Introduction
In [11 ] , Eymard introduced the Fourier and Fourier–Stieltjes algebras of a locally compact group G G G and initiated their systematic study. Denoted by A ( G ) A(G) A ( G ) and B ( G ) B(G) B ( G ) , respectively, these Banach algebras play a fundamental role in abstract harmonic analysis. When G G G is Abelian, the Fourier transform identifies A ( G ) A(G) A ( G ) and B ( G ) B(G) B ( G ) with the group algebra and the measure algebra of the dual group G ^ \widehat{G} G , respectively. In the non-Abelian setting, B ( G ) B(G) B ( G ) is naturally identified with the continuous dual of the full group C ∗ C^* C ∗ -algebra, while A ( G ) A(G) A ( G ) is the predual of the group von Neumann algebra. The Fourier and Fourier–Stieltjes algebras encode the properties of the underlying group. For instance, Eymard proved that the Gelfand spectrum of A ( G ) A(G) A ( G ) , namely the set of nonzero multiplicative linear functionals on A ( G ) A(G) A ( G ) , is homeomorphic to G G G via point evaluation.
Analogous function algebras, denoted A ( G ) A(\mathcal{G}) A ( G ) and B ( G ) B(\mathcal{G}) B ( G ) , have been defined when the underlying structure is not a group but a groupoid G \mathcal{G} G . Groupoids, first introduced by Brandt in [2 ] , are algebraic structures that generalize groups. They require only a partially defined product and do not necessarily have a unique identity element. Instead, they possess a unit space G ( 0 ) \mathcal{G}^{(0)} G ( 0 ) , whose elements serve as local identities. The theories of the Fourier and Fourier–Stieltjes algebras for locally compact groupoids were developed in recent years from multiple perspectives [7 , 24 , 26 , 27 , 28 , 29 ] . These theories largely differ in the choice between requiring continuity or measurability, and the choice of norm. In the special case where G \mathcal{G} G is a locally compact group, all of these theories reduce to the existing definitions for the Fourier and Fourier–Stieltjes algebras of groups. In this paper, we study the Fourier algebra of locally compact groupoids, adopting the approach of [27 ] by developing a continuous, rather than a measurable, theory. In this approach, a representation of a groupoid is defined to be a family of unitary maps between Hilbert spaces from a continuous Hilbert bundle, with coefficient functions being obtained using continuous sections of the Hilbert field.
While the Fourier and Fourier–Stieltjes algebras of locally compact groups have been studied extensively, their counterparts for locally compact groupoids remain comparatively unexplored. In particular, relatively little is known about their properties as function algebras or about the extent to which they reflect the structure of the underlying groupoid. The primary focus of this manuscript is the Fourier algebra of a class of groupoids known as HLS groupoids, continuing the investigation initiated in [7 ] . Motivated by analogous results for locally compact groups (see, for example, [5 , 12 , 14 , 15 , 19 ] ), we studied the restriction maps for the Fourier and Fourier–Stieltjes algebras of locally compact groupoids in [7 ] , establishing several general functorial properties. In particular, we proved that for every étale groupoid the restriction map from the Fourier algebra to the Fourier algebra of any isotropy subgroup is surjective, while the analogous statement for the Fourier–Stieltjes algebra fails in general. Complete descriptions of the Fourier and Fourier–Stieltjes algebras were also obtained for group bundles with discrete unit space.
Harmonic analysis on group bundles has been developed mainly in the Abelian setting through Pontryagin duality, fibrewise Fourier transforms, and dual Haar systems [13 , 22 , 30 , 33 ] , and in the compact setting through Peter–Weyl-type representation theory [9 ] . By contrast, little is known about the Fourier and Fourier–Stieltjes algebras of non-discrete, non-Abelian group bundles. HLS groupoids therefore provide a natural setting in which to investigate how the topology of a group bundle influences its Fourier algebra. Although HLS groupoids are group bundles, their unit space is no longer discrete, and the topology couples the finite fibers with the infinite fiber in a nontrivial way. Consequently, the techniques developed in [7 ] for discrete group bundles no longer apply here, and several basic structural questions concerning the Fourier algebra become substantially more delicate. On the other hand, the topology of HLS groupoids is sufficiently explicit to allow a detailed analysis. At the same time, HLS groupoids already exhibit several interesting phenomena absent in the settings of groups and discrete group bundles, including the failure of surjectivity of the Fourier–Stieltjes restriction map [7 ] and the failure of the converse to the groupoid amenability theorem [34 ] .
Main contributions
Our results fall into two categories. Most of the paper concerns HLS groupoids. The topology of an HLS groupoid admits a particularly clear description that simplifies the associated Fourier theory. As a result, we obtain concrete formulas and a structural description of the algebra, rather than existence statements alone. In the final section, we discuss the Fourier algebra of general étale groupoids. The general arguments are abstract and do not recover the finer information available for HLS groupoids.
Fourier algebra of HLS groupoids. Taking advantage of the special topological structure of HLS groupoids, we obtain new insights into their Fourier algebras and establish useful bounds for the Fourier norm (§ 3.1 ). In particular, we identify the continuous sections of the left regular Hilbert bundle with an explicit class of continuous functions on G \mathcal{G} G (Theorem 3.6). This concrete picture of the left regular representation underlies many subsequent results.
In our previous work [7 ] , we proved that the restriction map from the Fourier algebra of an étale groupoid to an isotropy subgroup is surjective. Here we strengthen that result for HLS groupoids by constructing explicit norm-preserving lifts of Fourier algebra functions from every fiber (Theorem 4.2 ). The construction preserves positive definiteness, and it allows elements on finitely many fibers to be lifted simultaneously.
In § 5 , we discuss two applications of the lifting result (Theorem 4.2 ). First, the explicit lifts let us describe the norm of an element in A ( G ) A(\mathcal{G}) A ( G ) completely in terms of its fibers (Theorems 5.1 and 5.2). Consequently, A ( G ) A(\mathcal{G}) A ( G ) embeds isometrically into the ℓ ∞ \ell^\infty ℓ ∞ -direct sum of the Fourier algebras of the isotropy subgroups of G \mathcal{G} G . Second, we prove a Leptin-type theorem for HLS groupoids: The Fourier algebra A ( G ) A(\mathcal{G}) A ( G ) has a bounded approximate identity if and only if the isotropy subgroup at infinity is amenable, and in that case the approximate identity can be chosen to be bounded by 1 (Theorem 5.5). Thus a Banach-algebraic property of A ( G ) A(\mathcal{G}) A ( G ) is governed entirely by the single infinite fiber.
Finally, we investigate the relationship between the Fourier algebra of an HLS groupoid and the underlying groupoid through its Gelfand spectrum. We prove that every nonzero multiplicative linear functional on the Fourier algebra is given by point evaluation, thereby showing that the Gelfand spectrum of the Fourier algebra of an HLS groupoid is naturally homeomorphic to the groupoid itself (Theorem 6.1 ).
Beyond HLS groupoids. In § 7 , we show that two of our results hold for every étale groupoid. Norm-preserving lifts exist for the restriction maps onto isotropy subgroups (Theorem 7.1), and the Gelfand spectrum of the Fourier algebra is homeomorphic to the groupoid (Theorem 7.2). The latter extends Eymard’s classical theorem for locally compact groups to étale groupoids, and answers a question of Paterson [27 Section 6] for the étale setting; the question remains open in general. These arguments, however, are abstract: the lifts are obtained from a non-constructive extension of vectors to continuous sections, and they come with no explicit formula. Our results on the fiberwise structure of the norm, the limiting behavior of the norm, and the bounded approximate identity all rely on the explicit form of the HLS lifts and on the HLS topology. We do not know analogues of these results for general étale groupoids. This is the main reason the HLS class is worth studying in its own right.
Notations and background
A groupoid is a set G \mathcal{G} G together with a subset G ( 2 ) ⊆ G × G \mathcal{G}^{(2)} \subseteq\mathcal{G} \times\mathcal{G} G ( 2 ) ⊆ G × G , a product map G ( 2 ) → G \mathcal{G}^{(2)} \to\mathcal{G} G ( 2 ) → G , ( a , b ) ↦ a b (a,b) \mapsto ab ( a , b ) ↦ ab and an involutive inverse map G → G \mathcal{G} \to\mathcal{G} G → G , a ↦ a − 1 a \mapsto a^{-1} a ↦ a − 1 that satisfies the following:
if ( a , b ) , ( b , c ) ∈ G ( 2 ) (a,b),(b,c) \in\mathcal{G}^{(2)} ( a , b ) , ( b , c ) ∈ G ( 2 ) , then ( a b , c ) , ( a , b c ) ∈ G ( 2 ) (ab,c),(a,bc) \in\mathcal{G}^{(2)} ( ab , c ) , ( a , b c ) ∈ G ( 2 ) and ( a b ) c = a ( b c ) (ab)c = a(bc) ( ab ) c = a ( b c ) ;
( b , b − 1 ) ∈ G ( 2 ) (b,b^{-1}) \in\mathcal{G}^{(2)} ( b , b − 1 ) ∈ G ( 2 ) for all b ∈ G b \in\mathcal{G} b ∈ G , and if ( a , b ) ∈ G ( 2 ) (a,b) \in\mathcal{G}^{(2)} ( a , b ) ∈ G ( 2 ) , then a − 1 ( a b ) = b a^{-1}(ab) = b a − 1 ( ab ) = b , and ( a b ) b − 1 = a (ab)b^{-1} = a ( ab ) b − 1 = a .
Groupoids come equipped with range and source maps, denoted r r r and s s s respectively which are defined as
r ( a ) : = a a − 1 and s ( a ) : = a − 1 a . r(a) := aa^{-1} \quad\text{and} \quad s(a) := a^{-1}a. r ( a ) := a a − 1 and s ( a ) := a − 1 a .
These maps can be used to define the unit space as G ( 0 ) : = r ( G ) = { a a − 1 : a ∈ G } \mathcal{G}^{(0)} := r(\mathcal{G}) = \{aa^{-1} : a \in\mathcal{G}\} G ( 0 ) := r ( G ) = { a a − 1 : a ∈ G } . Clearly, G ( 0 ) = s ( G ) \mathcal{G}^{(0)} = s(\mathcal{G}) G ( 0 ) = s ( G ) as well. For u ∈ G ( 0 ) u \in\mathcal{G}^{(0)} u ∈ G ( 0 ) , we set G u : = r − 1 ( { u } ) \mathcal{G}^{u} := r^{-1}(\{u\}) G u := r − 1 ({ u }) , G u : = s − 1 ( { u } ) \mathcal{G}_{u} := s^{-1}(\{u\}) G u := s − 1 ({ u }) , and G u u : = G u ∩ G u \mathcal{G}^{u}_{u} := \mathcal{G}_{u} \cap\mathcal{G}^{u} G u u := G u ∩ G u . These fibers do not typically form algebraic objects in their own right, however, the set G u u \mathcal{G}^{u}_{u} G u u always forms a group, called the *isotropy subgroup at u u u .
An important class of groupoids is a disjoint union of groups, known as a group bundle . Namely, for a collection { G i } i ∈ I \{G_i\}_{i\in I} { G i } i ∈ I of groups, define G : = ⨆ i ∈ I ( { i } × G i ) \mathcal{G} := \bigsqcup_{i\in I}(\{i\} \times G_i) G := ⨆ i ∈ I ({ i } × G i ) . Elements of G \mathcal{G} G are composable if they have identical indices in their first coordinates. The product and the inverse are defined as ( i , x ) ( i , y ) = ( i , x y ) (i,x)(i,y) = (i,xy) ( i , x ) ( i , y ) = ( i , x y ) and ( i , x ) − 1 = ( i , x − 1 ) (i,x)^{-1} = (i,x^{-1}) ( i , x ) − 1 = ( i , x − 1 ) respectively, where x y xy x y and x − 1 x^{-1} x − 1 are the product and inverse in the group G i G_i G i . Then G \mathcal{G} G becomes a groupoid with the unit space G ( 0 ) = { ( i , e G i ) : i ∈ I } \mathcal{G}^{(0)} = \{(i,e_{G_i}) : i \in I\} G ( 0 ) = {( i , e G i ) : i ∈ I } , where e G i e_{G_i} e G i is the neutral element of G i G_i G i .
A topological groupoid is a groupoid G \mathcal{G} G endowed with a topology in which the inverse map G → G \mathcal{G} \to\mathcal{G} G → G , γ ↦ γ − 1 \gamma\mapsto\gamma^{-1} γ ↦ γ − 1 and the product map G ( 2 ) → G \mathcal{G}^{(2)} \to\mathcal{G} G ( 2 ) → G , ( g , h ) ↦ g h (g,h) \mapsto gh ( g , h ) ↦ g h are continuous. Here, G ( 2 ) \mathcal{G}^{(2)} G ( 2 ) is equipped with the relative topology as a subset of G × G \mathcal{G} \times\mathcal{G} G × G . A locally compact (respectively Hausdorff ) groupoid is a topological groupoid whose topology is locally compact (respectively Hausdorff). Clearly, the range and source maps of a topological groupoid are continuous. If we take G \mathcal{G} G to be Hausdorff, then G ( 0 ) \mathcal{G}^{(0)} G ( 0 ) is also Hausdorff, and consequently, the fibers G u \mathcal{G}^{u} G u and G u \mathcal{G}_{u} G u are closed subsets of G \mathcal{G} G .
Locally compact groups are guaranteed Haar measures which allow a theory of integration to be developed. The analogous concept in the locally compact groupoid setting is called a Haar system . Roughly speaking, a Haar system is a family of positive, regular Borel measures defined on the fibers G u \mathcal{G}^{u} G u satisfying a natural translation-invariance property. Unlike locally compact groups, however, locally compact groupoids do not necessarily admit Haar systems. For a general discussion of Haar systems, see [7 Definition 2.3] or [25 Definition 2.2.2] .
Certain classes of groupoids, such as étale groupoids, admit canonically defined Haar systems. A locally compact, Hausdorff groupoid is called étale if the range map r : G → G ( 0 ) r:\mathcal{G}\to\mathcal{G}^{(0)} r : G → G ( 0 ) (or equivalently, the source map) is a local homeomorphism. All discrete groupoids are étale, and if G \mathcal{G} G is an étale groupoid, each G u \mathcal{G}^{u} G u is a discrete set in the relative topology. All étale groupoids have Haar systems consisting of counting measures, that is, each G u \mathcal{G}^{u} G u is equipped with its counting measure. Étale groupoids have received a great deal of attention (see, for example, [1 , 3 , 4 , 10 , 17 , 18 , 20 , 21 , 23 , 31 ] ). Unless otherwise specified, throughout this manuscript, we will assume our groupoids to be étale.
Remark 2.1. The terminology r -discrete is not uniform in the literature and is sometimes used synonymously with étale. Throughout this paper, we use only the latter term; see [35 ] for a discussion of the terminology.
Fourier and Fourier–Stieltjes algebras for groupoids
To extend the definition of Fourier and Fourier–Stieltjes algebras associated with locally compact groups to the realm of groupoids, we start with defining what a representation of a locally compact groupoid is. Roughly speaking, while a group requires unitary representations over Hilbert spaces, groupoids will require G \mathcal{G} G -Hilbert bundles.
Definition 2.2 (Continuous Banach field). [8 Definition 10.1.2] A continuous field of Banach spaces E \mathcal{E} E over a topological space T T T is a family { E t } t ∈ T \{E_t\}_{t\in T} { E t } t ∈ T of Banach spaces, with a set Γ ⊆ ∏ t ∈ T E t \Gamma\subseteq\prod_{t\in T}E_t Γ ⊆ ∏ t ∈ T E t such that:
Γ \Gamma Γ is a complex linear subspace of ∏ t ∈ T E t \prod_{t\in T}E_t ∏ t ∈ T E t ;
For every t ∈ T t\in T t ∈ T , the set { x ( t ) : x ∈ Γ } \{x(t):x\in\Gamma\} { x ( t ) : x ∈ Γ } is dense in E t E_t E t ;
For every x ∈ Γ x\in\Gamma x ∈ Γ , the function t ↦ ∥ x ( t ) ∥ E t t\mapsto\|x(t)\|_{E_t} t ↦ ∥ x ( t ) ∥ E t is continuous;
Let x ∈ ∏ t ∈ T E t x\in\prod_{t\in T}E_t x ∈ ∏ t ∈ T E t . If for every t ∈ T t\in T t ∈ T and every ϵ > 0 \epsilon>0 ϵ > 0 , there exists x ′ ∈ Γ x'\in\Gamma x ′ ∈ Γ such that ∥ x ( τ ) − x ′ ( τ ) ∥ E τ < ϵ \|x(\tau)-x'(\tau)\|_{E_\tau}<\epsilon ∥ x ( τ ) − x ′ ( τ ) ∥ E τ < ϵ for all τ \tau τ in some neighborhood of t t t , then x ∈ Γ x\in\Gamma x ∈ Γ .
We refer to elements of ∏ t ∈ T E t \prod_{t\in T}E_t ∏ t ∈ T E t as sections , and elements of Γ \Gamma Γ as continuous sections . In the case where each Banach space is also a Hilbert space, the definition above is referred to as a continuous Hilbert field .
In practice, continuous fields of Banach spaces are built as a ‘closure’ of a collection of sections satisfying (1), (2) and (3) of Theorem 2.2. This is done using the following result.
Proposition 2.3. [8 Proposition 10.2.3] Let { E t } t ∈ T \{E_t\}_{t\in T} { E t } t ∈ T be a family of Banach spaces indexed over a topological space T T T . Suppose that there exists a set Ξ ⊆ ∏ t ∈ T E t \Xi\subseteq\prod_{t\in T}E_t Ξ ⊆ ∏ t ∈ T E t of sections satisfying the first three conditions of Theorem 2.2. Then there is a unique subset Γ ⊆ ∏ t ∈ T E t \Gamma\subseteq\prod_{t\in T}E_t Γ ⊆ ∏ t ∈ T E t that satisfies all four conditions and contains Ξ \Xi Ξ .
The collection Ξ \Xi Ξ is referred to as the fundamental family of sections, and elements in Ξ \Xi Ξ are called the fundamental sections.
Moving forward, let H : = ( ( H t ) t ∈ T , Γ ) \mathcal{H}:=((\mathcal{H}_t)_{t\in T},\Gamma) H := (( H t ) t ∈ T , Γ ) be a continuous Hilbert field over the space T T T . Denote the norm of each Hilbert space H t \mathcal{H}_t H t as ∥ ⋅ ∥ t \|\cdot\|_t ∥ ⋅ ∥ t or ∥ ⋅ ∥ H t \|\cdot\|_{\mathcal{H}_t} ∥ ⋅ ∥ H t for t ∈ T t\in T t ∈ T . We define the section norm
∥ ξ ∥ Δ = sup t ∈ T ∥ ξ ( t ) ∥ t . \|\xi\|_{\Delta}=\sup_{t\in T}\|\xi(t)\|_t. ∥ ξ ∥ Δ = t ∈ T sup ∥ ξ ( t ) ∥ t .
Let Δ b ( H ) \Delta_b(\mathscr{H}) Δ b ( H ) denote the subspace of Γ \Gamma Γ consisting of sections whose ∥ ⋅ ∥ Δ \|\cdot\|_{\Delta} ∥ ⋅ ∥ Δ -norm is bounded. Equipped with pointwise operations and ∥ ⋅ ∥ Δ \|\cdot\|_{\Delta} ∥ ⋅ ∥ Δ -norm, the space Δ b ( H ) \Delta_b(\mathscr{H}) Δ b ( H ) is a Banach space ([27 pg 6] ). We further define Δ c ( H ) \Delta_c(\mathscr{H}) Δ c ( H ) to be the subspaces of Γ \Gamma Γ of continuous sections with compact support, and Δ 0 ( H ) \Delta_0(\mathscr{H}) Δ 0 ( H ) to be those sections in Γ \Gamma Γ that vanish at infinity under the section norm. All three of Δ b \Delta_b Δ b , Δ c \Delta_c Δ c , and Δ 0 \Delta_0 Δ 0 are C ( T ) C(T) C ( T ) -modules under pointwise multiplication as the module action ([8 Proposition 10.1.9] ).
We now give the definition of a groupoid representation as presented in [27 Section 3] .
Definition 2.4 (Groupoid representation). Let G \mathcal{G} G be a topological groupoid, and H : = { H u } u ∈ G ( 0 ) \mathscr{H} := \{ \mathscr{H}_u \}_{u\in\mathcal{G}^{(0)}} H := { H u } u ∈ G ( 0 ) , Γ \Gamma Γ be a continuous Hilbert field. A (continuous unitary) representation of G \mathcal{G} G on the continuous Hilbert field H \mathscr{H} H is a collection of unitary maps ϖ ( x ) : H s ( x ) → H r ( x ) \varpi(x):\mathscr{H}_{s(x)}\to\mathscr{H}_{r(x)} ϖ ( x ) : H s ( x ) → H r ( x ) , for each x ∈ G x\in\mathcal{G} x ∈ G , that satisfy the following:
For each ξ , η ∈ Δ b ( H ) \xi,\eta\in\Delta_b(\mathscr{H}) ξ , η ∈ Δ b ( H ) the map G → C \mathcal{G}\to\mathbb{C} G → C defined as x ↦ ⟨ ϖ ( x ) ξ ( s ( x ) ) , η ( r ( x ) ) ⟩ H r ( x ) x\mapsto\langle\varpi(x)\xi(s(x)),\eta(r(x))\rangle_{\mathscr{H}_{r(x)}} x ↦ ⟨ ϖ ( x ) ξ ( s ( x )) , η ( r ( x )) ⟩ H r ( x ) is continuous;
For every ( x , y ) ∈ G ( 2 ) (x,y)\in\mathcal{G}^{(2)} ( x , y ) ∈ G ( 2 ) , we have ϖ ( x y ) = ϖ ( x ) ϖ ( y ) \varpi(xy)=\varpi(x)\varpi(y) ϖ ( x y ) = ϖ ( x ) ϖ ( y ) .
A representation as above is also called a G \mathcal{G} G -Hilbert bundle over G ( 0 ) \mathcal{G}^{(0)} G ( 0 ) . We use these two terminologies interchangeably.
The map x ↦ ⟨ ϖ ( x ) ξ ( s ( x ) ) , η ( r ( x ) ) ⟩ H r ( x ) x\mapsto\langle\varpi(x)\xi(s(x)),\eta(r(x))\rangle_{\mathscr{H}_{r(x)}} x ↦ ⟨ ϖ ( x ) ξ ( s ( x )) , η ( r ( x )) ⟩ H r ( x ) given in Theorem 2.4 is referred to as the coefficient function of H \mathscr{H} H associated with ξ \xi ξ and η \eta η , and denoted by H ξ , η \mathscr{H}_{\xi,\eta} H ξ , η . To differentiate between coefficient functions of group representations, and coefficients of G \mathcal{G} G -Hilbert bundles, we will reserve lower case Greek letters and unbolded symbols for group representations, and accented text with bold symbols to refer to G \mathcal{G} G -Hilbert bundles and their sections. We note that it suffices to check condition (1) of Theorem 2.4 for ξ , η ∈ Ξ \xi,\eta\in\Xi ξ , η ∈ Ξ , the fundamental sections, that are ∥ ⋅ ∥ Δ \|\cdot\|_{\Delta} ∥ ⋅ ∥ Δ -bounded.
Example 2.5. Let G \mathcal{G} G be a locally compact étale groupoid. Each G u \mathcal{G}^u G u is a discrete space and is therefore naturally equipped with the counting measure. We define L 2 ( G ) L^2(\mathcal{G}) L 2 ( G ) to be the continuous Hilbert field associated to the family of Hilbert spaces { ℓ 2 ( G u ) } u ∈ G ( 0 ) \{\ell^2(\mathcal{G}^u)\}_{u\in\mathcal{G}^{(0)}} { ℓ 2 ( G u ) } u ∈ G ( 0 ) , with C c ( G ) C_c(\mathcal{G}) C c ( G ) serving as a fundamental family of sections via the identification
C c ( G ) ∋ f ⟼ { f u } u ∈ G ( 0 ) , C_c(\mathcal{G})\ni f\longmapsto\{f^u\}_{u\in\mathcal{G}^{(0)}}, C c ( G ) ∋ f ⟼ { f u } u ∈ G ( 0 ) , where f u = f ∣ G u f^u=f|_{\mathcal{G}^u} f u = f ∣ G u . It is straightforward to verify that C c ( G ) C_c(\mathcal{G}) C c ( G ) indeed forms a fundamental family of sections in the sense of Theorem 2.3, and hence determines the continuous structure uniquely. An equivalent description of L 2 ( G ) L^2(\mathcal{G}) L 2 ( G ) is given in [27 p. 10] : a section { F u } u ∈ G ( 0 ) \{F^u\}_{u\in\mathcal{G}^{(0)}} { F u } u ∈ G ( 0 ) is continuous if and only if the functions u ↦ ∥ F u ∥ ℓ 2 ( G u ) u\mapsto\|F^u\|_{\ell^2(\mathcal{G}^u)} u ↦ ∥ F u ∥ ℓ 2 ( G u ) is continuous and u ↦ ⟨ F u , g u ⟩ u\mapsto\langle F^u,g^u\rangle u ↦ ⟨ F u , g u ⟩ is continuous for all g ∈ C c ( G ) g\in C_c(\mathcal{G}) g ∈ C c ( G ) . (See also [7 Example 2.7] for the definition of L 2 ( G ) L^2(\mathcal{G}) L 2 ( G ) for general groupoids equipped with Haar systems.)
