thus, by c) of Lemma 17, 00 f (x) ++(g-1, x, 1 (v+ (g, s) f,h) s) h (g-lx) dx, and now part a) of the same lemmashowsthat (v+ (g, s) f, h) = (f, v (g-1,s') h). We iiow considerthe representation spaces HIa of the complementary series. These spaces are describedin the followinglemma.14 LEMMA 20. Let 0 <r < 2 and p =(1 -) . Then the innlerproduct (5. 5) is well definedfor f in Lp, and the completionHa of Lp with respect to the norm f j2 =- (f, f) is unitarilyequivalentto S9( via a mapping which coincides with the Fourier transformn on Lp. To provethis, suppose firstthat fC L1 n L2 and that F is its Fourier transform.By (6.14), whichis valid for 0 <,E < 1, and the dominatedconvergencetheoremwe obtain 700 .00 (7.5) l _00 14 F(x) 12 1 X 12uf-1 dx ac, f* 00 f (x) I X 1-26 dx. Lemmas 20, 21, and 22 are essentially restatements of known facts. REAL UNIMODULAR 33 GROUP. By Lemma 13 theleftside of (7. 5) is finiteforf C L., and by simpleapproximationarguments,it followsthat the rightside of (7. 5) exists and equals the left side for all f in L.. This showsthat the formula(5. 5) definesan innerproducton Lp. Now observethattheFouriertransform ofLp, 1 < p ? 2, includesthe characteristics functionsof finiteintervalsand theirlinear combinations. This observationtogetherwith (7. 5) establishesthe final statementof the lemmaand concludesthe proof. As a consequenceof Lemma 18 and Lemma 20, we obtainthe fact that the representations g-- v+(g,s) are definedonla dense linear subset of Ha, 0 <~ <K 2. Moreover,as the followinglemma shows,the operatorsv+(g, u) extenduniquelyto unitaryoperatorson Ha. Let 0 <, LEMMA 21. < 1 and p (7. 6) (1 -)-1. 11v+(g,a)f 11a Then for f in Lp, 1 f 11a In provingthis,we use the fact that g (x)- g (y) (x -y) (bx + d) -1(by+ d)-l, = which follows by straightforward computation. Then making the transformationsx-* g(x) and y-*g(y) we see that 00 f Ila42 11 00 jJ *00 - 00 f f(x)f(y) so0 00 00 -~ I x (g(x))f(g(y)) y 1-2adxdy + d 12a-21by+ d 12a-2dxdy 1-2 I bx X -y 00 =j v+(g,o )f Il12. Next we shall showthat thereexistsa uniformbound independentof g for the operatorsv+(g, s) in Ha ; s g + it, 0 <Ka< . In doing this, we considerthe lowertriangularsubgroupof G consistingof elementsg of the form = 9 [a ], a 7&?0 We make essentialuse of the fact that there are only two distinctdouble cosets of G modulo this subgroup. To be explicit,we introducethe group element i and provethe followingresult. 3 [ 1 0] 34 R. A. KUNZE AND E. M. STEIN. LEMMA 22. If g C G and is not lowertriangular,thereexist lower triangulargroupelementsg1 and g2 such that gg1jg2 9 (7. 8) We provethis by exhibitingsuch a decomposition.If g C G and is not lowertriangularwe may write b] [a b#0 Then as is easily verified 9 r1 0n Ldb-1 1L 1 b-i O0 a o -1 on bj In viewof thisresultand the factthatv? (g1g2,s) v? (g1,s) v? (g2, s) for all g1,g2 in G, it is natural to considerthe operators,v+(g, s), firstfor g in the lowertriangularsubgroupand then for g j. = LEMMA 1) + it, where0 < r-< I and -oo < t < oo. Then 23. Let s5= if g is lowertriangular,v+(g,s) has a unique unitaryextensionto all of Ha, and 15 2) _- A,(1 11v+ (j, s) 11,g + I t ). [ In proving 1), we supposethatg = (7. 9 ) v+(g, s) : --> f (x) a ], a #0. 12s-2f (a -2X + a-1c) and v-(g, s) =- sgn(a) v+(g, s). Furthermore, v+(g,s) relationstogetherwith Lemma 21 establishpart 1). operatorsv+(j,s) we findthat (7.10) (7.11) v+(j,s): v- (j, s): f(X) f (X) _ I -_Sgn(x)|I 128-2f( x Thenbydefinition - a |2itV+ (g, ; these Turning now to the JI) 12s-2f(_ JIX) Now with the aid of the operatorsmt+and mt-given by (6. 9) and (6. 10) we can writev+(j, s) =mt+v+ (j, a) and v-(j, s) =mt-v+ (j, a). Since v+(j, a) has a unitaryextension,it followsthat the bounds of the operatorsv+(j, s) are exactlythe boundsof mt+,mt-,considere-d as acting in H,. Now using the factthatH, is unitarilyequivalentto S9( and the definitions of Mt+,Mtwe obtain2) as a consequenceof Lemma 15. 16 The symbol 11-il1designates the bound of the operator on S. REAL UNIMODULAR 35 GROUP. Finally,using Lemma 22 and Lemma 23, we findthat (7.12) sup I v+(g,s) g -_A,(1 + I t ) < 1. 0 <l Because of (7. 12) we may,and shall fromnow on, assumethat the operators v+(g,s) are everywhere definedon H,. Since Ha and 5V, are unitarilyequivalentwe may transferthe representations9 -- v+(g,s) to S9( and obtainequivalentrepresentations g -*> V+(g, s). The operatorsV+(g, s) are obtainedas follows: Let 5,, 0 <a <I, be the unitarytransformation fromHa to S9( that coincideswiththe Fouriertransform (6. 6) on Lp, p = (1 -)'. In addition let be the Fourier trans- 52 formrestrictedto L2; we note that 52 is unitarybetweenL2 and X9i. We now defineV+(g,s) for s=-F+it by (7. 13) 0 <o_ V+(g,s)=5av+(g,s)5gr', . From (7. 12) and the definitions(7. 13), (7. 2) we obtainthe bounds (7. 14) Sup1 V+(g, 9 S) A, A( + t 0 < < 1. This resulttogetherwith (7. 1) implies (7.15) Sup 11<U<(g,1s).1 A( 1+ 9 t 0< < 1 Moreover,as the remarkat the end of ? 6 states,we may assume A, is boundedon any closed subintervalof (0, 1). Hence we have proved 5) of Theorem5, and conclusions2) and 3) followfrom (7.13). To show that (7.2) is a natural definitionwe consideronce again the class of functions0 introducedin ? 6. Recall that FC 0 if F is Cw and vanishesin a neighborhood of zero and outsidea compactset. LEMMA 24. Suppose F,HC 0 and that f,h are their Fourier transforms. Then forall s in the strip0 < Rs < 1, (7. 16) (v+(g, s) f,h) V (g, s)F, H). =( To provethiswe supposefirstof all that0 < Rs ? 1. Thenf,v+(g,s)f C Lp, (1 - )', and 1 < p 2. Our result, (7. 6), now followsfrom the p definition of Th(g, s) F and the Parseval formulafor Lp,Lp', whichis stated in Lemma 12. In case 1 <Rs < 1, VT(g,s) [T(g1,s')]'. V Thus (V-(g, s)F, H) (F, V7(g-1,s')fH.) By the resultjust established, (F, T7(g-1,s')H) = (f,v(g-1, s')h). 36 R. A. KUNZE AND E. M. STEIN. Now applyingLemma 19, we see that (f,v (g-1,s')h) (v (g, s)f, h). Thus (7. 16) also holdsfor 2 < Rs < 1, and henceforall s in 0 < Rs < 1. To prove that the representations g -* U' (g, s), definedby (7. 1), are continuous,it suffices to provethat the representations g -* VT(g, s) are; and forthis,it is sufficient by (7. 2) to considerthe case 0 < Rs 1. Now if f is continuousand has compactsupport,it may be shownthat for bounded functionsh, (g, x,s)f (g (x) ) h(x) dx f (x) h (x) dx as g -* e, e beingthe identityin G. Because the representations g -* v' (g, s), o <RS ? 