The C*-algebras of certain Lie groups Janne-Kathrin Günther
To cite this version:
Janne-Kathrin Günther. The C*-algebras of certain Lie groups. Algebraic Topology [math.AT]. Université de Lorraine, 2016. English. NNT : 2016LORR0118. tel-01446979
HAL Id: tel-01446979 https://tel.archives-ouvertes.fr/tel-01446979 Submitted on 26 Jan 2017
HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés.
AVERTISSEMENT
Ce document est le fruit d'un long travail approuvé par le jury de soutenance et mis à disposition de l'ensemble de la communauté universitaire élargie.
Il est soumis à la propriété intellectuelle de l'auteur. Ceci implique une obligation de citation et de référencement lors de l’utilisation de ce document.
D'autre part, toute contrefaçon, plagiat, reproduction illicite encourt une poursuite pénale.
Contact : [email protected]
LIENS
Code de la Propriété Intellectuelle. articles L 122. 4 Code de la Propriété Intellectuelle. articles L 335.2- L 335.10 http://www.cfcopies.com/V2/leg/leg_droi.php http://www.culture.gouv.fr/culture/infos-pratiques/droits/protection.htm PhD-FSTC-2016-41 Facult´edes Sciences et Technologies The Faculty of Sciences, Technology and Institut Elie Cartan de Lorraine Communication Ecole Doctorale IAEM Lorraine
DISSERTATION
Defense held on 22/09/2016 in Luxembourg to obtain the degree of
DOCTEUR DE L’UNIVERSITE´ DU LUXEMBOURG EN MATHEMATIQUES´ AND
DOCTEUR DE L’UNIVERSITE´ DE LORRAINE EN MATHEMATIQUES´
by Janne-Kathrin GUNTHER¨ Born on 4 March 1985 in Preetz (Germany)
THE C∗-ALGEBRAS OF CERTAIN LIE GROUPS
Dissertation defense committee Dr Martin OLBRICH, dissertation supervisor Professor, Universit´edu Luxembourg Dr Jean LUDWIG, dissertation supervisor Professor, Universit´ede Lorraine Dr Ingrid BELTITA, reporter Professor, Romanian Academy Dr Joachim HILGERT, reporter Professor, Universit¨at Paderborn Dr Anton THALMAIER, Chairman Professor, Universit´edu Luxembourg Dr Angela PASQUALE, Vice Chairman Professor, Universit´ede Lorraine
Contents
1 Introduction 1
2 General definitions and results 6 2.1 Nilpotent Lie groups ...... 6 2.1.1 Preliminaries ...... 6 2.1.2 Orbitmethod...... 7 2.1.3 The C∗-algebra C∗(G/U, χℓ) ...... 8 2.2 Generalresults ...... 9
3 The C∗-algebras of connected real two-step nilpotent Lie groups 11 3.1 Preliminaries ...... 11 3.1.1 Two-step nilpotent Lie groups ...... 11 3.1.2 Induced representations ...... 11 3.1.3 Orbitmethod...... 12 3.2 Conditions 1, 2 and 3(a) ...... 13 3.3 Condition 3(b) ...... 17 3.3.1 Introductiontothesetting ...... 17 3.3.2 Changing the Jordan-H¨older basis ...... 17 3.3.3 Definitions ...... 19 3.3.4 Formula for πk ...... 20 3.4 Condition 3(b) – First case ...... 22 3.5 Condition 3(b) – Second case ...... 23 3.5.1 Convergence of (πk)k N in G ...... 23 ∈ 3.5.2 Definition of νk ...... 25 3.5.3 Theorem – Second Case . .b ...... 29 V 3.5.4 Transition to πk k N ...... 37 3.6 Condition 3(b) – Third case∈ ...... 38 3.6.1 Convergence of (πk)k N in G ...... 38 ∈ 3.6.2 Definition of νk ...... 41 3.6.3 Theorem – Third Case . . .b ...... 46 V 3.6.4 Transition to πk k N ...... 50 3.7 Result for the connected real∈ two-step nilpotent Lie groups ...... 51 3.8 Example: The free two-step nilpotent Lie groups of 3 and 4 generators . . . . 51 3.8.1 The case n =3...... 52 3.8.2 The case n =4...... 56
4 The C∗-algebra of SL(2, R) 64 4.1 Preliminaries ...... 64 4.2 The spectrum of SL(2, R)...... 67 4.2.1 Introduction of the operator Ku ...... 67 4.2.2 Description of the irreducible unitary representations ...... 76 4.2.3 The SL(2, R)-representations applied to the Casimir operator . . . . . 79 4.2.4 The topology on SL(2, R)...... 81 4.2.5 Definition of subsets Γi ofthespectrum ...... 86 4.3 Condition 3(a) ...... d ...... 87 4.4 Condition 3(b) ...... 90 4.5 Result for the Lie group SL(2, R)...... 95
5 On the dual topology of the groups U(n) ⋉Hn 100 5.1 Preliminaries ...... 100 5.1.1 The coadjoint orbits of Gn ...... 101 5.1.2 The spectrum of U(n)...... 103 5.1.3 The spectrum and the admissible coadjoint orbits of Gn ...... 104 5.2 Convergence in the quotient space gn‡ /Gn ...... 107 5.3 The topology of the spectrum of Gn ...... 121 5.3.1 The representation π(µ,r) ...... 121 5.3.2 The representation π(λ,α) ...... 123 5.3.3 The representation τλ ...... 143 5.4 Results...... 143
6 R´esum´e´etendu de cette th`ese en fran¸cais 146 6.1 Introduction...... 146 6.2 D´efinitionsetr´esultatsg´en´eraux ...... 150 6.2.1 Les groupes de Lie nilpotents – Pr´eliminaires ...... 150 6.2.2 Les groupes de Lie nilpotents – La m´ethode des orbites ...... 151 6.2.3 Les groupes de Lie nilpotents – La C∗-alg`ebre C∗(G/U, χℓ) ...... 152 6.2.4 R´esultats g´en´eraux ...... 153 6.3 Les C∗-alg`ebres des groupes de Lie connexes r´eels nilpotents de pas deux . . 154 6.3.1 Pr´eliminaires – Les groupes de Lie nilpotents de pas deux ...... 154 6.3.2 Pr´eliminaires – La m´ethode des orbites ...... 154 6.3.3 Les Conditions 1, 2 et 3(a) ...... 156 6.3.4 Condition 3(b) – Introduction de la situation ...... 156 6.3.5 Condition 3(b) – Changement de la base de Jordan-H¨older ...... 157 6.3.6 Condition 3(b) – Premier cas ...... 159 6.3.7 Condition 3(b) – Deuxi`eme cas ...... 160 6.3.8 Condition 3(b) – Troisi`eme cas ...... 162 6.3.9 R´esultat pour les groupes de Lie connexes r´eels nilpotents de pas deux 162 6.4 La C∗-alg`ebre du groupe de Lie SL(2, R) ...... 163 6.4.1 Pr´eliminaires ...... 163 6.4.2 Le spectre de SL(2, R) – Introduction de l’op´erateur Ku ...... 165 6.4.3 Le spectre de SL(2, R) – Description des repr´esentations irr´eductibles unitaires...... 168 6.4.4 Le spectre de SL(2, R) – La topologie sur SL(2, R) ...... 170 6.4.5 Les conditions limites duales sous contrˆole normique ...... 171 6.4.6 R´esultat pour le groupe de Lie SL(2, R) . . .d ...... 172 6.5 Sur la topologie duale des groupes U(n) ⋉Hn ...... 173 6.5.1 Pr´eliminaires ...... 173 6.5.2 Le spectre et les orbites coadjointes admissibles de Gn ...... 174 6.5.3 Convergence dans l’espace quotient gn‡ /Gn et la topologie du spectre de Gn ...... 175 6.5.4 R´esultats ...... 177 7 Appendix 178
8 Acknowledgements 186
9 References 187 Abstract In this doctoral thesis, the C∗-algebras of the connected real two-step nilpotent Lie groups and the Lie group SL(2, R) are characterized. Furthermore, as a preparation for an analysis of its C∗-algebra, the topology of the spectrum of the semidirect product U(n)⋉Hn is described, where Hn denotes the Heisenberg Lie group and U(n) the unitary ∗ group acting by automorphisms on Hn. For the determination of the group C -algebras, the operator valued Fourier transform is used in order to map the respective C∗-algebra into the algebra of all bounded operator fields over its spectrum. One has to find the conditions that are satisfied by the image of this C∗-algebra under the Fourier transform and the aim is to characterize it through these conditions. In the present thesis, it is proved that both the C∗-algebras of the connected real two-step nilpotent Lie groups and the C∗-algebra of SL(2, R) fulfill the same conditions, namely the “norm controlled dual limit” conditions. Thereby, these C∗-algebras are described in this work and the “norm controlled dual limit” conditions are explicitly computed in both cases. The methods used for the two-step nilpotent Lie groups and the group SL(2, R) are completely different from each other. For the two-step nilpotent Lie groups, one regards their coadjoint orbits and uses the Kirillov theory, while for the group SL(2, R) one can accomplish the calculations more directly.
R´esum´e Dans la pr´esente th`ese de doctorat, les C∗-alg`ebres des groupes de Lie connexes r´eels nilpotents de pas deux et du groupe de Lie SL(2, R) sont caract´eris´ees. En outre, comme pr´eparation `aune analyse de sa C∗-alg`ebre, la topologie du spectre du produit semi-direct U(n) ⋉Hn est d´ecrite, o`u Hn d´enote le groupe de Lie de Heisenberg et U(n) le groupe ∗ unitaire qui agit sur Hn par automorphismes. Pour la d´etermination des C -alg`ebres de groupes, la transformation de Fourier `avaleurs op´erationnelles est utilis´ee pour appliquer chaque C∗-alg`ebre dans l’alg`ebre de tous les champs d’op´erateurs born´es sur son spectre. On doit trouver les conditions que satisfait l’image de cette C∗-alg`ebre sous la transfor- mation de Fourier et l’objectif est de la caract´eriser par ces conditions. Dans cette th`ese, il est d´emontr´eque les C∗-alg`ebres des groupes de Lie connexes r´eels nilpotents de pas deux et la C∗-alg`ebre de SL(2, R) satisfont les mˆemes conditions, des conditions appel´ees “limites duales sous contrˆole normique”. De cette mani`ere, ces C∗-alg`ebres sont d´ecrites dans ce travail et les conditions “limites duales sous contrˆole normique” sont explicitement calcul´ees dans les deux cas. Les m´ethodes utilis´ees pour les groupes de Lie nilpotents de pas deux et pour le groupe SL(2, R) sont tr`es diff´erentes l’une de l’autre. Pour les groupes de Lie nilpotents de pas deux, on regarde leurs orbites coadjointes et on utilise la th´eorie de Kirillov, alors que pour le groupe SL(2, R), on peut mener les calculs plus directement. Zusammenfassung In dieser Doktorarbeit werden die C∗-Algebren der zusammenh¨angenden reellen zweistufig nilpotenten Lie-Gruppen und der Lie-Gruppe SL(2, R) charakterisiert. Außer- dem wird die Topologie des Spektrums des semidirekten Produktes U(n) ⋉Hn in Vor- ∗ bereitung auf eine Analyse der zugeh¨origen C -Algebra beschrieben, wobei Hn die Heisenberg-Lie-Gruppe und U(n) die unit¨are Gruppe, die durch Automorphismen auf ∗ Hn wirkt, bezeichnen. F¨ur die Bestimmung der Gruppen-C -Algebren wird die opera- torwertige Fouriertransformation benutzt, um die jeweilige C∗-Algebra in die Algebra der beschr¨ankten Operatorenfelder ¨uber ihrem Spektrum abzubilden. Das Ziel ist, die Bedingungen zu finden, die das Bild dieser C∗-Algebra unter der Fouriertransforma- tion erf¨ullt und sie durch eben diese zu charakterisieren. In der vorliegenden Arbeit wird bewiesen, dass sowohl die C∗-Algebren der zusammenh¨angenden reellen zweistufig nilpotenten Lie-Gruppen, als auch die C∗-Algebra von SL(2, R) dieselben Bedingungen erf¨ullen, n¨amlich die “normkontrollierter dualer Limes”-Bedingungen. Dadurch werden diese C∗-Algebren beschrieben und die “normkontrollierter dualer Limes”-Bedingungen werden in beiden F¨allen explizit nachgewiesen. Die Methoden, die f¨ur die zweistufig nilpo- tenten Lie-Gruppen und die Gruppe SL(2, R) verwendet werden, sind dabei komplett unterschiedlich. F¨ur die zweistufig nilpotenten Lie-Gruppen betrachtet man deren koad- jungierte Bahnen und benutzt die Kirillov-Theorie, w¨ahrend man f¨ur die Gruppe SL(2, R) die Berechnungen direkter durchf¨uhren kann.
1 Introduction
In this doctoral thesis, the structure of the C∗-algebras of connected real two-step nilpotent Lie groups and the structure of the C∗-algebra of the Lie group SL(2, R) will be analyzed. Moreover, as a preparative work for an examination of its C∗-algebra, the topology of the spectrum of the semidirect product U(n) ⋉ Hn will be described, where Hn denotes the Heisenberg Lie group and U(n) the unitary group acting by automorphisms on Hn.
The research of C∗-algebras – an abstraction of algebras of bounded linear operators on Hilbert spaces – began in the 1930s due to a need in quantum mechanics and more precisely in order to serve as mathematical models for algebras of physical observables. Quantum systems are described with the help of self-adjoint operators on Hilbert spaces. Hence, algebras of bounded operators on these spaces were regarded. The term “C∗-algebra” was first introduced in the 1940s by I.Segal for describing norm-closed subalgebras of the algebra of bounded linear operators on a Hilbert space. By now, C∗-algebras represent an important tool in the theory of unitary representations of locally compact groups and in the mathematical description of quantum mechanics. Lie groups were introduced in the 1870s by S.Lie in the framework of the Lie theory in order to examine continuous symmetries in differential equations and their theory has been developed further in the 20th century. Today Lie groups are used in many mathematical fields and in theoretical physics, as for example in particle physics. As Lie groups and other symmetry groups hold an important position in physics, the examination of their group C∗-algebras – C∗-algebras that are constructed by building a completion of the space of all L1-functions on the respective groups – has been continued and advanced. This task represents the mission of this thesis. The method of analyzing a group C∗-algebra employed in this work has been initiated by G.M.Fell in the 1960s. It uses the non-abelian Fourier transform in order to write the C∗-algebras as algebras of operator fields and to study their elements in that way. A requirement to make use of this method and thus to understand these C∗-algebras is the knowledge of their spectrum and its topology which are unknown for many groups. Hence, this represents a major problem of Fell’s procedure. The advancement of this method that is adopted in this thesis has been elaborated in the last years (see [24] and [26]) and makes the study of a large class of group C∗-algebras possible.