The left regular representation of G \mathcal{G} G is defined on the continuous Hilbert field L 2 ( G ) L^2(\mathcal{G}) L 2 ( G ) as follows: For x ∈ G x\in\mathcal{G} x ∈ G , define
L x : ℓ 2 ( G s ( x ) ) → ℓ 2 ( G r ( x ) ) , L x δ z = δ x z . L_x:\ell^2(\mathcal{G}^{s(x)})\to\ell^2(\mathcal{G}^{r(x)}),\qquad L_x\delta_z=\delta_{xz}. L x : ℓ 2 ( G s ( x ) ) → ℓ 2 ( G r ( x ) ) , L x δ z = δ x z . This will turn L 2 ( G ) L^2(\mathcal{G}) L 2 ( G ) into a G \mathcal{G} G -Hilbert bundle, denoted Λ \Lambda Λ .
Definition 2.6. ([27 ] ) The Fourier–Stieltjes algebra of a locally compact groupoid G \mathcal{G} G is defined as
B ( G ) : = { H ξ , η : H is a G -Hilbert Bundle , ξ , η ∈ Δ b ( H ) } , B(\mathcal{G}):=\{\mathscr{H}_{\xi,\eta}:\mathscr{H}\text{ is a }\mathcal{G}\text{-Hilbert Bundle},\ \xi,\eta\in\Delta_b(\mathscr{H})\}, B ( G ) := { H ξ , η : H is a G -Hilbert Bundle , ξ , η ∈ Δ b ( H )} , equipped with pointwise operations and the norm defined as
∥ φ ∥ B ( G ) : = inf φ = H ξ , η ∥ ξ ∥ Δ ⋅ ∥ η ∥ Δ . \|\varphi\|_{B(\mathcal{G})}:=\inf_{\varphi=\mathscr{H}_{\xi,\eta}}\|\xi\|_{\Delta}\cdot\|\eta\|_{\Delta}. ∥ φ ∥ B ( G ) := φ = H ξ , η inf ∥ ξ ∥ Δ ⋅ ∥ η ∥ Δ . where the infimum is taken over all coefficient function representations of φ \varphi φ .
It was shown in [27 ] that B ( G ) B(\mathcal{G}) B ( G ) is a unital commutative Banach algebra. When G \mathcal{G} G admits a Haar system, the Fourier algebra A ( G ) A(\mathcal{G}) A ( G ) was also defined in [27 page 15] . Here, we recall this definition in the simpler setting of étale groupoids.
Definition 2.7. Let G \mathcal{G} G be a locally compact étale groupoid, and Λ \Lambda Λ denote the left regular representation of G \mathcal{G} G on L 2 ( G ) L^2(\mathcal{G}) L 2 ( G ) . Recall that C c ( G ) C_c(\mathcal{G}) C c ( G ) is the set of fundamental sections of L 2 ( G ) L^2(\mathcal{G}) L 2 ( G ) . Define
A c f ( G ) = { Λ f , g : f , g ∈ C c ( G ) } . A_{cf}(\mathcal{G})=\{\Lambda_{f,g}: f,g\in C_c(\mathcal{G})\}. A c f ( G ) = { Λ f , g : f , g ∈ C c ( G )} . Furthermore, let A c ( G ) A_c(\mathcal{G}) A c ( G ) be the algebra generated by A c f ( G ) A_{cf}(\mathcal{G}) A c f ( G ) under pointwise operations. The Fourier Algebra of G \mathcal{G} G is defined as
A ( G ) : = A c ( G ) ‾ ∥ ⋅ ∥ B ( G ) . A(\mathcal{G}):=\overline{A_c(\mathcal{G})}^{\|\cdot\|_{B(\mathcal{G})}}. A ( G ) := A c ( G ) ∥ ⋅ ∥ B ( G ) . It is easy to see that both A c f ( G ) A_{cf}(\mathcal{G}) A c f ( G ) and A c ( G ) A_c(\mathcal{G}) A c ( G ) are contained in C c ( G ) C_c(\mathcal{G}) C c ( G ) . By definition, A ( G ) A(\mathcal{G}) A ( G ) is a closed subalgebra of B ( G ) B(\mathcal{G}) B ( G ) and is therefore itself a Banach algebra. As in the group case, A ( G ) ⊆ C 0 ( G ) A(\mathcal{G})\subseteq C_0(\mathcal{G}) A ( G ) ⊆ C 0 ( G ) by [27 Proposition 14] . Moreover, A ( G ) A(\mathcal{G}) A ( G ) is an ideal of B ( G ) B(\mathcal{G}) B ( G ) by [27 Corollary 3] , and [27 Proposition 16(ii)] shows that A ( G ) A(\mathcal{G}) A ( G ) strongly separates points.
Next, we show that the two proposed definitions of A ( G ) A(\mathcal{G}) A ( G ) inspired by the above remark coincide for étale groupoids.2
Lemma 2.9. Let G \mathcal{G} G be a (locally compact Hausdorff) étale groupoid. Then, A c f ( G ) = A c ( G ) = C c ( G ) A_{cf}(\mathcal{G})=A_c(\mathcal{G})=C_c(\mathcal{G}) A c f ( G ) = A c ( G ) = C c ( G ) and
A ( G ) = B ( G ) ∩ C c ( G ) ‾ ∥ ⋅ ∥ B ( G ) . A(\mathcal{G})=\overline{B(\mathcal{G})\cap C_c(\mathcal{G})}^{\|\cdot\|_{B(\mathcal{G})}}. A ( G ) = B ( G ) ∩ C c ( G ) ∥ ⋅ ∥ B ( G ) .
Proof. By definition, A c f ( G ) ⊆ A c ( G ) ⊆ C c ( G ) A_{cf}(\mathcal{G})\subseteq A_c(\mathcal{G})\subseteq C_c(\mathcal{G}) A c f ( G ) ⊆ A c ( G ) ⊆ C c ( G ) . To prove the reverse inclusions, let f ∈ C c ( G ) f\in C_c(\mathcal{G}) f ∈ C c ( G ) be given, and note that s ( supp f ) s(\operatorname{supp}f) s ( supp f ) is a compact subset of G ( 0 ) \mathcal{G}^{(0)} G ( 0 ) , as the source map s s s is continuous. Choose a real-valued h ∈ C c ( G ( 0 ) ) h\in C_c(\mathcal{G}^{(0)}) h ∈ C c ( G ( 0 ) ) equal to 1 on s ( supp f ) s(\operatorname{supp}f) s ( supp f ) , and extend h h h to h ~ : G → C \widetilde{h}:\mathcal{G}\to\mathbb{C} h : G → C as follows:
h ~ ( x ) = { h ( x ) x ∈ G ( 0 ) , 0 otherwise . \widetilde{h}(x)=
\begin{cases}
h(x) & x\in\mathcal{G}^{(0)},\\
0 & \text{otherwise}.
\end{cases} h ( x ) = { h ( x ) 0 x ∈ G ( 0 ) , otherwise .
Note that G \mathcal{G} G is Hausdorff and étale, so the unit space G ( 0 ) \mathcal{G}^{(0)} G ( 0 ) is an open and closed subset of G \mathcal{G} G . Therefore, the function h ~ \widetilde{h} h belongs to C c ( G ) C_c(\mathcal{G}) C c ( G ) . Moreover, for every γ ∈ G \gamma\in\mathcal{G} γ ∈ G , we have
Λ h ~ , f ~ ( γ ) = ∑ t ∈ G r ( γ ) h ~ ( γ − 1 t ) f ( t ) = h ( γ − 1 γ ) f ( γ ) = h ( s ( γ ) ) f ( γ ) = f ( γ ) , \Lambda_{\widetilde{h},\widetilde{f}}(\gamma)=\sum_{t\in\mathcal{G}r(\gamma)}\widetilde{h}(\gamma^{-1}t)f(t)=h(\gamma^{-1}\gamma)f(\gamma)=h(s(\gamma))f(\gamma)=f(\gamma), Λ h , f ( γ ) = t ∈ G r ( γ ) ∑ h ( γ − 1 t ) f ( t ) = h ( γ − 1 γ ) f ( γ ) = h ( s ( γ )) f ( γ ) = f ( γ ) ,
where the second equality follows from the fact that t = γ t=\gamma t = γ is the only possible element of G \mathcal{G} G that satisfies γ − 1 t ∈ supp ( h ) ⊆ G ( 0 ) \gamma^{-1}t\in\operatorname{supp}(h)\subseteq\mathcal{G}^{(0)} γ − 1 t ∈ supp ( h ) ⊆ G ( 0 ) . Since both h ~ \widetilde{h} h and f ~ \widetilde{f} f are elements of C c ( G ) C_c(\mathcal{G}) C c ( G ) , we get Λ h ~ , f ~ ∈ A c f ( G ) \Lambda_{\widetilde{h},\widetilde{f}}\in A_{cf}(\mathcal{G}) Λ h , f ∈ A c f ( G ) . This establishes A c f ( G ) = A c ( G ) = C c ( G ) A_{cf}(\mathcal{G})=A_c(\mathcal{G})=C_c(\mathcal{G}) A c f ( G ) = A c ( G ) = C c ( G ) . Finally, we conclude
B ( G ) ∩ C c ( G ) ‾ ∥ ⋅ ∥ B ( G ) = B ( G ) ∩ A c ( G ) ‾ ∥ ⋅ ∥ B ( G ) = A c ( G ) ‾ ∥ ⋅ ∥ B ( G ) = A ( G ) . \overline{B(\mathcal{G})\cap C_c(\mathcal{G})}^{\|\cdot\|_{B(\mathcal{G})}}
=\overline{B(\mathcal{G})\cap A_c(\mathcal{G})}^{\|\cdot\|_{B(\mathcal{G})}}
=\overline{A_c(\mathcal{G})}^{\|\cdot\|_{B(\mathcal{G})}}
=A(\mathcal{G}). B ( G ) ∩ C c ( G ) ∥ ⋅ ∥ B ( G ) = B ( G ) ∩ A c ( G ) ∥ ⋅ ∥ B ( G ) = A c ( G ) ∥ ⋅ ∥ B ( G ) = A ( G ) .
Definition 2.10 (Restriction maps). For a locally compact groupoid G \mathcal{G} G and an element u ∈ G ( 0 ) u\in\mathcal{G}^{(0)} u ∈ G ( 0 ) , we define the map R B ( G ) u : B ( G ) → B ( G u ) R^{u}_{B(\mathcal{G})}:B(\mathcal{G})\to B(\mathcal{G}^{u}) R B ( G ) u : B ( G ) → B ( G u ) by φ ↦ φ ∣ G u \varphi\mapsto\varphi|_{\mathcal{G}^{u}} φ ↦ φ ∣ G u . An analogous restriction function is defined on A ( G ) A(\mathcal{G}) A ( G ) via R A ( G ) u : A ( G ) → A ( G u ) R^{u}_{A(\mathcal{G})}:A(\mathcal{G})\to A(\mathcal{G}^{u}) R A ( G ) u : A ( G ) → A ( G u ) , φ ↦ φ ∣ G u \varphi\mapsto\varphi|_{\mathcal{G}^{u}} φ ↦ φ ∣ G u .
The above restriction maps were introduced in [7 ] . Clearly, R A ( G ) u = R B ( G ) u ∣ A ( G ) R^u_{A(\mathcal{G})}=R^u_{B(\mathcal{G})}|_{A(\mathcal{G})} R A ( G ) u = R B ( G ) u ∣ A ( G ) . However, the two maps exhibit different fundamental properties. For example, when G \mathcal{G} G is étale, R A ( G ) u R^u_{A(\mathcal{G})} R A ( G ) u is surjective, while R B ( G ) u R^u_{B(\mathcal{G})} R B ( G ) u need not be. We therefore used the subscripts to distinguish their domains. In this paper, we consider only restrictions of elements of A ( G ) A(\mathcal{G}) A ( G ) and accordingly simplify our notation to R u : A ( G ) → A ( G u ) R_u:A(\mathcal{G})\to A(\mathcal{G}_u) R u : A ( G ) → A ( G u ) .
HLS groupoids and their Fourier algebras
In this section, we study a particular class of group bundles called the HLS3 groupoids, originally appearing in [16 ] . The definition of HLS groupoids is based on the notion of an approximating sequence of normal subgroups, which we now recall. An approximating sequence for a discrete group G G G is a sequence { K n } n ∈ N \{K_n\}_{n\in\mathbb{N}} { K n } n ∈ N of normal subgroups of G G G with finite index satisfying
⋂ n ∈ N K n = { e G } and K n ⊇ K n + 1 for n ∈ N . \bigcap_{n\in\mathbb{N}} K_n=\{e_G\}\quad\text{and}\quad K_n\supseteq K_{n+1}\quad\text{for }n\in\mathbb{N}. n ∈ N ⋂ K n = { e G } and K n ⊇ K n + 1 for n ∈ N .
The following lemma guarantees the existence of approximating sequence of normal subgroups for a large class of groups. This lemma is a generalization of [34 Lemma 2.8] , where a similar result for the group F 2 F_2 F 2 is proved. Recall that a residually finite group is one in which if x ≠ e G x\ne e_G x = e G then there exists a homomorphism φ \varphi φ from G G G to a finite group such that φ ( x ) ≠ e \varphi(x)\ne e φ ( x ) = e .
Lemma 3.1. Let G G G be a residually finite and finitely generated group. Then G G G has an approximating sequence.
Proof. Since G G G is finitely generated, there are only finitely many homomorphisms from G G G into groups of order at most n n n . Define K n K_n K n to be the intersection of the kernels of all such homomorphisms. Each K n K_n K n is then a finite-index normal subgroup of G G G , because finite intersections of finite-index subgroups again have finite index. The sequence { K n } \{K_n\} { K n } is nested by construction, and residual finiteness of G G G implies that the intersection of all the sets K n K_n K n , for n ∈ N n\in\mathbb{N} n ∈ N is trivial. Hence { K n } \{K_n\} { K n } is an approximating sequence for G G G .
Definition 3.2. Given a discrete group G G G and an approximating sequence { K n } n ∈ N \{K_n\}_{n\in\mathbb{N}} { K n } n ∈ N , let G n G_n G n denote the finite quotient group G / K n G/K_n G / K n , and q n : G → G n q_n:G\to G_n q n : G → G n be the canonical quotient map. We set G ∞ : = G G_\infty:=G G ∞ := G and q ∞ q_\infty q ∞ the identity map. The HLS groupoid associated to ( G , { K n } n ∈ N ) (G,\{K_n\}_{n\in\mathbb{N}}) ( G , { K n } n ∈ N ) is the group bundle
G : = ⋃ n ∈ N ∪ { ∞ } { n } × G n , \mathcal{G}:=\bigcup_{n\in\mathbb{N}\cup\{\infty\}}\{n\}\times G_n, G := n ∈ N ∪ { ∞ } ⋃ { n } × G n , where G \mathcal{G} G is equipped with the topology generated by the following open sets:
(i) The singleton { ( n , g ) } \{(n,g)\} {( n , g )} for each n ∈ N n\in\mathbb{N} n ∈ N and g ∈ G n g\in G_n g ∈ G n ;
(ii) The ray at ( ∞ , g ) (\infty,g) ( ∞ , g ) defined as ray n ( g ) : = { ( m , q m ( g ) ) : m ∈ N ∪ { ∞ } , m ≥ n } \operatorname{ray}_n(g):=\{(m,q_m(g)):m\in\mathbb{N}\cup\{\infty\},\,m\ge n\} ray n ( g ) := {( m , q m ( g )) : m ∈ N ∪ { ∞ } , m ≥ n } for fixed g ∈ G g\in G g ∈ G and n ∈ N n\in\mathbb{N} n ∈ N .
As any HLS groupoid G \mathcal{G} G is a group bundle, we will often think of each individual group G n G_n G n as a subset of G \mathcal{G} G via the identification q n ( x ) ∼ ( n , q n ( x ) ) q_n(x)\sim(n,q_n(x)) q n ( x ) ∼ ( n , q n ( x )) . HLS groupoids are locally compact, Hausdorff and étale. If the discrete group G G G is countable, then the associated HLS groupoid G \mathcal{G} G is second countable. The topology of an HLS groupoid gives a particularly simple description of continuous functions. Indeed, a function f : G → C f:\mathcal{G}\to\mathbb{C} f : G → C is continuous if and only if it is continuous along each ray, i.e., for every g ∈ G ∞ g\in G_\infty g ∈ G ∞ , one has
lim n → ∞ f ( n , q n ( g ) ) = f ( ∞ , g ) . (1) \lim_{n\to\infty} f(n,q_n(g))=f(\infty,g). \tag*{(1)} n → ∞ lim f ( n , q n ( g )) = f ( ∞ , g ) . (1)
The following result collects useful information about the topology of an HLS groupoid and the description of its compact subsets.
Proposition 3.3. Let G \mathcal{G} G be the HLS groupoid associated to the pair ( G , { K n } n ∈ N ) (G,\{K_n\}_{n\in\mathbb{N}}) ( G , { K n } n ∈ N ) . Then we have the following properties:
For every n ∈ N n\in\mathbb{N} n ∈ N and g ∈ G g\in G g ∈ G , the ray ray n ( g ) \operatorname{ray}_n(g) ray n ( g ) is an open and closed compact set.
A set C ⊆ G C\subseteq\mathcal{G} C ⊆ G is compact if and only if there exists a finite subset { g 1 , … , g n } \{g_1,\ldots,g_n\} { g 1 , … , g n } of G G G and a finite subset F F F of C C C such that
{ ( ∞ , g 1 ) , … , ( ∞ , g n ) } ⊆ C ∖ F ⊆ ⋃ 1 ≤ i ≤ n ray 1 ( g i ) . (2) \{(\infty,g_1),\ldots,(\infty,g_n)\}\subseteq C\setminus F\subseteq\bigcup_{1\leq i\leq n}\operatorname{ray}_1(g_i). \tag*{(2)} {( ∞ , g 1 ) , … , ( ∞ , g n )} ⊆ C ∖ F ⊆ 1 ≤ i ≤ n ⋃ ray 1 ( g i ) . (2) Let E E E be a finite set in G G G . There exists N ∈ N N\in\mathbb{N} N ∈ N such that ray N ( x ) ∩ ray N ( y ) = ∅ \operatorname{ray}_N(x)\cap\operatorname{ray}_N(y)=\varnothing ray N ( x ) ∩ ray N ( y ) = ∅ for every x ≠ y ∈ E x\ne y\in E x = y ∈ E .
Proof. Let { U α } α ∈ I \{U_\alpha\}_{\alpha\in I} { U α } α ∈ I be an open cover of ray n ( x ) \operatorname{ray}_n(x) ray n ( x ) . At least one set in the open cover, say U α x U_{\alpha_x} U α x , must contain ( ∞ , x ) (\infty,x) ( ∞ , x ) . By the definition of the topology of G \mathcal{G} G , there exists n ′ ∈ N n'\in\mathbb{N} n ′ ∈ N such that ray n ′ ( x ) ⊆ U α x \operatorname{ray}_{n'}(x)\subseteq U_{\alpha_x} ray n ′ ( x ) ⊆ U α x . The set ray n ( x ) ∖ ray n ′ ( x ) \operatorname{ray}_n(x)\setminus\operatorname{ray}_{n'}(x) ray n ( x ) ∖ ray n ′ ( x ) is finite (or possibly even empty), and can be covered by a finite number of sets in the open cover { U α } α ∈ I \{U_\alpha\}_{\alpha\in I} { U α } α ∈ I . This proves the compactness of the ray. As G \mathcal{G} G is Hausdorff and each ray n ( x ) \operatorname{ray}_n(x) ray n ( x ) is compact, they are also closed. This finishes the proof of Item 1.
To prove Item 2, let C C C be a subset of G \mathcal{G} G and consider the following open cover of C C C :
{ V g : = ray 1 ( g ) : ( ∞ , g ) ∈ C } ∪ { W g : = { ( n , g ) } : n ∈ N , g ∈ G n , ( n , g ) ∈ C } . \{V_g:=\operatorname{ray}_1(g):( \infty,g)\in C\}\cup\{W_g:=\{(n,g)\}:n\in\mathbb{N},\ g\in G_n,\ (n,g)\in C\}. { V g := ray 1 ( g ) : ( ∞ , g ) ∈ C } ∪ { W g := {( n , g )} : n ∈ N , g ∈ G n , ( n , g ) ∈ C } .
Any subcover of the above open cover should contain all sets of the form V g V_g V g . So, if C C C is compact, then C ∩ G ∞ C\cap G_\infty C ∩ G ∞ must be finite. In addition, to find a finite subcover of the above cover, C ∖ ( ⋃ ( ∞ , g ) ∈ C ∩ G ∞ V g ) C\setminus(\bigcup_{(\infty,g)\in C\cap G_\infty}V_g) C ∖ ( ⋃ ( ∞ , g ) ∈ C ∩ G ∞ V g ) must be finite as well. This proves the ‘only if’ direction.
To prove the ‘if’ direction of the statement, fix an element ( ∞ , g ) (\infty,g) ( ∞ , g ) in G \mathcal{G} G with g ∈ G g\in G g ∈ G . Note that every open set in G \mathcal{G} G containing ( ∞ , g ) (\infty,g) ( ∞ , g ) should include some ray at ( ∞ , g ) (\infty,g) ( ∞ , g ) as a subset. Suppose C C C is a subset of G \mathcal{G} G satisfying (3.2), and let { O α : α ∈ I } \{O_\alpha:\alpha\in I\} { O α : α ∈ I } be an open cover for C C C . Let 1 ≤ i ≤ n 1\leq i\leq n 1 ≤ i ≤ n . From the previous discussion, since ( ∞ , g i ) ∈ C (\infty,g_i)\in C ( ∞ , g i ) ∈ C , there exists k i ∈ N k_i\in\mathbb{N} k i ∈ N and α i ∈ I \alpha_i\in I α i ∈ I , such that ray k i ( g i ) ⊆ O α i \operatorname{ray}_{k_i}(g_i)\subseteq O_{\alpha_i} ray k i ( g i ) ⊆ O α i . Thus, { O α : α ∈ I } \{O_\alpha:\alpha\in I\} { O α : α ∈ I } has a finite subcover, and C C C is compact.
To prove Item 3, note that the set { x − 1 y : x , y ∈ E , x ≠ y } \{x^{-1}y:x,y\in E,\ x\ne y\} { x − 1 y : x , y ∈ E , x = y } is a finite set not including e G ∞ e_{G_\infty} e G ∞ . Since
⋂ n ∈ N K n = { e G ∞ } , \bigcap_{n\in\mathbb{N}}K_n=\{e_{G_\infty}\}, n ∈ N ⋂ K n = { e G ∞ } ,
for every z ≠ e G ∞ z\ne e_{G_\infty} z = e G ∞ there exists an index N z N_z N z such that z ∉ K N z z\notin K_{N_z} z ∈ / K N z . Let
N = max { N x − 1 y : x , y ∈ E , x ≠ y } . N=\max\{N_{x^{-1}y}:x,y\in E,\ x\ne y\}. N = max { N x − 1 y : x , y ∈ E , x = y } .
Since { K n } n ∈ N \{K_n\}_{n\in\mathbb{N}} { K n } n ∈ N is a nested sequence of normal subgroups, for each n ≥ N n\geq N n ≥ N , we have x − 1 y ∉ K n x^{-1}y\notin K_n x − 1 y ∈ / K n , and consequently, q n ( x ) ≠ q n ( y ) q_n(x)\ne q_n(y) q n ( x ) = q n ( y ) . Therefore, ray N ( x ) ∩ ray N ( y ) = ∅ \operatorname{ray}_N(x)\cap\operatorname{ray}_N(y)=\varnothing ray N ( x ) ∩ ray N ( y ) = ∅ .
:::
Fourier algebra of an HLS groupoid
Here, we present the simplified definitions related to the representations and Fourier algebra of HLS groupoids. To begin, we define N ∗ : = N ∪ { ∞ } \mathbb{N}^*:=\mathbb{N}\cup\{\infty\} N ∗ := N ∪ { ∞ } and equip it with the topology of one point compactification of N \mathbb{N} N .
Definition 3.4 (Theorem 2.2 simplified). A continuous Hilbert field indexed over N ∗ \mathbb{N}^* N ∗ is a family of Hilbert spaces { H n } n ∈ N ∗ \{\mathcal{H}_n\}_{n\in\mathbb{N}^*} { H n } n ∈ N ∗ together with a set Γ ⊆ ∏ n ∈ N ∗ H n \Gamma\subseteq\prod_{n\in\mathbb{N}^*}\mathcal{H}_n Γ ⊆ ∏ n ∈ N ∗ H n satisfying the following properties:
1′. Γ \Gamma Γ is a complex linear subspace of ∏ n ∈ N ∗ H n \prod_{n\in\mathbb{N}^*}\mathcal{H}_n ∏ n ∈ N ∗ H n ;
2′. For every n ∈ N ∗ n\in\mathbb{N}^* n ∈ N ∗ , the set { x ( n ) : x ∈ Γ } \{x(n):x\in\Gamma\} { x ( n ) : x ∈ Γ } is dense in H n \mathcal{H}_n H n ;
3′. For every x ∈ Γ x \in\Gamma x ∈ Γ , lim n → ∞ ∥ x ( n ) ∥ H n = ∥ x ( ∞ ) ∥ H ∞ \lim_{n\to\infty}\|x(n)\|_{\mathcal{H}_n}=\|x(\infty)\|_{\mathcal{H}_\infty} lim n → ∞ ∥ x ( n ) ∥ H n = ∥ x ( ∞ ) ∥ H ∞ ;
4′. Let x ∈ ∏ n ∈ N ∗ H n x\in\prod_{n\in\mathbb{N}^*}\mathcal{H}_n x ∈ ∏ n ∈ N ∗ H n . If, for every ϵ > 0 \epsilon>0 ϵ > 0 , there exists an x ′ ∈ Γ x'\in\Gamma x ′ ∈ Γ and an N ∈ N N\in\mathbb{N} N ∈ N such that ∥ x ′ ( n ) − x ( n ) ∥ H n < ϵ \|x'(n)-x(n)\|_{\mathcal{H}_n}<\epsilon ∥ x ′ ( n ) − x ( n ) ∥ H n < ϵ for every n ∈ [ N , ∞ ] n\in[N,\infty] n ∈ [ N , ∞ ] , then x ∈ Γ x\in\Gamma x ∈ Γ .