2, are uniformlyboundedon H, this is sufficientto insure their continuity. Hence the equivalent representations g9-->VT (g, s) are also con- tinuous. It remainsto proveconclusion4) whichrefersto the analyticityof the operatorsUe(g, s). For this purpose we prove a result which has some interestin its owInright. LEMMA 25. If g is a lowertriangularmatrix in G the operatorsU (g, s) are independentof s, 0 < Rs < 1. Let and chooseFE SV,. It then followsfrom (7. 9) and well knownproperties of the Fouriertransform that V (g,s): F (x) (7. 17) e'a I a1 We also obtain the relation V- (g, s) F E .94,we have by definition that Ue (g, s) F 28 F (a2X). sgn(a) V+(g, s). =W (s, 1)V (g, s) W(1 Startingnow with s) F. Hence by (7.17), V (g, s) W(1, s) F(x) eixacI a F(x) ex' e 128 1a2X and applyingW(s, 1) we get (7. 18) U+(g, s) a F (a2x). I-8F(a2X) REAL UNIMODULAR 37 GROUP. Similarly,we obtain the relationU (g, s) sgn(a) U+(g, s). Thus we have provedthe lemma. This resultshowsthat the inner products(U- (g, s) ,7) are constantas functionsof s, and henceanalytic,forany fixedlowertriangularg C G. Now if g is notlowertriangular,it has a decomposition g = g1jg2of thetype (7. 8). Since U+(g, s) = S) U+(j, U+(gl, S) U(g2,s whereU (gi,s), i -1, 2, are independentof s and have boundedinverses,it is sufficient to showthat (U+ (j, s) $, -) is analyticin s foreach pair of vectors Recall the uniformbound, (7.15a) for the representations in N9. e, -* g U+(g, s). Since the constantAa whichappearsis boundedas a function of o- over any closed subintervalof (0, 1), it is sufficient to prove that (U (j],s) e,-) is analyticfora dense collectionof vectorsin 51. Choosethis collectionto be theset 9) of functionswhichare C and vanishin a neighborhood of zero, and outside a compact set. Pick g = F and B = H in . Let F, (x) | x I'--3F(x) and put x1 J-it"H(x). H1s(x) It is theineasilv verifiedthat V T (j, s) F, HIs) (U`+(j, s),e7-) Denote the Fouriertransformns of F8, Hs by f8,h,. Then as F8, He belongto i), Lemma 24 applies,and we see that (U+(j, ,s) e, ) == (v+(j. s) fs,h.) Now using (7. 10), (7. 11) we obtain 00 (7.19) (7. 20) (U+(j, s)4r (U-U,s)e-)= ) - 1n f x 12s-2 fs(_ Jx)h8(x)dx. co sgn (x) oo I X 12s-9-8 llx)h (x) dx Since f. (x) = (2Tr) f eieyI y I- WF(y) dy, 00o and in view of the variousrestriction on F, we may concludethat f8(x) has the followingproperties:it is jointlycontinuousas a functionof x and s; it is analyticin s for each fixedx; and if s is restrictedto any compact subset of the strip, 0 < Rs < 1, negativepowerof I x 1. Since fB(x) decreases as I x | oo as fast as any 38 R. A. KUNZE AND E. M. STEIN. e00 hs(x) = (27rk-f Ie- yIs-l-H (y)dy it has the same properties. It is now a very straightforward matterthat (7. 19), (7. 20) can be obtainedas uniformlimitsof functionsanalyticin s. Hence the inner products (U+ (j, s) $,v ) are analyticin s. This concludes the proofof Theorem5. Conclusion 5) of Theorem 5 may be strengthenedas follows. Givenany E> 0, then COROLLARY. sup 11U(g, s) 1 _ A,a,e(1 + I t j) io-(14+e) 9 O<o-<1. for s =+it, In provingthe theoremwe made use of the estimategiven by (6. 11). If, however,we had used the estimategiven by Lemma 16, we would have obtainedthe above. We shall now Prove 1) of Thetorem6, whichassertsthat the representations U+( ,s) and U+( ,s') are contragredient. In orderto do this,we firstcombine(7. 3) and (6. 6) to obtain (7.21) V-(g,S) WV-(g-1, s)*Wa-1. =- It thenfollowsby definitionthat U+ (g, s') = W(S' 12) WaV+ (g-1, S) *Wa-lw (1, sI). Using the definitionsof W(s', 1),2 W4,togetherwith the fact that s'- ~~~~~~~~~~~~~ +cr Irts 1we findthat W(S 1 ) Wa WV (S5 1) Substitutinginto the above we obtain whichimplies, (7.22) U+ (g, St) Z_w (S, 2 ) V+9? )* 1 ) U+(g, s') U=U?(1, s). Hence we have provedpart 1). The second statementof Theorem6 is easily seen to follow fromthe fact that the representations g -* U+(g, u) are unitaryfor 0 <u < 1. In fact, suppose that g -* U (g, s) are any representations of G on S such that U (g, s') U(g1 U s)* 39 GROUP. REAL UNIMODULAR and forwhichthe innerproducts(U (g, s) e, -) are analyticin s. Then 1 -s) s) =- U(, U(, if and onlyif the representations U( , a) are unitary for eachla, 0 < cr< 1. To prove this, we observe that the condition U ( , s) = U ( ,1 - s) is, for0 < a < 1. U, 1 -) byanalyticity, equivalentto theconditionU(, a () U Hence the a' so that U(g',ar)* On the other hand, 1 U(g, 1 a). g E G. U (g-1,a) above is equivalent to the condition U (g, a) U( ,s) do not satisfy It is interestingto note that the representations 2). This is a reflectionof the knownfact that theyare not unitarywhen to showthat the representations af7/ 1. In orderto provethis,it is sufficient 0 < a < 2, are not unitary. Withoutgoing into detail,we remark v( (a), that this is a consequenceof the relation (7. 23) || v- (j, C) f ||a ao J || f || a" 00 00 J (sgn(x)sgn(y) -1) O00 00 which is valid for all f in Lp, p= (1 f(x) f(y) x -y 2"dxdy, a)1. We supposenow that S is a bounded operatorwith a boundedinverse such that SU( U(, s)S-1 1-s). Replacings by 1 - s and makingsimplecalculationswe findthat S2U- (> s) =UQ (, s) S2. We shall assumetheknownfactthattheunitaryrepresentations + it), U- (, t # 0, are irreducible. It then followsthat S2 is a scalar multiple of the identity. For lowertriangulargroup elementsg of the form 9 g= a-1 we knowthat and U- (g, s) ca 0] ?a] F (x) SU (g, s) =U a , a =/-Z 0, >eiac aF (a2x) (g, s) S. Setting a = 1 and then settingc =- 0 we find that S is the operationof K (a2x). multiplicityby a function,say K, with the propertythat K (x) Since S2 is a scalar multipleof the identity,we obtainthe additionalrelation (const.)sgn(x). (K(x))2= const.,which implies K(x) =const. or K(x) = 40 R. A. KUNZE AND E. M. STEIN. As thefirstalternativeholds if and onlyif U-(, u) is unitaryfor0 < we concludethat K (x) (const.)sgn(x). We shall now defineS by rf< 1 S: F(x) -sgn(x)F(x), (7.24) and prove that SU-(g,s)S-1= U-(g,1 -s) for all g in G. This may be showndirectlyfor all g; however,such a proofdoes not exhibitthe crux of the matter,which,as it turnsout, is the relation )-1 SU-(j, =-- U-(j, 1 -O). We thereforeproceed along different lines and firstof all recall that the operatorsU (g, s) are independentof s forlowertriangulargroupelementsg. For such g, the above relationbecomes SU- (g, s) = U(g, s) S. To verifythis,suppose 9 0] [al a 40. Then SU-(g, s): F(x) and -* sgn(x) eixac aF(a2x) U- (g, s) S: F (x) -> eixac a sgn (a2x) F (a2x). In view of the decomposition(7. 