In order to be able to understand C∗-algebras, the Fourier transform is an important tool. Denoting by the unitary dual or spectrum A of a C∗-algebra A the set of all equivalence classes of the irreducible unitary representations of A, the Fourier transform (a) =a ˆ of an F element a A is defined in the following way: Oneb chooses in every equivalence class γ A ∈ ∈ a representation (π , ) and defines γ Hγ b (a)(γ) := π (a) ( ), F γ ∈B Hγ where ( ) denotes the C -algebra of bounded linear operators on . Then, (a) is con- B Hγ ∗ Hγ F tained in the algebra of all bounded operator fields over A
l∞(A)= φ = φ(πγ) ( γ) φ :=b sup φ(πγ) op < γ A ∞ ∈B H ∈ b k k γ A k k ∞ n ∈ b o b
1 and the mapping : A l∞(A), a aˆ F → 7→ is an isometric -homomorphism. ∗ b Denoting by dx the Haar measure of the locally compact group G, one can define on L1(G) := L1(G, C) the convolution product in the following way ∗
1 1 f f ′(g) := f(x)f ′ x− g dx f,f ′ L (G) g G ∗ ∀ ∈ ∀ ∈ Z G and obtains a Banach algebra L1(G), . Moreover, one gets an isometric involution on this ∗ algebra by letting 1 1 f ∗(g):=∆ (g)− f g g G, G − ∀ ∈ ∆G being the modular function which is defined to be the positive function fulfilling
1 f tg− dt =∆ (g) f(t)dt f C (G) g G, G ∀ ∈ 0 ∀ ∈ Z Z G G where C0(G) is the space of all compactly supported continuous functions. For every irreducible unitary representation (π, ) of G, a representation (˜π, ) of L1(G) can H H be obtained as follows π˜(f) := f(g)π(g)dg f L1(G). ∀ ∈ GZ This representation turns out to be irreducible and unitary. 1 Now, the C∗-algebra of G is defined as the completion of the convolution algebra L (G) with 1 respect to the C∗-norm of L (G), i.e.
1 C∗(G) C∗(G) := L (G,dx)k·k with f C∗(G) := sup π(f) op, k k [π] G k k ∈ b where G is the spectrum of G, defined as above as the set of all equivalence classes of the irreducible unitary representations of G. Furthermore,b every irreducible unitary representation (˜π, ) of L1(G) can be uniquely written H as an integral in the above shown way and hence, one obtains an irreducible unitary represen- tation (π, ) of G. One therefore gets the following well-known result, that can be found in H [9], Chapter 13.9, and states that the spectrum of C∗(G) coincides with the spectrum of G: \ C∗(G)= G.
b The structure of the group C∗-algebras is already known for certain classes of Lie groups, such as the Heisenberg and the thread-like Lie groups (see [26]) and the ax + b-like groups (see [24]). Furthermore, the C∗-algebras of the 5-dimensional nilpotent Lie groups have been determined in [31], while those of all 6-dimensional nilpotent Lie groups have been characterized by H.Regeiba in [30]. In this thesis, a result shown by H.Regeiba and J.Ludwig (see [31], Theorem 3.5) shall be utilized for the characterization of the C∗-algebras mentioned at the beginning.
2 For the C∗-algebras of the two-step nilpotent Lie groups, the approach will partly be similar but more complex, to the one used for the characterization of the C∗-algebra of the Heisenberg Lie group (see [26]), which is also two-step nilpotent and thus serves as an example. In the case of the Heisenberg Lie group, the situation is a lot easier than for general two-step nilpotent Lie groups, as in this special case the appearing coadjoint orbits can only have two different dimensions, while in the general case, there are many different dimensions that can appear. Consequently, much more complicated cases emerge. The 5- and 6-dimensional nilpotent Lie groups, in turn, have a more complicated structure than the two-step nilpotent Lie groups. Here, non-flat coadjoint orbits appear, whereas in the two-step nilpotent case, one only has to deal with flat coadjoint orbits. In these cases of 5- and 6-dimensional nilpotent Lie groups, one therefore examines every group by itself, while for the two-step nilpotent Lie groups, one covers this whole class of groups without knowing their exact structure. As mentioned above, the approach for the group SL(2, R) will be completely different. Since one treats one single group which is explicitly given, in this case, one can carry out more concrete computations. For semisimple Lie groups in general, there is no explicit description of their group C∗-algebras given in literature. However, for those semisimple Lie groups whose spectrum is classified, the procedure of the determination of the group C∗-algebra used in this work might be successfully applied in a similar way. A description of the C∗-algebra of the Lie group SL(2, C) is given in [13], Theorem 5.3 and Theorem 5.4, and a characterization of reduced group C∗-algebras of semisimple Lie groups can be found in [34]. In the present doctoral thesis, the group C∗-algebras of connected real two-step nilpotent Lie groups and of SL(2, R) shall be described very explicitly. It will be shown that they are characterized by specific conditions that are called “norm controlled dual limit” conditions which will be given below. The main objective is to concretely compute these conditions and to construct the required “norm controls”. In an abstract existence result (see [2], Theorem 4.6), these conditions are shown to hold true for all simply connected connected nilpotent Lie groups. They are explicitly checked for all 5- and 6-dimensional nilpotent Lie groups (see [31]), for the Heisenberg Lie groups and the thread-like Lie groups (see [26]). The results about the group C∗-algebras of the connected real two-step nilpotent Lie groups have been published in the “Revista Matem´atica Complutense” as a joint article with J.Ludwig (see [17]). A further article about the group C∗-algebra of SL(2, R) has been submitted to the “Journal of Lie Theory” and can be found on arXiv (see [16]). Since for the determination of a group C∗-algebra the spectrum of the group and its topology are necessary, in this doctoral thesis, the topology of the spectrum of the groups G = U(n) ⋉H for n N = N 0 shall be analyzed as well. This work was started by n n ∈ ∗ \{ } M.Elloumi and J.Ludwig (see [11], Chapter 3) and their preprint will be elaborated in this thesis. Like for several other classes of Lie groups, for example the exponential solvable Lie groups and the Euclidean motion groups, the spectrum Gn can be parametrized by the space of all admissible coadjoint orbits in the dual gn∗ of the Lie algebra gn of Gn. The aim is to show that the bijection between thec space of all admissible coadjoint orbits equipped with the quotient topology and the spectrum Gn of Gn is a homeomorphism. This has already been shown for the classical motion groups SO(n) ⋉Rn for n 2 in [12]. ≥ c
3 For the groups G = U(n) ⋉H , this claim has been proved for n = 1. For general n N , it n n ∈ ∗ could be shown that the mapping between the spectrum Gn and the space of all admissible coadjoint orbits of Gn is continuous. The remaining parts of the proof of the claim are work in progress. c
In this section, the definition of a C∗-algebra with “norm controlled dual limits” will be given. The conditions required by this definition characterize group C∗-algebras. There- after, in Section 3, they will be shown to hold true for the two-step nilpotent Lie groups. After having completed this proof, an example of a two-step nilpotent Lie group will be examined. In Section 4, these conditions will then be proved in the case of the Lie group SL(2, R), in a very different approach than the one used for the two-step nilpotent Lie groups.
Definition 1.1. A C∗-algebra A is called C∗-algebra with norm controlled dual limits if it fulfills the following conditions:
- Condition 1: Stratification of the spectrum:
(a) There is a finite increasing family S S ... S = A of closed subsets of the 0 ⊂ 1 ⊂ ⊂ r spectrum A of A in such a way that for i 1, ,r the subsets Γ := S and ∈ { ··· } 0 0 Γi := Si Si 1 are Hausdorff in their relative topologies andb such that S0 consists of \ − all the charactersb of A. (b) For every i 0, ,r , there is a Hilbert space and for every γ Γ , there is a ∈ { ··· } Hi ∈ i concrete realization (π , ) of γ on the Hilbert space . γ Hi Hi - Condition 2: CCR C∗-algebra: A is a separable CCR (or liminal) C∗-algebra, i.e. a separable C∗-algebra such that the image of every irreducible representation (π, ) of A is contained in the algebra of compact H operators ( ) (which implies that the image equals ( )). K H K H - Condition 3: Changing of layers: Let a A. ∈ (a) The mappings γ (a)(γ) are norm continuous on the different sets Γ . 7→ F i (b) For any i 0, ,r and for any converging sequence contained in Γ with limit set ∈ { ··· } i outside Γi (thus in Si 1), there is a properly converging subsequence γ = (γk)k N, − as well as a constant C > 0 and for every k N an involutive linear mapping∈ ∈ ν˜k =ν ˜γ,k : CB(Si 1) ( i), which is bounded by C Si (uniformly in k), such − →B H k · k −1 that
lim (a)(γk) ν˜k (a) Si = 0. k F − F | −1 op →∞ Here, CB(Si 1) is the -algebra of all the uniformly bounded fields of operators − ∗ ψ(γ) ( l) , which are operator norm continuous on the subsets Γl ∈B H γ Γl,l=0, ,i 1 for every l 0, ∈ ,i ···1 −, provided with the infinity-norm ∈ { ··· − }
ψ Si−1 := sup ψ(γ) op. k k γ Si k k ∈ −1
4 Theorem 1.2 (Main result). The C∗-algebras of the connected real two-step nilpotent Lie groups G and of the Lie group G = SL(2, R) have norm controlled dual limits. \ In order to prove Theorem 1.2, concrete subsets Γi and Si of C∗(G) ∼= G will be defined and the mappings ν˜k k N will be constructed. ∈ b The norm controlled dual limit conditions completely characterize the structure of a C -algebra in the following sense: Taking the number r, the Hilbert spaces , the sets Γ ∗ Hi i and S for i 0,...,r and the mappings ν˜ required in the above definition, by [31], i ∈ { } k k N Theorem 3.5, one gets the result below for the∈ C -algebras of the two-step nilpotent Lie ∗ groups G and of G = SL(2, R).
Theorem. The C∗-algebra C∗(G) of a connected real two-step nilpotent Lie group G and the C∗-algebra of the Lie group G = SL(2, R), respectively, are isomorphic (under the Fourier transform) to the set of all operator fields ϕ defined over the spectrum G of the respective group such that:
1. ϕ(γ) ( ) for every i 1,...,r and every γ Γ . ∈K Hi ∈ { } ∈ b i 2. ϕ l (G). ∈ ∞ 3. The mappings γ ϕ(γ) are norm continuous on the different sets Γ . b 7→ i 4. For any sequence (γk)k N G going to infinity, lim ϕ(γk) op = 0. ∈ ⊂ k k k →∞ 5. For i 1,...,r and anyb properly converging sequence γ = (γk)k N Γi whose ∈ { } ∈ ⊂ limit set is contained in Si 1 (taking a subsequence if necessary) and for the mappings − ν˜k =ν ˜γ,k : CB(Si 1) ( i), one has − →B H
lim ϕ(γk) ν˜k ϕ Si = 0. k − | −1 op →∞
In the case of the Lie group G = SL(2, R), this result can still be simplified and more concretely described in Theorem 4.21.
5 2 General definitions and results
2.1 Nilpotent Lie groups 2.1.1 Preliminaries Let g be a real nilpotent Lie algebra. Fix a scalar product , on g and take on g the Campbell-Baker-Hausdorff multiplication h· ·i 1 1 1 u v = u + v + [u,v]+ u, [u,v] + v, [v,u] + ... u,v g, · 2 12 12 ∀ ∈ which is a finite sum, as g is nilpotent. This gives the simply connected connected Lie group G = (g, ) with Lie algebra g. The · exponential mapping exp : g G =(g, ) is in this case the identity mapping. → · The Haar measure of this group is a Lebesgue measure which is denoted by dx.
Convention 2.1. Throughout this work, all function spaces are spaces of complex valued functions.
For a vector space V , denote by the Schwartz space (V ) the space of all rapidly decreasing S smooth functions on V . Then, as G =(g, ), one can define (G) := (g). · 1 S S Define on g and thus on G a norm by := , 2 . If X ,...,X represents an orthonormal k·k |h· ·i| { 1 n} basis of g with respect to , and α = (α ,...,α ) Nn, define Xα := Xα1 Xαn in the h· ·i 1 n ∈ 1 ··· n universal enveloping algebra (g). Moreover, denote by f the derivative of f (G) in the U α′ ∈S direction of Xα for every α Nn. Then, the semi-norms on (G) can be given by ∈ S
2 N f = 1+ g f ′ (g) dg f (G) N N. k k(N) k k α ∀ ∈S ∀ ∈ α N Z | X|≤ G
The space (G) is dense in L1(G) and in turn, for a vector space V , a well-known result S states that the space f (V ) fˆ is compactly supported is dense in (V ). Hence, ∈ S | S f (G) fˆ is compactly supported is dense in (G) as well. ∈S | S Now, for a linear functional ℓ of g, consider the skew-bilinear form
B (X,Y ) := ℓ, [X,Y ] ℓ h i on g. Moreover, let
g(ℓ) := X g ℓ, [X, g] = 0 ∈ | h i { } be the radical of Bℓ and the stabilizer of the linear functional ℓ. Definition 2.2 (Polarization). A subalgebra p of g, that is subordinated to ℓ (i.e. that fulfills ℓ, [p, p] = 0 ) and that has h i { } the dimension 1 dim(p)= dim(g)+ dim(g(ℓ)) , 2 which means that p is maximal isotropic for Bℓ, is called a polarization of ℓ.