5′. For every n ∈ N n\in\mathbb{N} n ∈ N , we have { x ( n ) : x ∈ Γ } = H n \{x(n):x\in\Gamma\}=\mathcal{H}_n { x ( n ) : x ∈ Γ } = H n . In particular, H n \mathcal{H}_n H n embeds naturally in Γ \Gamma Γ in the following sense:
H ~ n : = { { x ( k ) } k ∈ N ∗ : x ( n ) ∈ H n and x ( k ) = 0 ∀ k ≠ n } ⊆ Γ . \widetilde{\mathcal{H}}_n:=\{\{x(k)\}_{k\in\mathbb{N}^*}:x(n)\in\mathcal{H}_n\text{ and }x(k)=0\ \forall k\ne n\}\subseteq\Gamma. H n := {{ x ( k ) } k ∈ N ∗ : x ( n ) ∈ H n and x ( k ) = 0 ∀ k = n } ⊆ Γ.
Note that the above definition is equivalent to Theorem 2.2 for an HLS groupoid. Indeed, since N \mathbb{N} N is a discrete subset of N ∗ \mathbb{N}^* N ∗ , the density condition given in Item 2′ implies that Condition 4 of Theorem 2.2 is equivalent to Items 4′ and 5′. Moreover, for every ξ ∈ Γ \xi\in\Gamma ξ ∈ Γ , the supremum ∥ ξ ∥ = sup n ∈ N ∗ ∥ ξ ( n ) ∥ H n \|\xi\|=\sup_{n\in\mathbb{N}^*}\|\xi(n)\|_{\mathcal{H}_n} ∥ ξ ∥ = sup n ∈ N ∗ ∥ ξ ( n ) ∥ H n is bounded due to Item 3′. This demonstrates that the set of all bounded sections matches the set of continuous sections, i.e., Δ b = Γ \Delta_b=\Gamma Δ b = Γ .
Definition 3.5 (Theorem 2.4 simplified). Let G \mathcal{G} G be an HLS groupoid. A continuous unitary representation ϖ \varpi ϖ of G \mathcal{G} G consists of a continuous Hilbert field { H n } n ∈ N ∗ , Γ \{\mathcal{H}_n\}_{n\in\mathbb{N}^*},\Gamma { H n } n ∈ N ∗ , Γ together with, for each n ∈ N ∗ n\in\mathbb{N}^* n ∈ N ∗ , a unitary representation π n : G → U ( H n ) \pi_n:\mathcal{G}\to\mathcal{U}(\mathcal{H}_n) π n : G → U ( H n ) such that the associated matrix coefficient functions vary continuously across the field. More precisely, for every ξ , η ∈ Γ \xi,\eta\in\Gamma ξ , η ∈ Γ , the map
ω ξ , η : G → C , ω ξ , η ( n , q n ( x ) ) = ⟨ π n ( q n ( x ) ) ξ ( n ) , η ( n ) ⟩ H n \omega_{\xi,\eta}:\mathcal{G}\to\mathbb{C},\qquad\omega_{\xi,\eta}(n,q_n(x))=\langle\pi_n(q_n(x))\xi(n),\eta(n)\rangle_{\mathcal{H}_n} ω ξ , η : G → C , ω ξ , η ( n , q n ( x )) = ⟨ π n ( q n ( x )) ξ ( n ) , η ( n ) ⟩ H n
is continuous.
Using Equation (1) , the above coefficient function is continuous precisely when for each x ∈ G ∞ x\in G_\infty x ∈ G ∞ , we have
lim n → ∞ ⟨ π n ( q n ( x ) ) ξ ( n ) , η ( n ) ⟩ H n = ⟨ π ∞ ( x ) ξ ( ∞ ) , η ( ∞ ) ⟩ H ∞ . \lim_{n\to\infty}\langle\pi_n(q_n(x))\xi(n),\eta(n)\rangle_{\mathcal{H}_n}=\langle\pi_\infty(x)\xi(\infty),\eta(\infty)\rangle_{\mathcal{H}_\infty}. n → ∞ lim ⟨ π n ( q n ( x )) ξ ( n ) , η ( n ) ⟩ H n = ⟨ π ∞ ( x ) ξ ( ∞ ) , η ( ∞ ) ⟩ H ∞ .
Let Λ \Lambda Λ be the left regular representation of the HLS groupoid G \mathcal{G} G with the collection of continuous sections Γ ⊆ ∏ n ∈ N ∗ ℓ 2 ( G n ) \Gamma\subseteq\prod_{n\in\mathbb{N}^*}\ell^2(G_n) Γ ⊆ ∏ n ∈ N ∗ ℓ 2 ( G n ) . Recall that Γ \Gamma Γ is formed via the unique closure of C c ( G ) C_c(\mathcal{G}) C c ( G ) (see [8 Proposition 10.2.3] and Theorem 2.5). Every continuous section { f n } n ∈ N ∗ \{f_n\}_{n\in\mathbb{N}^*} { f n } n ∈ N ∗ can be naturally identified with a function F : G → C F:\mathcal{G}\to\mathbb{C} F : G → C as follows:
F ( n , q n ( g ) ) = f n ( q n ( g ) ) , for g ∈ G ∞ . (3) F(n,q_n(g))=f_n(q_n(g)),\quad\text{for }g\in G_\infty. \tag*{(3)} F ( n , q n ( g )) = f n ( q n ( g )) , for g ∈ G ∞ . (3)
We now provide a concrete description of the continuous sections of the left regular representation for the case of HLS groupoids, and show that the coefficient functions associated with any pair of continuous sections in Γ \Gamma Γ belong to the groupoid Fourier algebra.
Proposition 3.6. Let G \mathcal{G} G be an HLS groupoid, and Γ \Gamma Γ the set of continuous sections of the left regular representation of G \mathcal{G} G . Consider the natural identification of sections in Γ \Gamma Γ with functions on G \mathcal{G} G given in (3) . Then, we have the following.
For every { f n } n ∈ N ∗ ∈ Γ \{f_n\}_{n\in\mathbb{N}^*}\in\Gamma { f n } n ∈ N ∗ ∈ Γ , the associated function F F F is continuous on G \mathcal{G} G .
Γ = { F ∈ C ( G ) : F ∣ G ∞ ∈ ℓ 2 ( G ∞ ) , ∥ F ∣ G n ∥ 2 → ∥ F ∣ G ∞ ∥ 2 } \Gamma=\{F\in C(\mathcal{G}):F|_{G_\infty}\in\ell^2(G_\infty),\ \|F|_{G_n}\|_2\to\|F|_{G_\infty}\|_2\} Γ = { F ∈ C ( G ) : F ∣ G ∞ ∈ ℓ 2 ( G ∞ ) , ∥ F ∣ G n ∥ 2 → ∥ F ∣ G ∞ ∥ 2 } . Here we have identified G n G_n G n , for n ∈ N ∗ n\in\mathbb{N}^* n ∈ N ∗ , with its image { n } × G n \{n\}\times G_n { n } × G n in G \mathcal{G} G .
For every F , H ∈ Γ F,H\in\Gamma F , H ∈ Γ , we have Λ F , H ∈ A ( G ) \Lambda_{F,H}\in A(\mathcal{G}) Λ F , H ∈ A ( G ) .
Proof. Let { f n } n ∈ N ∗ ∈ Γ \{f_n\}_{n\in\mathbb{N}^*}\in\Gamma { f n } n ∈ N ∗ ∈ Γ be given, and fix ϵ > 0 \epsilon>0 ϵ > 0 and g ∈ G ∞ g\in G_\infty g ∈ G ∞ . As described in Theorem 2.5, Γ \Gamma Γ is the closure of the space of the fundamental sections C c ( G ) C_c(\mathcal{G}) C c ( G ) . Hence, there exists h ∈ C c ( G ) h\in C_c(\mathcal{G}) h ∈ C c ( G ) and N 1 ∈ N N_1\in\mathbb{N} N 1 ∈ N such that
∥ f m − h m ∥ 2 < ϵ 3 for m ∈ [ N 1 , ∞ ] , \|f_m-h_m\|_2<\frac{\epsilon}{3}\quad\text{for }m\in[N_1,\infty], ∥ f m − h m ∥ 2 < 3 ϵ for m ∈ [ N 1 , ∞ ] ,
where h m = h ∣ G m h_m=h|_{G_m} h m = h ∣ G m . Consequently, for every m ∈ [ N 1 , ∞ ] m\in[N_1,\infty] m ∈ [ N 1 , ∞ ] and every g ∈ G ∞ g\in G_\infty g ∈ G ∞ , we have
∣ f ( m , q m ( g ) ) − h ( m , q m ( g ) ) ∣ < ϵ 3 . |f(m,q_m(g))-h(m,q_m(g))|<\frac{\epsilon}{3}. ∣ f ( m , q m ( g )) − h ( m , q m ( g )) ∣ < 3 ϵ .
Since h ∈ C c ( G ) h\in C_c(\mathcal{G}) h ∈ C c ( G ) , there exists N 2 ∈ N N_2\in\mathbb{N} N 2 ∈ N such that ∣ h ( n , q n ( g ) ) − h ( ∞ , g ) ∣ < ϵ 3 |h(n,q_n(g))-h(\infty,g)|<\frac{\epsilon}{3} ∣ h ( n , q n ( g )) − h ( ∞ , g ) ∣ < 3 ϵ for each n ≥ N 2 n\geq N_2 n ≥ N 2 . Applying an ϵ / 3 \epsilon/3 ϵ /3 -argument, for N ≥ max { N 1 , N 2 } N\geq\max\{N_1,N_2\} N ≥ max { N 1 , N 2 } , we have ∣ f ( n , q n ( g ) ) − f ( ∞ , g ) ∣ ≤ ϵ |f(n,q_n(g))-f(\infty,g)|\leq\epsilon ∣ f ( n , q n ( g )) − f ( ∞ , g ) ∣ ≤ ϵ . This proves 1.
To prove 2, note that the inclusion ‘⊂ \subset ⊂ ’ follows directly from part 1 and the definition of a continuous Hilbert field. Conversely, suppose that F : G → C F:\mathcal{G}\to\mathbb{C} F : G → C satisfies F ∈ C ′ ( G ) F\in C'(\mathcal{G}) F ∈ C ′ ( G ) , F ∣ G ∞ ∈ ℓ 2 ( G ∞ ) F|_{G_\infty}\in\ell^2(G_\infty) F ∣ G ∞ ∈ ℓ 2 ( G ∞ ) , and ∥ F ∣ G n ∥ 2 → ∥ F ∣ G ∞ ∥ 2 \|F|_{G_n}\|_2\to\|F|_{G_\infty}\|_2 ∥ F ∣ G n ∥ 2 → ∥ F ∣ G ∞ ∥ 2 . By the criterion recalled in Theorem 2.5, it suffices to show that, for every fundamental section H ∈ C c ( G ) H\in C_c(\mathcal{G}) H ∈ C c ( G ) , we have
⟨ F ∣ G n , H ∣ G n ⟩ ℓ 2 ( G n ) ⟶ ⟨ F ∣ G ∞ , H ∣ G ∞ ⟩ ℓ 2 ( G ∞ ) . \langle F|_{G_n},H|_{G_n}\rangle_{\ell^2(G_n)}\longrightarrow\langle F|_{G_\infty},H|_{G_\infty}\rangle_{\ell^2(G_\infty)}. ⟨ F ∣ G n , H ∣ G n ⟩ ℓ 2 ( G n ) ⟶ ⟨ F ∣ G ∞ , H ∣ G ∞ ⟩ ℓ 2 ( G ∞ ) .
Since H H H is compactly supported, there exists a finite set of elements of G ∞ G_\infty G ∞ , say z 1 , … , z m z_1,\ldots,z_m z 1 , … , z m , and a sufficiently large N N N , such that
supp ( H ∣ G ∖ ⋃ i = 1 m G i ) ⊆ ⋃ i = 1 m ray N ( z i ) . (4) \operatorname{supp}(H|_{\mathcal{G}\setminus\bigcup_{i=1}^{m}G_i})\subseteq\bigcup_{i=1}^{m}\operatorname{ray}_N(z_i). \tag*{(4)} supp ( H ∣ G ∖ ⋃ i = 1 m G i ) ⊆ i = 1 ⋃ m ray N ( z i ) . (4)
So, by continuity of F F F and H H H we get:
⟨ F ∣ G ∞ , H ∣ G ∞ ⟩ ℓ 2 ( G ∞ ) = ∑ i = 1 m F ( ∞ , z i ) H ( ∞ , z i ) ‾ = ∑ i = 1 m ( lim n → ∞ F ( n , q n ( z i ) ) ) ( lim n → ∞ H ( n , q n ( z i ) ) ‾ ) = lim n → ∞ ∑ i = 1 m F ( n , q n ( z i ) ) H ( n , q n ( z i ) ) ‾ = lim n → ∞ ⟨ F ∣ G n , H ∣ G n ⟩ ℓ 2 ( G n ) , \begin{aligned}
\langle F|_{G_\infty},H|_{G_\infty}\rangle_{\ell^2(G_\infty)}
&=\sum_{i=1}^{m}F(\infty,z_i)\overline{H(\infty,z_i)}\\
&=\sum_{i=1}^{m}\left(\lim_{n\to\infty}F(n,q_n(z_i))\right)\left(\overline{\lim_{n\to\infty}H(n,q_n(z_i))}\right)\\
&=\lim_{n\to\infty}\sum_{i=1}^{m}F(n,q_n(z_i))\overline{H(n,q_n(z_i))}\\
&=\lim_{n\to\infty}\langle F|_{G_n},H|_{G_n}\rangle_{\ell^2(G_n)},
\end{aligned} ⟨ F ∣ G ∞ , H ∣ G ∞ ⟩ ℓ 2 ( G ∞ ) = i = 1 ∑ m F ( ∞ , z i ) H ( ∞ , z i ) = i = 1 ∑ m ( n → ∞ lim F ( n , q n ( z i )) ) ( n → ∞ lim H ( n , q n ( z i )) ) = n → ∞ lim i = 1 ∑ m F ( n , q n ( z i )) H ( n , q n ( z i )) = n → ∞ lim ⟨ F ∣ G n , H ∣ G n ⟩ ℓ 2 ( G n ) ,
where the last equality follows from Equation (4) .
To prove 3, we first show that C c ( G ) C_c(\mathcal{G}) C c ( G ) is ∥ ⋅ ∥ Δ \|\cdot\|_\Delta ∥ ⋅ ∥ Δ -dense in Γ \Gamma Γ . Let F ∈ Γ F\in\Gamma F ∈ Γ and ϵ > 0 \epsilon>0 ϵ > 0 . Since Γ \Gamma Γ is the (local) closure of the fundamental sections C c ( G ) C_c(\mathcal{G}) C c ( G ) , there exist K ∈ C c ( G ) K\in C_c(\mathcal{G}) K ∈ C c ( G ) and N ∈ N N\in\mathbb{N} N ∈ N such that ∥ F ∣ G n − K ∣ G n ∥ 2 < ϵ \|F|_{G_n}-K|_{G_n}\|_2<\epsilon ∥ F ∣ G n − K ∣ G n ∥ 2 < ϵ for all n ∈ [ N , ∞ ] n\in[N,\infty] n ∈ [ N , ∞ ] . Since ⋃ n < N G n \bigcup_{n<N}G_n ⋃ n < N G n is a finite set, the function
K ′ ( n , x ) : = { F ( n , x ) , n < N , K ( n , x ) , N ≤ n ≤ ∞ , K'(n,x):=
\begin{cases}
F(n,x), & n<N,\\
K(n,x), & N\leq n\leq\infty,
\end{cases} K ′ ( n , x ) := { F ( n , x ) , K ( n , x ) , n < N , N ≤ n ≤ ∞ ,
belongs to C c ( G ) C_c(\mathcal{G}) C c ( G ) , and ∥ F − K ′ ∥ Δ = sup n ≥ N ∥ F ∣ G n − K ∣ G n ∥ 2 ≤ ϵ \|F-K'\|_\Delta=\sup_{n\geq N}\|F|_{G_n}-K|_{G_n}\|_2\leq\epsilon ∥ F − K ′ ∥ Δ = sup n ≥ N ∥ F ∣ G n − K ∣ G n ∥ 2 ≤ ϵ . This proves our claim.
Now let ϵ > 0 \epsilon>0 ϵ > 0 be arbitrary, and choose F ′ , H ′ ∈ C c ( G ) F',H'\in C_c(\mathcal{G}) F ′ , H ′ ∈ C c ( G ) with ∥ F − F ′ ∥ Δ < ϵ \|F-F'\|_\Delta<\epsilon ∥ F − F ′ ∥ Δ < ϵ and ∥ H − H ′ ∥ Δ < ϵ \|H-H'\|_\Delta<\epsilon ∥ H − H ′ ∥ Δ < ϵ . Clearly, Λ F ′ , H ′ ∈ A c ( G ) \Lambda_{F',H'}\in A_c(\mathcal{G}) Λ F ′ , H ′ ∈ A c ( G ) . Also, by the discussion after Theorem 3.4, we have Γ = Δ b \Gamma=\Delta_b Γ = Δ b and Λ F , H ∈ B ( G ) \Lambda_{F,H}\in B(\mathcal{G}) Λ F , H ∈ B ( G ) . The definition of the Fourier-Stieltjes norm gives that
∥ Λ F , H − Λ F ′ , H ′ ∥ B ( G ) ≤ ∥ F − F ′ ∥ Δ ∥ H ∥ Δ + ∥ F ′ ∥ Δ ∥ H − H ′ ∥ Δ ≤ ϵ ( ∥ H ∥ Δ + ∥ F ∥ Δ + ϵ ) , \|\Lambda_{F,H}-\Lambda_{F',H'}\|_{B(\mathcal{G})}\leq\|F-F'\|_\Delta\|H\|_\Delta+\|F'\|_\Delta\|H-H'\|_\Delta\leq\epsilon(\|H\|_\Delta+\|F\|_\Delta+\epsilon), ∥ Λ F , H − Λ F ′ , H ′ ∥ B ( G ) ≤ ∥ F − F ′ ∥ Δ ∥ H ∥ Δ + ∥ F ′ ∥ Δ ∥ H − H ′ ∥ Δ ≤ ϵ ( ∥ H ∥ Δ + ∥ F ∥ Δ + ϵ ) ,
which implies that Λ F , H ∈ A c ( G ) ‾ ∥ ⋅ ∥ B ( G ) = A ( G ) . □ \Lambda_{F,H}\in\overline{A_c(\mathcal{G})}^{\|\cdot\|_{B(\mathcal{G})}}=A(\mathcal{G}). \quad\Box Λ F , H ∈ A c ( G ) ∥ ⋅ ∥ B ( G ) = A ( G ) . □
Recall that for any étale groupoid G \mathcal{G} G , we have A c ( G ) = C c ( G ) A_c(\mathcal{G})=C_c(\mathcal{G}) A c ( G ) = C c ( G ) and A ( G ) = B ( G ) ∩ C c ( G ) ∥ ⋅ ∥ B ( G ) A(\mathcal{G})=B(\mathcal{G})\cap C_c(\mathcal{G})^{\|\cdot\|_{B(\mathcal{G})}} A ( G ) = B ( G ) ∩ C c ( G ) ∥ ⋅ ∥ B ( G ) (Theorem 2.9). When G \mathcal{G} G is an HLS groupoid, the representation of f ∈ C c ( G ) f\in C_c(\mathcal{G}) f ∈ C c ( G ) as a coefficient function in A c ( G ) A_c(\mathcal{G}) A c ( G ) takes a simpler form. For example, consider the case where f ∈ C c ( G ) f\in C_c(\mathcal{G}) f ∈ C c ( G ) satisfies supp ( f ) ⊆ 1 ray n ( x ) \operatorname{supp}(f)\subseteq\mathbf{1}_{\operatorname{ray}_n(x)} supp ( f ) ⊆ 1 ray n ( x ) for some x ∈ G ∞ x \in G_{\infty} x ∈ G ∞ and n ∈ N n \in\mathbb{N} n ∈ N . By Theorem 3.3 Item 1, the function g = 1 ray n ( e G ∞ ) g = \mathbf{1}_{\operatorname{ray}_{n}(e_{G_{\infty}})} g = 1 ray n ( e G ∞ ) belongs to C c ( G ) C_{c}(G) C c ( G ) .
Then, for m ∈ N ∗ m \in\mathbb{N}^{*} m ∈ N ∗ and z ∈ G ∞ z \in G_{\infty} z ∈ G ∞ , we have
Λ g , f ‾ ( q n , q m ( z ) ) = ∑ q m ( y ) ∈ G m g ( m , q m ( z − 1 y ) ) f ( m , q m ( y ) ) = { f ( m , q m ( z ) ) if m ≥ n , 0 if m < n = f ( m , q m ( z ) ) . \Lambda_{g,\overline{f}}(q_{n},q_{m}(z))=\sum_{q_{m}(y)\in G_{m}}g(m,q_{m}(z^{-1}y))f(m,q_{m}(y))=
\begin{cases}
f(m,q_{m}(z)) & \text{if }m\geq n,\\
0 & \text{if }m<n
\end{cases}
=f(m,q_{m}(z)). Λ g , f ( q n , q m ( z )) = q m ( y ) ∈ G m ∑ g ( m , q m ( z − 1 y )) f ( m , q m ( y )) = { f ( m , q m ( z )) 0 if m ≥ n , if m < n = f ( m , q m ( z )) .
Letting f = 1 ray n ( e G ∞ ) f=\mathbf{1}_{\operatorname{ray}_{n}(e_{G_{\infty}})} f = 1 ray n ( e G ∞ ) in the above discussion, we get the following easy corollary.
Corollary 3.7. For every x ∈ G ∞ x \in G_{\infty} x ∈ G ∞ and n ∈ N n \in\mathbb{N} n ∈ N , the characteristic function 1 ray n ( e G ∞ ) \mathbf{1}_{\operatorname{ray}_{n}(e_{G_{\infty}})} 1 ray n ( e G ∞ ) belongs to A ( G ) A(G) A ( G ) . In fact, we have Λ 1 ray n ( e G ∞ ) , 1 ray n ( e G ∞ ) = 1 ray n ( e G ∞ ) \Lambda_{\mathbf{1}_{\operatorname{ray}_{n}(e_{G_{\infty}})},\mathbf{1}_{\operatorname{ray}_{n}(e_{G_{\infty}})}=\mathbf{1}_{\operatorname{ray}_{n}(e_{G_{\infty}})}} Λ 1 ray n ( e G ∞ ) , 1 ray n ( e G ∞ ) = 1 ray n ( e G ∞ ) .
Remark 3.8. Note that restriction-type results depend on the subobject being a group rather than a groupoid. In particular, our results cannot be extended to restriction maps onto arbitrary subgroupoids of a given groupoid. Indeed, let G \mathcal{G} G be an HLS groupoid, and set G N : = G ∖ G ∞ \mathcal{G}_{N}:=\mathcal{G}\setminus G_{\infty} G N := G ∖ G ∞ . By Theorem 3.7, for x ∈ G x\in\mathcal{G} x ∈ G and n ∈ N n\in\mathbb{N} n ∈ N , we have 1 ray n ( x ) ∈ A ( G ) \mathbf{1}_{\operatorname{ray}_{n}}(x)\in A(\mathcal{G}) 1 ray n ( x ) ∈ A ( G ) . However,
1 ray n ( x ) ∣ G N ∉ A ( G N ) . \mathbf{1}_{\operatorname{ray}_{n}}(x)|_{\mathcal{G}_{N}}\notin A(\mathcal{G}_{N}). 1 ray n ( x ) ∣ G N ∈ / A ( G N ) .
To see this, note that G N \mathcal{G}_{N} G N is an étale groupoid, and hence A ( G N ) ⊆ C 0 ( G N ) A(\mathcal{G}_{N})\subseteq C_{0}(\mathcal{G}_{N}) A ( G N ) ⊆ C 0 ( G N ) . Consequently, the restriction map A ( G ) → A ( G N ) A(\mathcal{G})\to A(\mathcal{G}_{N}) A ( G ) → A ( G N ) is not well defined.
We end this section by establishing several results on the norms of the restricted functions.
Proposition 3.9. Let G = ⨆ n ∈ N ∗ G n \mathcal{G}=\bigsqcup_{n\in\mathbb{N}^{*}}G_{n} G = ⨆ n ∈ N ∗ G n be an HLS groupoid, and for each n ∈ N ∗ n\in\mathbb{N}^{*} n ∈ N ∗ , let R n : A ( G ) → A ( G n ) R_{n}:A(\mathcal{G})\to A(G_{n}) R n : A ( G ) → A ( G n ) denote the restriction map. Let u ∈ A ( G ) u\in A(\mathcal{G}) u ∈ A ( G ) .
sup n ∈ N ∗ ∥ R n ( u ) ∥ A ( G n ) ≤ ∥ u ∥ A ( G ) \sup_{n\in\mathbb{N}^{*}}\|R_{n}(u)\|_{A(G_{n})}\leq\|u\|_{A(\mathcal{G})} sup n ∈ N ∗ ∥ R n ( u ) ∥ A ( G n ) ≤ ∥ u ∥ A ( G ) .
∥ R ∞ ( u ) ∥ A ( G ∞ ) ≤ lim inf n ∥ R n ( u ) ∥ A ( G n ) \|R_{\infty}(u)\|_{A(G_{\infty})}\leq\liminf_{n}\|R_{n}(u)\|_{A(G_{n})} ∥ R ∞ ( u ) ∥ A ( G ∞ ) ≤ lim inf n ∥ R n ( u ) ∥ A ( G n ) .