8) it is thereforeseen to be sufficient to provethe relation SU-(j,s) = U(j, 1 -s)S; a, 0< u < Moreover,by analyticity, it is sufficient to provethis for s Now S U= (j, a) S- U (j,a') S-1 W(a',) W (1,"') if and onlyif W('2 ,') S U- Gj) U-(j, a') Thus using the fact that S commuteswith WI(V,c') sufficient to prove (7. 25) V-(j, ') SW(, U') V-(j, a)W(', r)S-1, 0 <a W(a',1 we see that it is < 2' In proving(7. 25) we use the followingconsiderations. The operation SW(o, a') is multiplicationby sgn (x) I x 12*i1. Going over to the Fourier this correspondsto convolutionby b /(27r)isgn(x) I X2f, where transform, REAL UNIMODULAR b= 41 GROUP. (2ou) sinvu. i/7rr (This fact may be establishedin the same way as (6. 13) was; for further discussionsee the proofof Lemma 15, ? 6.) Recalling the definitionof V-(j, u) in termsof the Fourier transform, it then sufficesto prove the following: the operation of convolutionby ba/(27r)lsgn(x)j x 1-2j, followedby the operationf(x) -> sgn(x)I x J-2f(- I/x) is equal to the operationf(x) -> sgn(x) I x 126-2f( 1/x) followedby convolution with ba/(27r)'sgn(x) Ix L2f. This leads to the verification 00 fsgn (x) sgn ( 1/x sgn(y)sgn(y -x) Iy =f f(y)dy y) I X 1-2 1 -/x _ y 1-2a 12-2 _00 y Ix- 1/y)dy. 2f( That this holds may be checkedby the obviouschangeof variables. The argumentabove needs to be made precise. We thereforeargue as follows. In proving(7. 25) it clearlysuffices to showthat (WV-(j, ')SW ( F, ')F,H) =-- (S W (r, u') VW(j,o) Let f be the Fouriertransformof F. for F, H C 0. f1(x) = (27r) f ei$y _00 (F), H) Put sgn(y)j y j21F(y) dy. Then by what has been said before, 4 f (x) = ba/(21r) Ix sgn (x-y) y 1-26f(y)dy. We defineh, and h, similarly;thus it followsthat hi(x) = bu/(27r) Sggn (x y) I x-y 1-2a h (y) dy. Since (SW(a, or')V-(j, a) (F), H) = (W(j, cr)(F), SW(o, ')H), in view of Lemma 24 to showthat it su-ffices (7. 26) Now, (v(j, a)f, hi) -00 (v-(j,')fl, =4 -ba/(27r)i h) = (v-(j,)f, hi). 00 sgn(y)I y I2T-2 f(- 1/y)hi(y)dy h(x)f(- 1/y)sgn(y)sgn(y - x) y 2- Xy 1-2odydx. 42 R. A. KUNZE AND E. M. STEIN. On the otherhand, (v(,'n,h) 00 ,*00 J boJ/(27r)f 00 h(x)f(y)sgn(x)sgn(-1/x-y) 00 j x-y I x 1t2o 1-2?dydx. If we makethechangeofvariablesy >-I/y in the firstdouble-integral, thenit is easilyverifiedthatthisfirstdouble-integral equals the seconddouble integral. This proves (7. 26) and concludesthe proofof Theorem6. CHAPTER III. THE FOURIER-LAPLACE TRANSFORM ON THE GROUP. 8. Hausdorff-Young theoremforthe group and certainof its implications. Let f C L, (G), and let us definethe Fourier transform of f on G as follows: ( 8.1)G +?iJ-(s)= U+(f,s) -( U+(g,s)f(g)dg, O< R(s) < 1, f CLI(G). U+( , s) is the analyticfamilyof representations which act on 91, and whichwerestudiedin ?? 5, 6 and 7. Because foreach fixeds, 0 < R(s) < 1, UT( , s) is a uniformlybounded representation, the integral appearing in (8. 1) is well defined. Moreover, if e, - C , then (51+(S)t) s)e,q) f(g)dg. jw_0=(U+(g An applicationof Fubini's theorem,and the analyticity of U+( , s) showsthat js(F (s)e,-q)ds-O , for any closed curve C in 0 < R(s) < 1. Thus the Fouriertransform 7+(s) is not onlywell-defined whenfC L,(G), and 0 < R (s) < 1, but is also an analyticoperator-valued functionof s in that strip. The resultsof this sectionshow,in a very preciseway, that one may obtain similar results for the Fourier transformof functionsin L,(G), 1 ? p < 2. These facts are contained in the followingtheoremtogether withits corollaries.16 THEOREM 7. Let 1 < p < 2, and q be its conjugateindexI/p + l/q = 1. 16The norms 11- jq, 1 < q ?< ?, are those introduced in ? 2. We recall that the " Hilbert-Schmidt" norm while 11O * 1i denotes the operator bound. 11 112is REAL UNIMODULAR 43 GROUP. There existsa measured/q,t,,(t) so that 00 (fIIq( (8. 2) oo a(C+dt)q,d,(t) it) )1/Q 1 f 11, + it, and 1/q <,c < I/p. For the measured/q,t, f simple,s (t) we have the followingestimate: Givenany 8 > 0, then: = djqu,(t) ? Aq,,, (1 + I t COROLLARY so that )1-Qi-i-Odt. 1. For eachfixedp, 1 < p < 2, thereexistsa u0, l/q < (o (f + 00 11 a|(cr < 1, + it) 1jqQdt)l/9 < Aq,7 11f 11 wheneveru0 <a < 1 - a0, and f is simple. COROLLARY 2. For each p, 1 ? p < 2, sup 11 5 (I + it) IojC Ap 11flIP,f simple. COROLLARY 3. For each p, 1? p < 2, l/q < R (s) < l/p, s + it, 11af (s) jj.,? Ap,o,t 11f j1p, if simple. COROLLARY 4. The Fourier transform, initiallydefinedfor f C L, nLp, has a unique boundedextensionto all of Lp(G), 1 p < 2, withthefollowing property:U+(f, *) is for each f C Lp(G) analyticin s, for l/q < R(s) < I/p. Moreover,the extensionsatisfies(8. 2) as well as the conditionsof Corollaries 1 through3. Remarks. A strictanalogue of the classical Hausdorff-Young theorem would have been a resultlike (8. 2), but onlyfor r= -1. The above results show,however, thatthesameconclusionholdsfora properstripwhichcontains the line uc in its interior. This, togetherwiththe analyticityof 5, has far-reachingconsequences. Once (8. 2) has been proved, the results of " type Corollaries2, 3, and 4 followby ratherstandard" Phragmen-Lindel6f arguments. It is possibleto obtainsomewhatstrongerversionsof Corollaries2 and 3 by replacingthe 11IIo operatornormby the norm 11IIq Since these latter resultsdo not seemto have any immediateapplications,we have not bothered to give theirproofs. A completeFourier analysis of an arbitraryfunction (in the class L2( G) ) necessitatestogetherwith the continuousprincipal series also the discreteprincipalseries. The discussionof the discreteprincipal series is much simpler,and is taken up in the next section. = 44 R. A. KUNZE AND E. M. STEIN. Proof of Theorem 7. Let us considerthe case U+, that of J-being entirelysimilar. On accountof the corollaryto Theorem5 (see ? 7) we may writedownthe followinginequality: (8. 3) (1 4+ sup t +4-it) IIoC AAie 11f 1ll, )-ii-la(+e) 0 <U <1, and c> 0. This inequalityfollowsfromthe above quotedcorollaryand the observation that it)11_sup 11U+(g,s) II'X1f 1ll ~~~~~~~~ 5l+ (a + We knowthat U+( ,+ it) is unitarilyequivalentto the representation v+, + it) of the continuousprincipal series. This series, however,is containedin the Plancherelformula (see ? 5"). Hence we may write down the followinginequality: (8. 4) (j 00 1 + (1 + I t I)-1dt)?