6 If p g is any subalgebra of g which is subordinated to ℓ, the linear functional ℓ defines a ⊂ unitary character χℓ of P := exp(p): 2πi ℓ,log(x) 2πi ℓ,x χ (x) := e− h i = e− h i x P. ℓ ∀ ∈ Definition 2.3 (The coadjoint action). Define for all g G the mappings ∈ 1 α : G G, x gxg− and Ad (g) := d(α ) : g g. g → 7→ g e → Then, Ad : G GL(g) is a group representation. The coadjoint action Ad ∗ can be defined → as follows:
1 Ad ∗(g): g∗ g∗ g G, Ad ∗(g)ℓ(X)= ℓ Ad g− X ℓ g∗ X g. → ∀ ∈ ∀ ∈ ∀ ∈ Ad ∗ : G GL(g ) is another group representation. → ∗ The space of coadjoint orbits Ad ∗(G)ℓ ℓ g is denoted by g /G. { | ∈ ∗} ∗ Furthermore, define the mapping
ad ∗(X): g∗ g∗ X g, ad ∗(X)ℓ(Y )= ℓ([Y,X]). → ∀ ∈ ad ∗ : g GL(g ) is a representation of the Lie algebra g. → ∗ Definition 2.4 (Induced representation). Let H be a closed subgroup of G and define
L2(G/H, χ ) := ξ : G C ξ measurable, ξ(gh)= χ (h)ξ(g) g G h H, ℓ → ℓ ∀ ∈ ∀ ∈
ξ 2 := ξ(g) 2dg˙ < , k k2 | | ∞ G/HZ where dg˙ is an invariant measure on G/H which always exists for nilpotent G. 2 L (G/H, χℓ) is a Hilbert space and one can define the induced representation G 1 2 ind χ (g)ξ(u) := ξ g− u g,u G ξ L (G/H, χ ). H ℓ ∀ ∈ ∀ ∈ ℓ This is a unitary representation. If the associated Lie algebra h of H is a polarization of ℓ, this representation is also irreducible.
2.1.2 Orbit method By the Kirillov theory (see [8], Section 2.2), for every representation class γ G, there exists ∈ an element ℓ g and a polarization p of ℓ in g such that γ = indGχ , where P := exp(p). ∈ ∗ P ℓ Moreover, if ℓ,ℓ′ g∗ are located in the same coadjoint orbit g∗/G band p and p′ ∈ O ∈ G G are polarizations in ℓ and ℓ′, respectively, the induced representations indP χℓ and indP ′ χℓ′ are equivalent and the Kirillov map which goes from the coadjoint orbit space g∗/G to the spectrum G of G G K : g∗/G G, Ad ∗(G)ℓ ind χ → 7→ P ℓ b is a homeomorphism (see [6] or [23], Chapter 3). Therefore, b g∗/G ∼= G as topological spaces. b
7 Definition 2.5 (Properly converging). Let T be a second countable topological space and suppose that T is not Hausdorff, which means that converging sequences can have many limit points. Denote by L((tk)k N) the collection of ∈ all the limit points of a sequence (tk)k N in T . A sequence (tk)k N is called properly converging ∈ ∈ if (tk)k N has limit points and if every subsequence of (tk)k N has the same limit set as (tk)k N. ∈ ∈ ∈ It is well-known that every converging sequence in T admits a properly converging subse- quence.
2.1.3 The C∗-algebra C∗(G/U, χℓ) Let u g be an ideal of g containing [g, g], U := exp(u) and let ℓ g such that ℓ, [g, u] = 0 ⊂ ∈ ∗ h i { } and such that u g(ℓ). Then, the character χℓ of the group U = exp(u) is G-invariant. One ⊂ 1 can thus define the involutive Banach algebra L (G/U, χℓ) as
1 1 L (G/U, χ ) := f : G C f measurable, f(gu)= χ (u− )f(g) g G u U, ℓ → ℓ ∀ ∈ ∀ ∈
f := f(g) dg˙ < . k k1 | | ∞ G/UZ
The convolution
1 f f ′(g) := f(x)f ′(x− g)dx˙ g G ∗ ∀ ∈ G/UZ
and the involution
1 f ∗(g) := f(g ) g G − ∀ ∈ are well-defined for f,f L1(G/U, χ ) and ′ ∈ ℓ
f f ′ f f ′ . k ∗ k1 ≤ k k1 k k1 Moreover, the linear mapping
p : L1(G) L1(G/U, χ ), p (F )(g) := F (gu)χ (u)du F L1(G) g G G/U → ℓ G/U ℓ ∀ ∈ ∀ ∈ UZ is a surjective -homomorphism between the algebras L1(G) and L1(G/U, χ ). ∗ ℓ Let I Gu,ℓ := [π] G π U = χℓ U π . ∈ | | H n o Then, Gu,ℓ is a closed subsetb of G, which canb be identified with the spectrum of the algebra L1(G/U, χ ). Indeed, it is easy to see that every irreducible unitary representation (π, ) ℓ Hπ whose equivalenceb class is locatedb in Gu,ℓ defines an irreducible representation (˜π, π) of the 1 H algebra L (G/U, χℓ) as follows: b π˜ p (F ) := π(F ) F L1(G). G/U ∀ ∈ 8 Similarly, if (˜π, ) is an irreducible unitary representation of L1(G/U, χ ), then [π] given by Hπ˜ ℓ π :=π ˜ p ◦ G/U
defines an element of Gu,ℓ. Let s g be a subspace of g such that g = g(ℓ) s. Since u contains [g, g], one has ⊂ ⊕ b G = χ π q (u + s)⊥ , u,ℓ q ⊗ ℓ | ∈ G letting πℓ := indP χℓ for a polarizationb p of ℓ and P := exp(p). 1 Denote by C∗(G/U, χℓ) the C∗-algebra of L (G/U, χℓ), whose spectrum can also be identified with Gu,ℓ. For π := indGχ , the Fourier transform defined by ℓ+q P ℓ+q F b (a)(q) := π (a) q (u + s)⊥ F ℓ+q ∀ ∈ then maps the C∗-algebra C∗(G/U, χℓ) onto the algebra C (u + s)⊥, ( π ) of the ∞ K H ℓ continuous mappings ϕ :(u + s) ( ) vanishing at infinity with values in the algebra ⊥ →K Hπℓ of compact operators on the Hilbert space of the representation π . Hπℓ ℓ If one restricts p to the Fr´echet algebra (G) L1(G), its image is the Fr´echet algebra G/U S ⊂ 1 (G/U, χ ) = f L (G/U, χ ) f smooth and for every subspace s′ g with g = s′ u S ℓ ∈ ℓ | ⊂ ⊕ and for S′ = exp(s′), f S′ (S′) . | ∈S
2.2 General results In this section, some general results shall be listed which are later needed in several proofs. Theorem 2.6 (Schur’s lemma). A unitary representation π of a topological group G on a Hilbert space is irreducible if and H only if the only bounded linear operators on commuting with π(g) for all g G are the H ∈ scalar operators. For the proof of this theorem see [20], Chapter I.3. Lemma 2.7. Let G be a Lie group and g the corresponding Lie algebra. Define the canonical projection pG going from g∗ to the space of coadjoint orbits g∗/G and equip the space g∗/G with the quotient 1 topology, i.e. a subset U in g∗/G is open if and only if pG− (U) is open in g∗. Then, a sequence ( k)k N of elements in g∗/G converges to the orbit g∗/G if and only if for any ℓ O ∈ O ∈ ∈ O there exists ℓk k for every k N such that ℓ = lim ℓk. ∈O ∈ k →∞ For the proof of this Lemma, see [23], Chapter 3.1.
Now, let G be a second countable locally compact group and (π, ) an irreducible Hπ unitary representation on the Hilbert space . A function of positive type of π is defined Hπ to be a linear functional π C Cξ : G , g π(g)ξ,ξ π , −→ 7−→ h iH where ξ is a cyclic vector in . Hπ
9 The spectrum G has a natural topology which can be characterized in the following way: Theorem 2.8 (Topology of the spectrum). b Let (πk, πk )k N be a family of irreducible unitary representations of G. Then, (πk)k N con- H ∈ ∈ verges to (π, π) in G if and only if for some non-zero (respectively for every) vector ξ in π, H π H for every k N there exists ξ such that the sequence C k = π ( )ξ ,ξ k πk ξk k N k k k π k N ∈ ∈H ∈ · H k ∈ converges uniformlyb on compacta to Cπ = π( )ξ,ξ . ξ π · H The topology of G can also be expressed by the weak convergence of the coefficient functions: Theorem 2.9. b Let πk, πk k N be a sequence of irreducible unitary representations of G. Then, πk, πk k N H ∈ H ∈ converges to(π, π) in G if and only if for some (respectively for every) unit vector ξ in π, H N H there is for every k a vector ξk π with ξk π 1 such that the sequence of linear ∈ ∈H k k kH k ≤ functionals Cπk convergesb weakly on some dense subspace of C (G) to Cπ. ξk k N ∗ ξ ∈ The proofs of the Theorems 2.8 and 2.9 can be found in [9], Theorem 13.5.2.
For the following corollary, let G be a Lie group, g its Lie algebra and (g) the uni- U versal enveloping algebra of g. Furthermore, for a unitary representation (π, ) of G, let Hπ := f G g π(g)f is smooth , the subspace of of smooth vectors. Hπ∞ ∈Hπ| ∋ 7→ ∈Hπ Hπ Corollary 2.10. Let πk, πk k N be a sequence of irreducible unitary representations of the Lie group G. If H ∈ πk, πk k N converges to (π, π) in G, then for each unit vector ξ in π∞, there exists for H ∈N H H every k a vector ξk ∞ such that the sequence dπk(D)ξk,ξk N converges to πk π k ∈ ∈ H b h H k ∈ dπ(D)ξ,ξ π for each D (g). h iH ∈U The proof below of this corollary originates from [11].
Proof: Let ξ a unit vector. It follows from [10], Theorem 3.3, that there are functions f ,...,f ∈Hπ∞ 1 s in C0∞(G), the space of all compactly supported C∞(G)-functions, and linearly independent vectors ξ ,...,ξ , such that ξ = π(f )ξ + ... + π(f )ξ . Since π is irreducible, for any 1 s ∈ Hπ 1 1 s s non-zero vector η π and every j 1...,s , there exist elements qj in the C∗-algebra of ∈ H s∈ { } G fulfilling ξj = π(qj)η. Hence, ξ = π(fj)π(qj)η. j=1 Now, choose for every k N vectorsPη such that Cπk converges weakly to Cπ. k πk ηk k N η ∈ s ∈ H ∈ Furthermore, for k N let ξk := πk(fj)πk(qj)ηk. Then, for D (g) it follows that ∈ j=1 ∈U P s s lim dπk(D)ξk,ξk π = lim πk(D fj)πk(qj)ηk, πk(fi)πk(qi)ηk k h iH k k ∗ →∞ →∞ π Xj=1 Xi=1 H k s s
= lim πk qi∗ fi∗ D fj qj ηk,ηk k ∗ ∗ ∗ ∗ π →∞ k Xi=1 Xj=1 D EH s s
= π qi∗ fi∗ D fj qj η,η = dπ(D)ξ,ξ π . ∗ ∗ ∗ ∗ π h iH i=1 j=1 H X X D E
10 3 The C∗-algebras of connected real two-step nilpotent Lie groups
In this section, the connected real two-step nilpotent Lie groups will be examined. In its first part, some preliminaries about two-step nilpotent Lie groups will be given which are needed in order to understand the setting and to prepare the proof of the norm controlled dual limit conditions listed in Definition 1.1. In Section 3.2, the Conditions 1, 2 and 3(a) will be verified and in Section 3.3, Condition 3(b) will be examined. Its proof is divided into three cases which are treated in the three following subsections. Subsequently, a result describing the C∗-algebras of connected real two-step nilpotent Lie groups will be given. After having finished the characterization of general connected real two-step nilpotent Lie groups, the example of the free two-step nilpotent Lie groups will be regarded.
The proof of Condition 1 of Definition 1.1 consists in the construction of the different layers of the spectrum of G and is based on a method of construction of a polarization developed in [27]. It is rather technical but not very long. Condition 2 follows directly from a result in [8] and Condition 3(a) is due to the construction of the above-mentioned polarization. The main work of the examination of the C∗-algebras of two-step nilpotent Lie groups consists in the proof of Property 3(b) of Definition 1.1 and in particular in the construction of the mappings (˜νk)k N. ∈ 3.1 Preliminaries 3.1.1 Two-step nilpotent Lie groups Let g be a real Lie algebra which is nilpotent of step two. This means that
[g, g] := span [X,Y ] X,Y g | ∈ is contained in the center of g. Fix a scalar product , on g like for general nilpotent Lie algebras. The Campbell-Baker- h· ·i Hausdorff multiplication is then of the form 1 u v = u + v + [u,v] u,v g · 2 ∀ ∈ and one gets again the simply connected connected Lie group G =(g, ) with Lie algebra g. · As g is two-step nilpotent, [g, g] g(ℓ) and thus g(ℓ) is an ideal of g. ⊂ Furthermore, for ℓ g , every maximal isotropic subspace p of g for B containing [g, g] is a ∈ ∗ ℓ polarization of ℓ.
3.1.2 Induced representations G The induced representation σℓ,p = indP χℓ for a polarization p of ℓ and P := exp(p) can be described in the following way: Since p contains [g, g] and even the center z of g, one can write g = s p and p = t z for two ⊕ ⊕ subspaces t and s of g. The quotient space G/P is then homeomorphic to s and the Lebesgue measure ds on s defines an invariant Borel measure dg˙ on G/P . The group G acts by the left 2 translation σℓ,p on the Hilbert space L (G/P, χℓ).