Suppose u ∈ A ( G ) ∩ C c ( G ) u\in A(\mathcal{G})\cap C_{c}(\mathcal{G}) u ∈ A ( G ) ∩ C c ( G ) . If R ∞ ( u ) = 0 R_{\infty}(u)=0 R ∞ ( u ) = 0 then lim n → ∞ ∥ R n ( u ) ∥ A ( G n ) = 0 \lim_{n\to\infty}\|R_{n}(u)\|_{A(G_{n})}=0 lim n → ∞ ∥ R n ( u ) ∥ A ( G n ) = 0 .
Proof. Item 1 follows from [7 Proposition 3.4(iii)] , as each restriction map R n R_{n} R n is contractive. To prove Item 2, we view each R n ( u ) R_{n}(u) R n ( u ) as a function on G ∞ G_{\infty} G ∞ . Namely, for n ∈ N n\in\mathbb{N} n ∈ N , define
u ~ n : G ∞ → C , u ~ n ( x ) = u ( n , q n ( x ) ) = R n ( u ) ( q n ( x ) ) . \widetilde{u}_{n}:G_{\infty}\to\mathbb{C},\qquad\widetilde{u}_{n}(x)=u(n,q_{n}(x))=R_{n}(u)(q_{n}(x)). u n : G ∞ → C , u n ( x ) = u ( n , q n ( x )) = R n ( u ) ( q n ( x )) .
Claim 3.10. We have u ~ n ∈ B ( G ∞ ) \widetilde{u}_{n}\in B(G_{\infty}) u n ∈ B ( G ∞ ) and ∥ u ~ n ∥ B ( G ∞ ) ≤ ∥ R n ( u ) ∥ A ( G n ) \|\widetilde{u}_{n}\|_{B(G_{\infty})}\leq\|R_{n}(u)\|_{A(G_{n})} ∥ u n ∥ B ( G ∞ ) ≤ ∥ R n ( u ) ∥ A ( G n ) . Consequently, { u ~ n } n ≥ N \{\widetilde{u}_{n}\}_{n\geq N} { u n } n ≥ N is a bounded sequence in B ( G ∞ ) B(G_{\infty}) B ( G ∞ ) .
Proof of claim. Let R n ( u ) = ⟨ λ G n ( ⋅ ) ξ n , η n ⟩ R_{n}(u)=\langle\lambda_{G_{n}}(\cdot)\xi_{n},\eta_{n}\rangle R n ( u ) = ⟨ λ G n ( ⋅ ) ξ n , η n ⟩ be a representation of R n ( u ) R_{n}(u) R n ( u ) as a coefficient function of the left regular representation of G n G_{n} G n with ξ n , η n ∈ ℓ 2 ( G n ) \xi_{n},\eta_{n}\in\ell^{2}(G_{n}) ξ n , η n ∈ ℓ 2 ( G n ) and ∥ R n ( u ) ∥ A ( G n ) = ∥ ξ n ∥ 2 ∥ η n ∥ 2 \|R_{n}(u)\|_{A(G_{n})}=\|\xi_{n}\|_{2}\|\eta_{n}\|_{2} ∥ R n ( u ) ∥ A ( G n ) = ∥ ξ n ∥ 2 ∥ η n ∥ 2 . It is easy to see that λ G n ∘ q n \lambda_{G_{n}}\circ q_{n} λ G n ∘ q n is a unitary representation of G ∞ G_{\infty} G ∞ , and for every x ∈ G ∞ x\in G_{\infty} x ∈ G ∞ , u ~ n ( x ) = ⟨ ( λ G n ∘ q n ) ( x ) ξ n , η n ⟩ \widetilde{u}_{n}(x)=\langle(\lambda_{G_{n}}\circ q_{n})(x)\xi_{n},\eta_{n}\rangle u n ( x ) = ⟨( λ G n ∘ q n ) ( x ) ξ n , η n ⟩ . So u ~ n ∈ B ( G ∞ ) \widetilde{u}_{n}\in B(G_{\infty}) u n ∈ B ( G ∞ ) . Moreover,
∥ u ~ n ∥ B ( G ∞ ) ≤ ∥ ξ n ∥ 2 ∥ η n ∥ 2 = ∥ R n ( u ) ∥ A ( G n ) . \|\widetilde{u}_{n}\|_{B(G_{\infty})}\leq\|\xi_{n}\|_{2}\|\eta_{n}\|_{2}=\|R_{n}(u)\|_{A(G_{n})}. ∥ u n ∥ B ( G ∞ ) ≤ ∥ ξ n ∥ 2 ∥ η n ∥ 2 = ∥ R n ( u ) ∥ A ( G n ) .
This finishes the proof of the claim, since we have already proved that the sequence { ∥ R n ( u ) ∥ A ( G n ) } n \{\|R_{n}(u)\|_{A(G_{n})}\}_{n} { ∥ R n ( u ) ∥ A ( G n ) } n is bounded above.
As u u u is a continuous function on G \mathcal{G} G , using (1) , we get u ~ n → R ∞ ( u ) \widetilde{u}_{n}\to R_{\infty}(u) u n → R ∞ ( u ) pointwise. By Eymard’s bounded convergence theorem for B ( G ∞ ) B(G_{\infty}) B ( G ∞ ) , the sequence { u ~ n } n ∈ N \{\widetilde{u}_{n}\}_{n\in\mathbb{N}} { u n } n ∈ N converges to R ∞ ( u ) R_{\infty}(u) R ∞ ( u ) in the weak* topology induced by C ∗ ( G ∞ ) C^{*}(G_{\infty}) C ∗ ( G ∞ ) . Since the norm is weak* lower semi-continuous, we have
∥ R ∞ ( u ) ∥ A ( G ∞ ) ≤ lim inf n ∥ u ~ n ∥ B ( G ∞ ) ≤ lim inf n ∥ R n ( u ) ∥ A ( G n ) . (5) \|R_{\infty}(u)\|_{A(G_{\infty})}\leq\liminf_{n}\|\widetilde{u}_{n}\|_{B(G_{\infty})}\leq\liminf_{n}\|R_{n}(u)\|_{A(G_{n})}. \tag*{(5)} ∥ R ∞ ( u ) ∥ A ( G ∞ ) ≤ n lim inf ∥ u n ∥ B ( G ∞ ) ≤ n lim inf ∥ R n ( u ) ∥ A ( G n ) . (5)
where we used Theorem 3.10 in the last inequality. This proves Item 2.
To prove Item 3, note that since u ∈ C c ( G ) u \in C_c(\mathcal{G}) u ∈ C c ( G ) , there are finitely many points of the form ( ∞ , x 1 ) , … , ( ∞ , x m ) (\infty,x_1),\ldots,(\infty,x_m) ( ∞ , x 1 ) , … , ( ∞ , x m ) contained in supp ( u ) \operatorname{supp}(u) supp ( u ) . Moreover, applying Theorem 3.3, Items 2 and 3, for sufficiently large N N N , we have
supp ( u ∣ G ∖ ⋃ i = 1 N G i ) ⊆ ⋃ i = 1 m ray N ( x i ) . (6) \operatorname{supp}\left(u\big|_{\mathcal{G}\setminus\bigcup_{i=1}^{N}G_i}\right)\subseteq\bigcup_{i=1}^{m}\operatorname{ray}_{N}(x_i). \tag*{(6)} supp ( u G ∖ ⋃ i = 1 N G i ) ⊆ i = 1 ⋃ m ray N ( x i ) . (6)
where ⊔ \sqcup ⊔ denotes the union of disjoint sets. So for n ∈ [ N , ∞ ) n\in[N,\infty) n ∈ [ N , ∞ ) , each function R n ( u ) R_n(u) R n ( u ) is supported in { ( n , q n ( x 1 ) ) , … , ( n , q n ( x m ) ) } \{(n,q_n(x_1)),\ldots,(n,q_n(x_m))\} {( n , q n ( x 1 )) , … , ( n , q n ( x m ))} . On the other hand, since u u u is continuous on G \mathcal{G} G , for every x ∈ G ∞ x\in G_\infty x ∈ G ∞ we have
u ( n , q n ( x ) ) ⟶ u ( ∞ , x ) as n → ∞ . u(n,q_n(x))\longrightarrow u(\infty,x)\quad\text{as }n\to\infty. u ( n , q n ( x )) ⟶ u ( ∞ , x ) as n → ∞.
The above convergence holds simultaneously for x 1 , … , x m x_1,\ldots,x_m x 1 , … , x m . So, ∥ R n ( u ) ∥ ∞ → 0 \lVert R_n(u)\rVert_\infty\to0 ∥ R n ( u ) ∥ ∞ → 0 as n → ∞ n\to\infty n → ∞ . Thus, for n ≥ N n\geq N n ≥ N we have
∥ R n ( u ) ∥ A ( G n ) = ∥ ∑ i = 1 m u ( n , q n ( x i ) ) δ q n ( x i ) ∥ A ( G n ) ≤ ∑ i = 1 m ∣ u ( n , q n ( x i ) ) ∣ ≤ m ∥ R n ( u ) ∥ s u p , \lVert R_n(u)\rVert_{A(G_n)}
=\left\lVert\sum_{i=1}^{m}u(n,q_n(x_i))\delta_{q_n(x_i)}\right\rVert_{A(G_n)}
\leq\sum_{i=1}^{m}\left|u(n,q_n(x_i))\right|
\leq m\lVert R_n(u)\rVert_{\mathrm{sup}}, ∥ R n ( u ) ∥ A ( G n ) = i = 1 ∑ m u ( n , q n ( x i )) δ q n ( x i ) A ( G n ) ≤ i = 1 ∑ m ∣ u ( n , q n ( x i )) ∣ ≤ m ∥ R n ( u ) ∥ sup ,
where we used the fact that the Fourier norm of a point-mass function is bounded above by 1. So, ∥ R n ( u ) ∥ A ( G n ) → 0 \lVert R_n(u)\rVert_{A(G_n)}\to0 ∥ R n ( u ) ∥ A ( G n ) → 0 as desired. □ \square □
Functorial properties of the Fourier algebras of HLS groupoids
In this section, we exploit the particular structure of HLS groupoids to strengthen general functorial results for the Fourier algebra of étale groupoids in the setting of HLS groupoids. Namely, for each n ∈ N ∗ n\in\mathbb{N}^* n ∈ N ∗ , we construct explicit norm-preserving lifts for the restriction map R n : A ( G ) → A ( G n ) R_n:A(\mathcal{G})\to A(G_n) R n : A ( G ) → A ( G n ) (Theorem 4.2 ).
When n ∈ N n\in\mathbb{N} n ∈ N , there is a natural embedding from each A ( G n ) A(G_n) A ( G n ) into A ( G ) A(\mathcal{G}) A ( G ) which also maintains the norm.
Lemma 4.1. Let n ∈ N n\in\mathbb{N} n ∈ N . Define ι n : A ( G n ) → A ( G ) \iota_n:A(G_n)\to A(\mathcal{G}) ι n : A ( G n ) → A ( G ) by
( ι n u ) ( m , q m ( x ) ) : = { u ( q n ( x ) ) m = n , 0 m ≠ n . (\iota_nu)(m,q_m(x)):=\begin{cases}
u(q_n(x)) & m=n,\\
0 & m\ne n.
\end{cases} ( ι n u ) ( m , q m ( x )) := { u ( q n ( x )) 0 m = n , m = n . Then, ι n \iota_n ι n is an algebra homomorphism, which is also an isometry. Moreover, ι n ( A ( G n ) ) \iota_n(A(G_n)) ι n ( A ( G n )) is an ideal in A ( G ) A(\mathcal{G}) A ( G ) .
Proof. Clearly ι n \iota_n ι n is both linear and multiplicative. If u ∈ A ( G n ) u\in A(G_n) u ∈ A ( G n ) , then we may choose f 1 , f 2 ∈ ℓ 2 ( G n ) f_1,f_2\in\ell^2(G_n) f 1 , f 2 ∈ ℓ 2 ( G n ) such that ∥ u ∥ A ( G n ) = ∥ f 1 ∥ 2 ∥ f 2 ∥ 2 \lVert u\rVert_{A(G_n)}=\lVert f_1\rVert_2\lVert f_2\rVert_2 ∥ u ∥ A ( G n ) = ∥ f 1 ∥ 2 ∥ f 2 ∥ 2 . Moreover, by defining F i : G → C F_i:\mathcal{G}\to\mathbb{C} F i : G → C via F i ( n , q n ( x ) ) = f i ( q n ( x ) ) F_i(n,q_n(x))=f_i(q_n(x)) F i ( n , q n ( x )) = f i ( q n ( x )) and F i ( m , q m ( x ) ) = 0 F_i(m,q_m(x))=0 F i ( m , q m ( x )) = 0 if m ≠ n m\ne n m = n , we see that F i ∈ C c ( G ) F_i\in C_c(\mathcal{G}) F i ∈ C c ( G ) and thus defines continuous and bounded sections of Λ \Lambda Λ . The norm ∥ F i ∥ Δ \lVert F_i\rVert_\Delta ∥ F i ∥ Δ is precisely ∥ f i ∥ 2 \lVert f_i\rVert_2 ∥ f i ∥ 2 , and so we have
∥ ι n ( u ) ∥ A ( G ) = ∥ Λ F 1 , F 2 ∥ A ( G ) ≤ ∥ F 1 ∥ Δ ∥ F 2 ∥ Δ = ∥ f 1 ∥ 2 ∥ f 2 ∥ 2 = ∥ u ∥ A ( G n ) . \lVert\iota_n(u)\rVert_{A(\mathcal{G})}
=\lVert\Lambda_{F_1,F_2}\rVert_{A(\mathcal{G})}
\leq\lVert F_1\rVert_\Delta\lVert F_2\rVert_\Delta
=\lVert f_1\rVert_2\lVert f_2\rVert_2
=\lVert u\rVert_{A(G_n)}. ∥ ι n ( u ) ∥ A ( G ) = ∥ Λ F 1 , F 2 ∥ A ( G ) ≤ ∥ F 1 ∥ Δ ∥ F 2 ∥ Δ = ∥ f 1 ∥ 2 ∥ f 2 ∥ 2 = ∥ u ∥ A ( G n ) . By [7 Proposition 3.4] , the restriction map R n R_n R n is contractive, and so
∥ u ∥ A ( G n ) = ∥ R n ( ι n ( u ) ) ∥ A ( G n ) ≤ ∥ ι n ( u ) ∥ A ( G ) \lVert u\rVert_{A(G_n)}
=\lVert R_n(\iota_n(u))\rVert_{A(G_n)}
\leq\lVert\iota_n(u)\rVert_{A(\mathcal{G})} ∥ u ∥ A ( G n ) = ∥ R n ( ι n ( u )) ∥ A ( G n ) ≤ ∥ ι n ( u ) ∥ A ( G ) demonstrating ι n \iota_n ι n is an isometry.
Let u ∈ A ( G n ) u\in A(G_n) u ∈ A ( G n ) and v ∈ A ( G ) v\in A(\mathcal{G}) v ∈ A ( G ) be arbitrary. We want to show that ι n ( u ) v ∈ ι n ( A ( G n ) ) \iota_n(u)v\in\iota_n(A(G_n)) ι n ( u ) v ∈ ι n ( A ( G n )) . For every x ∈ G ∞ x\in G_\infty x ∈ G ∞ , we have
( ι n ( u ) v ) ( m , q m ( x ) ) = { 0 m ≠ n , u ( q n ( x ) ) v ( n , q n ( x ) ) m = n , (\iota_n(u)v)(m,q_m(x))=\begin{cases}
0 & m\ne n,\\
u(q_n(x))v(n,q_n(x)) & m=n,
\end{cases} ( ι n ( u ) v ) ( m , q m ( x )) = { 0 u ( q n ( x )) v ( n , q n ( x )) m = n , m = n , which gives ι n ( u ) v = ι n ( u R n ( v ) ) \iota_n(u)v=\iota_n(uR_n(v)) ι n ( u ) v = ι n ( u R n ( v )) , which finishes the proof. □ \square □
In view of the above lemma, we will regard each ι n \iota_n ι n as an embedding of A ( G n ) A(G_n) A ( G n ) into A ( G ) A(\mathcal{G}) A ( G ) .
Theorem 4.2. Let G \mathcal{G} G be the HLS groupoid associated to the pair ( G , { K n } n ∈ N ) (G,\{K_n\}_{n\in\mathbb{N}}) ( G , { K n } n ∈ N ) . Then for each n ∈ N ∪ { ∞ } n\in\mathbb{N}\cup\{\infty\} n ∈ N ∪ { ∞ } and u ∈ A ( G n ) u\in A(G_n) u ∈ A ( G n ) , there exists v ∈ A ( G ) v\in A(\mathcal{G}) v ∈ A ( G ) such that R n ( v ) = u R_n(v)=u R n ( v ) = u and ∥ v ∥ A ( G ) = ∥ u ∥ A ( G n ) \|v\|_{A(\mathcal{G})}=\|u\|_{A(G_n)} ∥ v ∥ A ( G ) = ∥ u ∥ A ( G n ) . In particular, the map R n : A ( G ) → A ( G n ) R_n:A(\mathcal{G})\to A(G_n) R n : A ( G ) → A ( G n ) is a quotient map.
Proof. Recall that we set G ∞ : = G G_\infty:=G G ∞ := G , and denote by q n : G → G / K n q_n:G\to G/K_n q n : G → G / K n the canonical quotient map. If n < ∞ n<\infty n < ∞ , the restriction map R n R_n R n is a quotient map, since any f ∈ ℓ 2 ( G n ) f\in\ell^2(G_n) f ∈ ℓ 2 ( G n ) embeds isometrically as an element of C c ( G ) ⊆ Γ C_c(G)\subseteq\Gamma C c ( G ) ⊆ Γ as described in the proof of Theorem 4.1. So, we will only focus on the map R ∞ R_\infty R ∞ and show that every L f , g ∈ A ( G ∞ ) L_{f,g}\in A(G_\infty) L f , g ∈ A ( G ∞ ) with f , g ∈ ℓ 2 ( G ∞ ) f,g\in\ell^2(G_\infty) f , g ∈ ℓ 2 ( G ∞ ) has a pre-image under R ∞ R_\infty R ∞ whose norm is equal to ∥ L f , g ∥ A ( G ∞ ) \|L_{f,g}\|_{A(G_\infty)} ∥ L f , g ∥ A ( G ∞ ) .
To begin, fix f ∈ ℓ 2 ( G ∞ ) f\in\ell^2(G_\infty) f ∈ ℓ 2 ( G ∞ ) , and assume that f ≥ 0 f\geq0 f ≥ 0 pointwise. For n ∈ N n\in\mathbb{N} n ∈ N define the following functions:
f n : G / K n ⟶ C f n ( q n ( x ) ) : = ∥ f ∣ x K n ∥ 2 , x ∈ G (7) \begin{aligned}
f_n:G/K_n&\longrightarrow\mathbb{C}\\
f_n(q_n(x))&:=\|f|_{xK_n}\|_2,\qquad x\in G \tag*{(7)}
\end{aligned} f n : G / K n f n ( q n ( x )) ⟶ C := ∥ f ∣ x K n ∥ 2 , x ∈ G (7)
Next define
F : G ⟶ C F ( n , q n ( x ) ) : = { f n ( q n ( x ) ) n < ∞ f ( x ) n = ∞ (8) \begin{aligned}
F:\mathcal{G}&\longrightarrow\mathbb{C}\\
F(n,q_n(x))&:=\begin{cases}
f_n(q_n(x)) & n<\infty\\
f(x) & n=\infty
\end{cases} \tag*{(8)}
\end{aligned} F : G F ( n , q n ( x )) ⟶ C := { f n ( q n ( x )) f ( x ) n < ∞ n = ∞ (8)
Claim 4.3. Suppose f ∈ ℓ 2 ( G ∞ ) f\in\ell^2(G_\infty) f ∈ ℓ 2 ( G ∞ ) is pointwise non-negative. If supp ( f ) \operatorname{supp}(f) supp ( f ) is compact, then supp ( F ) \operatorname{supp}(F) supp ( F ) is compact.
Proof of Theorem 4.3: Since supp ( f ) ⊆ G ∞ \operatorname{supp}(f)\subseteq G_\infty supp ( f ) ⊆ G ∞ is compact and G ∞ G_\infty G ∞ is discrete, supp ( f ) = { z 1 , … , z m } \operatorname{supp}(f)=\{z_1,\ldots,z_m\} supp ( f ) = { z 1 , … , z m } for some m ∈ N m\in\mathbb{N} m ∈ N . Suppose for n ∈ N n\in\mathbb{N} n ∈ N , F ( n , q n ( x ) ) > 0 F(n,q_n(x))>0 F ( n , q n ( x )) > 0 . We then have ∥ f ∣ x K n ∥ 2 > 0 \|f|_{xK_n}\|_2>0 ∥ f ∣ x K n ∥ 2 > 0 and so x K n ∩ supp ( f ) ≠ ∅ xK_n\cap\operatorname{supp}(f)\ne\varnothing x K n ∩ supp ( f ) = ∅ . From this we see q n ( x ) = q n ( z i ) q_n(x)=q_n(z_i) q n ( x ) = q n ( z i ) for some z i ∈ supp ( f ) z_i\in\operatorname{supp}(f) z i ∈ supp ( f ) , which implies ( n , q n ( x ) ) ∈ ran ( z i ) (n,q_n(x))\in\operatorname{ran}(z_i) ( n , q n ( x )) ∈ ran ( z i ) . Therefore, supp ( F ) ⊆ ⋃ i = 1 m ran ( z i ) \operatorname{supp}(F)\subseteq\bigcup_{i=1}^m\operatorname{ran}(z_i) supp ( F ) ⊆ ⋃ i = 1 m ran ( z i ) , and F F F is compactly supported by Theorem 3.3 Item 2.
Claim 4.4. Suppose f ∈ ℓ 2 ( G ∞ ) f\in\ell^2(G_\infty) f ∈ ℓ 2 ( G ∞ ) is pointwise non-negative. Then F F F is continuous on G \mathcal{G} G , that is, for all x ∈ G ∞ x\in G_\infty x ∈ G ∞ ,
lim n → ∞ F ( n , q n ( x ) ) = f ( x ) . \lim_{n\to\infty}F(n,q_n(x))=f(x). n → ∞ lim F ( n , q n ( x )) = f ( x ) .
Proof of Theorem 4.4: Let x ∈ G ∞ x\in G_\infty x ∈ G ∞ and ϵ > 0 \epsilon>0 ϵ > 0 be arbitrary. First note that
F ( n , q n ( x ) ) 2 = f n ( q n ( x ) ) 2 = ∥ f ∣ x K n ∥ 2 2 = ∑ z ∈ K n ∣ f ( x z ) ∣ 2 = ∣ f ( x ) ∣ 2 + ∑ z ∈ K n ∖ { e G ∞ } ∣ f ( x z ) ∣ 2 . (9) F(n,q_n(x))^2=f_n(q_n(x))^2=\|f|_{xK_n}\|_2^2=\sum_{z\in K_n}|f(xz)|^2=|f(x)|^2+\sum_{z\in K_n\setminus\{e_{G_\infty}\}}|f(xz)|^2. \tag*{(9)} F ( n , q n ( x ) ) 2 = f n ( q n ( x ) ) 2 = ∥ f ∣ x K n ∥ 2 2 = z ∈ K n ∑ ∣ f ( x z ) ∣ 2 = ∣ f ( x ) ∣ 2 + z ∈ K n ∖ { e G ∞ } ∑ ∣ f ( x z ) ∣ 2 . (9)
Since ∥ f ∥ 2 2 < ∞ \|f\|_2^2<\infty ∥ f ∥ 2 2 < ∞ , there exists a finite set E ⊆ G ∞ E\subseteq G_\infty E ⊆ G ∞ such that ∑ z ∈ G ∞ ∖ ( E ∪ { x } ) ∣ f ( z ) ∣ 2 < ϵ \sum_{z\in G_\infty\setminus(E\cup\{x\})}|f(z)|^2<\epsilon ∑ z ∈ G ∞ ∖ ( E ∪ { x }) ∣ f ( z ) ∣ 2 < ϵ . This implies that, for every x x x ,
∑ y ∈ K n ∖ ( x − 1 E ∪ { e G ∞ } ) ∣ f ( x y ) ∣ 2 = ∑ z ∈ x K n ∖ ( E ∪ { x } ) ∣ f ( z ) ∣ 2 ≤ ∑ z ∈ G ∞ ∖ ( E ∪ { x } ) ∣ f ( z ) ∣ 2 < ϵ . \sum_{y\in K_n\setminus(x^{-1}E\cup\{e_{G_\infty}\})}|f(xy)|^2
=\sum_{z\in xK_n\setminus(E\cup\{x\})}|f(z)|^2
\leq\sum_{z\in G_\infty\setminus(E\cup\{x\})}|f(z)|^2<\epsilon. y ∈ K n ∖ ( x − 1 E ∪ { e G ∞ }) ∑ ∣ f ( x y ) ∣ 2 = z ∈ x K n ∖ ( E ∪ { x }) ∑ ∣ f ( z ) ∣ 2 ≤ z ∈ G ∞ ∖ ( E ∪ { x }) ∑ ∣ f ( z ) ∣ 2 < ϵ .
