Abstract.

This article presents the construction of a bundle of C*-algebras with a compact base space by means of localization processes resorting to inductive limits. This construction achieved in terms of the Stone-Čech compactification provides a solution to the compactness issue raised by P. Bertozzini, R. Conti and N. Pitiwan in Discrete Non-Commutative Gelfand-Naimark Duality, where the lack of compactness in the base space was an obstacle for the foreseen applications to quantum mechanics.

keywords:
bundles of C*-algebras; Stone-Čech compactification; localization process.
MSC:
54D35; 54H10; 46L05; 46M20.

1. Introduction

Bundles of C*-algebras have become an important tool in various areas, particularly in quantum mechanics, where the structure provided by bundles of C*-algebras can model observables and state spaces. However, one significant issue, as noted by the authors in [2], is the lack of compactness in the base space of some of those bundles, which can hinder applications and theoretical considerations.

In this paper, we propose a solution to this problem by constructing a bundle of C*-algebras whose base space is compact. This is obtained by means of the Stone-Čech compactification of the base space and localization processes that define the new fibers over the remaining points of the compactification while preserving the initial fibers, provided that the set of global and bounded sections of the original bundle is full and the base space is completely regular. Furthermore, the construction is such that the resulting bundle preserves the continuity of the C*-algebra operations over the whole compactified base space.

The localization process we follow is based on inductive limits, providing a procedure that not only respects the original bundle structure but could also enhance its applicability.

2. Preliminaries

Let E and T be topological spaces and p:E⟶T be a surjective function, not necessarily continuous. For each t∈T, the set Et=p−1⁢(t)={a∈E:p⁢(a)=t} is called the fiber over t. A local selection for p is a function σ:Q⟶E such that Q⊂T is an open set and p∘σ is the identity map Q. A local section for p is by definition a continuous local selection. Let Q⊂T be an open set. Set

ΓQ⁢(p):={σ:Q⟶E⁢ such that ⁢σ⁢ is continuous and ⁢p∘σ=I⁢dQ}.

If Q=T, σ is a global section and one writes Γ⁢(p) instead of ΓT⁢(p). A set of sections Σ is said to be full if for each x∈E there exists σ∈Σ such that σ⁢(p⁢(x))=x.

Let E∨E:={(u,v)∈E×E:p⁢(u)=p⁢(v)}. A function d:E∨E⟶ℝ is called a metric for p provided that the restriction of d to Et×Et is a metric of Et, for each t∈T.

Definition 2.1.

Let d be a metric for p, σ a local selection, and ε>0, the set 𝒯ε⁢(σ)={u∈E:d⁢(u,σ⁢(p⁢(u)))<ε} is called the ε-tube around σ.

Definition 2.2.

Let E and T be topological spaces, p:E⟶T a surjective function and d a metric for p. The triplet (E,p,T) is a bundle of metric spaces or a metric bundle provided that:

  1. (1)

    For each u∈E and ε>0, there exists a local section σ such that u∈𝒯ε⁢(σ).

  2. (2)

    The tubes, around all local sections for p, form a base for the topology of E.

Remark 2.3.
  1. (1)

    Let (E,p,T) be a bundle of metric spaces. The function p is continuous, indeed, consider u∈E and an open neighborhood Q of p⁢(u). For ε>0 and any local section σ such that u∈𝒯ε⁢(σ), let Q0=Q∩Dom ⁢σ, then u∈𝒯ε(σ↾Q0) and p(𝒯ε(σ↾Q0))=Q0⊂Q, as desired.

  2. (2)

    Note that whenever (E,p,T) is a bundle of metric spaces and (α,β)∈Σ×Σ, then the function s⟼d⁢(α⁢(s),β⁢(s)):Dom ⁢α∩Dom ⁢β⟶ℝ is upper semicontinuous. In fact, for each ε>0, d⁢(α⁢(s),β⁢(s))<ε implies that α⁢(s)∈𝒯ε⁢(β) and on account of the continuity of α and the fact that the tube 𝒯ε⁢(β) is open, there exists an open neighborhood Q of s in T, such that α⁢(r)∈𝒯ε⁢(β), for every r∈Q. Then d⁢(α⁢(r),β⁢(r))<ε, for each r∈Q. It follows that the function s⟼d⁢(α⁢(s),β⁢(s)) is upper semicontinuous. This motivates the following result.

Theorem 2.4 (Existence Theorem cf.[12]).

Let T be a topological space and p:E⟶T be a surjective function. Consider a set Σ of selections for p and a metric d for p. Assume the following:

  1. a)

    For each u∈E and ε>0, there exists α∈Σ such that u∈𝒯ε⁢(α).

  2. b)

    For each (α,β)∈Σ×Σ the function s⟼d⁢(α⁢(s),β⁢(s)):Dom ⁢α∩Dom ⁢β⟶ℝ is upper semicontinuous.

Then there exists a topology 𝒮 over E such that:

  1. 1)

    The topology 𝒮 has a base consisting of all the sets of the form 𝒯ε⁢(αQ), where ε>0 and αQ is the restriction to an open subset Q⊂Dom ⁢α of a selection α∈Σ.

  2. 2)

    Each α∈Σ is a section.

  3. 3)

    (E,p,T) is a metric bundle.