11 Now, if one uses the coordinates G = s p, one can identify the Hilbert spaces L2(G/P, χ ) · ℓ and L2(s)= L2(s,ds): Let U : L2(s,ds) L2(G/P, χ ) be defined by ℓ → ℓ U (ϕ)(S Y ) := χ ( Y )ϕ(S) Y p S s ϕ L2(s). ℓ · ℓ − ∀ ∈ ∀ ∈ ∀ ∈
Then, Uℓ is a unitary operator and one can transform the representation σℓ,p into a represen- 2 tation πℓ,p on the space L (s):
π := U ∗ σ U . (1) ℓ,p ℓ ◦ ℓ,p ◦ ℓ
Furthermore, one can express the representation σℓ,p in the following way:
1 1 σ (S Y )ξ(R) = ξ Y − S− R ℓ,p · 1 1 = ξ (R S) Y + [R,S] [R S,Y ] − · − 2 − 2 − 2πi ℓ, Y + 1 [R,S] 1 [R S,Y ] 2 = e h − 2 − 2 − iξ(R S) R,S s Y p ξ L (G/P, χ ) − ∀ ∈ ∀ ∈ ∀ ∈ ℓ and hence,
2πi ℓ, Y + 1 [R,S] 1 [R S,Y ] 2 π (S Y )ϕ(R) = e h − 2 − 2 − iϕ(R S) R,S s Y p ϕ L (s). (2) ℓ,p · − ∀ ∈ ∀ ∈ ∀ ∈ 3.1.3 Orbit method In the case of two-step nilpotent Lie groups, for every ℓ g and for every g G =(g, ), the ∈ ∗ ∈ · element Ad ∗(g)ℓ can be computed to be
Ad ∗(g)ℓ = Ig∗ + ad ∗(g) ℓ ℓ + g(ℓ)⊥. ∈ Therefore, as ad ∗(g)ℓ = g(ℓ)⊥,
:= Ad ∗(G)ℓ = Ad ∗(g)ℓ g G = ℓ + g(ℓ)⊥ ℓ g∗, (3) Oℓ { | ∈ } ∀ ∈ i.e. only flat orbits appear. Hence,
g∗/G = Ad ∗(G)ℓ ℓ g∗ = ℓ + g(ℓ)⊥ ℓ g∗ . { | ∈ } | ∈ Thus, by the Kirillov theory, one gets
G = g∗/G = ℓ + g(ℓ)⊥ ℓ g∗ ∼ | ∈ as topological spaces. b
Now, let [πk] k N G be a properly converging sequence in G with limit set L [πk] k N . ∈ ⊂ ∈ Let g∗/Gbe the coadjoint orbit of some [π] L [πk] k N , k the coadjoint orbit of O ∈ b ∈ ∈b O [πk] for every k N and let ℓ . Then, by Lemma 2.7, there exists for every k N an ∈ ∈ O ∈ element ℓk k, such that lim ℓk = ℓ in g∗. One can assume that, passing to a subsequence ∈O k →∞ if necessary, the sequence (g(ℓk))k N converges in the topology of the Grassmannian to a ∈
12 subalgebra u of g(ℓ) and that there exists a number d N, such that dim( ) = 2d for every ∈ Ok k N. Then, it follows from (3) that ∈ L(( k)k N)= lim ℓk + g(ℓk)⊥ = ℓ + u⊥ = ℓ + u⊥ g∗/G. (4) O ∈ k O ⊂ →∞ Since in the two-step nilpotent case g(ℓ ) contains [g, g] for every k N, the subspace u also k ∈ contains [g, g]. Hence, as the Kirillov mapping is a homeomorphism and u⊥ consists only of characters of g, the limit set L [πk] k N of the sequence [πk] k N in G is the “affine” subset ∈ ∈ G L [πk] k N = χq indP χℓ q u⊥ b ∈ ⊗ | ∈ for a polarization p of ℓ and P := exp( p).
The observations above lead to the following proposition. Proposition 3.1. There are three different types of possible limit sets of the sequence ( k)k N of coadjoint orbits: O ∈ 1. dim( ℓ) = 2d: In this case, the limit set L(( k)k N) is the singleton ℓ = ℓ + g(ℓ)⊥, O O ∈ O i.e. u = g(ℓ).
N 2. dim( ℓ) = 0: Here, L(( k)k ) = ℓ + u⊥ and the limit set L [πk] k N consists only O O ∈ ∈ of characters, i.e. q [g, g] = 0 for all q L(( k)k N). { } O ∈ O ∈ 3. The dimension of the orbit is strictly larger than 0 and strictly smaller than 2d. In Oℓ this case, 0 < dim( ) < 2d for every L(( k)k N) and O O∈ O ∈ G N L(( k)k )= q + ℓ, i.e. L [πk] k N = χq indP χℓ O ∈ O ∈ ⊗ q u⊥ q u⊥ ∈[ [∈ for a polarization p of ℓ and P := exp(p).
3.2 Conditions 1, 2 and 3(a) Now, to start with the proof of the conditions listed in Definition 1.1, the families of sets (Si)i 0,...,r and (Γi)i 0,...,r are going to be defined and the Properties 1, 2 and 3(a) of Definition∈{ 1.1} are going∈{ to be} verified. In order to be able to define the required families (Γi)i 0,...,r and (Si)i 0,...,r , one needs to ∈{ } ∈{ } construct a polarization pV of ℓ g in the following way: ℓ ∈ ∗ Let H ,...,H be a Jordan-H¨older basis of g, in such a way that g := span H ,...,H { 1 n} i { i n} for i 0,...,n is an ideal in g. Since g is two-step nilpotent, one can first choose a basis ∈ { } Hn˜, ,Hn of [g, g] and then add the vectors H1, ,Hn˜ 1 to obtain a basis of g. Let { ··· } ··· − IPuk := i n g(ℓ) g = g(ℓ) g ℓ { ≤ | ∩ i ∩ i+1} be the Pukanszky index set for ℓ g . The number of elements IPuk of IPuk is the dimension ∈ ∗ ℓ ℓ of the orbit of ℓ. Oℓ Moreover, if one denotes by gi(ℓ gi ) the stabilizer of ℓ gi in gi, | | n V pℓ := gi(ℓ gi ) | Xi=1 13 is the Vergne polarization of ℓ in g. Its construction will now be analyzed by a method developed in [27], Section 1. So, let ℓ g . Then, choose the largest index j (ℓ) 1,...,n such that H g(ℓ) and let ∈ ∗ 1 ∈ { } j1(ℓ) 6∈ Y V,ℓ := H . Moreover, choose the index k (ℓ) 1,...,n such that ℓ, [H ,H ] = 0 1 j1(ℓ) 1 ∈ { } k1(ℓ) j1(ℓ) 6 and ℓ, [H ,H ] = 0 for all i>k (ℓ) and let XV,ℓ := H . i j1(ℓ) 1 1 k1(ℓ) Next, let g1,ℓ := U g ℓ, U, Y V,ℓ = 0 . Then, g1,ℓ is an ideal in g which does not contain ∈ | 1 XV,ℓ, and g = RXV,ℓ g1,ℓ. Now, the Jordan-H¨older basis will be changed, taking out H . 1 1 k1(ℓ) ⊕ 1,ℓ 1,ℓ 1,ℓ 1,ℓ 1,ℓ Consider the Jordan-H¨older basis H ,...,H ,H ,...,Hn of g , where 1 k1(ℓ) 1 k1(ℓ)+1 − V,ℓ V,ℓ ℓ, Hi,Y X H1,ℓ := H i>k (ℓ) and H1,ℓ := H 1 1 i Then, choose the largest index j2(ℓ) 1,...,k1(ℓ) 1,k1(ℓ) + 1,...,n in such a way that 1,ℓ 1,ℓ V,ℓ 1,ℓ∈ { − } Hj (ℓ) g (ℓ g1,ℓ ) and let Y2 := Hj (ℓ). Again, choose k2(ℓ) 1,...,k1(ℓ) 1,k1(ℓ)+1,...,n 2 6∈ | 2 ∈ { − } 1,ℓ 1,ℓ 1,ℓ 1,ℓ in such a way that ℓ, [H ,H ] = 0 and ℓ, [H ,H ] = 0 for all i>k2(ℓ) and set k2(ℓ) j2(ℓ) 6 i j2(ℓ) XV,ℓ := H1,ℓ . 2 k2(ℓ) Iterating this procedure, one gets sets Y V,ℓ,...,Y V,ℓ and XV,ℓ,...,XV,ℓ for d 0,..., [ n ] 1 d 1 d ∈ { 2 } with the properties pV = span Y V,ℓ,...,Y V,ℓ g(ℓ) ℓ 1 d ⊕ and ℓ, XV,ℓ,Y V,ℓ = 0, ℓ, XV,ℓ,Y V,ℓ = 0 i = j 1,...,d and i i 6 i j ∀ 6 ∈ { } ℓ, Y V,ℓ ,YV,ℓ = 0 i,j 1,...,d . i j ∀ ∈ { } Now, let J(ℓ) := j (ℓ), ,j (ℓ) and K(ℓ) := k (ℓ), ,k (ℓ) . { 1 ··· d } { 1 ··· d } Then, IPuk = J(ℓ) ˙ K(ℓ) and j (ℓ) >...>j (ℓ). ℓ ∪ 1 d Puk The index sets Iℓ , J(ℓ) and K(ℓ) are the same on every coadjoint orbit (see [27]) and can therefore also be denoted by IPuk, J( ) and K( ) if ℓ is located in the coadjoint orbit . O O O O Now, for the parametrization of g∗/G and thus of G and for the choice of the concrete realization of a representation required in Property 1(b) of Definition 1.1, let g /G.A O ∈ ∗ theorem of L.Pukanszky (see [29], Part II, Chapter I.3b or [27], Corollary 1.2.5) states that Puk there exists one unique ℓ such that ℓ (Hi) = 0 for all i I . Choose this ℓ , let O ∈ O O ∈ O P V := exp(pV ) and define the irreducible unitary representation O ℓO ℓO V G π := ind V χ ℓO P ℓO ℓO 2 V 2 d associated to the orbit and acting on L G/P ,χℓ = L (R ). O ℓO O ∼ 14 Next, one has to construct the demanded sets Γ for i 0,...,r . i ∈ { } For this, define for a pair of sets (J, K) such that J, K 1, ,n , J = K and J K = ⊂ { ··· } | | | | ∩ ∅ the subset (g∗/G)(J,K) of g∗/G by (g∗/G) := g∗/G (J, K)=(J( ),K( )) . (J,K) {O ∈ | O O } Moreover, let := (J, K) J, K 1,...,n , J K = , J = K , (g∗/G) = M { | ⊂ { } ∩ ∅ | | | | (J,K) 6 ∅} and Puk (g∗/G)2d := g∗/G I = 2d . O∈ O n o Then, (g∗/G)2d = (g∗/G)(J,K) (J,K): J,K 1,...,n , J = K =[d,⊂{ J K=} | | | | ∩ ∅ and g∗/G = (g∗/G)2d = (g∗/G)(J,K). d 0,...,[ n ] (J,K) ∈{ [ 2 } [∈M Now, an order on the set shall be introduced. M First, if J = K = d, J = K = d and d V Γ := π (g∗/G) and iJK ℓO O∈ (J,K) n o SiJK := Γi. i 0,...,iJK ∈{ [ } Then, obviously, the family (Si)i 0,...,r is an increasing family in G. Furthermore, the set S is closed∈{ for every} i 0,...,r . This can easily be deduced from the i ∈ { } definition of the index sets J(ℓ) and K(ℓ). The indices jm(ℓ) andbkm(ℓ) for m 1,...,d m∈1,ℓ { } are chosen such that they are the largest to fulfill a condition of the type ℓ, [H − , ] = 0 jm(ℓ) · 6 m 1,ℓ or ℓ, [H − , ] = 0, respectively. For a more detailed proof of the closure of the sets km(ℓ) · 6 (Si)i 0,...,r , see Lemma 7.1. ∈{ } In addition, the sets Γi are Hausdorff. For this, let i = iJK for (J, K) and ( k)k N ∈ M O ∈ in (g /G) be a sequence of orbits such that the sequence πV converges in Γ , ∗ (J,K) ℓO k N i k ∈ i.e. ( k)k N converges in (g∗/G)(J,K) and thus has a limit point in (g∗/G)(J,K). If now ∈ O k O k ℓk →∞ ℓ , then by (4), it follows that the limit u of the sequence (g(ℓk))k N is equal O ∋ −→ ∈O ∈ 15 to g(ℓ). Therefore, the sequence ( ) N and thus also the sequence πV have unique k k ℓO k N O ∈ k ∈ limits and hence, Γi is Hausdorff. Moreover, one can still observe that for d = 0 the choice J = K = represents the only ∅ possibility to get J = K = d. So, the pair ( , ) is the first element in the above defined | | | | ∅ ∅ order and therefore corresponds to 0. Thus, V Puk Γ0 = πℓ I = , O O ∅ n o which is equivalent to the fact that g(ℓ )= g which again is equivalent to the fact that every V O π Γ0 is a character. Hence, S0 = Γ0 is the set of all characters of G, as demanded. ℓO ∈ Since one can identify G/P V with Rd by means of the subspace s = span XV,ℓO ,...,XV,ℓO , ℓO ℓO 1 d 2 V V one can thus identify the Hilbert space L G/P ,χℓ , that the representation π acts on, ℓO O ℓO with the Hilbert space L2(Rd) for every (g /G) as in (1). Therefore, one can take for O ∈ ∗ 2d all J, K 1,...,n with J = K = d the Hilbert space L2(Rd) as the desired Hilbert space ∈ { V 2} d | | | | i and π ,L (R ) thus represents the demanded concrete realization. H JK ℓO Hence, the first condition of Definition 1.1 is fulfilled. For the proof of the Properties 2 and 3(a), a proposition will be shown. Proposition 3.2. For every a C (G) and every (J, K) with J = K = d 0,..., [ n ] , the mapping ∈ ∗ ∈M | | | | ∈ { 2 } Γ L2(Rd), γ (a)(γ) iJK → 7→ F is norm continuous and the operator (a)(γ) is compact for all γ Γ . F ∈ iJK Proof: The compactness follows directly from a general theorem which can be found in [8], Chapter 4.2 or [29], Part II, Chapter II.5 and states that the C∗-algebra C∗(G) of every connected nilpotent Lie group G fulfills the CCR condition, i.e. the image of every irreducible representation of C∗(G) is a compact operator. n Next, let d 0,..., [ 2 ] and (J, K) such that J = K = d. ∈ { } ∈M | | V | | First, one has to observe that the polarization pℓ is continuous in ℓ on the set ℓ ′ ′ (g∗/G)(J,K) with respect to the topology of the Grassmannian. This can O | O ∈ V,ℓ V,ℓ be seen by the construction of the vectors Y1 ,...,Yd . Now, let ( k)k N be a sequence of orbits in (g∗/G)(J,K) and (g∗/G)(J,K) such that ∈ k O k O ∈ V →∞ V →∞ πℓ πℓ and let a C∗(G). Then, ℓ k ℓ and by the observation above, Ok −→ O ∈ O −→ O the associated sequence of polarizations pV converges to the polarization pV . By the ℓO k N ℓO k ∈ V k V proof of Theorem 2.3 in [31], thus, πℓ (a ) →∞ πℓ (a) in the operator norm. Ok −→ O Since C∗(G) is obviously separable, this proposition proves the desired Properties 2 and 3(a) of Definition 1.1 and hence, it remains to show Property 3(b). 