Without loss of generality, assume x ∉ E x\notin E x ∈ / E . Let x − 1 E = { y 1 , … , y m } x^{-1}E=\{y_1,\ldots,y_m\} x − 1 E = { y 1 , … , y m } . As { K n } n ∈ N \{K_n\}_{n\in\mathbb{N}} { K n } n ∈ N is a nested sequence, ⋂ n ∈ N K n = { e G ∞ } \bigcap_{n\in\mathbb{N}}K_n=\{e_{G_\infty}\} ⋂ n ∈ N K n = { e G ∞ } , and e G ∞ ∉ x − 1 E e_{G_\infty}\notin x^{-1}E e G ∞ ∈ / x − 1 E , there exists N N N such that for all n ≥ N n\geq N n ≥ N , y i ∉ K n y_i\notin K_n y i ∈ / K n for 1 ≤ i ≤ m 1\leq i\leq m 1 ≤ i ≤ m . So, x − 1 E ∩ K n = ∅ x^{-1}E\cap K_n=\varnothing x − 1 E ∩ K n = ∅ for n ≥ N n\geq N n ≥ N , and
∑ z ∈ K n ∖ { e G ∞ } ∣ f ( x z ) ∣ 2 = ∑ z ∈ K n ∖ ( x − 1 E ∪ { e G ∞ } ) ∣ f ( x z ) ∣ 2 < ϵ . (10) \sum_{z\in K_n\setminus\{e_{G_\infty}\}}|f(xz)|^2
=\sum_{z\in K_n\setminus(x^{-1}E\cup\{e_{G_\infty}\})}|f(xz)|^2<\epsilon. \tag*{(10)} z ∈ K n ∖ { e G ∞ } ∑ ∣ f ( x z ) ∣ 2 = z ∈ K n ∖ ( x − 1 E ∪ { e G ∞ }) ∑ ∣ f ( x z ) ∣ 2 < ϵ . (10)
Putting Equation (9) and Equation (10) together, we have f ( x ) 2 ≤ F ( n , q n ( x ) ) 2 ≤ f ( x ) 2 + ϵ f(x)^2 \leq F(n,q_n(x))^2 \leq f(x)^2+\epsilon f ( x ) 2 ≤ F ( n , q n ( x ) ) 2 ≤ f ( x ) 2 + ϵ . This implies that lim n → ∞ F 2 ( n , q n ( x ) ) = f 2 ( x ) \lim_{n\to\infty}F^2(n,q_n(x))=f^2(x) lim n → ∞ F 2 ( n , q n ( x )) = f 2 ( x ) . Since f f f was assumed to be pointwise positive, we conclude that F ( n , q n ( x ) ) → F ( ∞ , x ) = f ( x ) F(n,q_n(x))\to F(\infty,x)=f(x) F ( n , q n ( x )) → F ( ∞ , x ) = f ( x ) as n → ∞ n\to\infty n → ∞ .
Note that given an arbitrary f ∈ ℓ 2 ( G ∞ ) f\in\ell^2(G_\infty) f ∈ ℓ 2 ( G ∞ ) , we may decompose f f f into f = f 1 − f 2 + i ( f 3 − f 4 ) f=f_1-f_2+i(f_3-f_4) f = f 1 − f 2 + i ( f 3 − f 4 ) where f i ≥ 0 f_i\geq0 f i ≥ 0 pointwise for 1 ≤ i ≤ 4 1\leq i\leq4 1 ≤ i ≤ 4 by taking the positive and negative parts of the real and imaginary parts of f f f . For each f i f_i f i , construct F i F_i F i as in Equation (8) and define
F : G → C , F = F 1 − F 2 + i ( F 3 − F 4 ) . (11) F:\mathcal{G}\to\mathbb{C},\qquad F=F_1-F_2+i(F_3-F_4). \tag*{(11)} F : G → C , F = F 1 − F 2 + i ( F 3 − F 4 ) . (11)
Since each F i F_i F i is continuous (by Theorem 4.4), the function F F F is continuous. Applying Theorem 4.3, if we take f ∈ C c ( G ∞ ) f\in C_c(G_\infty) f ∈ C c ( G ∞ ) then F ∈ C c ( G ) F\in C_c(\mathcal{G}) F ∈ C c ( G ) as well.
Claim 4.5. For every f ∈ ℓ 2 ( G ∞ ) f\in\ell^2(G_\infty) f ∈ ℓ 2 ( G ∞ ) , the associated function F F F is a continuous section of the left regular bundle Λ \Lambda Λ .
Proof of Theorem 4.5: Let F F F be the function on G \mathcal{G} G associated with f ∈ ℓ 2 ( G ∞ ) f\in\ell^2(G_\infty) f ∈ ℓ 2 ( G ∞ ) . The function F ∈ C ( G ) F\in C(\mathcal{G}) F ∈ C ( G ) can be viewed as a section { F n } n ∈ N ∪ { ∞ } \{F^n\}_{n\in\mathbb{N}\cup\{\infty\}} { F n } n ∈ N ∪ { ∞ } , where F n = F ∣ G n F^n=F|_{G_n} F n = F ∣ G n , associated with the left regular bundle (Theorem 2.5). For each n ∈ N n\in\mathbb{N} n ∈ N , let S n S_n S n denote a set of representatives of quotients in G n G_n G n .
First suppose that f ≥ 0 f\geq0 f ≥ 0 . Using the definition of continuous sections, we need to prove the following two items.
Note that
∥ F n ∥ 2 2 = ∑ x ∈ S n ∣ f n ( q n ( x ) ) ∣ 2 = ∑ x ∈ S n ∥ f ∣ x K n ∥ 2 2 = ∥ f ∥ 2 2 . \|F^n\|_2^2=\sum_{x\in S_n}|f_n(q_n(x))|^2=\sum_{x\in S_n}\|f|_{xK_n}\|_2^2=\|f\|_2^2. ∥ F n ∥ 2 2 = x ∈ S n ∑ ∣ f n ( q n ( x )) ∣ 2 = x ∈ S n ∑ ∥ f ∣ x K n ∥ 2 2 = ∥ f ∥ 2 2 .
Constant functions are always continuous, and so the map n ↦ ∥ F n ∥ 2 n\mapsto\|F^n\|_2 n ↦ ∥ F n ∥ 2 is continuous.
We show that F F F is locally close to the fundamental sections of Λ \Lambda Λ , i.e., elements of C c ( G ) C_c(\mathcal{G}) C c ( G ) . Let ϵ > 0 \epsilon>0 ϵ > 0 be given, and choose h ∈ C c ( G ∞ ) h\in C_c(G_\infty) h ∈ C c ( G ∞ ) with ∥ h − f ∥ 2 < ϵ \|h-f\|_2<\epsilon ∥ h − f ∥ 2 < ϵ and h ≥ 0 h\geq0 h ≥ 0 . Construct H H H from h h h via Equation (8) . By Theorem 4.3 and Theorem 4.4, H ∈ C c ( G ) H\in C_c(\mathcal{G}) H ∈ C c ( G ) . Note that
∥ F n − H n ∥ 2 2 = ∑ x ∈ S n ∥ f ∣ x K n ∥ 2 2 − ∥ h ∣ x K n ∥ 2 2 ≤ ∑ x ∈ S n ∥ f ∣ x K n − h ∣ x K n ∥ 2 2 = ∥ f − h ∥ 2 2 , \|F^n-H^n\|_2^2=\sum_{x\in S_n}\|f|_{xK_n}\|_2^2-\|h|_{xK_n}\|_2^2\leq\sum_{x\in S_n}\|f|_{xK_n}-h|_{xK_n}\|_2^2=\|f-h\|_2^2, ∥ F n − H n ∥ 2 2 = x ∈ S n ∑ ∥ f ∣ x K n ∥ 2 2 − ∥ h ∣ x K n ∥ 2 2 ≤ x ∈ S n ∑ ∥ f ∣ x K n − h ∣ x K n ∥ 2 2 = ∥ f − h ∥ 2 2 ,
which implies ∥ F n − H n ∥ 2 < ϵ \|F^n-H^n\|_2<\epsilon ∥ F n − H n ∥ 2 < ϵ for all n ∈ N n\in\mathbb{N} n ∈ N .
Next, consider a general f ∈ ℓ 2 ( G ∞ ) f\in\ell^2(G_\infty) f ∈ ℓ 2 ( G ∞ ) , and consider the decomposition f = f 1 − f 2 + i ( f 3 − f 4 ) f=f_1-f_2+i(f_3-f_4) f = f 1 − f 2 + i ( f 3 − f 4 ) , f i ≥ 0 f_i\geq0 f i ≥ 0 , and let F = F 1 − F 2 + i ( F 3 − F 4 ) F=F_1-F_2+i(F_3-F_4) F = F 1 − F 2 + i ( F 3 − F 4 ) be defined as in Equation (11) . By the previous case, each one of F 1 , F 2 , F 3 F_1,F_2,F_3 F 1 , F 2 , F 3 and F 4 F_4 F 4 are continuous sections of Λ \Lambda Λ . Since continuous sections form a complex linear space (Theorem 2.2), the linear combination F = F 1 − F 2 + i ( F 3 − F 4 ) F=F_1-F_2+i(F_3-F_4) F = F 1 − F 2 + i ( F 3 − F 4 ) is a continuous section as well. This finishes the proof of the claim.
We have now demonstrated that for f ∈ ℓ 2 ( G ∞ ) f\in\ell^2(G_\infty) f ∈ ℓ 2 ( G ∞ ) , constructing F F F as in Equation (11) yields a continuous section of Λ \Lambda Λ with F ∞ = f F^\infty=f F ∞ = f . Next, we show that this construction is norm-preserving in the following sense.
Claim 4.6. For every f ∈ ℓ 2 ( G ∞ ) f\in\ell^2(G_\infty) f ∈ ℓ 2 ( G ∞ ) , the associated continuous section F F F satisfies ∥ F ∥ Λ = ∥ f ∥ 2 \|F\|_\Lambda=\|f\|_2 ∥ F ∥ Λ = ∥ f ∥ 2 .
Proof of Theorem 4.6: Let f = f 1 − f 2 + i ( f 3 − f 4 ) f=f_1-f_2+i(f_3-f_4) f = f 1 − f 2 + i ( f 3 − f 4 ) be the decomposition of f f f into the positive and negative parts of the real and imaginary components of f f f , and F = F 1 − F 2 + i ( F 3 − F 4 ) F=F_1-F_2+i(F_3-F_4) F = F 1 − F 2 + i ( F 3 − F 4 ) be defined as in Equation (8) . For 1 ≤ i ≤ 4 1\leq i\leq4 1 ≤ i ≤ 4 , we have
∥ F i n ∥ ℓ 2 ( G n ) 2 = ∑ x ∈ S n ∣ F i n ( q n ( x ) ) ∣ 2 = ∑ x ∈ S n ∥ f i ∣ x K n ∥ 2 2 = ∥ f i ∥ 2 2 . (12) \|F_i^n\|_{\ell^2(G_n)}^2=\sum_{x\in S_n}|F_i^n(q_n(x))|^2=\sum_{x\in S_n}\|f_i|_{xK_n}\|_2^2=\|f_i\|_2^2. \tag*{(12)} ∥ F i n ∥ ℓ 2 ( G n ) 2 = x ∈ S n ∑ ∣ F i n ( q n ( x )) ∣ 2 = x ∈ S n ∑ ∥ f i ∣ x K n ∥ 2 2 = ∥ f i ∥ 2 2 . (12)
Moreover, since by definition each function F i n F_i^n F i n is pointwise non-negative, we have
∥ F n ∥ ℓ 2 ( G n ) 2 = ∥ F 1 n − F 2 n + i ( F 3 n − F 4 n ) ∥ ℓ 2 ( G n ) 2 = ∥ F 1 n − F 2 n ∥ ℓ 2 ( G n ) 2 + ∥ F 3 n − F 4 n ∥ ℓ 2 ( G n ) ≤ ∥ F 1 n ∥ ℓ 2 ( G n ) 2 + ∥ F 2 n ∥ ℓ 2 ( G n ) 2 + ∥ F 3 n ∥ ℓ 2 ( G n ) 2 + ∥ F 4 n ∥ ℓ 2 ( G n ) 2 . \begin{aligned}
\|F^n\|_{\ell^2(G_n)}^2=\|F_1^n-F_2^n+i(F_3^n-F_4^n)\|_{\ell^2(G_n)}^2=\|F_1^n-F_2^n\|_{\ell^2(G_n)}^2+\|F_3^n-F_4^n\|_{\ell^2(G_n)} \\
&\leq\lVert F_1^n\rVert_{\ell^2(G_n)}^2+\lVert F_2^n\rVert_{\ell^2(G_n)}^2+\lVert F_3^n\rVert_{\ell^2(G_n)}^2+\lVert F_4^n\rVert_{\ell^2(G_n)}^2.
\end{aligned} ∥ F n ∥ ℓ 2 ( G n ) 2 = ∥ F 1 n − F 2 n + i ( F 3 n − F 4 n ) ∥ ℓ 2 ( G n ) 2 = ∥ F 1 n − F 2 n ∥ ℓ 2 ( G n ) 2 + ∥ F 3 n − F 4 n ∥ ℓ 2 ( G n ) ≤ ∥ F 1 n ∥ ℓ 2 ( G n ) 2 + ∥ F 2 n ∥ ℓ 2 ( G n ) 2 + ∥ F 3 n ∥ ℓ 2 ( G n ) 2 + ∥ F 4 n ∥ ℓ 2 ( G n ) 2 .
Using Equation (12) , we have
∥ F n ∥ ℓ 2 ( G n ) 2 ≤ ∥ F 1 n ∥ ℓ 2 ( G n ) 2 + ∥ F 2 n ∥ ℓ 2 ( G n ) 2 + ∥ F 3 n ∥ ℓ 2 ( G n ) 2 + ∥ F 4 n ∥ ℓ 2 ( G n ) 2 = ∥ f 1 ∥ ℓ 2 ( G ∞ ) 2 + ∥ f 2 ∥ ℓ 2 ( G ∞ ) 2 + ∥ f 3 ∥ ℓ 2 ( G ∞ ) 2 + ∥ f 4 ∥ ℓ 2 ( G ∞ ) 2 = ∥ f 1 − f 2 ∥ ℓ 2 ( G ∞ ) 2 + ∥ f 3 − f 4 ∥ ℓ 2 ( G ∞ ) 2 = ∥ f ∥ ℓ 2 ( G ∞ ) 2 , \begin{aligned}
\lVert F^n\rVert_{\ell^2(G_n)}^2
&\leq\lVert F_1^n\rVert_{\ell^2(G_n)}^2+\lVert F_2^n\rVert_{\ell^2(G_n)}^2+\lVert F_3^n\rVert_{\ell^2(G_n)}^2+\lVert F_4^n\rVert_{\ell^2(G_n)}^2\\
&=\lVert f_1\rVert_{\ell^2(G_\infty)}^2+\lVert f_2\rVert_{\ell^2(G_\infty)}^2+\lVert f_3\rVert_{\ell^2(G_\infty)}^2+\lVert f_4\rVert_{\ell^2(G_\infty)}^2\\
&=\lVert f_1-f_2\rVert_{\ell^2(G_\infty)}^2+\lVert f_3-f_4\rVert_{\ell^2(G_\infty)}^2\\
&=\lVert f\rVert_{\ell^2(G_\infty)}^2,
\end{aligned} ∥ F n ∥ ℓ 2 ( G n ) 2 ≤ ∥ F 1 n ∥ ℓ 2 ( G n ) 2 + ∥ F 2 n ∥ ℓ 2 ( G n ) 2 + ∥ F 3 n ∥ ℓ 2 ( G n ) 2 + ∥ F 4 n ∥ ℓ 2 ( G n ) 2 = ∥ f 1 ∥ ℓ 2 ( G ∞ ) 2 + ∥ f 2 ∥ ℓ 2 ( G ∞ ) 2 + ∥ f 3 ∥ ℓ 2 ( G ∞ ) 2 + ∥ f 4 ∥ ℓ 2 ( G ∞ ) 2 = ∥ f 1 − f 2 ∥ ℓ 2 ( G ∞ ) 2 + ∥ f 3 − f 4 ∥ ℓ 2 ( G ∞ ) 2 = ∥ f ∥ ℓ 2 ( G ∞ ) 2 ,
where we used the fact that f 1 f_1 f 1 and f 2 f_2 f 2 (respectively f 3 f_3 f 3 and f 4 f_4 f 4 ) have disjoint support in the penultimate equality. Since F ∞ = f F^\infty=f F ∞ = f , the section norm of F F F can be computed as follows:
∥ F ∥ Δ = sup n ∈ N ∪ { ∞ } ∥ F n ∥ ℓ 2 ( G n ) = ∥ f ∥ ℓ 2 ( G ∞ ) . (13) \lVert F\rVert_\Delta=\sup_{n\in\mathbb{N}\cup\{\infty\}}\lVert F^n\rVert_{\ell^2(G_n)}=\lVert f\rVert_{\ell^2(G_\infty)}. \tag*{(13)} ∥ F ∥ Δ = n ∈ N ∪ { ∞ } sup ∥ F n ∥ ℓ 2 ( G n ) = ∥ f ∥ ℓ 2 ( G ∞ ) . (13)
Let u ∈ A ( G ∞ ) u\in A(G_\infty) u ∈ A ( G ∞ ) be arbitrary. Choose f , g ∈ ℓ 2 ( G ∞ ) f,g\in\ell^2(G_\infty) f , g ∈ ℓ 2 ( G ∞ ) such that u = L f , g u=L_{f,g} u = L f , g and ∥ u ∥ A ( G ∞ ) = ∥ f ∥ 2 ∥ g ∥ 2 \lVert u\rVert_{A(G_\infty)}=\lVert f\rVert_2\lVert g\rVert_2 ∥ u ∥ A ( G ∞ ) = ∥ f ∥ 2 ∥ g ∥ 2 [11 Lemma 2.14] . Let F , G F,G F , G be the sections constructed for f , g f,g f , g as in Equation (8) and Equation (11) . It is clear that Λ F , G ∈ A ( G ) \Lambda_{F,G}\in A(G) Λ F , G ∈ A ( G ) and R ∞ ( Λ F , G ) = L f , g R_\infty(\Lambda_{F,G})=L_{f,g} R ∞ ( Λ F , G ) = L f , g . Since the restriction map R ∞ R_\infty R ∞ is contractive [7 Proposition 3.4 (iii)] , we get
∥ u ∥ A ( G ∞ ) = ∥ R n ( Λ F , G ) ∥ A ( G ∞ ) ≤ ∥ Λ F , G ∥ A ( G ) . \lVert u\rVert_{A(G_\infty)}=\lVert R_n(\Lambda_{F,G})\rVert_{A(G_\infty)}\leq\lVert\Lambda_{F,G}\rVert_{A(G)}. ∥ u ∥ A ( G ∞ ) = ∥ R n ( Λ F , G ) ∥ A ( G ∞ ) ≤ ∥ Λ F , G ∥ A ( G ) .
On the other hand,
∥ Λ F , G ∥ A ( G ) ≤ ∥ F ∥ Δ ∥ G ∥ Δ = ∥ f ∥ 2 ∥ g ∥ 2 = ∥ u ∥ A ( G ∞ ) . \lVert\Lambda_{F,G}\rVert_{A(G)}\leq\lVert F\rVert_\Delta\lVert G\rVert_\Delta=\lVert f\rVert_2\lVert g\rVert_2=\lVert u\rVert_{A(G_\infty)}. ∥ Λ F , G ∥ A ( G ) ≤ ∥ F ∥ Δ ∥ G ∥ Δ = ∥ f ∥ 2 ∥ g ∥ 2 = ∥ u ∥ A ( G ∞ ) .
So ∥ Λ F , G ∥ A ( G ) = ∥ u ∥ A ( G ∞ ) \lVert\Lambda_{F,G}\rVert_{A(G)}=\lVert u\rVert_{A(G_\infty)} ∥ Λ F , G ∥ A ( G ) = ∥ u ∥ A ( G ∞ ) , and the restriction map is a quotient map. □ \square □
Remark 4.7. Here, we list two easy consequences of Theorem 4.2 .
Positive-definite functions admit positive-definite norm-preserving lifts. Indeed, suppose u ∈ A ( G ∞ ) u\in A(G_\infty) u ∈ A ( G ∞ ) is positive-definite, and let f ∈ ℓ 2 ( G ∞ ) f\in\ell^2(G_\infty) f ∈ ℓ 2 ( G ∞ ) be such that
u ( x ) = ⟨ λ G ∞ ( x ) f , f ⟩ , ∥ f ∥ 2 2 = u ( e G ∞ ) = ∥ u ∥ A ( G ∞ ) . u(x)=\langle\lambda_{G_\infty}(x)f,f\rangle,\qquad\lVert f\rVert_2^2=u(e_{G_\infty})=\lVert u\rVert_{A(G_\infty)}. u ( x ) = ⟨ λ G ∞ ( x ) f , f ⟩ , ∥ f ∥ 2 2 = u ( e G ∞ ) = ∥ u ∥ A ( G ∞ ) .
Let F F F be the continuous section lifted from f f f as described in Equation (8) and Equation (11) . Then v = Λ F , F v=\Lambda_{F,F} v = Λ F , F is a positive-definite element of A ( G ) A(G) A ( G ) satisfying R ∞ ( v ) = u R_\infty(v)=u R ∞ ( v ) = u and ∥ v ∥ A ( G ) = ∥ u ∥ A ( G ∞ ) \lVert v\rVert_{A(G)}=\lVert u\rVert_{A(G_\infty)} ∥ v ∥ A ( G ) = ∥ u ∥ A ( G ∞ ) .
The lifting property proved in Theorem 4.2 can be extended to finitely many fibers. More precisely, let S ⊆ N ∗ S\subseteq\mathbb{N}^* S ⊆ N ∗ be a nonempty finite set. For each n ∈ S n\in S n ∈ S , let u n ∈ A ( G n ) u_n\in A(G_n) u n ∈ A ( G n ) be given. Then there exists v ∈ A ( G ) v\in A(G) v ∈ A ( G ) such that
R n ( v ) = u n and ∥ v ∥ A ( G ) = max n ∈ S ∥ u n ∥ A ( G n ) ∀ n ∈ S . (14) R_n(v)=u_n\quad\text{and}\quad\lVert v\rVert_{A(G)}=\max_{n\in S}\lVert u_n\rVert_{A(G_n)}\quad\forall n\in S. \tag*{(14)} R n ( v ) = u n and ∥ v ∥ A ( G ) = n ∈ S max ∥ u n ∥ A ( G n ) ∀ n ∈ S . (14)
To see this, for each n ∈ S n\in S n ∈ S , let f n ∈ ℓ 2 ( G n ) f_n\in\ell^2(G_n) f n ∈ ℓ 2 ( G n ) be chosen such that u n ( ⋅ ) = ⟨ λ G n ( ⋅ ) f n , f n ⟩ u_n(\cdot)=\langle\lambda_{G_n}(\cdot)f_n,f_n\rangle u n ( ⋅ ) = ⟨ λ G n ( ⋅ ) f n , f n ⟩ with ∥ u n ∥ A ( G n ) = ∥ f n ∥ 2 2 \lVert u_n\rVert_{A(G_n)}=\lVert f_n\rVert_2^2 ∥ u n ∥ A ( G n ) = ∥ f n ∥ 2 2 . If ∞ ∈ S \infty\in S ∞ ∈ S , let F = { F n } n ∈ N ∗ F=\{F_n\}_{n\in\mathbb{N}^*} F = { F n } n ∈ N ∗ denote the continuous section that lifts f ∞ f_\infty f ∞ (described in Equation (8) and Equation (11) ); otherwise, let F F F denote the constant 0 section. We now define the new continuous section H = { H n } n ∈ N ∗ H=\{H_n\}_{n\in\mathbb{N}^*} H = { H n } n ∈ N ∗ by adjusting F F F to fit the rest of S S S as follows:
H n = { f n n ∈ S ∖ { ∞ } , F n otherwise . H_n=
\begin{cases}
f_n & n\in S\setminus\{\infty\},\\
F_n & \text{otherwise}.
\end{cases} H n = { f n F n n ∈ S ∖ { ∞ } , otherwise .
It is easy to see that v = Λ H , H v=\Lambda_{H,H} v = Λ H , H satisfies the conditions of Equation (14) .
Applications: Fourier norm estimates and bounded approximate identity
The constructions in Theorem 4.2 enable us to extend the result of Theorem 3.9, Item 3 to arbitrary elements of A ( G ) A(\mathcal{G}) A ( G ) . This extension yields a formula expressing the Fourier norm of u ∈ A ( G ) u \in A(\mathcal{G}) u ∈ A ( G ) in terms of the norms of its restrictions to the fibers. We also characterize the HLS groupoids G \mathcal{G} G for which A ( G ) A(\mathcal{G}) A ( G ) admits a bounded approximate identity.
Proposition 5.1. For every u ∈ A ( G ) u \in A(\mathcal{G}) u ∈ A ( G ) , we have lim n → ∞ ∥ R n ( u ) ∥ A ( G n ) = ∥ R ∞ ( u ) ∥ A ( G ∞ ) \lim_{n\to\infty}\lVert R_n(u)\rVert_{A(\mathcal{G}_n)}=\lVert R_\infty(u)\rVert_{A(\mathcal{G}_\infty)} lim n → ∞ ∥ R n ( u ) ∥ A ( G n ) = ∥ R ∞ ( u ) ∥ A ( G ∞ ) . Consequently, we can identify the kernel of the restriction map R ∞ R_\infty R ∞ as follows:
ker ( R ∞ ) = { u ∈ A ( G ) : ∥ R n ( u ) ∥ A ( G n ) → 0 } . \operatorname{ker}(R_\infty)=\{u\in A(\mathcal{G}):\lVert R_n(u)\rVert_{A(\mathcal{G}_n)}\to0\}. ker ( R ∞ ) = { u ∈ A ( G ) : ∥ R n ( u ) ∥ A ( G n ) → 0 } . We denote the subspace defined on the right hand side by c 0 − ⨁ n ∈ N A ( G n ) c_0-\bigoplus_{n\in\mathbb{N}} A(\mathcal{G}_n) c 0 − ⨁ n ∈ N A ( G n ) .
Proof. First, fix u ∈ A ( G ) ∩ C c ( G ) u\in A(\mathcal{G})\cap C_c(\mathcal{G}) u ∈ A ( G ) ∩ C c ( G ) . Applying Theorem 4.2 to R ∞ ( u ) R_\infty(u) R ∞ ( u ) , there exists v ∈ A ( G ) v\in A(\mathcal{G}) v ∈ A ( G ) such that
R ∞ ( v ) = R ∞ ( u ) . R_\infty(v)=R_\infty(u). R ∞ ( v ) = R ∞ ( u ) .