-<A 1f 112. it) 1122t2(1 + Here we have used the semi-trivialobservationthat, t2(1 + I t I) -1 Attanhrt, -00 < t < 00. We shall apply Theorem3 to inequalities (8.3) and (8.4) above. We argueas follows. Assumethatcris givenand l/q < f < i/p. We assumefirst thatUr< 1. Let a be a fixedreal numberwith 0 < a < a < but otherwise arbitrary.Rewrite(8. 3) with acinsteadof u. It becomes 27 (8.5) sup -x0< t< x0 Our given p, I < p Now if l/q (8. 6) < < ( -T) l/p= u < +(a+it) I + It withc 1-, or ||- Aa,e || f II (a -)(I+ 2, determines a parameter Tr 0 and + T/2 =-T/2, <T < 1, with /q =T/2. therealwaysexistsan a, 0 < a < u, so that (-) I +PT, (+ 3=)r The above relation determinesac uniquely,which ac we now fix. In applying Theorem 3 to (8. 4) and (8. 5) we make the followingfurther identifications: (8.7) r c= a b Now the resultof Theorem3 is ( I 1 - +(E) 45 REAL TJNIMODULAR GROUP. (8. 8) ( 1+( + it) 11 qq(I + I td|) dt) llq C AE,T 11f Ilp whenever f is simple. A straightforwardcalculation leads to (8.9) dq=1- u-1 I q( ( >O) +E), . Now given any 8 > 0, we can choose an E> 0, small enough so that (8.10) dq I I-Iu- q - Substituting this value of dq in (8. 8) proves (8. 2), wheneverf is simple. The consideration of the case 1 < o-< i/p is carried out in the same manner once one defines a5+ (1 a1+(s) The consideration of 5-(s) of Theorem 5. s). is analogous to 5+(s). This concludes the proof Proof of Corollary 1. Consider the quantity 1-q . This 2-a is the exponent that occurs in the measure d1tq, Recall that 8 was 0(t). arbitrary,except 8 > 0. Notice that if q is fixed we can make the quantity non-negative by choosing 8 small enough and a sufficientlyclose to 1. However cr is also restricted by l/q <Ka < I/p. Thus it is clear that we can realize the conditions of the corollary if we take UO 1- - lq). max (llq, Hence for this choice of uo, the corollary is proved. The proofs of the other corollaries necessitate the following lemma which is along very classical lines. LEMMA 26. Let ? (s) be a (numerical-valued) function analytic in an open region which contains the strip a ?R Suppose that for some c> and furthermore,for some q, q> O(It), as ItIoo, 1, . 00 so< I((a+it) Let a< a< 0 sup I4(o+it)I L , (s)? y < 8. Is dt C- 1I, I (,3 + it) 1Ivdt _- I. 46 R. A. KUNZE AND E. M. STEIN. Conclusion: sup ? + it) (e -x <t<xo _A. A dependson , /3,Y, and q, but does not otherwisedependon c or c. Proof. Let p be the index conjugate to q, l/p + i/p 1. Choose 4 to be a continuousfunctionon (- oo, oo) which vanishes outside a finite interval,and satisfies = (8.11l) z 00 I + (t) |Pdt_ 1, but let 4) be arbitraryotherwise. Define ,(ai + it) by b1a+ it) J _00 a+ it + a _f it,) 0(ti) dti, Then it is easy to verifythat ID (s) is analyticin an open regionwhich containsa ?< R (s) < f8; that sup I 1(oI + it)I = O(I t c), as I tI a-'a',B ; and in view of the assumptionson c1 and (8. 11) that sup I ii (a + it) I _ 1, and -xo<t<xo sup I (I (/ + it) I1. -xo<t<xo We are now in a positionto apply the classical Phragmen-Lindel6f principle to D1,7 The conclusionis that I is bounded by 1 in the entire strip a < R (s) ?fA. In particular, I <DI(0f) I - a -<f-A Going back to the definitionof cIv, we obtain 00 Consideringthe arbitrariness of 4 (exceptfor condition(8. 11)) the converse of lld1der'sinequalityshows: ^00 (8. 12) 1D | (,T + it) Iq dt_ , if a f_. For functionswhichare analyticin a stripand satisfya uniformestimate like (8.12) thereis a knownvariantof Cauchy'sintegralformula. It is 17 See e.g. Titchmarsh [22; p. 181]. REAL UNIMODULAR 47 GROUP. +it) (,ID (a,+ (8. 13) it,) / (a+it, -y - t) )dt, ^00 2..,'( -y-it it, + it, ) / p+ ( ) ) dti, a< Ky < P. In Paley and Wiener ([18] pp. 3-5), (8.13) is demonstratedunder to q 2 in (8.12). However,the proof in the assumptioncorresponding the generalcase, q 1, is no different. If one applies Hldider'sinequalityto each of the integralsin (8. 13) one obtains: > (8. 14) sup where Aopeq 21 [ ( < [Aa + J(y+it) -oo<t<x yq, 00 dtl ( _( a) Aaj~~~~~~~~~~~~~qso + 2 t2 ) p/2 ) Il/p + dt/(- (_ (J (1/p + 1/q )2 + t2)p/2) /p] ) A simplecalculationshows, (8. 15) Aaf,>yq-_C[ (_/ ,)-11q+ -llq /a and with c some absoluteconstant. This concludesthe proofof the lemma. Proof of Corollary2. Let us assumefor simplicitythat 11f IIP 1. Consideringthe indexo, definedin Corollary1, chooseuo < a1 < 1, and keep u1 fixedthroughout the rest of this argument. Now by the choiceof cr1 (and the normalizationimposedon f) we have ^00 *1a- U,f (8. 16) - + it) + 11qq dt_<-Aq, it) Jqq dt? Aq, withA independentof f, for some appropriateA. Choosee and -qto be twovectorsin 4, subjectto the restriction14 v 1, but otherwisearbitrary. Now I I <11a|(s) 11)'11 a+(S)JJq. I (9 (s)t) Hence if we let 1, R. A. KUNZE 48 then (8.17) AN)D E. M. STEIN. f 00 00 fI4 (rl+it)lqdt<1 +it)fqdt?1. 14D(1-al However @F(s) is clearlyanalytic in an open region containing 1c?R (s) < 1 - oh-it is analyticin 0 < R (s) < 1. Moreover,it satisfiesthe growth conditionspecifiedin Lemma 11, withc 1. We also noticea, < 1 < 1 -a,. We thenconclude: sup JC(j+it)j -Xo<t<o A 11,q. Going back to our definitionthis leads to sup 2(5?(+it)4,q)1?<3q. -ox<t<ox Notice that Al,q and hence Bq is independentof 4,,q. Taking the sup over ?1, all 4,, 11411 ?1 we obtain -X sup <t< oO II+(?+it)I?< Bq. If we now dropthe normalization11 f IIP- 1, we obtainthe conclusionof Corollary2. This concludesthe proof. Proof of Corollary3. The proofis similarto that of Corollary2 but is somewhatmorecomplicated. Let f be a simple function. We use inequality (8. 2) which we have alreadyprovedforsuch f. We fixsome8 > 0, and assumemomentarily that f IIP=1. Let us call 11 wherewe choose1/q < <1K Then (8. 2) becomes I r0 ll5+(of1 + it) J]q(1 + ItI)A Goo dt ? (Aq,ci1JX)q. Choose4,fE C , with11411?1,11 ? 1, and let (s) = Then * (a + it) ( 1([a|(r+ ( F+(s), 4,). it) llo_1 ( + it) 1q- The above then becomes I OD IO *((J1 + it) I2 (l + I t )| dt _<(Aq, al,X) q. REAL UNIMODULAR 49 GROUP. Since the formula (8. 2) is symmetricin and 1- oi 01, one also obtains 00 o We let I*(l o1 + it) Iq(1 +I I 1) Xdt<(Aq, 01 x)2q. (D(s)=- cl (2 + s) x/,g(s) . If we choosecl as appropriateconstant(dependingon q, ul, and A) thenthe aboveinequalitiesbecome I f + it),7dt --<I, 00 and fHGo,(1 -, + it) Iq dt? 1. Moreoverit is an easy matterto verifythat '1 (s) satisfiesthe growthconditionspecifiedin Lemma 11. We may thus conclude,(see (8. 