16 3.3 Condition 3(b) 3.3.1 Introduction to the setting For simplicity, in the following, the representations will be identified with their equivalence classes. Let d 0,..., [ n ] and (J, K) with J = K = d. Furthermore, fix i = i 0,...,r . ∈ { 2 } ∈M | | | | JK ∈ { } V V Let π N = π N be a sequence in Γi whose limit set is located outside Γi. k k ℓOk k Since every∈ converging sequence∈ has a properly converging subsequence, it will be assumed V that πk k N is properly converging and the transition to a subsequence will be omitted. ∈ Hence, the sequence πV takes on the role of the subsequence (γ ) N in Condition 3(b) k k N k k of Definition 1.1. Therefore,∈ for every k N, one has to construct∈ an involutive linear ∈ mappingν ˜k : CB(Si 1) ( i), bounded by C Si and fulfilling − →B H k · k −1 V lim (a) πk ν˜k (a) Si−1 = 0, k F − F | op →∞ with a constant C > 0 independent of k. In order to do so, regard the sequence of coadjoint orbits ( k)k N corresponding to the V O ∈ sequence πk k N. It is contained in (g∗/G)(J,K) and in particular, every k has the same ∈ O dimension 2d. Moreover, it converges properly to a set of orbits L(( k)k N). V O ∈ V In addition, since Si is closed, the limit set L πk k N of the sequence πk k N is contained ∈ ∈ in Si 1 = Γj and therefore, for every element L(( k)k N) there exists a pair − j 0,...,i 1 O ∈ O ∈ ∈{ S − } V (J ,K ) < (J, K) such that πℓ ΓiJ K or equivalently, (g∗/G)(JO,KO). O O O ∈ O O O∈ Now, the Lie algebra g will be examined and divided into different parts. With their help, a new sequence of representations (πk)k N which are equivalent to the representations V ∈ πk k N will be defined. Then, in the second and third case mentioned in Proposition 3.1, ∈ (πk)k N will be analyzed and for this new sequence, mappings (νk)k N with certain properties ∈ ∈ similar to the ones required in Condition 3(b) of Definition 1.1 will be constructed. At the end, the requested convergence will be deduced from the convergence of (πk)k N together V ∈ with the equivalence of the representations πk and πk . 3.3.2 Changing the Jordan-H¨older basis ˜ ˜ N ˜ N Let ℓ L(( k)k ). Then, there exists a sequence ℓk k N in ( k)k such that ∈ O ∈ O ∈ ∈ O ∈ ℓ˜= lim ℓ˜k. k →∞ ˜ ˜ ⊥ ˜ Since one is interested in the orbits k = ℓk + g ℓk , one can change the sequence ℓk k N O ∈ to a sequence (ℓk)k N by letting ℓk(A) = 0 for every A g ℓ˜k ⊥ = g(ℓk)⊥. ∈ ∈ Thus, one obtains another converging sequence (ℓk)k N in ( k)k N whose limit ℓ is located ∈ O ∈ in an orbit L(( k)k N). O∈ O ∈ In the following, this sequence (ℓk)k N will be investigated and with its help, the above- mentioned splitting of g will be accomplished.∈ 17 As above, one can suppose that the subalgebras (g(ℓk))k N converge to a subalgebra u, whose corresponding Lie group exp(u) is denoted by U. These∈ subalgebras g(ℓ ) for k N can be k ∈ written as g(ℓ )=[g, g] s , k ⊕ k where sk [g, g]⊥. In addition, let nk,0 be the kernel of ℓk [g,g] and sk,0 the kernel of ℓk s for ⊂ | | k all k N. One can assume that s = s and choose T s of length 1 orthogonal to s . ∈ k,0 6 k k ∈ k k,0 The case s = s for k N, being easier, will be omitted. k,0 k ∈ Similarly, choose Zk [g, g] of length 1 orthogonal to nk,0. One can easily see that such a Zk ∈ V exists: If ℓk [g,g] = 0, then πℓ is a character and thus contained in S0 = Γ0. This happens | Ok only in the case i = 0 and then the whole sequence πV is contained in S . But S = Γ ℓO k N 0 0 0 k ∈ is closed and thus, πV cannot have a limit set outside Γ , as regarded in the setting ℓO k N 0 k ∈ of Condition 3(b) of Definition 1.1. Furthermore, let r = g(ℓ ) g. k k ⊥ ⊂ One can assume that, passing to a subsequence if necessary, lim Zk =: Z, lim Tk =: T and k k →∞ →∞ lim rk =: r exist. k →∞ Now, new polarizations pk of ℓk are going to be constructed in order to define the represenations (πk)k N with their help. ∈ The restriction to rk of the skew-form Bk defined by B (V,W ) := ℓ , [V,W ] , V,W g k h k i ∀ ∈ is non-degenerate on rk and there exists an invertible endomorphism Sk of rk such that x,S (x′) = B (x,x′) x,x′ r . h k i k ∀ ∈ k S is skew-symmetric, i.e. St = S , and with the help of Lemma 7.2, one can decompose r k k − k k into an orthogonal direct sum d j rk = Vk Xj=1 of two-dimensional S -invariant subspaces. Choose an orthonormal basis Xk,Y k of V j. k { j j } k Then, [Xk,Xk] n i,j 1,...,d , i j ∈ k,0 ∀ ∈ { } [Y k,Y k] n i,j 1,...,d and i j ∈ k,0 ∀ ∈ { } [Xk,Y k]= δ ckZ mod n i,j 1,...,d , i j i,j j k k,0 ∀ ∈ { } k R k where 0 = cj and sup cj < for every j 1,...,d . 6 ∈ k N ∞ ∈ { } ∈ k Again, by passing to a subsequence if necessary, the sequence (cj )k N converges for every ∈ j 1,...,d to some cj R. ∈ { k }k ∈ k k Since Xj ,Yj rk and ℓk(A) = 0 for every A rk, ℓk(Xj )= ℓk(Yj ) = 0 for all j 1,...,d . ∈ ∈ k k ∈ { } Furthermore, one can suppose that the sequences (Xj )k N, (Yj )k N converge in g to vectors ∈ ∈ Xj,Yj which form a basis modulo u in g. 18 It follows that k k k ℓk, [Xj ,Yj ] = cj λk, where k λ = ℓ ,Z →∞ ℓ,Z =: λ. k h k ki −→ h i As Z was chosen orthogonal to n , λ = 0 for every k N. k k,0 k 6 ∈ Now, let p := span Y k, ,Y k, g(ℓ ) k { 1 ··· d k } and Pk := exp(pk). Then, pk is a polarization of ℓk. Furthermore, define the representation πk as G πk := indPk χℓk . V Since πk as well as πk are induced representations of polarizations and of the characters χℓk and χℓ , where ℓk and ℓ lie in the same coadjoint orbit k, the two representations are Ok Ok O equivalent, as observed in Section 2.1.2. Let ak := nk,0 + sk,0. Then, ak is an ideal of g on which ℓk is 0. Therefore, the nor- mal subgroup exp(ak) is contained in the kernel of the representation πk. Moreover, let a := lim ak. k →∞ N k k In addition, let p ,p ˜ 1,...,p and let A1, ,Ap˜ denote an orthonormal basis of ∈ ∈ { } { ··· k} k nk,0, the part of ak which is located inside [g, g], and Ap˜+1, ,Ap an orthonormal basis k k { ··· } of sk,0, the part of ak outside [g, g]. Then, A1, ,Ap is an orthonormal basis of ak and as k { ··· } above, one can assume that lim Aj = Aj exists for all j 1,...,p . k ∈ { } Now, for every k N, one can→∞ take as an orthonormal basis for g the set of vectors ∈ Xk, ,Xk,Y k, ,Y k,T ,Z ,Ak, ..., Ak { 1 ··· d 1 ··· d k k 1 p} as well as the set X , ,X ,Y , ,Y ,T,Z,A , ..., A . { 1 ··· d 1 ··· d 1 p} This gives the following Lie brackets: k k k [Xi ,Yj ] = δi,jcj Zk mod ak, k k [Xi ,Xj ] =0 mod ak and k k [Yi ,Yj ] =0 mod ak. The vectors Zk and Tk are central modulo ak. 3.3.3 Definitions Before starting the analysis of (πk)k N, some notations have to be introduced. ∈ Choose for j 1,...,d the Schwartz functions η (R) such that η 2 R = 1 and ∈ { } j ∈ S k jkL ( ) η ∞ R 1. k jkL ( ) ≤ Furthermore, for x ,...,x ,y ,...,y ,t,z,a , ,a R, write 1 d 1 d 1 ··· p ∈ d d k k (x)k := (x1,...,xd)k := xjXj , (y)k := (y1,...,yd)k := yjYj , (t)k := tTk, Xj=1 Xj=1 19 p˜ p k k (z)k := zZk, (˙a)k := (a1,...,ap˜)k := ajAj , (¨a)k := (ap˜+1,...,ap)k := ajAj Xj=1 j=˜Xp+1 p k and (a)k := (˙a, a¨)k =(a1,...,ap)k = ajAj , Xj=1 where ( ,..., ) is defined to be the d-,p ˜-, (p p˜)- or the p-tuple with respect to the bases · · k − Xk,...,Xk , Y k,...,Y k , Ak, ..., Ak , Ak , ..., Ak and Ak, ..., Ak , respectively, and let { 1 d } { 1 d } { 1 p˜} { p˜+1 p} { 1 p} (g)k := (x1,...,xd,y1,...,yd,t,z,a1,...,ap)k := ((x)k, (y)k, (t)k, (z)k, (˙a)k, (¨a)k) = ((x)k, (h)k) d d p k k k = xjXj + yjYj + tTk + zZk + ajAj , Xj=1 Xj=1 Xj=1 where (h) is in the polarization p and the (2d+2+p)-tuple ( ,..., ) is regarded with respect k k · · k to the basis Xk,...,Xk,Y k,...,Y k,T ,Z ,Ak, ..., Ak . { 1 d 1 d k k 1 p} Moreover, define the limits d d (x) := (x1,...,xd) := lim (x)k = xjXj, (y) := (y1,...,yd) := lim (y)k = yjYj, ∞ ∞ k ∞ ∞ k →∞ →∞ Xj=1 Xj=1 p˜ (t) := lim (t)k = tT, (z) := lim (z)k = zZ, (˙a) := (a1,...,ap˜) := lim (˙a)k = ajAj, ∞ k ∞ k ∞ ∞ k →∞ →∞ →∞ Xj=1 p (¨a) := (ap˜+1,...,ap) := lim (¨a)k = ajAj, ∞ ∞ k →∞ j=˜Xp+1 p (a) := (˙a, a¨) =(a1,...,ap) := lim (a)k = ajAj and ∞ ∞ ∞ k →∞ Xj=1 d d p (g) := (x,y,t,z, a,˙ a¨) := lim (g)k = xjXj + yjYj + tT + zZ + ajAj. ∞ ∞ k →∞ Xj=1 Xj=1 Xj=1 3.3.4 Formula for πk Now, the representations (πk)k N can be computed. Let f L1(G). ∈ ∈ d k k k R k With ρk := ℓk,Tk , c := (c1,...,cd) and for s1,...,sd and (s)k := (s1,...,sd)k = sjXj , h i ∈ j=1 k k P where again ( ,..., )k is the d-tuple with respect to the basis X1 ,...,Xd , as in (2), the · · 2 { } representation πk acts on L (G/Pk,χℓk ) in the following way: 20 1 π (g) ξ (s) = ξ (g)− (s) k k k k · k 1 = ξ (s) (x) (h) + [ (x) (h) , (s) ] k − k − k 2 − k − k k 1 = ξ (s) (x) (h) + [(s) , (h) ] [(x) , (h) +(s) ] k − k · − k k k − 2 k k k 2πi ℓ , (h) +[(s) ,(h) ] 1 [(x) ,(h) +(s) ] = e h k − k k k − 2 k k k iξ (s x) − k 1 1 2πi ℓk, (y)k (t)k (z)k (˙a)k (¨a)k+[(s)k (x )k,(y)k] [(x)k,(s)k] = e h − − − − − − 2 − 2 iξ (s x) − k d k 1 2πi tρk zλk+ P λkcj (sj 2 xj )yj − − j=1 − = e ξ (s x)k (5) k 1 − 2πi( tρk zλk+λkc ((s)k (x)k)(y)k) = e − − − 2 ξ (s x)k , − k since ℓk(Y ) = 0 for all j 1,...,d . j ∈ { } Rd Rd R R Rp˜ Rp p˜ R2d+2+p 2 Rd Rd Now, identify G with − ∼= , let ξ L ( ) and s . × × × × × 2 d ∈ ∈ Moreover, identify πk with a representation acting on L (R ) which will also be called πk. To stress the dependence on k of the function f L1(G) fixed above, denote by f L1(R2d+2+p) ∈ k ∈ the function f applied to an element in the k-basis: d d p k k k fk(g) := f((g)k)= f xjXj + yjYj + tTk + zZk + ajAj . ! Xj=1 Xj=1 Xj=1 Denote by f for f (G) and α N2d+2+p the derivative of f in the direction of α,k′ ∈ S ∈ (Xk)α1 (Xk)αd (Y k)αd+1 (Y k)α2d (T )α2d+1 (Z )α2d+2 (Ak)α2d+3 (Ak)α2d+2+p . Then, 1 ··· d 1 ··· d k k 1 ··· p 2 N f = 1+ g f ′ (g) dg. k kk(N) k k α,k α N 2dZ+2+p | X|≤ R Since the k-basis of G converges to the -basis, one can estimate the expressions f (g) ∞ α,k′ for g R2d+2+p and α N by constants independent of k and therefore gets an estimate ∈ | | ≤ for fk (N) that does not depend on k. k k ˆ2,3,4,5,6 Now, denoting by fk the Fourier transform in the 2nd, 3rd, 4th, 5th and 6th variable, πk(f)ξ(s) = fk(g)πk(g)ξ(s)dg Rd Rd R ZR Rp˜ Rp−p˜ × × × × × 2πi tρ zλ +λ ck(s 1 x)y = f (x,y,t,z, a,˙ a¨)e − k− k k − 2 ξ(s x)d(x,y,t,z, a,˙ a¨) k − Rd Rd R ZR Rp˜ Rp−p˜ × × × × × 2πi tρ zλ + 1 λ ck(s+x)y = f (s x,y,t,z, a,˙ a¨)e − k− k 2 k ξ(x)d(x,y,t,z, a,˙ a¨) k − Rd Rd R ZR Rp˜ Rp−p˜ × × × × × λ ck = fˆ2,3,4,5,6 s x, k (s + x), ρ ,λ , 0, 0 ξ(x)dx. (6) k − − 2 k k RZd 21 3.4 Condition 3(b) – First case First, consider the case that L(( k)k N) consists of one single limit point , the first case O ∈ O described in Proposition 3.1. In this case, for every k N, ∈ 2d = dim( ) = dim( ). Ok O Thus, the regarded situation occurs if and only if λ = 0 and c = 0 for every j 1,...,d . 6 j 6 ∈ { } The first case is the easiest case. After a few observations, here the operators (˜νk)k N can be defined immediately. Moreover, the claims of Condition 3(b) of Definition 1.1 can∈ be shown easily. In this first case, the transition to the representations (πk)k N and the definitions and computations accomplished in the Sections 3.3.3 and 3.3.4 are not∈ needed. Consider again the sequence (ℓk)k N chosen above which converges to ℓ . As the dimen- ∈ ∈ O sions of the orbits k and are the same, there exists a subsequence of (ℓk)k N (which will O O V ∈ also be denoted by (ℓk)k N for simplicity) such that p := lim pℓ is a polarization of ℓ, but ∈ k k →∞ V not necessarily the Vergne polarization. Moreover, define P := exp(p)= lim Pℓ and let k k →∞ G π := indP χℓ. V G Now, if one identifies the Hilbert spaces V and π of the representations π = ind V χℓ πℓ ℓk P k H k H ℓk and π with L2(Rd), from [31], Theorem 2.3, one can conclude that V G G k π (a) π(a) = ind V χ (a) ind χ (a) →∞ 0 ℓk op P ℓk P ℓ − ℓk − op −→ for every a C∗(G ). ∈ V G Since π and π = ind V χℓ are both induced representations of polarizations and of the same ℓ Pℓ character χℓ, they are equivalent and hence, there exists a unitary intertwining operator 2 Rd 2 Rd V F : πV = L ( ) π = L ( ) such that F πℓ (a)= π(a) F a C∗(G). H ℓ ∼ →H ∼ ◦ ◦ ∀ ∈ V V G V G Furthermore, the two representations π = π = ind V χℓO and π = ind V χℓk are k ℓOk Pℓ k ℓk Pℓ Ok k equivalent for every k N because ℓ and ℓk are located in the same coadjoint orbit k and ∈ Ok O pV and pV are polarizations. Thus, there exist further unitary intertwining operators ℓOk ℓk 2 d 2 d V V V R V R Fk : π = L ( ) π = L ( ) such that Fk πk (a)= πℓk (a) Fk a C∗(G). H k ∼ →H ℓk ∼ ◦ ◦ ∀ ∈ Now, define the required operatorsν ˜ for every k N as k ∈ V ν˜k(ϕ) := Fk∗ F ϕ πℓ F ∗ Fk ϕ CB(Si 1), ◦ ◦ ◦ ◦ ∀ ∈ − V V which is reasonable since πℓ is a limit point of the sequence πk k N and hence contained ∈ in Si 1, as seen in Section 3.3.1. − As ϕ πV (L2(Rd)) and F and F are intertwining operators and thus bounded, the ℓ ∈ B k image ofν ˜ is contained in (L2(Rd)), as requested. k B 22 Next, it needs to be shown thatν ˜ is bounded. By the definition of , one has for every k k · kSi−1 ϕ CB(Si 1), ∈ − V ν˜ (ϕ) = F ∗ F ϕ π F ∗ F k k kop k ◦ ◦ ℓ ◦ ◦ k op V F ∗ op F op ϕ π F ∗ op F op ≤ k k k k k ℓ op k k k kk = ϕ πV ℓ op ϕ Si−1. ≤ k k In addition, one can easily observe thatν ˜k is involutive. For every ϕ CB(Si 1), ∈ − V V ν˜ (ϕ)∗ = F ∗ F ϕ π F ∗ F ∗ = F ∗ F ϕ∗ π F ∗ F =ν ˜ (ϕ∗). k k ◦ ◦ ℓ ◦ ◦ k k ◦ ◦ ℓ ◦ ◦ k k Now, the last thing to check is the required convergence of Condition 3(b). For all a C (G), ∈ ∗ V V V πk (a) ν˜k (a) Si = πk (a) Fk∗ F (a) Si πℓ F ∗ Fk − F | −1 op − ◦ ◦ F | −1 ◦ ◦ op V V = π (a) F ∗ F π (a) F ∗ F k − k ◦ ◦ ℓ ◦ ◦ k op V = F ∗ π (a) F F ∗ π(a) F k ◦ ℓk ◦ k − k ◦ ◦ k op V = F ∗ π π (a) F k ◦ ℓk − ◦ k op k πV (a ) π(a) →∞ 0. ≤ ℓk − op −→ V N Therefore, the representations πk k N and the constructed (˜νk)k fulfill Condition 3(b) of Definition 1.1 and thus, in this case,∈ the claim is shown. ∈ 3.5 Condition 3(b) – Second case In the second case of Condition 3(b) of Definition 1.1 described in Proposition 3.1, the situ- ation that λ = 0 or c = 0 for every j 1,...,d is going to be considered. j ∈ { } In this case, k ℓ , [Xk,Y k] = ckλ →∞ c λ = 0 j 1,...,d , k j j j k −→ j ∀ ∈ { } k while cj λk = 0 for every k and every j 1,...,d . 6 ˜ ∈ { } ˜ Then, ℓ [g,g] = 0 and since ℓ [g,g] obviously also vanishes for all ℓ u⊥, every L(( k)k N) | | ∈ O ∈ O ∈ vanishes on [g, g], which means that the associated representation is a character. Therefore, every limit orbit in the set L(( k)k N) has the dimension 0. O O ∈ 3.5.1 Convergence of (πk)k N in G ∈ In order to better understand the sequence of representations (πk)k N and to get an idea of b ∈ how to construct the required linear mappings (νk)k N, in this section, the convergence in G ∈ of (πk)k N is going to be analyzed. This shall serve as a motivation for the construction of ∈ the mappings (νk)k N which will be carried out subsequently. b ∈ 23 As in Calculation (6) in Section 3.3.4, identify G again with R2d+2+p. From now on, this identification will be used most of the time. Only in some cases where one applies ℓk or ℓ and thus it is important to know whether one is using the basis depending on k or the limit basis, the calculation will be done in the above defined bases ( )k or ( ) . · · ∞ Now, adapt the methods developed in [26] to this given situation. Let s =(s ,...,s ), α =(α , ,α ), β =(β , ,β ) Rd and define 1 d 1 ··· d 1 ··· d ∈ d 1 1 β 2πiαs k 4 k j ηk,α,β(s)= ηk,α,β(s1, ,sd) := e λkc ηj λkc 2 sj + ··· j | j | λ ck j=1 k j Y for the Schwartz functions η for j 1,...,d chosen at the beginning of Section 3.3.3. This j ∈ { } definition is reasonable because λ ck = 0 for every k and every j 1,...,d . k j 6 ∈ { } k Next, the convergence of (πk)k N to χℓ+ℓα,β in G will be shown. For this, let cα,β be the coefficient function defined by∈ b ck (g) := π (g)η ,η g G = R2d+2+p α,β h k k,α,β k,α,βi ∀ ∈ ∼ and ℓα,β the linear functional ℓ (g)= ℓ (x,y,t,z,a) := αx + βy g =(x,y,t,z,a) G = R2d+2+p. α,β α,β ∀ ∈ ∼ Since the Hilbert space of the character χ is one-dimensional, one can choose 1 as a ℓ+ℓα,β { } basis and gets the following computation: Using Relation (5), for ρ := ℓ,T and g =(x,y,t,z, a,˙ a¨) R2d+2+p, h i ∈ k cα,β(g) = πk(g)ηk,α,β(s)ηk,α,β(s)ds RZd d k 1 2πi tρk zλk+ P λkcj (sj 2 xj )yj − − j=1 − = e ηk,α,β(s1 x1,...,sd xd) − − RZd ηk,α,β(s1,...,sd)d(s1,...,sd) d k 1 2πi tρk zλk+ P λkcj (sj 2 xj )yj − − j − 2πiαx = e =1 e− RZd d 1 1 β k 4 k 2 j λkc ηj λkc sj xj + j j − λ ck j=1 k j Y d 1 1 β k 4 k 2 j λkc ηj λkc sj + d(s1,...,sd) j j λ ck j=1 k j Y 24 d 1 k 1 2πi(tρk+zλk) 2πiαx 2πi 2 xj yj cj λk k 2 = e− e− e− λkcj j=1 Y k 1 β 1 β 2πiyj sj cj λk k 2 j k 2 j e ηj λkcj sj xj + k ηj λkcj sj + k dsj − λkc λkc ZR j j ! d 1 k 1 2πi(tρk+zλk) 2πiαx 2πi 2 xj yj cj λk k 2 = e− e− e− λkcj j=1 Y βj k 2πiyj sj c λk 1 1 λ ck j e − k j η λ ck 2 (s x ) η λ ck 2 (s ) ds j k j j − j j k j j j ZR d 1 k 2πi(tρ +zλ ) 2πiαx 2πi xj yj c λk = e− k k e− e− 2 j jY=1 k k 1 1 2πiyj sj sgn(λ c ) λ c 2 βj k e k j | k j | − η s λ c 2 x η (s )ds j j − k j j j j j ZR d k 2πitρ 2πizλ 2πiαx 2πiyj βj →∞ e− e− e− e− η (s )η (s )ds −→ j j j j j jY=1 ZR ηj 2=1 k k 2πitρ 2πizλ 2πiαx 2πiyβ = e− e− e− e− = χℓ+ℓα,β (g) = χℓ+ℓα,β (g)1, 1 , where this convergence holds uniformly on compacta. 