Moreover, Theorem 4.3 implies that v ∈ C c ( G ) v\in C_c(\mathcal{G}) v ∈ C c ( G ) . Thus, u − v ∈ C c ( G ) u-v\in C_c(\mathcal{G}) u − v ∈ C c ( G ) and R ∞ ( u − v ) = 0 R_\infty(u-v)=0 R ∞ ( u − v ) = 0 . Applying Theorem 3.9 (3) to u − v u-v u − v , we obtain lim n → ∞ ∥ R n ( u ) − R n ( v ) ∥ A ( G n ) = 0 \lim_{n\to\infty}\lVert R_n(u)-R_n(v)\rVert_{A(\mathcal{G}_n)}=0 lim n → ∞ ∥ R n ( u ) − R n ( v ) ∥ A ( G n ) = 0 . Consequently,
lim n → ∞ ∣ ∥ R n ( u ) ∥ A ( G n ) − ∥ R n ( v ) ∥ A ( G n ) ∣ = 0. (15) \lim_{n\to\infty}\left|\lVert R_n(u)\rVert_{A(\mathcal{G}_n)}-\lVert R_n(v)\rVert_{A(\mathcal{G}_n)}\right|=0. \tag*{(15)} n → ∞ lim ∥ R n ( u ) ∥ A ( G n ) − ∥ R n ( v ) ∥ A ( G n ) = 0. (15)
Next, we show that lim n → ∞ ∥ R n ( v ) ∥ A ( G n ) = ∥ R ∞ ( v ) ∥ A ( G ∞ ) \lim_{n\to\infty}\lVert R_n(v)\rVert_{A(\mathcal{G}_n)}=\lVert R_\infty(v)\rVert_{A(\mathcal{G}_\infty)} lim n → ∞ ∥ R n ( v ) ∥ A ( G n ) = ∥ R ∞ ( v ) ∥ A ( G ∞ ) . First, note that ∥ v ∥ A ( G ) = ∥ R ∞ ( v ) ∥ A ( G ∞ ) \lVert v\rVert_{A(\mathcal{G})}=\lVert R_\infty(v)\rVert_{A(\mathcal{G}_\infty)} ∥ v ∥ A ( G ) = ∥ R ∞ ( v ) ∥ A ( G ∞ ) (Theorem 4.2 ). So, by Theorem 3.9 (1), lim sup n → ∞ ∥ R n ( v ) ∥ A ( G n ) ≤ ∥ R ∞ ( v ) ∥ A ( G ∞ ) \limsup_{n\to\infty}\lVert R_n(v)\rVert_{A(\mathcal{G}_n)}\leq\lVert R_\infty(v)\rVert_{A(\mathcal{G}_\infty)} lim sup n → ∞ ∥ R n ( v ) ∥ A ( G n ) ≤ ∥ R ∞ ( v ) ∥ A ( G ∞ ) . On the other hand, Theorem 3.9 (2) gives
∥ R ∞ ( v ) ∥ A ( G ∞ ) ≤ lim inf n → ∞ ∥ R n ( v ) ∥ A ( G n ) . \lVert R_\infty(v)\rVert_{A(\mathcal{G}_\infty)}\leq\liminf_{n\to\infty}\lVert R_n(v)\rVert_{A(\mathcal{G}_n)}. ∥ R ∞ ( v ) ∥ A ( G ∞ ) ≤ n → ∞ lim inf ∥ R n ( v ) ∥ A ( G n ) .
It follows that lim sup n → ∞ ∥ R n ( v ) ∥ A ( G n ) ≤ ∥ R ∞ ( v ) ∥ A ( G ∞ ) ≤ lim inf n → ∞ ∥ R n ( v ) ∥ A ( G n ) \limsup_{n\to\infty}\lVert R_n(v)\rVert_{A(\mathcal{G}_n)}\leq\lVert R_\infty(v)\rVert_{A(\mathcal{G}_\infty)}\leq\liminf_{n\to\infty}\lVert R_n(v)\rVert_{A(\mathcal{G}_n)} lim sup n → ∞ ∥ R n ( v ) ∥ A ( G n ) ≤ ∥ R ∞ ( v ) ∥ A ( G ∞ ) ≤ lim inf n → ∞ ∥ R n ( v ) ∥ A ( G n ) , and therefore
lim n → ∞ ∥ R n ( v ) ∥ A ( G n ) = ∥ R ∞ ( v ) ∥ A ( G ∞ ) . (16) \lim_{n\to\infty}\lVert R_n(v)\rVert_{A(\mathcal{G}_n)}=\lVert R_\infty(v)\rVert_{A(\mathcal{G}_\infty)}. \tag*{(16)} n → ∞ lim ∥ R n ( v ) ∥ A ( G n ) = ∥ R ∞ ( v ) ∥ A ( G ∞ ) . (16)
Combining Equation (15) and Equation (16) , together with the fact that R ∞ ( v ) = R ∞ ( u ) R_\infty(v)=R_\infty(u) R ∞ ( v ) = R ∞ ( u ) , we get that
lim n → ∞ ∥ R n ( u ) ∥ A ( G n ) = ∥ R ∞ ( u ) ∥ A ( G ∞ ) \lim_{n\to\infty}\lVert R_n(u)\rVert_{A(\mathcal{G}_n)}=\lVert R_\infty(u)\rVert_{A(\mathcal{G}_\infty)} n → ∞ lim ∥ R n ( u ) ∥ A ( G n ) = ∥ R ∞ ( u ) ∥ A ( G ∞ )
for every u ∈ A ( G ) ∩ C c ( G ) u\in A(\mathcal{G})\cap C_c(\mathcal{G}) u ∈ A ( G ) ∩ C c ( G ) .
Finally, let u ∈ A ( G ) u\in A(\mathcal{G}) u ∈ A ( G ) and fix ϵ > 0 \epsilon>0 ϵ > 0 . Choose v ∈ A ( G ) ∩ C c ( G ) v\in A(\mathcal{G})\cap C_c(\mathcal{G}) v ∈ A ( G ) ∩ C c ( G ) such that ∥ u − v ∥ A ( G ) < ϵ 3 \lVert u-v\rVert_{A(\mathcal{G})}<\frac{\epsilon}{3} ∥ u − v ∥ A ( G ) < 3 ϵ . Since each restriction map is contractive, for every n ∈ N ∗ n\in\mathbb{N}^* n ∈ N ∗ we have
∣ ∥ R n ( u ) ∥ A ( G n ) − ∥ R n ( v ) ∥ A ( G n ) ∣ ≤ ∥ R n ( u − v ) ∥ A ( G n ) ≤ ∥ u − v ∥ A ( G ) < ϵ 3 . \left|\lVert R_n(u)\rVert_{A(\mathcal{G}_n)}-\lVert R_n(v)\rVert_{A(\mathcal{G}_n)}\right|\leq\lVert R_n(u-v)\rVert_{A(\mathcal{G}_n)}\leq\lVert u-v\rVert_{A(\mathcal{G})}<\frac{\epsilon}{3}. ∥ R n ( u ) ∥ A ( G n ) − ∥ R n ( v ) ∥ A ( G n ) ≤ ∥ R n ( u − v ) ∥ A ( G n ) ≤ ∥ u − v ∥ A ( G ) < 3 ϵ .
By the first part of the proof, there exists N ∈ N N\in\mathbb{N} N ∈ N such that, for every n ≥ N n\geq N n ≥ N ,
∣ ∥ R n ( v ) ∥ A ( G n ) − ∥ R ∞ ( v ) ∥ A ( G ∞ ) ∣ < ϵ 3 . \left|\lVert R_n(v)\rVert_{A(\mathcal{G}_n)}-\lVert R_\infty(v)\rVert_{A(\mathcal{G}_\infty)}\right|<\frac{\epsilon}{3}. ∥ R n ( v ) ∥ A ( G n ) − ∥ R ∞ ( v ) ∥ A ( G ∞ ) < 3 ϵ .
Therefore, for every n ≥ N n\geq N n ≥ N ,
∣ ∥ R n ( u ) ∥ A ( G n ) − ∥ R ∞ ( u ) ∥ A ( G ∞ ) ∣ ≤ ∣ ∥ R n ( u ) ∥ A ( G n ) − ∥ R n ( v ) ∥ A ( G n ) ∣ + ∣ ∥ R n ( v ) ∥ A ( G n ) − ∥ R ∞ ( v ) ∥ A ( G ∞ ) ∣ + ∣ ∥ R ∞ ( v ) ∥ A ( G ∞ ) − ∥ R ∞ ( u ) ∥ A ( G ∞ ) ∣ \begin{aligned}
\left|\lVert R_n(u)\rVert_{A(\mathcal{G}_n)}-\lVert R_\infty(u)\rVert_{A(\mathcal{G}_\infty)}\right|
&\leq\left|\lVert R_n(u)\rVert_{A(\mathcal{G}_n)}-\lVert R_n(v)\rVert_{A(\mathcal{G}_n)}\right| \\
&\quad+\left|\lVert R_n(v)\rVert_{A(\mathcal{G}_n)}-\lVert R_\infty(v)\rVert_{A(\mathcal{G}_\infty)}\right| \\
&\quad+\left|\lVert R_\infty(v)\rVert_{A(\mathcal{G}_\infty)}-\lVert R_\infty(u)\rVert_{A(\mathcal{G}_\infty)}\right|
\end{aligned} ∥ R n ( u ) ∥ A ( G n ) − ∥ R ∞ ( u ) ∥ A ( G ∞ ) ≤ ∥ R n ( u ) ∥ A ( G n ) − ∥ R n ( v ) ∥ A ( G n ) + ∥ R n ( v ) ∥ A ( G n ) − ∥ R ∞ ( v ) ∥ A ( G ∞ ) + ∥ R ∞ ( v ) ∥ A ( G ∞ ) − ∥ R ∞ ( u ) ∥ A ( G ∞ )
< ε <\varepsilon < ε .
Hence, ∥ R n ( u ) ∥ A ( G n ) ⟶ ∥ R ∞ ( u ) ∥ A ( G ∞ ) \|R_n(u)\|_{A(G_n)} \longrightarrow\|R_\infty(u)\|_{A(G_\infty)} ∥ R n ( u ) ∥ A ( G n ) ⟶ ∥ R ∞ ( u ) ∥ A ( G ∞ ) .
To prove the second statement, first observe that c 0 - ⨁ n ∈ N A ( G n ) ⊆ ker ( R ∞ ) c_0\text{-}\bigoplus_{n\in\mathbb{N}} A(G_n) \subseteq\ker(R_\infty) c 0 - ⨁ n ∈ N A ( G n ) ⊆ ker ( R ∞ ) . For the reverse inclusion, let u ∈ ker ( R ∞ ) u\in\ker(R_\infty) u ∈ ker ( R ∞ ) . By the first statement of the proposition,
lim n → ∞ ∥ R n ( u ) ∥ A ( G n ) = ∥ R ∞ ( u ) ∥ A ( G ∞ ) = 0. \lim_{n\to\infty}\|R_n(u)\|_{A(G_n)}=\|R_\infty(u)\|_{A(G_\infty)}=0. n → ∞ lim ∥ R n ( u ) ∥ A ( G n ) = ∥ R ∞ ( u ) ∥ A ( G ∞ ) = 0.
Thus, the sequence { R n ( u ) } n ∈ N \{R_n(u)\}_{n\in\mathbb{N}} { R n ( u ) } n ∈ N belongs to c 0 - ⨁ n ∈ N A ( G n ) c_0\text{-}\bigoplus_{n\in\mathbb{N}} A(G_n) c 0 - ⨁ n ∈ N A ( G n ) , and we are done. □ \square □
Corollary 5.2. For every u ∈ A ( G ) u\in A(G) u ∈ A ( G ) , we have ∥ u ∥ A ( G ) = sup n ∈ N ∗ ∥ R n ( u ) ∥ A ( G n ) \|u\|_{A(G)}=\sup_{n\in\mathbb{N}^*}\|R_n(u)\|_{A(G_n)} ∥ u ∥ A ( G ) = sup n ∈ N ∗ ∥ R n ( u ) ∥ A ( G n ) .
Proof. By Theorem 3.9 [1 ] , ∥ u ∥ A ( G ) ≥ sup n ∈ N ∗ ∥ R n ( u ) ∥ A ( G n ) \|u\|_{A(G)}\geq\sup_{n\in\mathbb{N}^*}\|R_n(u)\|_{A(G_n)} ∥ u ∥ A ( G ) ≥ sup n ∈ N ∗ ∥ R n ( u ) ∥ A ( G n ) . So we only need to show the reverse inequality.
First assume that u ∈ A ( G ) u\in A(G) u ∈ A ( G ) satisfies R ∞ ( u ) = 0 R_\infty(u)=0 R ∞ ( u ) = 0 . By Theorem 5.1, lim n → ∞ ∥ R n ( u ) ∥ A ( G n ) = 0 \lim_{n\to\infty}\|R_n(u)\|_{A(G_n)}=0 lim n → ∞ ∥ R n ( u ) ∥ A ( G n ) = 0 . For each n ∈ N n\in\mathbb{N} n ∈ N we may choose f n , h n ∈ ℓ 2 ( G n ) f_n,h_n\in\ell^2(G_n) f n , h n ∈ ℓ 2 ( G n ) such that
R n ( u ) = ⟨ λ G n ( ⋅ ) f n , h n ⟩ ℓ 2 ( G n ) and ∥ f n ∥ 2 = ∥ h n ∥ 2 = ∥ R n ( u ) ∥ A ( G n ) 1 / 2 . R_n(u)=\langle\lambda_{G_n}(\cdot)f_n,h_n\rangle_{\ell^2(G_n)}
\quad\text{and}\quad
\|f_n\|_2=\|h_n\|_2=\|R_n(u)\|_{A(G_n)}^{1/2}. R n ( u ) = ⟨ λ G n ( ⋅ ) f n , h n ⟩ ℓ 2 ( G n ) and ∥ f n ∥ 2 = ∥ h n ∥ 2 = ∥ R n ( u ) ∥ A ( G n ) 1/2 .
Set f ∞ = g ∞ = 0 f_\infty=g_\infty=0 f ∞ = g ∞ = 0 , and let F = { f n } n ∈ N ∗ F=\{f_n\}_{n\in\mathbb{N}^*} F = { f n } n ∈ N ∗ and H = { h n } n ∈ N ∗ H=\{h_n\}_{n\in\mathbb{N}^*} H = { h n } n ∈ N ∗ . By Theorem 3.6 [2 ] both F F F and H H H belong to Γ \Gamma Γ , and by Theorem 3.6 [3 ] , we have Λ F , H ∈ A ( G ) \Lambda_{F,H}\in A(G) Λ F , H ∈ A ( G ) . Since R n ( Λ F , H ) = R n ( u ) R_n(\Lambda_{F,H})=R_n(u) R n ( Λ F , H ) = R n ( u ) for every n ∈ N ∗ n\in\mathbb{N}^* n ∈ N ∗ , we have Λ F , H = u \Lambda_{F,H}=u Λ F , H = u . Therefore,
∥ u ∥ A ( G ) ≤ ∥ F ∥ Δ ∥ H ∥ Δ = ( sup n ∈ N ∥ f n ∥ 2 ) ( sup n ∈ N ∥ h n ∥ 2 ) = sup n ∈ N ∥ R n ( u ) ∥ A ( G n ) . \|u\|_{A(G)}\leq\|F\|_\Delta\|H\|_\Delta
=\left(\sup_{n\in\mathbb{N}}\|f_n\|_2\right)\left(\sup_{n\in\mathbb{N}}\|h_n\|_2\right)
=\sup_{n\in\mathbb{N}}\|R_n(u)\|_{A(G_n)}. ∥ u ∥ A ( G ) ≤ ∥ F ∥ Δ ∥ H ∥ Δ = ( n ∈ N sup ∥ f n ∥ 2 ) ( n ∈ N sup ∥ h n ∥ 2 ) = n ∈ N sup ∥ R n ( u ) ∥ A ( G n ) .
This proves the desired statement for the case where R ∞ ( u ) = 0 R_\infty(u)=0 R ∞ ( u ) = 0 .
Next, let u ∈ A ( G ) u\in A(G) u ∈ A ( G ) be arbitrary and put S : = sup n ∈ N ∗ ∥ R n ( u ) ∥ A ( G n ) S:=\sup_{n\in\mathbb{N}^*}\|R_n(u)\|_{A(G_n)} S := sup n ∈ N ∗ ∥ R n ( u ) ∥ A ( G n ) , which is finite by Theorem 3.9 [1 ] . For each n ∈ N ∗ n\in\mathbb{N}^* n ∈ N ∗ choose ξ n , η n ∈ ℓ 2 ( G n ) \xi_n,\eta_n\in\ell^2(G_n) ξ n , η n ∈ ℓ 2 ( G n ) such that
R n ( u ) = ⟨ λ G n ( ⋅ ) ξ n , η n ⟩ and ∥ ξ n ∥ 2 2 = ∥ η n ∥ 2 2 = ∥ R n ( u ) ∥ A ( G n ) ≤ S . R_n(u)=\langle\lambda_{G_n}(\cdot)\xi_n,\eta_n\rangle
\quad\text{and}\quad
\|\xi_n\|_2^2=\|\eta_n\|_2^2=\|R_n(u)\|_{A(G_n)}\leq S. R n ( u ) = ⟨ λ G n ( ⋅ ) ξ n , η n ⟩ and ∥ ξ n ∥ 2 2 = ∥ η n ∥ 2 2 = ∥ R n ( u ) ∥ A ( G n ) ≤ S .
Let F = { f n } n ∈ N ∗ F=\{f_n\}_{n\in\mathbb{N}^*} F = { f n } n ∈ N ∗ , H = { h n } n ∈ N ∗ H=\{h_n\}_{n\in\mathbb{N}^*} H = { h n } n ∈ N ∗ be the continuous sections associated with ξ ∞ \xi_\infty ξ ∞ and η ∞ \eta_\infty η ∞ defined by Equation (8) and Equation (11) (Theorem 4.4). Clearly, f ∞ = ξ ∞ f_\infty=\xi_\infty f ∞ = ξ ∞ , h ∞ = η ∞ h_\infty=\eta_\infty h ∞ = η ∞ , and by Equation (13) ,
∥ f n ∥ 2 ≤ ∥ ξ ∞ ∥ 2 and ∥ h n ∥ 2 ≤ ∥ η ∞ ∥ 2 ( n ∈ N ∗ ) . (17) \|f_n\|_2\leq\|\xi_\infty\|_2
\quad\text{and}\quad
\|h_n\|_2\leq\|\eta_\infty\|_2
\qquad(n\in\mathbb{N}^*). \tag*{(17)} ∥ f n ∥ 2 ≤ ∥ ξ ∞ ∥ 2 and ∥ h n ∥ 2 ≤ ∥ η ∞ ∥ 2 ( n ∈ N ∗ ) . (17)
Let v : = Λ F , H ∈ A ( G ) v:=\Lambda_{F,H}\in A(G) v := Λ F , H ∈ A ( G ) . Clearly, R ∞ ( u − v ) = 0 R_\infty(u-v)=0 R ∞ ( u − v ) = 0 , and by Theorem 5.1, ∥ R n ( u − v ) ∥ A ( G n ) → 0 \|R_n(u-v)\|_{A(G_n)}\to0 ∥ R n ( u − v ) ∥ A ( G n ) → 0 . Let ε > 0 \varepsilon>0 ε > 0 be arbitrary, and fix N ∈ N N\in\mathbb{N} N ∈ N such that
∥ R n ( u − v ) ∥ A ( G n ) < ε for all n ≥ N . (18) \|R_n(u-v)\|_{A(G_n)}<\varepsilon
\qquad\text{for all }n\geq N. \tag*{(18)} ∥ R n ( u − v ) ∥ A ( G n ) < ε for all n ≥ N . (18)
Next, we adjust the continuous sections F F F and H H H so that the corresponding coefficient function matches u u u on the first N − 1 N-1 N − 1 fibers. Namely, we define F ′ = { f n ′ } n ∈ N ∗ F'=\{f'_n\}_{n\in\mathbb{N}^*} F ′ = { f n ′ } n ∈ N ∗ and H ′ = { h n ′ } n ∈ N ∗ H'=\{h'_n\}_{n\in\mathbb{N}^*} H ′ = { h n ′ } n ∈ N ∗ by
f n ′ : = { ξ n , n < N , f n , n ∈ [ N , ∞ ] , h n ′ : = { η n , n < N , h n , n ∈ [ N , ∞ ] . f'_n:=
\begin{cases}
\xi_n, & n<N,\\
f_n, & n\in[N,\infty],
\end{cases}
\qquad
h'_n:=
\begin{cases}
\eta_n, & n<N,\\
h_n, & n\in[N,\infty].
\end{cases} f n ′ := { ξ n , f n , n < N , n ∈ [ N , ∞ ] , h n ′ := { η n , h n , n < N , n ∈ [ N , ∞ ] .
Clearly, F ′ F' F ′ and H ′ H' H ′ are continuous sections, as they eventually match F F F and H H H respectively. Let w : = Λ F ′ , H ′ w:=\Lambda_{F',H'} w := Λ F ′ , H ′ . By Theorem 3.6 [3 ] w ∈ A ( G ) w\in A(G) w ∈ A ( G ) , and by Equation (17) ,
∥ w ∥ A ( G ) ≤ ∥ F ′ ∥ Δ ∥ H ′ ∥ Δ ≤ S . (19) \|w\|_{A(G)}\leq\|F'\|_\Delta\|H'\|_\Delta\leq S. \tag*{(19)} ∥ w ∥ A ( G ) ≤ ∥ F ′ ∥ Δ ∥ H ′ ∥ Δ ≤ S . (19)
Next, we estimate ∥ u − w ∥ A ( G ) \lVert u-w\rVert_{A(G)} ∥ u − w ∥ A ( G ) . By construction of w w w , we have R ∞ ( u − w ) = 0 R_{\infty}(u-w)=0 R ∞ ( u − w ) = 0 . So, by the first part of the proof, we have
∥ u − w ∥ A ( G ) = sup n ∈ N ∗ ∥ R n ( u − w ) ∥ A ( G n ) = sup n ∈ [ N , ∞ ] ∥ R n ( u ) − R n ( w ) ∥ A ( G n ) = sup n ∈ [ N , ∞ ] ∥ R n ( u ) − R n ( v ) ∥ A ( G n ) , \lVert u-w\rVert_{A(G)}=\sup_{n\in\mathbb{N}^{*}}\lVert R_n(u-w)\rVert_{A(G_n)}
=\sup_{n\in[N,\infty]}\lVert R_n(u)-R_n(w)\rVert_{A(G_n)}
=\sup_{n\in[N,\infty]}\lVert R_n(u)-R_n(v)\rVert_{A(G_n)}, ∥ u − w ∥ A ( G ) = n ∈ N ∗ sup ∥ R n ( u − w ) ∥ A ( G n ) = n ∈ [ N , ∞ ] sup ∥ R n ( u ) − R n ( w ) ∥ A ( G n ) = n ∈ [ N , ∞ ] sup ∥ R n ( u ) − R n ( v ) ∥ A ( G n ) ,
so by Equation (18) , we have ∥ u − w ∥ A ( G ) ≤ ϵ \lVert u-w\rVert_{A(G)}\leq\epsilon ∥ u − w ∥ A ( G ) ≤ ϵ . Putting this fact together with Equation (19) completes the proof, as for each ϵ > 0 \epsilon>0 ϵ > 0 , we have
∥ u ∥ A ( G ) ≤ ∥ u − w ∥ A ( G ) + ∥ w ∥ A ( G ) ≤ S + ϵ . \lVert u\rVert_{A(G)}\leq\lVert u-w\rVert_{A(G)}+\lVert w\rVert_{A(G)}\leq S+\epsilon. ∥ u ∥ A ( G ) ≤ ∥ u − w ∥ A ( G ) + ∥ w ∥ A ( G ) ≤ S + ϵ .
□ \square □
Remark 5.3. For a locally compact group, the norm of an element of A ( G ) A(G) A ( G ) can be obtained using only coefficient functions of the left regular representation. An analogous result for groupoids would, in particular, allow the Fourier algebra to be defined without first introducing the Fourier–Stieltjes algebra. We were not able to extend this group property to groupoids (or even HLS groupoids) in general. However, as an easy corollary of the previous result, we observe that for every u ∈ A ( G ) u\in A(G) u ∈ A ( G ) with R ∞ ( u ) = 0 R_{\infty}(u)=0 R ∞ ( u ) = 0 , we have
∥ u ∥ A ( G ) = inf { ∥ F ∥ Δ ∥ G ∥ Δ : u = Λ F , G } . \lVert u\rVert_{A(G)}=\inf\{\lVert F\rVert_{\Delta}\lVert G\rVert_{\Delta}:u=\Lambda_{F,G}\}. ∥ u ∥ A ( G ) = inf {∥ F ∥ Δ ∥ G ∥ Δ : u = Λ F , G } .
Indeed, in the proof of Theorem 5.2, we constructed sections F , H ∈ Γ F,H\in\Gamma F , H ∈ Γ such that Λ F , H = u \Lambda_{F,H}=u Λ F , H = u , and
∥ u ∥ A ( G ) ≤ ∥ F ∥ Δ ∥ H ∥ Δ = sup n ∈ N ∥ R n ( u ) ∥ A ( G n ) ≤ ∥ u ∥ A ( G ) , \lVert u\rVert_{A(G)}\leq\lVert F\rVert_{\Delta}\lVert H\rVert_{\Delta}=\sup_{n\in\mathbb{N}}\lVert R_n(u)\rVert_{A(G_n)}\leq\lVert u\rVert_{A(G)}, ∥ u ∥ A ( G ) ≤ ∥ F ∥ Δ ∥ H ∥ Δ = n ∈ N sup ∥ R n ( u ) ∥ A ( G n ) ≤ ∥ u ∥ A ( G ) ,
where the final inequality follows from contractility of restriction maps R n R_n R n . Thus, equality holds throughout, proving the desired statement.