15)), |I(?+it) o-O ?C2[I -1l/q] < or < u1 + 1 +a7 [| 1 l/q] 1 -'r, Goingback to the definitions of 1Dand I the above becomes 12 + s X/q I5r(S)e5,l) I< s-a= o + 9J rl|l/a+| C3[I|v it, a, < af < Il-l- 1 1r|11a]5 0-al. Notice that the right-hand side is independent of @, and 7. If we rememberthat $ and v are arbitraryexcept 11J 11 1I,1111_ 1, and we take the sup of the left-handside, droppingthe restriction11 f II 1, we then obtain 2 + s IX/q115?(s) <C3[1 -i -l/q] f (8. 18) -l1/q + I 1 C-1 I 1/ IIPI wheres-o=u+ it, <i of< 1- ar, C3 l/q <o C3 (q, al0 < 12, A). Notice that this formulaactually holds for everys in the open strip l/q < R(s) < i/p. In fact,for such an s, we need onlychoosea o-, so that ul < u < 1 - , and 1/q < o- < 1. If we now fix our A and s, it is clear that (8. 18) implies Corollary3. Proof of Corollary4. It is clear from Corollary3, that whenever l/q < R(s) < l/p, 5- (s) has a unique boundedextensionto all of L.(G). Inequality (8. 18) shows that the bounds are uniformwhenevers is restrictedto a compactsubsetof l/q < R(s) < i/p. But we know that f; is analyticin the strip 0<1R(s) <1, when fcL1(G)n Lp (G). 4 50 R. A. KUNZE AND E. M. STEIN. Hence a simple limitingargumentalso shows that 91 is analytic in 1/q <R(s) < l/p, for each fixedfELp(G). Other limiting arguments (which we will not give) show that the extension9Jt to all of Lp also satisfiesthe inequalities(8. 2) and thosecontained in Corollaries1, 2, and 3. This concludesour discussionof Corollary4. 9. The discreteseries. We now intendto investigatethe formof the Hausdorff-Young theoremfor our group, so far as it involvesthe discrete series. As contrastedwith the case of the continuousseries consideredabove, we do not concernourselveswith an analyticstructurein the discreteseries. This lack is mitigatedby the fact that in the Plancherelformulafor the group,elementsof the discreteseriesoccurwithweightsboundedaway from zero. We beginby provingthe followingtheorem. THEOREM (9.1) 8. Let 1 p<2, (7(k+ k=O and 1/q+1/p= 1. Then D+ 7,)11q + (k +1) D- (f,7) ll 1,q)l wheneverfE L1 (G) n Lp(G). f II Proof. We considerthe measurespace M, definedas follows:The points of M are the pairs (k, ?), wherek runs overthe non-negative integers,and the secondcomponentis either+ or - as indicated. On M we definethe measuredm as follows:The point (k, +) is assigned the measurek +I; the point (k,-) is assignedthe measurelk+ 1. We let 9& denote a separable infinite-dimensional Hilbert space. In accordancewiththe discussionof ? 2 we considerfunctionsfromM to bounded operatorson S. In view of the discretenessof M, all such functionsare automaticallymeasurable. We now definea mappingfromsimplefunctionson G to operatorvalued functionson M. The mapping,whichwe denoteby T, is given by T: f- F={F (c, + )}, and with D?)(f,k) ifD?(g, k)f (g)dg. As explicitlygiven,the representations D+ ( ., 7c) act on different Hilbert REAL UNIMODULAR GROUP. 51 spaces. However,since all separableinfinitedimensionalHilbert spaces are unitarilyequivalent,we may assumethat we deal with appropriateunitarily all of whichact on our given S4. equivalentrepresentations, Using the definitionsof ? 2, (9. 1) becomes 1IT(f) II (9. 2) f 1lq. 11 This is whatwe mustprove Observethat by definition, T(f) 11 co 1122= kFO (k + i) IID+(f, k) 1122 + (k + 1) 1D-(f, k) 1122. Hence, in view of the Plancherelformulafor G, (see ? 5), we have (9.3) T(f) 112Cli 11 f 112. Notice also that = sup 11D (f, k) ||x 11T(f) JIM k,+ while 11D (f, since D( 7,) k) lloo_ || f l1, is unitary. We therefore have, 1IT(f) 11X,11 f Ii. (9. 4) We now use the general interpolationtheoremof ? 3. In the present case the operatorT is independentof z, and so a fortiorisatisfiesthe conditionsof analyticityand admissiblegrowth. We let (p0,qo) (2, 2), and (p,, q,) (1,oo). Then it is easily verifiedthat l/p + 1/q 1, and that we may chooseany p, 1 _ p _ 2, by an appropriatechoice of ,0 <-- T < 1. It is apparentthat in the presentcase A, (y) 1 because of (9. 3), and also A1(y) =1 because of (9.4). The resultof Theorem3 is = = 11 T(f) IIq?_AT1If IIP Since Ao(y) =A1(y) =1, it followsthat logA,==0, and henceA,=1. Thus we have demonstrated(9. 2), and therefore(9. 1), wheneverf is a simplefunction. The extensionof the inequalityto all L1 (G) n Lp (G) followsby standard limitingarguments. This concludesthe proofof the theorem. The followingcorollaryis basic forour applicationsof the abovetheorem. 52 R. A. KUNZE AND E. M. STEIN. COROLLARY. The mapping f-* D+(f,k7) has a unique extensiont to all of Lv(G), and this extensionsatisfiesthe following: (9. 5) sup 1jD-(f, 7c)110? 21-1/P1I f whenever1 _ p ? 2. Proof. We considerfirstthe case when f C L1 n L,. obtain (k + 1lD+ (fnk7) 1qqCk Hence, k=O (7c+1) 11D+ (f|7) Using (9. 1) we llq qC-- 11f IlPq. D1 (f,7k)llJqC 1/(lk+ -1) 11 f ?IP f jlj_ C 2 11 A similarargumentfor D- (f,7k) showsthat sup 11D (f,k) llqC 2/q 11f IP- 21-1/P f 11p. 11 kC,I Since theoperatornormsused aboveare non-increasing withq, (see (2. 2)), we conclude (9. 5), wheneverfC L1 n Lp. In view of the inequality just proved it follows that the mapping f-*D+ (f,k) has a unique extensionto Lp which again satisfies(9. 5). CHAPTER IV. APPLICATIONS. 10. Boundedness of convolutionoperator. We are now in a position to obtain an importantapplicationof the analysis of the previoussections. We shall findit convenientto adopt a slight change in our notation. In this sectionlettersx,y,z, - - will denoteelementsof the group G, and f,g,h, * functionson the group. We recall the operationof convolutionof two functionsf and g, defined as follows . (f * g) (x) dy f f(xy-1)g (y) dy, Haar measure. Now if f C L2, and g E Lp, 1 ? p ? 2, then by Young's inequality (see [23]), f*g is well definedand is in Lr, where1/r=J+1/p -1. TIIEOREM h E L2, and (10.1) 9. Let fE L2, and gLEL h 11 112 C 1? p < 2; if h f11211LIP, A 11 f*g, then REAL UNIMODULAR 53 GROUP. whereAp does not depend on f or g. Ilence the operationof convolution bya functiong E Lp, 1 ? p < 2, is a boundedoperatoron L2. Remarks. Inequality (10. 1) fails when p = 2. This is not surprising for manyreasons; we indicateone such reason. Inequality (10. 1) is essentially a statementof the fact that the Fouriertransformof a functiong in LP, 1 p < 2, is uniformlybounded. But a functionin L2 may be given by appropriatelyassigningits Fourier transform,and this may be done so that the Fourier transformis not uniformlybounded. The statementwhich correspondsto (10. 