3.5.2 Definition of νk Finally, the linear mappings (νk)k N which are needed for the sequence (πk)k N to fulfill conditions similar to those required∈ in 3(b) are going to be defined and analyzed.∈ For 0 =η ˜ L2(G/P ,χ ) = L2(Rd), let 6 ∈ k ℓk ∼ P : L2(Rd) Cη,˜ ξ η˜ ξ, η˜ . η˜ → 7→ h i Pη˜ is the orthogonal projection onto the space Cη˜. Let h C (G/U, χ ). Again, identify G/U with Rd Rd = R2d and as already introduced ∈ ∗ ℓ × ∼ in Section 3.3.4 in order to show the dependence on k, here, the utilization of the limit basis will be expressed by an index if necessary: ∞ h (x,y) := h((x,y) ). ∞ ∞ 2d Now, hˆ can be regarded as a function in C (ℓ+u⊥) = C (R ) and, using this identification, ∞ ∞ ∼ ∞ define the linear operator d(˜x, y˜) νk(h) := hˆ (˜x, y˜)Pη . ∞ k,x,˜ y˜ d k RZ2d λkcj j=1 Q 25 Then, the following proposition holds: Proposition 3.3. 1. For all k N and h (G/U, χ ), the integral defining ν (h) converges in the operator ∈ ∈S ℓ k norm. 2. The operator ν (h) is compact and ν (h) h ∗ . k k k kop ≤ k kC (G/U,χℓ) 3. ν is involutive, i.e. ν (h) = ν (h ) for every h C (G/U, χ ). k k ∗ k ∗ ∈ ∗ ℓ Proof: 1) Let h (G/U, χ ) = (R2d). Since ∈S ℓ ∼ S P = η 2 k ηk,α,β kop k k,α,βk2 d 2 1 1 β k 2 2πiαj sj k 2 j = λkcj e ηj λkcj sj + k dsj λkc j=1 ZR j Y d 1 1 k 2 k = λ c 2 η (s ) λ c − 2 ds k j | j j | k j j j=1 ZR Y d = η 2 = 1, k jk2 jY=1 one can estimate the operator norm of νk(h) as follows: d(˜x, y˜) νk(h) op = hˆ (˜x, y˜) Pη ∞ k,x,˜ y˜ d k k k RZ2d λkc op j j=1 Q ˆ d(˜x, y˜) h L1(R2d) hˆ (˜x, y˜) = . ∞ d d ≤ k k Z2d R λkcj λkcj j=1 j=1 Q Q Therefore, the convergence of the integral νk(h) in the operator norm is shown for h (R2d) = (G/U, χ ). ∈S ∼ S ℓ 2) First, let h (G/U, χ ) = (R2d). ∈S ℓ ∼ S Define for s =(s ,...,s ) Rd, 1 d ∈ d 1 1 β k 4 k 2 j ηk,β(s) := λkc ηj λkc sj + . j j λ ck j=1 k j Y Then, 2πiαs ηk,α,β(s)= e ηk,β(s) and thus, one has for ξ (Rd) and s Rd, ∈S ∈ 26 d(˜x, y˜) νk(h)ξ(s) = hˆ (˜x, y˜) ξ,ηk,x,˜ y˜ ηk,x,˜ y˜(s) ∞ d k Z2d R λkcj j=1 Q ˆ d(˜x, y˜) = h (˜x, y˜) ξ(r)ηk,x,˜ y˜(r)dr ηk,x,˜ y˜(s) ∞ d 2d d k RZ RZ λkcj j=1 Q ˆ 2πixr˜ 2πixs˜ d(˜x, y˜) = h (˜x, y˜) ξ(r)e− ηk,y˜(r)dr e ηk,y˜(s) ∞ d 2d d k RZ RZ λkcj j=1 Q ˆ 2πix˜(s r) dxd˜ y˜ = h (˜x, y˜)e − ξ(r)ηk,y˜(r)dr ηk,y˜(s) ∞ d k RZd RZd RZd λkcj j=1 dy˜ Q = hˆ2 (s r, y˜)ξ(r)η (r)η (s) dr, (7) k,y˜ k,y˜ d ∞ − k RZd RZd λkcj j=1 Q where, similar as above, hˆ2 denotes the Fourier transform of hˆ in the second variable. ∞ Hence, as the kernel function∞ dy˜ h (s,r) := hˆ2 (s r, y˜)η (r)η (s) K k,y˜ k,y˜ d ∞ − k RZd λkcj j=1 Q of ν (h) is contained in (R2d), the operator ν (h) is compact. k S k Now, it will be shown that νk(h) op hˆ . k k ≤ ∞ Again, for ξ (Rd) one has ∈S ν (h)ξ 2 k k k2 2 dy˜ = hˆ2 (s r, y˜)ξ(r)η (r)η (s) dr ds k,y˜ k,y˜ d ∞ − k RZd RZd RZd λkc j j=1 Q 2 dy˜ = hˆ2 ( , y˜) (ξη )(s)η (s) ds k,y˜ k,y˜ d ∞ · ∗ k RZd RZd λkc j j=1 Cauchy Q − d Schwarz, 1 2 dy˜ λ ck 2 hˆ2 ( , y˜) (ξη )(s) ds k j k,y˜ d ηj ≤∞ 1 ∞ · ∗ k 2 k k ≤ j=1 Zd Zd Y R R λkcj j=1 Q 27 d Plancherel ˆ 2 1 2 h 3 ξηk,y˜ 2dy˜ ≤ ∞ k 2 k k j=1 λkcj Zd Y R d ˆ 2 1 2 2 = h 3 ξ(s) ηk,y˜(s) dsdy˜ ∞ k 2 | | | | j=1 λkcj Zd Zd Y R R d 2 2 1 1 y˜ ˆ k 2 j = h k ηj λkcj sj + k dy˜1 dy˜d ∞ λkc ··· ··· λkc ··· j=1 j ZR ZR ZR ZR j Y ξ(s ,...,s ) 2ds ds | 1 d | 1 ··· d d 2 2 2 = hˆ ηj(˜yj) dy˜1 dy˜d ξ(s1,...,sd) ds1 dsd ∞ ··· ··· | | ··· | | ··· j=1 ZR ZR ZR ZR Y ηj 2=1 2 k k ˆ 2 = h ξ 2. ∞k k Thus, since (R d) is dense in L2(Rd), S νk(h) op = sup νk(h)(ξ) 2 hˆ k k ξ L2(Rd), k k ≤ ∞ ∈ ξ 2=1 k k for h (R2d) = (G/U, χ ). Therefore, with the density of (G/U, χ ) in C (G/U, χ ), ∈ S ∼ S ℓ S ℓ ∗ ℓ one gets the compactness of the operator ν (h) for h C (G/U, χ ), as well as the desired k ∈ ∗ ℓ inequality ˆ ∗ νk(h) op h = h C (G/U,χℓ). k k ≤ ∞ k k 3) For h C (G/U, χ ) = C (R2d), one has ∈ ∗ ℓ ∼ ∗ d(˜x, y˜) ∗ νk(h)∗ = hˆ (˜x, y˜) Pη ∞ k,x,˜ y˜ d k ! RZ2d λkcj j=1 Q P =P ∗ d(˜x, y˜ ) = hˆ (˜x, y˜) Pη ∞ k,x,˜ y˜ d k RZ2d λkcj j=1 Q d(˜ x, y˜) = h (˜x, y˜) P = ν (h∗). ∗ ηk,x,˜ y˜ d k ∞ k RZ2d λkcj c j=1 Q This proposition firstly shows that the image of the operator νk is located in (L2(Rd)) = ( ) as required in Condition 3(b) of Definition 1.1. Secondly, the B B Hi proposition gives the boundedness and the involutivity of the linear mappings νk for every k N. For the analysis of the sequence (πk)k N, it remains to show a convergence, from ∈ ∈ V which the convergence for the original sequence (πk )k N demanded in Condition 3(b) can be deduced. ∈ 28 3.5.3 Theorem – Second Case Theorem 3.4. Define as in Section 2.1.3 p : L1(G) L1(G/U, χ ), p (f)(˜g) := f(˜gu)χ (u)du g˜ G f L1(G) G/U → ℓ G/U ℓ ∀ ∈ ∀ ∈ UZ and canonically extend pG/U to a mapping going from C∗(G) to C∗(G/U, χℓ). Furthermore, let a C (G). Then, ∈ ∗ lim πk(a) νk pG/U (a) = 0. k − op →∞ Proof: First, it is well-known that pG/U is a surjective homomorphism from C∗(G) to C∗(G/U, χℓ) and from L1(G) to L1(G/U, χ ) as well as from (G) to (G/U, χ ). ℓ S S ℓ Furthermore, for u =(t,z, a,˙ a¨) U = span T,Z,A1, ..., Ap˜,Ap˜+1, ..., Ap , ∞ ∈ { } 2πi ℓ,(t,z,a,˙ a¨)∞ 2πi(tρ+zλ) χℓ(u)= e− h i = e− and therefore, identifying U again with R R Rp˜ Rp p˜ and L1(G/U, χ ) with L1(R2d), for × × × − ℓ f L1(G) = L1(R2d+2+p) andg ˜ = (˜x, y,˜ 0, 0, 0, 0) R2d, one has ∈ ∼ ∈ ˜ ˜ 2πi(tρ˜ +˜zλ) ˜ ˜ pG/U (f) (˜g) = f (˜x, y,˜ t,˜ z,˜ a,˙ a¨)e− d(0, 0, t,˜ z,˜ a,˙ a¨) ∞ ∞ R R RZp˜ Rp−p˜ × × × = fˆ3,4,5,6(˜x, y,˜ ρ, λ, 0, 0), (8) ∞ where f (˜x, y,˜ ρ, λ, 0, 0) = f((˜x, y,˜ ρ, λ, 0, 0) ). ∞ ∞ R2d+2+p Now, let f (G) ∼= ( ) such that its Fourier transform in [g, g] has a com- ∈ S R2dS+2+p pact support in G ∼= . If one writes the elements g of G as g = (x,y,t,z, a,˙ a¨) like above, where x span X ... span X , y span Y ... span Y , t span T , ∈ { 1}× × { d} ∈ { 1}× × { d} ∈ { } z span Z , a˙ span A ... span A , a¨ span A ... span A ∈ { } ∈ { 1}× × { p˜} ∈ { p˜+1}× × { p} or x span Xk ... span Xk , y span Y k ... span Y k , t span T , ∈ { 1 }× × { d } ∈ { 1 }× × { d } ∈ { k} z span Z , a˙ span Ak ... span Ak , a¨ span Ak ... span Ak , ∈ { k} ∈ { 1}× × { p˜} ∈ { p˜+1}× × { p} respectively, this means that the partial Fourier transform fˆ4,5 has a compact support in G, since [g, g] = span Z ,Ak, ..., Ak = span Z,A , ..., A . { k 1 p˜} { 1 p˜} Moreover, let ξ L2(Rd) and s Rd and define ∈ ∈ d 1 1 k 4 k 2 ηk,0(s) := λkcj ηj λkcj (sj) . j=1 Y (Compare the definition of ηk,β in the last proof.) 29 To prove the theorem, the expression π (f) ν p (f) ξ is going to be regarded, using k − k G/U Results (6) from Section 3.3.4 and (7) from the proof of the proposition above. The obtained integrals are going to be divided into five parts and the norm of each part is going to be estimated separately. For this, G will again be identified with R2d+2+p and therefore, f will be replaced by the functions fk and f , respectively, which are defined as above. ∞ π (f) ν p (f) ξ(s) k − k G/U k (6),(7) λc = fˆ2,3,4,5,6 s r, k (s + r), ρ ,λ , 0, 0 ξ(r)dr k − − 2 k k RZd \ 2 dy˜ p (f) (s r, y˜)ξ(r)η (r)η (s) dr G/U k,y˜ k,y˜ d − ∞ − k RZd RZd λkcj j=1 k Q ηk,0 2=1 λ c k k ˆ2,3,4,5,6 k = fk s r, (s + r), ρk,λk, 0, 0 ξ(r)ηk,0(˜y)ηk,0(˜y)dydr˜ (8) − − 2 RZd RZd dy˜ fˆ2,3,4,5,6(s r, y,˜ ρ, λ, 0, 0)ξ(r)η (r)η (s) dr k,y˜ k,y˜ d − ∞ − k RZd RZd λkcj j=1 k Q λ c = fˆ2,3,4,5,6 s r, k (s + r), ρ ,λ , 0, 0 ξ(r)η (˜y)η (˜y)dydr˜ k − − 2 k k k,0 k,0 RZd RZd ˆ2,3,4,5,6 k f s r, λkc (˜y s), ρ, λ, 0, 0 ξ(r)ηk,0(˜y + r s)ηk,0(˜y)dydr.˜ − ∞ − − − Zd Zd R R As described above, the integrals just obtained are now going to be divided into five parts. To do so, the functions qk, uk, vk, nk and wk will be defined. λ ck q (s, y˜) := ξ(r)η (˜y + r s) fˆ2,3,4,5,6 s r, k (s + r), ρ ,λ , 0, 0 k k,0 − k − − 2 k k RZd λ ck fˆ2,3,4,5,6 s r, k (s + r), ρ, λ , 0, 0 dr, − k − − 2 k λ ck u (s, y˜) := ξ(r)η (˜y + r s) fˆ2,3,4,5,6 s r, k (s + r), ρ, λ , 0, 0 k k,0 − k − − 2 k RZd λ ck fˆ2,3,4,5,6 s r, k (s + r), ρ, λ, 0, 0 dr, − k − − 2 λ ck v (s, y˜) := ξ(r)η (˜y + r s) fˆ2,3,4,5,6 s r, k (s + r), ρ, λ, 0, 0 k k,0 − k − − 2 RZd fˆ2,3,4,5,6 s r, λ ck(˜y s), ρ, λ, 0, 0 dr, − k − k − 30 n (s, y˜) := ξ(r)η (˜y + r s) fˆ2,3,4,5,6 s r, λ ck(˜y s), ρ, λ, 0, 0 k k,0 − k − k − Zd R 2,3,4,5,6 k fˆ s r, λkc (˜y s), ρ, λ, 0, 0 dr − ∞ − − and w (s) := ξ(r)η (˜y) η (˜y) η (˜y + r s) k k,0 k,0 − k,0 − Zd Zd R R λ ck fˆ2,3,4,5,6 s r, k (s + r), ρ ,λ , 0, 0 drdy.