We finish this section by presenting another interesting applications of Theorem 4.2 and Theorem 5.2 as follows. We first recall one more definition from Banach algebra theory.
Definition 5.4. Let A A A be a Banach algebra. A left (respectively right ) approximate identity for A A A is a net { e α } α ∈ I \{e_{\alpha}\}_{\alpha\in I} { e α } α ∈ I in A A A such that
lim α ∥ e α a − a ∥ = 0 ( respectively lim α ∥ a e α − a ∥ = 0 ) \lim_{\alpha}\lVert e_{\alpha}a-a\rVert=0\qquad\left(\text{respectively }\lim_{\alpha}\lVert ae_{\alpha}-a\rVert=0\right) α lim ∥ e α a − a ∥ = 0 ( respectively α lim ∥ a e α − a ∥ = 0 ) for all a ∈ A a\in A a ∈ A .
A net that is both a left and a right approximate identity is called a (two-sided) approximate identity . An approximate identity { e α } α ∈ I \{e_{\alpha}\}_{\alpha\in I} { e α } α ∈ I is bounded if there is a constant C ≥ 1 C\geq1 C ≥ 1 with
sup α ∈ I ∥ e α ∥ ≤ C . \sup_{\alpha\in I}\lVert e_{\alpha}\rVert\leq C. α ∈ I sup ∥ e α ∥ ≤ C . In this case { e α } α ∈ I \{e_{\alpha}\}_{\alpha\in I} { e α } α ∈ I is called a bounded approximate identity with bound C C C .
Corollary 5.5 (Leptin’s theorem for HLS groupoids). Let G : = ⨆ n ∈ N ∗ G n G:=\bigsqcup_{n\in\mathbb{N}^{*}}G_n G := ⨆ n ∈ N ∗ G n be an HLS groupoid. Then A ( G ) A(G) A ( G ) has a bounded approximate identity if and only if G ∞ G_{\infty} G ∞ is amenable. Moreover, in this case A ( G ) A(G) A ( G ) has a bounded approximate identity bounded by 1.
Proof. Suppose that { e α } α ∈ A \{e_{\alpha}\}_{\alpha\in A} { e α } α ∈ A is a bounded approximate identity for A ( G ) A(G) A ( G ) . Since R ∞ R_{\infty} R ∞ is a contractive algebra homomorphism, { R ∞ ( e α ) } α ∈ A \{R_{\infty}(e_{\alpha})\}_{\alpha\in A} { R ∞ ( e α ) } α ∈ A is a bounded approximate identity for A ( G ∞ ) A(G_{\infty}) A ( G ∞ ) , and G ∞ G_{\infty} G ∞ is amenable by Leptin’s theorem [19 ] . This proves the forward direction.
To prove the backward direction, suppose that G ∞ G_\infty G ∞ is amenable. By Leptin’s theorem [19 ] , A ( G ∞ ) A(G_\infty) A ( G ∞ ) has a bounded approximate identity { φ β } β ∈ B \{\varphi_\beta\}_{\beta\in B} { φ β } β ∈ B with ∥ φ β ∥ A ( G ∞ ) ≤ 1 \|\varphi_\beta\|_{A(G_\infty)}\leq1 ∥ φ β ∥ A ( G ∞ ) ≤ 1 for all β ∈ B \beta\in B β ∈ B . By Theorem 4.2 , for each β ∈ B \beta\in B β ∈ B there exists v β ∈ A ( G ) v_\beta\in A(G) v β ∈ A ( G ) such that
R ∞ ( v β ) = φ β and ∥ v β ∥ A ( G ) = ∥ φ β ∥ A ( G ∞ ) ≤ 1. R_\infty(v_\beta)=\varphi_\beta\quad\text{and}\quad\|v_\beta\|_{A(G)}=\|\varphi_\beta\|_{A(G_\infty)}\leq1. R ∞ ( v β ) = φ β and ∥ v β ∥ A ( G ) = ∥ φ β ∥ A ( G ∞ ) ≤ 1.
Next, we define a new net of functions in A ( G ) A(G) A ( G ) indexed over the directed set B × N B\times\mathbb{N} B × N that will act as a bounded approximate identity for A ( G ) A(G) A ( G ) . For β ∈ B \beta\in B β ∈ B and N ∈ N N\in\mathbb{N} N ∈ N , define
e β , N : = v β + ∑ n = 1 N − 1 ι n ( 1 G n − R n ( v β ) ) , e_{\beta,N}:=v_\beta+\sum_{n=1}^{N-1}\iota_n(1_{G_n}-R_n(v_\beta)), e β , N := v β + n = 1 ∑ N − 1 ι n ( 1 G n − R n ( v β )) ,
where ι n : A ( G n ) → A ( G ) \iota_n:A(G_n)\to A(G) ι n : A ( G n ) → A ( G ) is the embedding of Theorem 4.1, and 1 G n 1_{G_n} 1 G n denotes the constant function on G n G_n G n attaining 1 1 1 everywhere. Since G n G_n G n is finite, we know 1 G n ∈ A ( G n ) 1_{G_n}\in A(G_n) 1 G n ∈ A ( G n ) and ∥ 1 G n ∥ A ( G n ) = 1 \|1_{G_n}\|_{A(G_n)}=1 ∥ 1 G n ∥ A ( G n ) = 1 . So, e β , N ∈ A ( G ) e_{\beta,N}\in A(G) e β , N ∈ A ( G ) . Moreover, applying Theorem 5.2, we get
∥ e β , N ∥ A ( G ) = max { max n ∈ [ 1 , N − 1 ] ∥ 1 G n ∥ A ( G n ) , sup n ∈ [ N , ∞ ] ∥ R n ( v β ) ∥ A ( G n ) } ≤ 1 , \|e_{\beta,N}\|_{A(G)}=\max\left\{\max_{n\in[1,N-1]}\|1_{G_n}\|_{A(G_n)},\sup_{n\in[N,\infty]}\|R_n(v_\beta)\|_{A(G_n)}\right\}\leq1, ∥ e β , N ∥ A ( G ) = max { n ∈ [ 1 , N − 1 ] max ∥ 1 G n ∥ A ( G n ) , n ∈ [ N , ∞ ] sup ∥ R n ( v β ) ∥ A ( G n ) } ≤ 1 ,
since ∥ v β ∥ A ( G ) ≤ 1 \|v_\beta\|_{A(G)}\leq1 ∥ v β ∥ A ( G ) ≤ 1 . So { e β , N } ( β , N ) ∈ B × N \{e_{\beta,N}\}_{(\beta,N)\in B\times\mathbb{N}} { e β , N } ( β , N ) ∈ B × N is a bounded net of functions in A ( G ) A(G) A ( G ) , with bound 1 1 1 .
To complete the proof, it only remains to show that this net is an approximate identity for A ( G ) A(G) A ( G ) . Let u ∈ A ( G ) u\in A(G) u ∈ A ( G ) be arbitrary. First note that by Theorem 5.2 and the definition of e β , N e_{\beta,N} e β , N , for every β ∈ B \beta\in B β ∈ B , and N ∈ N N\in\mathbb{N} N ∈ N we have
∥ u e β , N − u ∥ A ( G ) = sup n ∈ [ N , ∞ ] ∥ R n ( u e β , N − u ) ∥ A ( G n ) . (20) \|u e_{\beta,N}-u\|_{A(G)}=\sup_{n\in[N,\infty]}\|R_n(u e_{\beta,N}-u)\|_{A(G_n)}. \tag*{(20)} ∥ u e β , N − u ∥ A ( G ) = n ∈ [ N , ∞ ] sup ∥ R n ( u e β , N − u ) ∥ A ( G n ) . (20)
Let ϵ > 0 \epsilon>0 ϵ > 0 be arbitrary. Since { φ β } β ∈ B \{\varphi_\beta\}_{\beta\in B} { φ β } β ∈ B is an approximate identity for A ( G ∞ ) A(G_\infty) A ( G ∞ ) , there exists β 0 ∈ B \beta_0\in B β 0 ∈ B such that for all β ≥ β 0 \beta\geq\beta_0 β ≥ β 0 and every N ∈ N N\in\mathbb{N} N ∈ N , we have
∥ R ∞ ( u e β , N − u ) ∥ A ( G ∞ ) = ∥ R ∞ ( u ) φ β − R ∞ ( u ) ∥ A ( G ∞ ) < ϵ 2 , (21) \|R_\infty(u e_{\beta,N}-u)\|_{A(G_\infty)}
=\|R_\infty(u)\varphi_\beta-R_\infty(u)\|_{A(G_\infty)}
<\frac{\epsilon}{2}, \tag*{(21)} ∥ R ∞ ( u e β , N − u ) ∥ A ( G ∞ ) = ∥ R ∞ ( u ) φ β − R ∞ ( u ) ∥ A ( G ∞ ) < 2 ϵ , (21)
where in the first equality we used the fact that φ β = R ∞ ( v β ) = R ∞ ( e β , N ) \varphi_\beta=R_\infty(v_\beta)=R_\infty(e_{\beta,N}) φ β = R ∞ ( v β ) = R ∞ ( e β , N ) . Applying Theorem 5.1 to u e β , 1 − u u e_{\beta,1}-u u e β , 1 − u , there exists N 0 ∈ N N_0\in\mathbb{N} N 0 ∈ N such that for all n ≥ N 0 n\geq N_0 n ≥ N 0 , we have
∥ R n ( u e β , 1 − u ) ∥ A ( G n ) < ϵ . (22) \|R_n(u e_{\beta,1}-u)\|_{A(G_n)}<\epsilon. \tag*{(22)} ∥ R n ( u e β , 1 − u ) ∥ A ( G n ) < ϵ . (22)
Combining Equation (20) , Equation (21) and Equation (22) , we get ∥ u e β , N − u ∥ A ( G ) < ϵ \|u e_{\beta,N}-u\|_{A(G)}<\epsilon ∥ u e β , N − u ∥ A ( G ) < ϵ for all N ≥ N 0 N\geq N_0 N ≥ N 0 and β ≥ β 0 \beta\geq\beta_0 β ≥ β 0 , so { u β , n } ( β , n ) ∈ B × N \{u_{\beta,n}\}_{(\beta,n)\in B\times\mathbb{N}} { u β , n } ( β , n ) ∈ B × N forms an approximate identity for A ( G ) A(G) A ( G ) .
Gelfand spectrum of the Fourier algebra
Let G G G be a locally compact group. The Gelfand spectrum of A ( G ) A(G) A ( G ) , denoted σ ( A ( G ) ) \sigma(A(G)) σ ( A ( G )) , is the set of all nonzero multiplicative linear functionals on A ( G ) A(G) A ( G ) . A classical result of Eymard shows that
σ ( A ( G ) ) ≅ G \sigma(A(G))\cong G σ ( A ( G )) ≅ G
homeomorphically [11 Theorem 3.34] . In [27 ] , Paterson remarks that it is unknown whether the analogous statement holds for arbitrary groupoids. The discrete groupoid case was established in [6 ] . In this section, we answer Paterson’s question for HLS groupoids.
Theorem 6.1. Let G : = ⨆ n ∈ N ∗ G n \mathcal{G}:=\bigsqcup_{n\in\mathbb{N}^*}G_n G := ⨆ n ∈ N ∗ G n be an HLS groupoid. Then σ ( A ( G ) ) \sigma(A(\mathcal{G})) σ ( A ( G )) is homeomorphic to G \mathcal{G} G .
Proof. Note that G \mathcal{G} G is a locally compact and Hausdorff space, and A ( G ) A(\mathcal{G}) A ( G ) is a subalgebra of C 0 ( G ) C_0(\mathcal{G}) C 0 ( G ) that strongly separates points of G \mathcal{G} G , i.e., for every distinct x , y ∈ G x,y \in\mathcal{G} x , y ∈ G there exists f ∈ A ( G ) f \in A(\mathcal{G}) f ∈ A ( G ) with f ( x ) ≠ f ( y ) f(x) \ne f(y) f ( x ) = f ( y ) . Thus, by [32 Theorem 3.2.4] , the natural embedding G → σ ( A ( G ) ) \mathcal{G} \to\sigma(A(\mathcal{G})) G → σ ( A ( G )) , given by point evaluation, is a homeomorphism into σ ( A ( G ) ) \sigma(A(\mathcal{G})) σ ( A ( G )) . What remains is to demonstrate that any multiplicative, linear functional on A ( G ) A(\mathcal{G}) A ( G ) must correspond to point evaluation for a given groupoid element.
To that end, let Φ : A ( G ) → C \Phi: A(\mathcal{G}) \to\mathbb{C} Φ : A ( G ) → C be a non-zero, bounded, multiplicative linear functional. For each n ∈ N n \in\mathbb{N} n ∈ N , let ι n : A ( G n ) ↪ A ( G ) \iota_n : A(\mathcal{G}_n) \hookrightarrow A(\mathcal{G}) ι n : A ( G n ) ↪ A ( G ) be the embedding as in Theorem 4.1. First, we observe that Φ ∣ ι n ( A ( G n ) ) ≠ 0 \Phi|_{\iota_n(A(\mathcal{G}_n))} \ne0 Φ ∣ ι n ( A ( G n )) = 0 for at most one n ∈ N n \in\mathbb{N} n ∈ N . Indeed, if there were distinct n , m n,m n , m such that Φ ( u ) ≠ 0 \Phi(u) \ne0 Φ ( u ) = 0 and Φ ( v ) ≠ 0 \Phi(v) \ne0 Φ ( v ) = 0 for u ∈ ι n ( A ( G n ) ) u \in\iota_n(A(\mathcal{G}_n)) u ∈ ι n ( A ( G n )) and v ∈ ι m ( A ( G m ) ) v \in\iota_m(A(\mathcal{G}_m)) v ∈ ι m ( A ( G m )) , then since u v = 0 uv=0 uv = 0 , we necessarily have
0 = Φ ( 0 ) = Φ ( u v ) = Φ ( u ) Φ ( v ) 0 = \Phi(0) = \Phi(uv) = \Phi(u)\Phi(v) 0 = Φ ( 0 ) = Φ ( uv ) = Φ ( u ) Φ ( v )
a contradiction.
To prove the result, we consider two cases:
Claim 6.2. If u ∈ A ( G ) u \in A(\mathcal{G}) u ∈ A ( G ) with u ∣ G n 0 = 0 u|_{\mathcal{G}_{n_0}}=0 u ∣ G n 0 = 0 , then Φ ( u ) = 0 \Phi(u)=0 Φ ( u ) = 0 .
Proof of Theorem 6.2: Recall that v ∈ A ( G n 0 ) v \in A(\mathcal{G}_{n_0}) v ∈ A ( G n 0 ) was chosen such that Φ ( ι n 0 ( v ) ) ≠ 0 \Phi(\iota_{n_0}(v)) \ne0 Φ ( ι n 0 ( v )) = 0 . Clearly u ι n 0 ( v ) = 0 u\iota_{n_0}(v)=0 u ι n 0 ( v ) = 0 , so we have
0 = Φ ( 0 ) = Φ ( ι n 0 ( v ) u ) = Φ ( ι n 0 ( v ) ) Φ ( u ) . 0 = \Phi(0) = \Phi(\iota_{n_0}(v)u) = \Phi(\iota_{n_0}(v))\Phi(u). 0 = Φ ( 0 ) = Φ ( ι n 0 ( v ) u ) = Φ ( ι n 0 ( v )) Φ ( u ) .
Since Φ ( ι n 0 ( v ) ) \Phi(\iota_{n_0}(v)) Φ ( ι n 0 ( v )) is non-zero, we must have that Φ ( u ) = 0 \Phi(u)=0 Φ ( u ) = 0 , demonstrating the claim.
To conclude this case, we decompose f ∈ A ( G ) f \in A(\mathcal{G}) f ∈ A ( G ) in the form f = f − ι n 0 ( R n 0 ( f ) ) + ι n 0 ( R n 0 ( f ) ) f=f-\iota_{n_0}(R_{n_0}(f))+\iota_{n_0}(R_{n_0}(f)) f = f − ι n 0 ( R n 0 ( f )) + ι n 0 ( R n 0 ( f )) . So,
Φ ( f ) = Φ ( f − ι n 0 ( R n 0 ( f ) ) ) + Φ ( ι n 0 ( R n 0 ( f ) ) ) = Φ ( f − ι n 0 ( R n 0 ( f ) ) ) + ψ ( R n 0 ( f ) ) . \Phi(f) = \Phi(f-\iota_{n_0}(R_{n_0}(f)))+\Phi(\iota_{n_0}(R_{n_0}(f))) = \Phi(f-\iota_{n_0}(R_{n_0}(f)))+\psi(R_{n_0}(f)). Φ ( f ) = Φ ( f − ι n 0 ( R n 0 ( f ))) + Φ ( ι n 0 ( R n 0 ( f ))) = Φ ( f − ι n 0 ( R n 0 ( f ))) + ψ ( R n 0 ( f )) .
The first term is 0 by Theorem 6.2 and the second one is f ( n 0 , x 0 ) f(n_0,x_0) f ( n 0 , x 0 ) , proving that Φ \Phi Φ is just the evaluation functional at the point ( n 0 , x 0 ) ∈ G (n_0,x_0) \in\mathcal{G} ( n 0 , x 0 ) ∈ G . This proves the result for Case 1.
Case 2: Suppose that for every n ∈ N n \in\mathbb{N} n ∈ N , Φ ∣ ι n ( A ( G n ) ) = 0 \Phi|_{\iota_n(A(\mathcal{G}_n))}=0 Φ ∣ ι n ( A ( G n )) = 0 . Let 1 \mathbb{1} 1 denote the constant function 1 on G \mathcal{G} G , and 1 n : G → C \mathbb{1}_n:\mathcal{G}\to\mathbb{C} 1 n : G → C be the characteristic function of G n \mathcal{G}_n G n . Let u ∈ A ( G ) u \in A(\mathcal{G}) u ∈ A ( G ) be an arbitrary element. It is easy to see that ι k ( R k ( u ) ) = 1 k u \iota_k(R_k(u))=\mathbb{1}_k u ι k ( R k ( u )) = 1 k u . Using Theorem 4.1 and the fact that each R k : A ( G ) → A ( G k ) R_k:A(\mathcal{G})\to A(\mathcal{G}_k) R k : A ( G ) → A ( G k ) is a homomorphism, for every n ≥ 1 n \ge1 n ≥ 1 , the functions 1 n \mathbb{1}_n 1 n and u n = ∑ k = 1 n ι k ( R k ( u ) ) u_n=\sum_{k=1}^{n}\iota_k(R_k(u)) u n = ∑ k = 1 n ι k ( R k ( u )) belong to A ( G ) A(\mathcal{G}) A ( G ) . Since Φ ∣ ι k ( A ( G k ) ) = 0 \Phi|_{\iota_k(A(\mathcal{G}_k))}=0 Φ ∣ ι k ( A ( G k )) = 0 for k ∈ N k \in\mathbb{N} k ∈ N , we have
Φ ( u ) = Φ ( u − u n ) + Φ ( ∑ k = 1 n ι k ( R k ( u ) ) ) = Φ ( u − u n ) , ∀ n ∈ N . (23) \Phi(u)=\Phi(u-u_n)+\Phi\left(\sum_{k=1}^{n}\iota_k(R_k(u))\right)=\Phi(u-u_n),\quad\forall n\in\mathbb{N}. \tag*{(23)} Φ ( u ) = Φ ( u − u n ) + Φ ( k = 1 ∑ n ι k ( R k ( u )) ) = Φ ( u − u n ) , ∀ n ∈ N . (23)
Since A ( G ) ∩ C c ( G ) A(\mathcal{G})\cap C_c(\mathcal{G}) A ( G ) ∩ C c ( G ) is dense in A ( G ) A(\mathcal{G}) A ( G ) and Φ \Phi Φ is nonzero, there exists a compactly supported v ∈ A ( G ) v \in A(\mathcal{G}) v ∈ A ( G ) satisfying Φ ( v ) ≠ 0 \Phi(v)\ne0 Φ ( v ) = 0 . By Theorem 3.3, Items 2 and 3, there exist x 1 , … , x m ∈ G ∞ x_1,\ldots,x_m\in\mathcal{G}_\infty x 1 , … , x m ∈ G ∞ and N ∈ N N\in\mathbb{N} N ∈ N such that
supp ( v − v N ) ⊆ ⋃ i = 1 m ray N ( x i ) , \operatorname{supp}(v-v_N)\subseteq\bigcup_{i=1}^{m}\operatorname{ray}_N(x_i), supp ( v − v N ) ⊆ i = 1 ⋃ m ray N ( x i ) ,
where v N = ∑ n = 1 N 1 n v v_N=\sum_{n=1}^{N}\mathbb{1}_n v v N = ∑ n = 1 N 1 n v . Using (23) , we have Φ ( v − v N ) ≠ 0 \Phi(v-v_N)\ne0 Φ ( v − v N ) = 0 . So, replacing v v v by v − v N v-v_N v − v N if necessary, there exists v ∈ A ( G ) v\in A(\mathcal{G}) v ∈ A ( G ) satisfying the following:
v ∈ A ( G ) ∩ C c ( G ) , Φ ( v ) ≠ 0 , supp ( v ) ⊆ ⋃ i = 1 m ray N ( x i ) . (24) v\in A(\mathcal{G})\cap C_c(\mathcal{G}),\quad\Phi(v)\ne0,\quad\operatorname{supp}(v)\subseteq\bigcup_{i=1}^{m}\operatorname{ray}_N(x_i). \tag*{(24)} v ∈ A ( G ) ∩ C c ( G ) , Φ ( v ) = 0 , supp ( v ) ⊆ i = 1 ⋃ m ray N ( x i ) . (24)
Moreover, v = v 1 ray N ( x 1 ) + ⋯ + v 1 ray N ( x m ) v=v\mathbb{1}_{\operatorname{ray}_N(x_1)}+\cdots+v\mathbb{1}_{\operatorname{ray}_N(x_m)} v = v 1 ray N ( x 1 ) + ⋯ + v 1 ray N ( x m ) , and each summand belongs to A ( G ) A(\mathcal{G}) A ( G ) by Theorem 3.7. Since Φ ( v 1 ray N ( x 1 ) ) + ⋯ + Φ ( v 1 ray N ( x m ) ) ≠ 0 \Phi(v\mathbb{1}_{\operatorname{ray}_N(x_1)})+\cdots+\Phi(v\mathbb{1}_{\operatorname{ray}_N(x_m)})\ne0 Φ ( v 1 ray N ( x 1 ) ) + ⋯ + Φ ( v 1 ray N ( x m ) ) = 0 , without loss of generality we can assume that Φ ( v 1 ray N ( x 1 ) ) ≠ 0 \Phi(v\mathbb{1}_{\operatorname{ray}_N(x_1)})\ne0 Φ ( v 1 ray N ( x 1 ) ) = 0 . Again replacing v v v by v 1 ray N ( x 1 ) v\mathbb{1}_{\operatorname{ray}_N(x_1)} v 1 ray N ( x 1 ) if necessary, we can assume that v v v is supported in the single ray ray N ( x 1 ) \operatorname{ray}_N(x_1) ray N ( x 1 ) .
Claim 6.3. Φ ( 1 ray 1 ( x 1 ) ) = 1 \Phi(\mathbf{1}_{\operatorname{ray}_1(x_1)})=1 Φ ( 1 ray 1 ( x 1 ) ) = 1 and Φ ( 1 ray 1 ( x ) ) = 0 \Phi(\mathbf{1}_{\operatorname{ray}_1(x)})=0 Φ ( 1 ray 1 ( x ) ) = 0 for x ≠ x 1 x\ne x_1 x = x 1 . Consequently, if u ∈ A ( G ) ∩ C c ( G ) u\in A(\mathcal{G})\cap C_c(\mathcal{G}) u ∈ A ( G ) ∩ C c ( G ) and ( ∞ , x 1 ) ∉ supp ( u ) (\infty,x_1)\notin\operatorname{supp}(u) ( ∞ , x 1 ) ∈ / supp ( u ) then Φ ( u ) = 0 \Phi(u)=0 Φ ( u ) = 0 .
Proof of Theorem 6.3: Since v v v is supported in ray 1 ( x 1 ) \operatorname{ray}_1(x_1) ray 1 ( x 1 ) , we have that
0 ≠ Φ ( v ) = Φ ( v 1 ray 1 ( x 1 ) ) = Φ ( v ) Φ ( 1 ray 1 ( x 1 ) ) , 0\ne\Phi(v)=\Phi(v\mathbf{1}_{\operatorname{ray}_1(x_1)})=\Phi(v)\Phi(\mathbf{1}_{\operatorname{ray}_1(x_1)}), 0 = Φ ( v ) = Φ ( v 1 ray 1 ( x 1 ) ) = Φ ( v ) Φ ( 1 ray 1 ( x 1 ) ) ,
which implies that Φ ( 1 ray 1 ( x 1 ) ) = 1 \Phi(\mathbf{1}_{\operatorname{ray}_1(x_1)})=1 Φ ( 1 ray 1 ( x 1 ) ) = 1 . Moreover, for every x ∈ G ∞ x\in G_\infty x ∈ G ∞ with x ≠ x 1 x\ne x_1 x = x 1 , there exits K ∈ N K\in\mathbb{N} K ∈ N for which we have ray K ( x ) ∩ ray K ( x 1 ) = ∅ \operatorname{ray}_K(x)\cap\operatorname{ray}_K(x_1)=\varnothing ray K ( x ) ∩ ray K ( x 1 ) = ∅ , implying that ( v − v K ) 1 ray K ( x 1 ) = 0 (v-v_K)\mathbf{1}_{\operatorname{ray}_K(x_1)}=0 ( v − v K ) 1 ray K ( x 1 ) = 0 . So,
0 = Φ ( ( v − v K ) 1 ray K ( x 1 ) ) = Φ ( v − v K ) Φ ( 1 ray K ( x ) ) = Φ ( v ) Φ ( 1 ray 1 ( x ) ) , 0=\Phi((v-v_K)\mathbf{1}_{\operatorname{ray}_K(x_1)})=\Phi(v-v_K)\Phi(\mathbf{1}_{\operatorname{ray}_K(x)})=\Phi(v)\Phi(\mathbf{1}_{\operatorname{ray}_1(x)}), 0 = Φ (( v − v K ) 1 ray K ( x 1 ) ) = Φ ( v − v K ) Φ ( 1 ray K ( x ) ) = Φ ( v ) Φ ( 1 ray 1 ( x ) ) ,
where we used (23) in the last equality. Since Φ ( v ) ≠ 0 \Phi(v)\ne0 Φ ( v ) = 0 , we must have Φ ( 1 ray 1 ( x ) ) = 0 \Phi(\mathbf{1}_{\operatorname{ray}_1(x)})=0 Φ ( 1 ray 1 ( x ) ) = 0 .