1) when G is, for example,a non-compactabelian groupis false,as long as p # 1. This is so even in the case when G is the additive group of the real-line. We postponefurther discussionsof these mattersto the next section. to prove inequality(10. 1) for a dense class of Proof. It is sufficient assume that f and g are in L1 (in additionto the fact functions,and so we also in L2 and Lp). that f and g are respectively f * g, and x Noticethat if h then U$ Uh = U (10. 2) Here U, = is any (say unitary) representation, UV. f(x)Ux,dx, with similar definitionsfor U,, and Uh. Moreoverby (2. 13) and (10. 2) we obtain U7I 11 (10. 3) 112?-11 U, 11211Ug 110 We apply (10.3) successivelyto the cases when U U-(,2 + it), (the continuousprincipalseries), and U D+(, ck), (the discreteseries). For the continuousprincipalserieswe apply Corollary2 of Theorem7, (withg in place of f) ; forthe discreteserieswe similarlyapplythe corollary of Theorem8. The result for the continuousseries is (10.4) U(h, 11 g IIP, 4+ it) 112? Ap 11U+(f, 21+ it) 11211 1?S p < 2, withA. independentof t. The resultforthe discreteseriesis (10. 5) 1ID-(h, 7C)112 ? 21-1/P1i D+(f, 7k)112 1j9 I, 1? p ? 2. h I12and 11f 112via thePlancherelformula,(see ? 5). Finally,we calculate11 54 R. A. KUNZE AND E. M. STEIN. It is to be notedthatin computingthe requirednorms,it makesno difference whetherwe use the representations v@( , j + it), or the unitarilyequivalent representationsU-( , 1 + it). Using (10.4) and (10.5) we then easily obtain 1Ih11 2C AP11 2 119 IIPi f 11 1C~P < 2. This proves (10. 1), and hencethe theorem. From theabovetheorem,and withthe use of variousdevices,it is possible to prove otherinequalitieslike (10. 1). All of these have in commonthe remarkablepropertythat they hold for the group we are consideringand also forcompactgroups,butfail in the simplestnon-compact abelianinstances. We shall limitourselvesto the proofof onlyone moresuch result. and COROLLARY. Let fE L2, andgE L2. If h = f * g, then hE Lq, 2 < q< IIh 1q1 Aq 11 f 11211g (10. 6) oo, 112, WhereAq does not dependon f or g. Remark. By the results of the next section it will be seen that this corollaryand the theoremfromwhichit is derivedare essentiallyequivalent results. Proof. Let le L1 n Lp, where i/p + 1/q= 1, but let k be arbitrary otherwise. Then, f h(x)kl(x)dx IC k(x) ff(xy-1l)g(y)dydx g(y) fk(x) where I =* *k, with f* (x) f (xy-1)dxdy= f g(y)l(y)dy, f(x-1). Hence, IJh(x)kc(x)dx g- X (y)lI(y)dy I 1 11211 g112 11 However,by our theorem 1l1 112CAp 11A2 11 f 1k IIIP, since 1 < p < 2. Thus we have I rh(x)kc(x)dxI?AD II 112 11 f112 a 11 11kIp. REAL UNIMODULAR 55 GROUP. k II Now take the sup of the left-handside over7k,such that 11 resultis hlIq? Ap11f 1121g11j2, 11 1. The and the corollaryis proved. 11. Characterizationof unitaryrepresentationsof G. Let g -> U. be (not necessarilyirreducible)on a Hilbert space 54. a unitaryrepresentation of the We now introducetwo notionswhichare basic for our characterization of G. representations Definition. +(g) is an entryfunction,if <<>(g)=--(Uv4,r), (1.1) ~,q) S. Definition. g -> Ug is extendableto Lp(G), if for some fixedp, p 1, fIIP,everyfE L1(G) nLpL(G), C A 11 UfIIX 11 (11. 2) withA independentof f. to notethe followingfacts. Theorem9, whichdealt with It is interesting the boundednessof the operationof convolution,can be restatedby saying is extendableto Lp(G) foreveryp, 1 ? p < 2. thatthe regularrepresentation We may furthernote that the corollaryto Theorem9 statesthat everyentry is in Lq (G), for everyq > 2. functionof the regularrepersentation As a preliminarymatter,we obtain the followingrelationbetweenthe notionsdefinedabove. 27.18 The representation g-*>Ug is extendableto Lp(G), if and onlyif for everypair X, E 9, the entryfunction4, definedin (11.1), lies 1. in Lq (G), where1/p+ 1/q LEMMA Assumefirstthat Ug is extendableto Lp. Let fE L,l nLp. fa(g)f(g) dg J (Ug., r1)f(g)dy U( fUof(g)dg<q)=-(Uf, q). Thus by (11.2) 18 Ths l (g)emma hol fr a l c a1171ct c g 18This lemmaholdsforany locallycompactgroup. p p11 11 11 Then 56 R. A. KUNZE AND E. M. STEIN. We now limitourselvesto thosef's forwhich 11 f II 1, and we take the sup of the left-hand A 1111 side. We thenobtain1pqI ?1qC-Az and thus 1l71 11, 4 C Lq. This provesthe implicationin one direction. To provethe converse we shall use the closed-graphtheoremseveraltimes. We argue as follows. For fixedq, consider themapping * (Ug$ q) =p(g) as a mappingfromX to Lq (G). By the assumptionsof the lemma,it is clear that this mappingis everywhere definedon X ; obviouslyit is linear. We nextnoticethat it is closed. For suposethat 4,n -: , and (9) O)n <q) -> 0o(9) in Lq norm. (Ug$ni However,pn(g) --'p(g) (Ug$ q), foreverygE G. Thus p(g) ===4(g) a.e., and .j (g) -* (g) in Lq norm. This showsthat the mappingis closed. Hence, (11.3) qf:-:: AX I Similarly, (11.4) j 11 11 BCq1f1 Now let f be any functionin Lp(G). We proposeto defineUf. We shall do this by defining(Uf$,-), foreverypair E, N. In fact set (U,) X) = (g)f(g)dg, wherep(g) - (Ugj, q). Since c C Lq, f C L., and l/p + l/q = 1, the integral is well-defined,, by HSlder's inequality. Holder's inequality, (11. 3), and (11.4) furthershow: (11.5) and (11.6) f IIp, I(U,,) ICAX1111 11 X11 fII.P (Ue,,))IB_Bl 11 11 Now (11. 6) showsthat the vectorU,. is well-defined for every C1i E Moreover,(11. 5) and a simpleargument,provethat UJis a closed operator. Hence, usingthe closedgraphtheorem,we obtainthat U,,foreach f e LP(G), is a boundedoperatoron 69 (to itself). Finally, considerthe mapping f- Uf, whichis a mappingfromL, (G) to 6 -Banach space of boundedoperators REAL UNIMODULAR GROUP. 57 on 9 withusual norm. We have just seen that this mappingis everywhere defined. It is clear fromthe definitionthat this mapping is linear. We shall next see that it is closed. In fact,assumef,,-> f in Lp norm,and that UO in the operatornorm. Then Ufn-* ( UtntN foreveryt and E N. Hence ) -->(-Uo*- 1) By (11. 