˜ k − − 2 k k Then, π (f) ν p (f) ξ(s) = q (s, y˜)η (˜y)dy˜ + u (s, y˜)η (˜y)dy˜ k − k G/U k k,0 k k,0 Zd Zd R R + vk(s, y˜)ηk,0(˜y)dy˜ + nk(s, y˜)ηk,0(˜y)dy˜ + wk(s). RZd RZd In order to show that k π (f) ν p (f) →∞ 0, k − k G/U op −→ it suffices to prove that there are κ , γ , δ , ω and ǫ which are tending to 0 for k , k k k k k → ∞ such that q κ ξ , u γ ξ , v δ ξ , n ω ξ and w ǫ ξ . k kk2 ≤ kk k2 k kk2 ≤ kk k2 k kk2 ≤ kk k2 k kk2 ≤ kk k2 k kk2 ≤ kk k2 First, regard the last factor of the function qk: λ ck λ ck fˆ2,3,4,5,6 s r, k (s + r), ρ ,λ , 0, 0 fˆ2,3,4,5,6 s r, k (s + r), ρ, λ , 0, 0 k − − 2 k k − k − − 2 k 1 λ ck = (ρ ρ) ∂ fˆ2,3,4,5,6 s r, k (s + r), ρ + t(ρ ρ),λ , 0, 0 dt. k − 3 k − − 2 k − k Z0 Since f is a Schwartz function, one can find a constant C1 > 0 (depending on f) such that λ ck λ ck fˆ2,3,4,5,6 s r, k (s + r), ρ ,λ , 0, 0 fˆ2,3,4,5,6 s r, k (s + r), ρ, λ , 0, 0 k − − 2 k k − k − − 2 k C ρ ρ 1 . ≤ | k − | (1 + s r )2d k − k 31 Hence, one gets the following estimate for qk: q 2 = q (s, y˜) 2d(s, y˜) k kk2 | k | RZ2d C 2 ξ(r)η (˜y + r s) ρ ρ 1 dr d(s, y˜) ≤ k,0 − | k − | (1 + s r )2d Z2d Zd R R k − k Cauchy ξ(r)η (˜y + r s) 2 − 2 2 k,0 − C1 ρk ρ d dr Schwarz≤ | − | (1 + s r ) ! Z2d Zd R R k − k 1 2 d dr d(s, y˜) (1 + s r ) ! RZd k − k 2 2 ξ(r) 2 = C′ ρ ρ | | η (˜y + r s) d(r,s, y˜) 1| k − | (1 + s r )2d k,0 − Z3d R k − k ηk,0 2=1 k k 2 2 C′′ ρ ρ ξ ≤ 1 | k − | k k2 for matching constants C1′ > 0 and C1′′ > 0 depending on f. Thus, for κk := C1′′ ρk ρ , k k | − | κ →∞ 0 since ρ →∞ ρ, and k −→ k −→ p q κ ξ . k kk2 ≤ kk k2 k As λ →∞ λ, the estimation for the function u can be done analogously. k −→ k Now, regard vk. Like for qk and uk, one has λ ck fˆ2,3,4,5,6 s r, k (s + r), ρ, λ, 0, 0 fˆ2,3,4,5,6 s r, λ ck(˜y s), ρ, λ, 0, 0 k − − 2 − k − k − 1 = λ ck (r s) (r s +y ˜) k 2 − − − 1 1 ∂ fˆ2,3,4,5,6 s r, λ ck(˜y s)+ tλ ck (r s) (r s +y ˜) , ρ, λ, 0, 0 dt, · 2 k − k − k 2 − − − Z0 where is the scalar product, and hence there exists again a constant C depending on f such · 3 that λ ck fˆ2,3,4,5,6 s r, k (s + r), ρ, λ, 0, 0 fˆ2,3,4,5,6 s r, λ ck(˜y s), ρ, λ, 0, 0 k − − 2 − k − k − C λ ck(r s) + ck(r s +y ˜) 3 . ≤ | k| − − (1 + s r )2d+1 k − k 32 Therefore, definingη ˜ (t) := t η (t), one gets a similar estimation for v : j k k j k v 2 k kk2 = v (s, y˜) 2d(s, y˜) | k | RZ2d ξ(r)η (˜y + r s) λ ck(r s) + ck(r s +y ˜) ≤ k,0 − | k| − − Z2d Zd R R C 2 3 dr d(s, y˜) (1 + s r )2d+1 k − k Cauchy 2 − ξ(r) 2 2 C3′ | | 2d+2 ηk,0(˜y + r s) λk Schwarz≤ (1 + s r ) − | | Z3d R k − k 2 ck(r s) + ck(r s +y ˜) d(r,s, y˜) − − 2 ξ(r) 2 2 k 2 2C′ | | η (˜y + r s) λ c (r s +y ˜) d(r,s, y˜) ≤ 3 (1 + s r )2d+2 k,0 − | k| − Z3d R k − k 2 ξ(r) 2 2 k 2 + 2C′ | | η (˜y + r s) λ c (r s) d(r,s, y˜) 3 (1 + s r )2d+2 k,0 − | k| − Z3d R k − k 2 2 ξ(r) k 2 2 2C3′ λk | | 2d+2 d(r, s) c y˜ ηk,0(˜y) dy˜ ≤ | | (1 + s r ) ! Z2d Zd R k − k R 2 k 2 ξ(r) 2 + 2C′ λ c | | η (˜y + r s) d(r,s, y˜) 3 k (1 + s r )2d k,0 − Z3d R k − k 2 k ξ(r) 2C3′ λkc | | 2d+2 d(r, s) ≤ (1 + s r ) ! Z2d R k − k d 1 1 2 1 2 k 2 k 2 k 2 λkcj λkcj y˜j ηj λkcj (˜yj) dy˜j j=1 ZR Y 2 k 2 ξ(r) 2 + 2C′ λ c | | η (˜y + r s) d(r,s, y˜) 3 k (1 + s r )2d k,0 − Z3d R k − k d ηk,0 2=1 k k k 2 2 k 2 2 2C′ λ c ξ η˜ (˜y ) dy˜ + 2C′ λ c ξ ≤ 3 k k k2 | j j | j 3 k k k2 j=1 ZR Y d 2 k k 2 = C′′ η˜ + λ c λ c ξ 3 k jk2 k k k k2 j=1 Y with constants C3′ > 0 and C3′′ > 0, again depending on f. 33 1 d k 2 k Now, since λ ck →∞ 0, δ := C η˜ 2 + λ ck λ ck fulfills δ →∞ 0 and k −→ k 3′′ k jk2 k k k −→ j=1 Q v δ ξ . k kk2 ≤ kk k2 For the estimation of nk, the fact that the Fourier transform in [g, g], i.e. in the 4th and 5th variable, fˆ4,5 =: f˜ has a compact support will be needed. Therefore, let the support of f be located in the compact set K K K K K K Rd Rd R R Rp˜ Rp p˜ 1 × 2 × 3 × 4 × 5 × 6 ⊂ × × × × × − and let K := K K K Rd R Rp p˜. 2 × 3 × 6 ⊂ × × − Furthermore, since a Fourier transform is independent of the choice of the basis, the value of f˜ on [g, g] = span Z ,Ak, ..., Ak = span Z,A , ..., A expressed in the k-basis and its value { k 1 p˜} { 1 p˜} expressed in the limit basis are the same: f( , , , (z, a˙)k, )= f( , , , (z, a˙) , ). · · · · · · · ∞ · So, in the course of this proof, the limit basis will be chosen for the representation of the 4th and 5th position of an element g. Then, ˆ2,3,4,5,6 k ˆ2,3,4,5,6 k fk s r, λkc (˜y s), ρ, λ, 0, 0 f s r, λkc (˜y s), ρ, λ, 0, 0 − − − ∞ − − k 2πi(λkc (˜y s)y+ρt) = f˜k(s r,y,t,λ, 0, a¨) f˜ (s r,y,t,λ, 0, a¨) e− − d(y,t, a¨) − − ∞ − Rd RZ Rp−p˜ × × = f˜((s r,y,t)k, (λ, 0) , (¨a)k) f˜((s r,y,t,λ, 0, a¨) ) − ∞ − − ∞ Rd RZ Rp−p˜ × × 2πi(λ ck(˜y s)y+ρt) e− k − d(y,t, a¨) k 2πi(λkc (˜y s)y+ρt) = f˜((s r,y,t)k, (λ, 0) , (¨a)k) f˜((s r,y,t,λ, 0, a¨) ) e− − d(y,t, a¨). − ∞ − − ∞ KZ Furthermore, f˜((s r,y,t)k, (λ, 0) , (¨a)k) f˜((s r,y,t,λ, 0, a¨) ) − ∞ − − ∞ d d p = f˜ (s r )Xk + y Y k + tT + λZ + a Ak i − i i i i k i i Xi=1 Xi=1 i=˜Xp+1 d d p f˜ (s r )X + y Y + tT + λZ + a A − i − i i i i i i Xi=1 Xi=1 i=˜Xp+1 d d p = (s r )(Xk X )+ y (Y k Y )+ t(T T )+ a (Ak A ) i − i i − i i i − i k − i i − i Xi=1 Xi=1 i=˜Xp+1 1 d d p ∂f˜ (s r )X + y Y + tT + λZ + a A · i − i i i i i i Z0 Xi=1 Xi=1 i=˜Xp+1 d d p + t˜ (s r )(Xk X )+ y (Y k Y )+ t(T T )+ a (Ak A ) dt.˜ i − i i − i i i − i k − i i − i Xi=1 Xi=1 i=˜Xp+1 34 k k k k k k k Since (X1 ,...,Xd ,Y1 ,...,Yd ,Tk,Zk,A1, ..., Ap) →∞ (X1,...,Xd,Y1,...,Yd,T,Z,A1, ..., Ap), k −→ there exist ωi →∞ 0 for i 1,..., 4 and a constant C > 0 depending on f such that k −→ ∈ { } 4 f˜((s r,y,t)k, (λ, 0) , (¨a)k) f˜((s r,y,t,λ, 0, a¨) ) − ∞ − − ∞ C s r ω1 + y ω2 + t ω3 + a¨ ω4 4 . ≤ k − k k k k k | | k k k k (1 + s r )2d+1 k − k Now, with the help of the two computations above, n 2 can be estimated: k kk2 n 2 k kk2 = n (s, y˜) 2d(s, y˜) | k | RZ2d = ξ(r)η (˜y + r s) fˆ2,3,4,5,6 s r, λ ck(˜y s), ρ, λ, 0, 0 k,0 − k − k − Z2d Zd R R 2 2,3,4,5,6 k fˆ s r, λkc (˜y s), ρ, λ, 0, 0 dr d(s, y˜) − ∞ − − ξ(r)η (˜y + r s) s r ω1 + y ω 2 + t ω3 + a¨ ω4 ≤ k,0 − k − k k k k k | | k k k k Z2d Zd Z R R K 2 C4 2πi(λ ck(˜y s)y+ρt) e− k − d(y,t, a¨)dr d(s, y˜) (1 + s r )2d+1 k − k 1 2 = C2 ξ(r) η (˜y + r s) s r ω5 + ω6 dr d(s, y˜) 4 (1 + s r )2d+1 | | k,0 − k − k k k Z2d Zd R R k − k Cauchy 2 − 2 1 2 2 5 6 C4 2d+2 ξ(r) ηk,0(˜y + r s) s r ωk + ωk dr Schwarz≤ (1 + s r ) | | − k − k Z2d Zd R R k − k 1 dr d(s, y˜) (1 + s r )2d RZd k − k 7 1 2 2 C′ ω ξ(r) η (˜y + r s) drd(s, y˜) ≤ 4 k (1 + s r )2d | | k,0 − Z2d Zd R R k − k ηk,0 2=1 7 1 2 k =k C′ ω ξ(r) drds 4 k (1 + s r )2d | | RZd RZd k − k 7 2 = C′′ω ξ 4 kk k2 i k with constants C4′ > 0 and C4′′ > 0 depending on f and ωk →∞ 0 for i 5,..., 7 . Thus, k −→ ∈ { } ω := C ω7 fulfills ω →∞ 0 and k 4′′ k k −→ q n ω ξ . k kk2 ≤ kk k2 35 Last, it still remains to examine wk: η (˜y) η (˜y + r s) k,0 − k,0 − d 1 = (r s ) ∂ η (˜y + t(r s))dt j − j j k,0 − Xj=1 Z0 d 1 d 1 1 = (r s ) λ ck 4 η λ ck 2 (˜y + t(r s )) j − j k i i k i i i − i j=1 Z i=1 X 0 Yi=j 6 3 1 k 4 k 2 λkcj ∂ηj λkcj (˜yj + t(rj sj)) dt. − ! Since f and the functions (ηj)j 1,...,d are Schwartz functions, one can find a constant C5 ∈{ } depending on (ηj)j 1,...,d such that ∈{ } λ ck η (˜y) η (˜y + r s) fˆ2,3,4,5,6 s r, k (s + r), ρ ,λ , 0, 0 k,0 − k,0 − k − − 2 k k d d 1 1 C5 r s λ ck 4 λ ck 2 . ≤ k − k k i k j (1 + r s )2d+1 j=1 i=1 X Y k − k Now, one has the following estimation for w , which is again similar to the above ones: k kk2 w 2 = w (s) 2ds k kk2 | k | RZd d d 1 1 ξ(r) η (˜y) r s λ ck 4 λ ck 2 ≤ | | k,0 k − k k i k j Zd Zd Zd j=1 i=1 R R R X Y C 2 5 drdy˜ ds (1 + r s )2d+1 k − k d d Cauchy 1 ξ(r) 2 − k 2 k 2 2 C5′ λkci λkcj | | 2d+2 ηk,0(˜y) r s d(r, y,s˜ ) Schwarz≤ (1 + r s ) k − k j=1 i=1 Z3d X Y R k − k d d ηk,0 2=1 1 k k k k 2 C′′ λ c 2 λ c ξ , ≤ 5 k i k j k k2 j=1 i=1 X Y where the constants C5′ > 0 and C5′′ > 0 depend on the functions (ηj)j 1,...,d . Therefore, for 1 ∈{ } d d 1 2 k ǫ := C λ ck 2 λ ck , the desired properties ǫ →∞ 0 and k 5′′ k i k j k −→ j=1 i=1 P Q w ǫ ξ k kk2 ≤ kk k2 are fulfilled. 36 Thus, for those f (R2d+2+p) = (G) whose Fourier transform in [g, g] has a com- ∈ S ∼ S pact support, k →∞ πk(f) νk pG/U (f) op = sup πk(f) νk pG/U (f) (ξ) 2 0. − ξ L2(Rd) − −→ ∈ ξ 2=1 k k Because of the density in L1(G) and thus in C (G) of the set of Schwartz functions f (G) ∗ ∈S whose partial Fourier transform has a compact support, the claim is true for general elements a C (G) by Corollary 7.6 in the appendix. ∈ ∗ V 3.5.4 Transition to πk k N ∈ Finally, from the results shown above for the sequence (πk)k N, the assertion for the original V ∈ sequence πk k N shall be deduced. ∈ Since for every k N the two representations π and πV are equivalent, there exist unitary ∈ k k intertwining operators 2 Rd 2 Rd V Fk : πV = L ( ) πk = L ( ) such that Fk πk (a)= πk(a) Fk a C∗(G). H k ∼ →H ∼ ◦ ◦ ∀ ∈ V V N Furthermore, since the limit set L πk k N of the sequence πk )k is contained in Si 1, ∈ ∈ − as discussed in Section 3.3.1, by identifying G with the set of coadjoint orbits g∗/G, one can restrict an operator field ϕ CB(Si 1) to L(( k)k N) = ℓ + u⊥ and obtains an element in ∈ − O ∈ CB(ℓ + u⊥). Thus, as (a) L(( ) N) a bC∗(G) = C (L(( k)k N)) = C (ℓ + u⊥), one F | Ok k∈ | ∈ ∞ O ∈ ∞ can define the -isomorphism ∗ R2d R2d τ : C ( ) = C (ℓ + u⊥) C∗(G/U, χℓ) = C∗( ), (a) L(( ) N) pG/U (a). ∞ ∼ ∞ → ∼ F | Ok k∈ 7→ Now, for k N, defineν ˜ as ∈ k ν˜k(ϕ) := Fk∗ (νk τ) ϕ L(( ) ) Fk ϕ CB(Si 1). ◦ ◦ | Ok˜ k˜∈N ◦ ∀ ∈ − Since the image of ν is in (L2(Rd)) and F is an intertwining operator and thus bounded, k B k the image ofν ˜ is contained in (L2(Rd)) as well. k B Moreover, the operatorν ˜k is bounded: From the boundedness of νk (see Proposition 3.3) and using that τ is an isomorphism, one gets for every ϕ CB(Si 1), ∈ − ν˜k(ϕ) op = Fk∗ (νk τ) ϕ L(( ) ) Fk k k ◦ ◦ | Ok˜ k˜∈N ◦ op (νk τ) ϕ L(( ) ) ≤ ◦ | Ok˜ k˜∈N op τ ϕ L(( ) ) ≤ | Ok˜ k˜∈N C∗(R2 d)