Next, let u ∈ A ( G ) ∩ C c ( G ) u\in A(\mathcal{G})\cap C_c(\mathcal{G}) u ∈ A ( G ) ∩ C c ( G ) be such that ( ∞ , x 1 ) ∉ supp ( u ) (\infty,x_1)\notin\operatorname{supp}(u) ( ∞ , x 1 ) ∈ / supp ( u ) . Applying Theorem 3.3 again, there exists N ′ ∈ N N'\in\mathbb{N} N ′ ∈ N such that u − u N ′ u-u_{N'} u − u N ′ is supported in finitely many disjoint rays whose intersection with the infinite fiber G ∞ G_\infty G ∞ lies inside supp ( u ) \operatorname{supp}(u) supp ( u ) . That is, there exist z 1 , … , z t ∈ G ∞ z_1,\ldots,z_t\in G_\infty z 1 , … , z t ∈ G ∞ so that
supp ( u − u N ′ ) ⊆ ⨆ i = 1 t ray N ′ ( z i ) and ( ∞ , z i ) ∈ supp ( u ) . \operatorname{supp}(u-u_{N'})\subseteq\bigsqcup_{i=1}^{t}\operatorname{ray}_{N'}(z_i)\quad\text{and}\quad(\infty,z_i)\in\operatorname{supp}(u). supp ( u − u N ′ ) ⊆ i = 1 ⨆ t ray N ′ ( z i ) and ( ∞ , z i ) ∈ supp ( u ) .
Consequently u − u N ′ = ( u − u N ′ ) ( 1 ray 1 ( z 1 ) + ⋯ + 1 ray 1 ( z t ) ) u-u_{N'}=(u-u_{N'})(\mathbf{1}_{\operatorname{ray}_1(z_1)}+\cdots+\mathbf{1}_{\operatorname{ray}_1(z_t)}) u − u N ′ = ( u − u N ′ ) ( 1 ray 1 ( z 1 ) + ⋯ + 1 ray 1 ( z t ) ) and z i ≠ x 1 z_i\ne x_1 z i = x 1 . So, using the first part of the claim, we have
Φ ( u ) = Φ ( u N ′ ) + Φ ( u − u N ′ ) ( Φ ( 1 ray 1 ( z 1 ) ) + ⋯ + Φ ( 1 ray 1 ( z t ) ) ) = 0 , \Phi(u)=\Phi(u_{N'})+\Phi(u-u_{N'})(\Phi(\mathbf{1}_{\operatorname{ray}_1(z_1)})+\cdots+\Phi(\mathbf{1}_{\operatorname{ray}_1(z_t)}))=0, Φ ( u ) = Φ ( u N ′ ) + Φ ( u − u N ′ ) ( Φ ( 1 ray 1 ( z 1 ) ) + ⋯ + Φ ( 1 ray 1 ( z t ) )) = 0 ,
which finishes the proof of the claim.
Claim 6.4. If u ∈ A ( G ) ∩ C c ( G ) u\in A(\mathcal{G})\cap C_c(\mathcal{G}) u ∈ A ( G ) ∩ C c ( G ) and ( ∞ , x 1 ) ∈ supp ( u ) (\infty,x_1)\in\operatorname{supp}(u) ( ∞ , x 1 ) ∈ supp ( u ) then Φ ( u ) = u ( ∞ , x 1 ) \Phi(u)=u(\infty,x_1) Φ ( u ) = u ( ∞ , x 1 ) .
Proof of Theorem 6.4: Let u ~ = u − u ( ∞ , x 1 ) 1 ray 1 ( x 1 ) \widetilde{u}=u-u(\infty,x_1)\mathbf{1}_{\operatorname{ray}_1(x_1)} u = u − u ( ∞ , x 1 ) 1 ray 1 ( x 1 ) . Clearly, u ~ \widetilde{u} u is a compactly supported element of A ( G ) A(\mathcal{G}) A ( G ) as well. Assuming ( ∞ , x 1 ) ∉ supp ( u ~ ) (\infty,x_1)\notin\operatorname{supp}(\widetilde{u}) ( ∞ , x 1 ) ∈ / supp ( u ) , by Theorem 6.3 we have Φ ( u ~ ) = 0 \Phi(\widetilde{u})=0 Φ ( u ) = 0 . This finishes the proof of the current claim under this assumption, since
Φ ( u ) = u ( ∞ , x 1 ) Φ ( 1 ray 1 ( x 1 ) ) = u ( ∞ , x 1 ) . \Phi(u)=u(\infty,x_1)\Phi(\mathbf{1}_{\operatorname{ray}_1(x_1)})=u(\infty,x_1). Φ ( u ) = u ( ∞ , x 1 ) Φ ( 1 ray 1 ( x 1 ) ) = u ( ∞ , x 1 ) .
Next, suppose ( ∞ , x 1 ) ∈ supp ( u ~ ) (\infty,x_1)\in\operatorname{supp}(\widetilde{u}) ( ∞ , x 1 ) ∈ supp ( u ) . By the definition of u ~ \widetilde{u} u , we have u ~ ( ∞ , x 1 ) = 0 \widetilde{u}(\infty,x_1)=0 u ( ∞ , x 1 ) = 0 . Since u ~ \widetilde{u} u is compactly supported, there exist M ∈ N M\in\mathbb{N} M ∈ N and y 1 , … , y s ∈ G ∞ y_1,\ldots,y_s\in G_\infty y 1 , … , y s ∈ G ∞ satisfying
y 1 = x 1 and supp ( u ~ − u ~ M ) ⊆ ⨆ i = 1 s ray M ( y i ) . (25) y_1=x_1\quad\text{and}\quad\operatorname{supp}(\widetilde{u}-\widetilde{u}_M)\subseteq\bigsqcup_{i=1}^{s}\operatorname{ray}_M(y_i). \tag*{(25)} y 1 = x 1 and supp ( u − u M ) ⊆ i = 1 ⨆ s ray M ( y i ) . (25)
Therefore, u ~ − u ~ M = ( u ~ − u ~ M ) ( ∑ i = 1 s 1 ray M ( y i ) ) \widetilde{u}-\widetilde{u}_M=(\widetilde{u}-\widetilde{u}_M)(\sum_{i=1}^{s}\mathbf{1}_{\operatorname{ray}_M(y_i)}) u − u M = ( u − u M ) ( ∑ i = 1 s 1 ray M ( y i ) ) , and applying Equation (23) and Theorem 6.3 again we have
Φ ( u ~ ) = Φ ( u ~ − u ~ M ) = Φ ( ( u ~ − u ~ M ) 1 ray M ( x 1 ) ) . (26) \Phi(\widetilde{u})=\Phi(\widetilde{u}-\widetilde{u}_M)=\Phi((\widetilde{u}-\widetilde{u}_M)\mathbf{1}_{\operatorname{ray}_M(x_1)}). \tag*{(26)} Φ ( u ) = Φ ( u − u M ) = Φ (( u − u M ) 1 ray M ( x 1 ) ) . (26)
For each m ≥ M m\ge M m ≥ M , define w m : = ( u ~ − u ~ m ) 1 ray m ( x 1 ) w_m:=(\widetilde{u}-\widetilde{u}_m)\mathbf{1}_{\operatorname{ray}_m(x_1)} w m := ( u − u m ) 1 ray m ( x 1 ) , and note that w m ∈ A ( G ) w_m\in A(\mathcal{G}) w m ∈ A ( G ) . Since w m w_m w m is supported in ray m ( x 1 ) \operatorname{ray}_m(x_1) ray m ( x 1 ) , we can write it as a coefficient function as follows:
w m = Λ ξ , η , where ξ n = { ( u ( n , q n ( x 1 ) ) − u ( ∞ , x 1 ) ) δ q n ( x 1 ) n ≥ m , 0 n < m , η n = { δ q n ( e ∞ ) n ≥ m , 0 n < m . w_m=\Lambda_{\xi,\eta},\quad\text{where}\quad
\xi_n=
\begin{cases}
(u(n,q_n(x_1))-u(\infty,x_1))\delta_{q_n(x_1)} & n\ge m,\\
0 & n<m,
\end{cases}
\quad
\eta_n=
\begin{cases}
\delta_{q_n(e_\infty)} & n\ge m,\\
0 & n<m.
\end{cases} w m = Λ ξ , η , where ξ n = { ( u ( n , q n ( x 1 )) − u ( ∞ , x 1 )) δ q n ( x 1 ) 0 n ≥ m , n < m , η n = { δ q n ( e ∞ ) 0 n ≥ m , n < m .
Clearly η ∈ Δ \eta\in\Delta η ∈ Δ . The fact that ξ ∈ Δ \xi\in\Delta ξ ∈ Δ follows from the continuity of u u u at ( ∞ , x 1 ) (\infty,x_1) ( ∞ , x 1 ) . From the above representation of w m w_m w m , we get
∥ w m ∥ A ( G ) ≤ ∥ ξ ∥ Δ ∥ η ∥ Δ = sup n ≥ m ∣ u ( n , q n ( x 1 ) ) − u ( ∞ , x 1 ) ∣ . \lVert w_m\rVert_{A(\mathcal{G})}\le\lVert\xi\rVert_\Delta\lVert\eta\rVert_\Delta=\sup_{n\ge m}|u(n,q_n(x_1))-u(\infty,x_1)|. ∥ w m ∥ A ( G ) ≤ ∥ ξ ∥ Δ ∥ η ∥ Δ = n ≥ m sup ∣ u ( n , q n ( x 1 )) − u ( ∞ , x 1 ) ∣.
Combining (23) and (26) , we get
∣ Φ ( u ~ ) ∣ = lim m → ∞ ∣ Φ ( w m ) ∣ ≤ lim n → ∞ ∣ u ( n , q n ( x 1 ) ) − u ( ∞ , x 1 ) ∣ = 0. |\Phi(\widetilde{u})|=\lim_{m\to\infty}|\Phi(w_m)|\le\lim_{n\to\infty}|u(n,q_n(x_1))-u(\infty,x_1)|=0. ∣Φ ( u ) ∣ = m → ∞ lim ∣Φ ( w m ) ∣ ≤ n → ∞ lim ∣ u ( n , q n ( x 1 )) − u ( ∞ , x 1 ) ∣ = 0.
This completes the proof of the claim, as by Theorem 6.3, we have Φ ( 1 r a y 1 ( x 1 ) ) = 1 \Phi(\mathbf{1}_{\mathrm{ray}_1}(x_1))=1 Φ ( 1 ray 1 ( x 1 )) = 1 .
Putting Theorem 6.3 and Theorem 6.4 together concludes Case 2. So far we have proved that there exist n 1 ∈ N ∪ { ∞ } n_1 \in\mathbb{N} \cup\{\infty\} n 1 ∈ N ∪ { ∞ } and x 1 ∈ G n 1 x_1 \in G_{n_1} x 1 ∈ G n 1 satisfying the following:
∀ u ∈ A ( G ) ∩ C c ( G ) , Φ ( u ) = u ( n 1 , q n 1 ( x 1 ) ) . \forall u \in A(\mathcal{G}) \cap C_c(\mathcal{G}), \quad\Phi(u)=u(n_1,q_{n_1}(x_1)). ∀ u ∈ A ( G ) ∩ C c ( G ) , Φ ( u ) = u ( n 1 , q n 1 ( x 1 )) .
It is easy to see that the same equality holds for every element of A ( G ) A(\mathcal{G}) A ( G ) . Indeed, let u ∈ A ( G ) u \in A(\mathcal{G}) u ∈ A ( G ) and consider a sequence { f n } n ⊆ A ( G ) ∩ C c ( G ) \{f_n\}_n \subseteq A(\mathcal{G}) \cap C_c(\mathcal{G}) { f n } n ⊆ A ( G ) ∩ C c ( G ) with ∥ u − f n ∥ A ( G ) → 0 \|u-f_n\|_{A(\mathcal{G})} \to0 ∥ u − f n ∥ A ( G ) → 0 as n → ∞ n \to\infty n → ∞ . By continuity of Φ \Phi Φ , we have
Φ ( u ) = lim n → ∞ Φ ( f n ) = lim n → ∞ f n ( n 1 , q n 1 ( x 1 ) ) = u ( n 1 , q n 1 ( x 1 ) ) , \Phi(u)=\lim_{n\to\infty}\Phi(f_n)=\lim_{n\to\infty}f_n(n_1,q_{n_1}(x_1))=u(n_1,q_{n_1}(x_1)), Φ ( u ) = n → ∞ lim Φ ( f n ) = n → ∞ lim f n ( n 1 , q n 1 ( x 1 )) = u ( n 1 , q n 1 ( x 1 )) ,
where the last equality holds because convergence in A ( G ) A(\mathcal{G}) A ( G ) implies pointwise convergence.
Extending beyond HLS groupoids
Some of the results established in this paper extend beyond HLS groupoids to arbitrary étale groupoids. In this short section, we present these extensions. First, we prove an analogue of Theorem 4.2 for all étale groupoids. Unlike the proof of Theorem 4.2 , however, the argument given here is non-constructive. Second, we show that the spectrum of A ( G ) A(\mathcal{G}) A ( G ) is homeomorphic to G \mathcal{G} G under the map given by point evaluation. This generalizes the result of Theorem 6.1 and answers a question raised by Paterson in [27 Section 6] for the case of étale groupoids. However, Paterson’s question remains open in full generality.
Proposition 7.1. Let G \mathcal{G} G be a (locally compact Hausdorff) étale groupoid, and consider a ∈ G ( 0 ) a \in\mathcal{G}^{(0)} a ∈ G ( 0 ) . Let R a R_a R a denote the restriction map A ( G ) → A ( G a a ) A(\mathcal{G}) \to A(\mathcal{G}_a^a) A ( G ) → A ( G a a ) . Then for each u ∈ A ( G a a ) u \in A(\mathcal{G}_a^a) u ∈ A ( G a a ) , there exists v ∈ A ( G ) v \in A(\mathcal{G}) v ∈ A ( G ) such that R a ( v ) = u R_a(v)=u R a ( v ) = u and ∥ v ∥ A ( G ) = ∥ u ∥ A ( G a a ) \|v\|_{A(\mathcal{G})}=\|u\|_{A(\mathcal{G}_a^a)} ∥ v ∥ A ( G ) = ∥ u ∥ A ( G a a ) .
Proof. Consider a representation of u u u as a coefficient function ⟨ λ G ( ⋅ ) f , g ⟩ \langle\lambda_{\mathcal{G}}(\cdot)f,g\rangle ⟨ λ G ( ⋅ ) f , g ⟩ , where f , g ∈ ℓ 2 ( G a a ) f,g \in\ell^2(\mathcal{G}_a^a) f , g ∈ ℓ 2 ( G a a ) satisfying ∥ u ∥ A ( G a a ) = ∥ f ∥ 2 ∥ g ∥ 2 \|u\|_{A(\mathcal{G}_a^a)}=\|f\|_2\|g\|_2 ∥ u ∥ A ( G a a ) = ∥ f ∥ 2 ∥ g ∥ 2 . Since G a a \mathcal{G}_a^a G a a is a subset of the discrete set G a \mathcal{G}^a G a , we may extend f f f and g g g to functions f ~ \widetilde{f} f and g ~ \widetilde{g} g in ℓ 2 ( G a ) \ell^2(\mathcal{G}^a) ℓ 2 ( G a ) by setting them equal to 0 on G a \ G a a \mathcal{G}^a\backslash\mathcal{G}_a^a G a \ G a a . This extension provides an isometric embedding of ℓ 2 ( G a a ) \ell^2(\mathcal{G}_a^a) ℓ 2 ( G a a ) into ℓ 2 ( G a ) \ell^2(\mathcal{G}^a) ℓ 2 ( G a ) . The vectors f ~ \widetilde{f} f and g ~ \widetilde{g} g can in turn be extended to continuous sections, say F F F and G G G , respectively, on the unit space ([8 Proposition 10.1.10] ). In particular, F F F and G G G are continuous sections of the Hilbert bundle associated with the left regular representation of G \mathcal{G} G , as described in Theorem 2.5. Moreover, these extensions may be chosen to have the same supremum norm as the original fiber vectors. Indeed, given an extension F = { F b } b ∈ G ( 0 ) F=\{F^b\}_{b\in\mathcal{G}^{(0)}} F = { F b } b ∈ G ( 0 ) of f ~ \widetilde{f} f , replace it by the section obtained by multiplying F b F^b F b by
b ⟼ min { 1 , ∥ F a ∥ 2 ∥ F b ∥ 2 } , b \longmapsto\min\left\{1,\frac{\|F^a\|_2}{\|F^b\|_2}\right\}, b ⟼ min { 1 , ∥ F b ∥ 2 ∥ F a ∥ 2 } ,
where the multiplier is defined to be 1 whenever F b = 0 F^b=0 F b = 0 . By [8 Proposition 10.1.9] , the resulting vector field remains continuous. It agrees with F a F^a F a , while its norm is bounded above by ∥ F a ∥ 2 \|F^a\|_2 ∥ F a ∥ 2 at every fiber; hence ∥ F ∥ Δ = sup b ∈ G ( 0 ) ∥ F b ∥ 2 = ∥ F a ∥ 2 = ∥ f ~ ∥ 2 \|F\|_\Delta=\sup_{b\in\mathcal{G}^{(0)}}\|F^b\|_2=\|F^a\|_2=\|\widetilde{f}\|_2 ∥ F ∥ Δ = sup b ∈ G ( 0 ) ∥ F b ∥ 2 = ∥ F a ∥ 2 = ∥ f ∥ 2 . The same argument applies to g ~ \widetilde{g} g .
Let v v v be the coefficient function associated with F , G F,G F , G , i.e., v = Λ F , G v=\Lambda_{F,G} v = Λ F , G . For arbitrary y ∈ G a a y\in\mathcal{G}_a^a y ∈ G a a , we have
v ( y ) = ⟨ L y F a , G a ⟩ ℓ 2 ( G a a ) = ∑ z ∈ G a a F a ( y − 1 z ) G a ( z ) ‾ = ∑ z ∈ G a a f ( y − 1 z ) g ( z ) ‾ = u ( y ) . v(y)=\langle L_yF^a,G^a\rangle_{\ell^2(\mathcal{G}_a^a)}
=\sum_{z\in\mathcal{G}_a^a}F^a(y^{-1}z)\overline{G^a(z)}
=\sum_{z\in\mathcal{G}_a^a}f(y^{-1}z)\overline{g(z)}=u(y). v ( y ) = ⟨ L y F a , G a ⟩ ℓ 2 ( G a a ) = z ∈ G a a ∑ F a ( y − 1 z ) G a ( z ) = z ∈ G a a ∑ f ( y − 1 z ) g ( z ) = u ( y ) .
So R a ( v ) = u R_a(v)=u R a ( v ) = u . Since each restriction map is contractive, we get in particular that ∥ u ∥ A ( G ) ≤ ∥ v ∥ A ( G ) \|u\|_{A(\mathcal{G})}\leq\|v\|_{A(\mathcal{G})} ∥ u ∥ A ( G ) ≤ ∥ v ∥ A ( G ) . On the other hand, ∥ v ∥ A ( G ) ≤ ∥ F ∥ Δ ∥ G ∥ Δ = ∥ f ∥ 2 ∥ g ∥ 2 = ∥ u ∥ A ( G ) \|v\|_{A(\mathcal{G})}\leq\|F\|_\Delta\|G\|_\Delta=\|f\|_2\|g\|_2=\|u\|_{A(\mathcal{G})} ∥ v ∥ A ( G ) ≤ ∥ F ∥ Δ ∥ G ∥ Δ = ∥ f ∥ 2 ∥ g ∥ 2 = ∥ u ∥ A ( G ) . So ∥ u ∥ A ( G ) = ∥ v ∥ A ( G ) \|u\|_{A(\mathcal{G})}=\|v\|_{A(\mathcal{G})} ∥ u ∥ A ( G ) = ∥ v ∥ A ( G ) .
Proposition 7.2. Let G \mathcal{G} G be a (locally compact Hausdorff) étale groupoid with its counting Haar system. Then σ ( A ( G ) ) \sigma(A(\mathcal{G})) σ ( A ( G )) is homeomorphic to G \mathcal{G} G .
Proof. Recall that A ( G ) ⊆ C 0 ( G ) A(\mathcal{G}) \subseteq C_0(\mathcal{G}) A ( G ) ⊆ C 0 ( G ) , and A ( G ) A(\mathcal{G}) A ( G ) strongly separates points of G \mathcal{G} G . So, the natural embedding G → σ ( A ( G ) ) \mathcal{G} \to\sigma(A(\mathcal{G})) G → σ ( A ( G )) , given by point evaluation, is a homeomorphism into σ ( A ( G ) ) \sigma(A(\mathcal{G})) σ ( A ( G )) ([32 Theorem 3.2.4] ). It only remains to prove that this embedding is surjective.
Next, consider an arbitrary character Φ : A ( G ) → C \Phi: A(\mathcal{G}) \to\mathbb{C} Φ : A ( G ) → C . By Theorem 2.9, we have C c ( G ) ⊆ A ( G ) C_c(\mathcal{G}) \subseteq A(\mathcal{G}) C c ( G ) ⊆ A ( G ) . We will prove that Φ ∣ C c ( G ) \Phi|_{C_c(\mathcal{G})} Φ ∣ C c ( G ) is contractive, i.e., we prove
∣ Φ ( f ) ∣ ≤ ∥ f ∥ s u p for every f ∈ C c ( G ) . (27) |\Phi(f)| \le\|f\|_{\mathrm{sup}} \quad\text{for every } f \in C_c(\mathcal{G}). \tag*{(27)} ∣Φ ( f ) ∣ ≤ ∥ f ∥ sup for every f ∈ C c ( G ) . (27)
For f ∈ C c ( G ) f \in C_c(\mathcal{G}) f ∈ C c ( G ) and c ∈ C c \in\mathbb{C} c ∈ C satisfying ∣ c ∣ > ∥ f ∥ s u p |c| > \|f\|_{\mathrm{sup}} ∣ c ∣ > ∥ f ∥ sup , define h : G → C h : \mathcal{G} \to\mathbb{C} h : G → C as
h ( x ) = 1 c + f ( x ) c ( c − f ( x ) ) , h(x) = \frac{1}{c} + \frac{f(x)}{c(c-f(x))}, h ( x ) = c 1 + c ( c − f ( x )) f ( x ) ,
and note that the summand f ( x ) c ( c − f ( x ) ) \frac{f(x)}{c(c-f(x))} c ( c − f ( x )) f ( x ) belongs to C c ( G ) C_c(\mathcal{G}) C c ( G ) , and thus, to A ( G ) A(\mathcal{G}) A ( G ) . Let A 1 A_1 A 1 denote the unitization of A ( G ) A(\mathcal{G}) A ( G ) , and Φ ~ \widetilde{\Phi} Φ be the unique extension of Φ \Phi Φ to a character on A 1 A_1 A 1 . It is easy to see that h ( c 1 − f ) = 1 h(c1-f)=1 h ( c 1 − f ) = 1 , so Φ ~ ( h ) Φ ~ ( c 1 − f ) = 1 \widetilde{\Phi}(h)\widetilde{\Phi}(c1-f)=1 Φ ( h ) Φ ( c 1 − f ) = 1 . Therefore, Φ ~ ( f ) ≠ c \widetilde{\Phi}(f) \ne c Φ ( f ) = c , which implies that (27) holds. Consequently, Φ \Phi Φ can be extended to a character on C 0 ( G ) C_0(\mathcal{G}) C 0 ( G ) , and it is evaluation at some x ∈ G x \in\mathcal{G} x ∈ G .
Remark 7.3. The conclusion of Theorem 7.2 holds for any locally compact Hausdorff groupoid G \mathcal{G} G equipped with a Haar system that satisfies C c ( G ) ⊆ A ( G ) C_c(\mathcal{G}) \subseteq A(\mathcal{G}) C c ( G ) ⊆ A ( G ) . In particular, this includes all étale groupoids.
Acknowledgments
The first author thank the organizers of the Northeastern Analysis Meeting (NEAM 2025) , held at the University of Delaware on November 1–2, 2025, for financial support that enabled JD’s visit to the University of Delaware, during which the authors initiated discussions leading to this work. The second author acknowledges support from National Science Foundation Grant DMS-2408008 and CCF-2427965 during the preparation of this article. The authors acknowledge Simons Foundation Travel Support for Mathematicians that funded their collaboration. This material is also based upon work supported by the Swedish Research Council under grant no. 2021-06594 while M.G. was in residence at Institut Mittag-Leffler in Djursholm, Sweden during the semester “Operator algebras and Quantum information theory” Spring, 2026. The authors used ChatGPT (OpenAI) to assist with improving the language and presentation of the manuscript in some places. All mathematical content, except Theorem 7.1 and Theorem 7.2, was developed and verified by the authors. The statements of Theorem 7.1 and Theorem 7.2 and an outline of their proof were suggested by ChatGPT. The authors wrote the proofs (in the mathematical style of the paper) and independently verified their correctness.