5) it followsthat v)- ( Uwf4nt7) ( Ugt (Uf$,B) (Uo4, ). = Thus Uf U,. Thereforethe mappingf-* Uf is closed. A finalapplication of the closed graph theoremgives IIUfJ C-A JfIIp. This showsthat g -> U. is extendableto L., and the lemema is completely proved. We noticethat the lemma provesthat Theorem9 and its corollaryare equivalentpropositions.It is to be observedthat the identityrepresentation (on the one-dimensional space) is not extendableto Lp if p 7 1. Thus there are very simple representations which are not extendableto Lp if p # 1. We make one furtherremarkbeforewe proceed. Every entryfunctionis automaticallyin L>>(G). Hence a simple argumentshows that if it is in Lq,(G), it is also in Lq (G), whereq > qo. Thereforethe lemmaleads to the factthat if a representation is extendableto Lp,0(G),and po > 1, thenit also is extendableto L. (G), for 1 ? p < po. We are now in a positionto give our characterization of the irreducible unitaryrepresentations of the 2 X 2 real unimodulargroup G. THEOREM 10. Let g -e U, be an irreducibleunitaryrepresentation of G. Assume U is not the identityrepresentation.Theen (a) U is unitarilyequivalentto an elementof the discreteseriesif and onlyif U is extendableto L2 (G). (b) U is unitarilyequivalentto an elementof the continuousprincipal seriesif and onlyif U is extendableto everyL. (G), 1 ? p < 2, but is not extendableto L2(G). (c) U is unitarilyequivalentto the element of the complementary series correspondingto the parametera-, 0 <a- < j, if and only if U is 5 8 R. A. KUNZE for 1 extendable to Lp(G) AND E. p <17(1 M. STEIN. but is not extendable to -r), (G;). Ll/pi_a) COROLLARY. Let g ---U be an irreducible unitaryrepresentation different fromthe identityrepresentation.Then U is unitarilyequivalentto (1) an elementof the discreteseries, (2) an elementof the continuousprincipal series, or, (3) 0 <,v (2') the element of the complementaryseries corresponding to o, every entry function is in L2(G), < 1, if and only if respectively (1') q > 2, but not every entry function every entry function is in Lq(G), is in L2(G), every entry function is in Lq(G), or, (3') q>1/a, but not ev7eryentry fuuction is in L/la(G). Before we pass to the proof of these facts we should like to clarify the difference of notation that we have adopted for the representations of G and that which is used in Bargmann's paper. The parameter a, 0 < a- < 1, which series corres- we have used to identify the elements of the complementary ponds to Bargmann's parameter 1- metrization of the discrete series. a. There is also a difference in para- We have called elements of the discrete series those which appear as discrete summands the Plancherel formula of the group. This (with non-zero measure) exhausts Bargmann's series, except for the representations which he labels D+1/2, and D-1/2. in discrete In our notation these two elements occur as follows. The representation g -* U (g, i) of the continuous principal series is not irreducible. It splits into the direct sum of D+1/2 and D-1/2. Thus in our notation we count D+1/2, and D-1/2 as elements of the continuous principal series. It is with these definitions in mind that the above theorem and corollary are stated. Now to the proof. It is known that every irreducible unitary represen- tation of the group is, except for the trivial representation, up to the unitary equivalence, either an element of the discrete series, the continuous principal series, or the complementary series. By the corollary of Theorem 8, it follows that elements of the discrete series are extendable to L2(G). By Corollary 2 of Theorem 7 it follows that elements of the continuous principal series are extendable to Lp ( G), 1 ? p < 2. If we consider the representation g -* U (g, i) we see that it is also extendable to Lp, for 1 < p < 2. However, this representation splits into two irreducible representations (which we have counted among the continuous principal series). A simple argument shows that each of these pieces is then also extendable to Lp(G), 1 < p < 2. Finally, Corollary 3 of Theorem 7 implies that the element corresponding to a, 0 < a < j, is extendable to Lp, 1 < p < 1/(1-a). We must now show REAL UNIMODULAR 59 GROUP. thatelementsof the continuousprincipalseriesare not extendableto L2 (G), series correspondingto a is not and the element of the complementary (G). We considerfirstthe continuousprincipalseries. extendableto L1/(,-of) to exhibitan entryfunctionwhich is not in By Lemma 27 it is sufficient L2 (U). We considerthe parametrization of the group given by Bargmann with 0 < y <oo, 0?,uc 27r,and 0 ? v< 2r. In this case Haar measure becomes(2,r)2 dydptdv, (see Bargmann (10. 14)). We considerthe "prin" corresponding to thisrepresentation.In Bargmann's cipal sphericalfunction which has the asymptoticexpansion,as y-o o, notationthis is WOO (y), We also have Woo(y) ,-2y-1,1(PSoo (it, o) ytt ) . I,8)"(it, 0) 12= (coth7rt)/47rt, or (tanh-rt)/47rt, dependingon whetherwe are dealing with U+(g, + it) These or U-(g, 1+it), (see Bargmann (11.4), (11.7a1), and (11.7b)). Except for this case we asymptoticrelationsare valid exceptfor U-(g,j). can easilysee that the elementWoo(y) is not in L2 (0, oo; dy), becauseof the factory-1. Thus the corresponding principalsphericalfunctionsare not in L2(G). In considering therepresentation U (g, I) we recallthatit splitsintoD+j and D-T (in Bargmann'snotation). It is also demonstrated by Bargmann to theserepresentations are asympthatthe sphericalfunctionscorresponding Thus in to times these are also not totic constant L2(G). The compley-. in mentaryseriesis dealt withsimilarly. Taking into accountour difference notation, we have /l3o(1 - a, 0) y4, as asymptotic expression (as y -* oo) for the principalsphericalfunctioncorespondingto o-, (see Bargmann (11. 5)). Now clearlythis functionis not in L,Iy(0, oo; dy). Hence the principal is not sphericalfunctionis not in L/1a(G), and thus the representation extendableto L1/(,-u)(G). If we recall Lemma 27, we see that Theorem10 and its corollaryare completelydemonstrated. We now pass to the considerationof not necessarilyirreducibleunitary of our group on a representations.Let g -> Ug be such a representation separableHilbertspace 59. Using the von Neuman reductiontheory[17], and followingSegal [19], we may decomposethe representation as follows. f The Hilbert space 5tlmay be writtenas a directintegral xdo(A) the representation of Hilbertspaces .94X. With respectto this decomposition, g -* U0 may be decomposedinto {U\,}, where g-9 UP is irreducibleand unitary,for a. e. A. 60 R. A. KUNZE AND E. M. STEIN. We do not wish to go into the backgroundof these facts, or into the sensein whichthis reductionis unique. Aside fromthe simplemanipulative factswhichwe shall use, we shall also use the followingfact: Let A be an operatoron X which can be decomposedwith respectto the above decompositionof X4 into the direct integral of the V9X's. We write A - {AX}. A 110 esssup 11 AX10 Then 11 x Our theoremis the following:It may be viewed as an extensionand clarification of Theorem9 and its corollary. 11. Let g-9 U. be a unitary representationof G on 69. Considerits reductioninto a directintegralof irreducibleunitaryrepresenconditionthat (except for a set tationsg -* Uxg. A necessaryand sufficient of measurezero) everyUXgbe unitarilyequivalentto elementsof the discrete g - Us,be extendable or continuousprincipalseriesis that the representation the conditionis equivalent to Lp(G) for everyp, 1 ? p < 2. Alternatively, g -> Ug be in withrequiringthat everyentryfunction.of the representation everyLq(G), 2 < q. THEOREM Proof. Assumefirstthat,disregardinga set of measurezero,everyUP is equivalentto eitherelementsof the discreteseries or of the continuous principalseries. Let f C L1 (G) n Lp(G). Then U- Now 11Uf esssup11Uxf { U",}. 2 of Theorem7, and Becauseof Corollary x the corollaryto Theorem8, we obtain esssup 11Uxfll_A x p < 2; 1 fII, we have disregardedthe set of measurezero which does not correspondto eitherthe discreteor continuousprincipalseries. Hence, 11 Uf11CAP 11f IIP,5 1<p < 25 and g-* Ug is extendableto everyLp(G), 1 ? p < 2. To provethe converse, we argue as follows. Let {fn} be a denumerablecollectionof functionson G Now whichlie in everyL.p(G), and are dense in everyLp(G), 1?p<oo. have we to be can extended Since g-* Ug L.,(G), Ufn -Uxf. Thus, I1IUfnJ1o:IX_APIIfnIp5 essUx sup II flo 1 C-p < 2. fnIIP, Ag 11 1' P < 2 REAL UNIMODULAR GROUP. (1 to the above Let En be the exceptionalset of measurezero corresponding esssup. U En; then E is still of measurezero. Now Let E C Ap || fnIIP, sup 11Ulf. IIOO 1 C p < 2. Owingto the densenessof the collection{f4),we obtain P, f 11 11FIf IIo'< Ap 11 1C5;P < 2~,A E fE L, n LpI ThereforeUxgcan be extendedto Lp, 1 ? p < 2, foreveryAf E. By Theorem 10, and the fact that E is a set of measurezero we obtainthat almostevery belongsto eitherthe discreteor continouusprincipalseries. Using Lemma UXg 27 we obtainthe alternatecondition. This concludesthe proofof Theorem11. Let us now considerthe additivegroupof the line. We shall showthat the analogues of Theorem9 fails for everyp & 1, and that the analogue of the corollaryof Theorem9 fails if q#0oo. In fact let f(x) = (log(2+- I xI)) -, - oo<x<oo. Then by the use of Theorem124 of Titchmarsh[21] it may be shownf is the Fouriertransformof a positivefunctionwhichis in L1 (- oo, oo). A simpleapplication of the Planchereltheoremthen showsthat f is the convolutionof two functions in L2(- oo, oo). However,clearly,f Lq(- oo, so), if q oo. Thus the analogue of the corollaryof Theorem9 fails. Because of Lemma 27 appliedto theregularrepresentation on L2 (-oo, oo), we see thatthe analogue 9 of Theorem fails if p 7A 1. Let us considerthe problemof whetherTheorem9 would hold for our group in the case p- 2. This, clearly,is equivalentto requiringthat the regularrepresentation of the groupis extendableto L2(G). By an argument like that in the proofof Theorem11, it would then followthat the regular representation can be writtenas a directsum of representations equivalentto representations of the discreteseries. This, of course,is not true. MASSACIUSETTS INSTITUTE OF TECHNOLOGY. REFERENCES. [1] V. Bargmann, " Representations of the Lorentz group," Annals of Mathematics, vol. 48 (1947), pp. 568-640. [2] S. Bochner, Vorlesungen uiberFouriersche Integral, New York, 1948. 62 R. A. KUNZE AND E. M. STEIN. [3] A. P. Calderon and A. Zygmund, "On the theorem of Hausdorff-Young and its extension," Contributions to Fourier Analysis, Princeton, 1950. [4] J. Dixmier, " Formes lin6aires sur un anneau d'operateurs," Bulletin de la SociWt6Math6matique de France, vol. 81 (1953), pp. 9-39. [5] L. Ehrenpreis and F. I. Mautner, " Uniformlybounded representationsof groups," Proceedings of the National Academy of Sciences, vol. 41 (1955), pp. 231233. [6] , " Some properties of the Fourier transformon semi-simpleLie groups, I," Annals of Mathematics, vol. 61 (1955), pp. 406-439. , " Some properties of the Fourier transformon semi-simple Lie groups, II," [7] Transactions of the American Mathematical Society, vol. 84, No. 1 (1957), pp. 1-55. [8] R. Godement, " Sur les relations d'orthogonalite de V. Bargmann I," Comptes Rendus Hebdomadaires des Seances de l'Academie des Sciences, vol. 225 (1947), pp. 521-523. , " Sur les relations d'orthogonalite de V. Bargmann II," ibid., vol. 225 [9] (1947), pp. 657-659. [10] P. Halmos, Measure Theory, New York, 1950. [11] G. H. Hardy and J. E. Littlewood, " Some more theoremsconcerningFourier series and Fourier power series," Duke Mathematical Journal, vol. 2 (1936), pp. 355-381. [12] and G. Polya, Inequalities, Cambridge, 1934. [13] Harish-Chandra, " Plancherel formula for the 2 X 2 real unimodular group," Proceedings of the National Academy of Sciences, vol. 38 (1952), pp. 337342. [14] " Representations of semi-simple Lie groups VI," American Journal of Mathematics, vol. 78 (1956), pp. 564-628. [15] E. Hille and R. S. Phillips, Functional analysis and semi-groups, Revised ed., American Mathematical Society Colloquium Publications, vol. XXXI, Providence, 1957. [16] R. A. Kunze, "L, Fourier transforms on locally compact unimodular groups," Transactions of the American Mathematical Society, vol. 89, No. 2 (1958), pp. 519-540. [17] J. von-Neumann, " On rings of operators. Reduction theory," Annals of Mathematics, vol. 50 (1949, pp. 401-485. [18] R. E. A. C. Paley and N. Wiener, Fourier transforms in the complex domain, American Mathematical Society Colloquium Publications, vol. XIX, New York, 1934. [19] I. E. Segal, Decompositions of operator algebras. I," Memoirs of the American Mathematical Society, No. 9, New York, 1951. [20] E. M. Stein, " Interpolation of linear operators," Transactions of the American Mathematical Society, vol. 83 1956), pp. 482-492. [21] E. C. Titchmarsh,Introduction to the theoryof Fourier integrals, Oxford, 1937. [22] , The theoryof functions,2nd ed., Oxford, 1939. [23] A. Weil, L'integration dans les groupes topologiques et ses applications, Paris, 1938. ,
© Copyright 2026 Paperzz