3. Bundles of C*-algebras

Definition 3.1.

Let (E,p,T) be a metric bundle whose fibers are normed spaces and whose metric d is defined by d⁢(x,y)=‖x−y‖. The bundle (E,p,T) is called a bundle of normed spaces if the functions

  1. (1)

    (x,y)⟼x+y:E∨E⟶E and

  2. (2)

    (α,x)⟼α⁢x:ℂ×E⟶E

are continuous.

Remark 3.2.

If (E,p,T) is a bundle of normed spaces and τ is a global section for p, by the item (2) of the definition above, the function σ0:T⟶E defined by σ0⁢(t)=0⁢τ⁢(t)=0 is continuous. In other words, the function that applies t∈T to the sum identity of Et is in fact a global section.

Definition 3.3.

Let (E,p,T) be a bundle of normed spaces whose fibers are C*-algebras. If the functions

  1. (1)

    (x,y)⟼x⁢y:E∨E⟶E and

  2. (2)

    x⟼x∗:E⟶E

are continuous, (E,p,T) is called a bundle of C*-algebras.

If for every t∈T, p−1⁢(t) has a unit element ζ⁢(t) and t⟼ζ⁢(t):T⟶E is continuous, then (E,p,T) is called a bundle of C*-algebras with a unit element.

Remark 3.4.

Let (E,p,T) be a bundle of C*-algebras.

  1. (1)

    Remarks 2.3 (2) and 3.2 guarantee that the function t⟼‖σ⁢(t)‖:T⟶ℝ is upper semicontinuous, for each section σ for p.

  2. (2)

    Let Γb⁢(p) be the set of all bounded global sections of (E,p,T) such that

    1. (a)

      the operations are pointwise defined and

    2. (b)

      the norm is defined by ||σ||=sup{||σ(t)||:t∈T},

    then Γb⁢(p) is a C*-algebra [5].

    Indeed, the completeness of Γb⁢(p) can be viewed in [4], Proposition 1.18, p. 7 and 1.13, p. 53.

4. Compactification of Bundles of C*-algebras

Let (E,p,T) be a bundle of C*-algebras with a completely regular, Hausdorff base space T. Denote β⁢(T) the Stone-Čech compactification of T and let e:T⟶β⁢(T) be the canonical immersion. For each s∈β⁢(T), define the equivalence relation Rs in the set Γb⁢(p) of bounded global sections, by σ⁢Rs⁢τ, if and only if, infV∈𝒱⁢(s)supt∈e−1⁢(V)‖σ⁢(t)−τ⁢(t)‖=0. Let [σ]s be the equivalence class of σ modulo Rs, E^s={[σ]s:σ∈Γb⁢(p)}, E^=∐s∈β⁢(T)E^s (that is, the disjoint union of the family {E^s:s∈β⁢(T)}) and p^:E^⟶β⁢(T) defined by p^⁢([σ]s)=s.

For each σ∈Γb⁢(p), consider the selection σ^ for p^ defined by

s⟼σ^⁢(s)=[σ]s:β⁢(T)⟶E^.

The Existence Theorem secures that (E^,p^,β⁢(T)), with the metric induced by the norm defined by ‖|[σ]s|‖s=infV∈𝒱⁢(s)supt∈e−1⁢(V)‖σ⁢(t)||t, is a bundle of metric spaces with compact base β⁢(T). Our aim is to establish that such bundle is in fact a bundle of C∗-algebras.

Remark 4.1.
  1. (1)

    Let σ,τ∈Γb⁢(p) and t∈T. On account of the upper semicontinuity of the function r↦‖σ⁢(r)−τ⁢(r)‖:T⟶ℝ, it follows that:

    σ⁢Rt⁢τ ⇔infV∈𝒱⁢(t)supr∈e−1⁢(V)‖σ⁢(r)−τ⁢(r)‖=0,
    ⇔ for each ⁢ε>0, there exists ⁢V∈𝒱⁢(t)⁢ such that
    supr∈e−1⁢(V)‖σ⁢(r)−τ⁢(r)‖<ε,
    ⇔‖σ⁢(t)−τ⁢(t)‖<ε, for each ⁢ε>0, (the implication ⟸
    requires the upper semicontinuity assumption)
    ⇔σ⁢(t)=τ⁢(t).

    Then [σ]t can be identified with σ⁢(t) and σ^ turns out to be an extension of σ. Furthermore, in case Γb⁢(p) happens to be full, Et can be identified with E^t.

  2. (2)

    Arguing as before, one finds that for each t∈T and σ∈Γb⁢(p), ‖σ⁢(t)‖≤‖|σ^⁢(t)|‖. Then

    ‖σ‖ =supt∈T‖σ⁢(t)‖
    ≤supt∈T‖|σ^⁢(t)|‖
    ≤sups∈β⁢(T)‖|σ^⁢(s)|‖=‖|σ^|‖.

    Moreover, for each s∈β⁢(T),

    ‖|[σ]s|‖s =infV∈𝒱⁢(s)supt∈e−1⁢(V)‖σ⁢(t)||t
    ≤supt∈e−1⁢(β⁢(T))‖σ⁢(t)||t
    =supt∈T‖σ⁢(t)||t=‖σ‖.

    It follows that sups∈η⁢(T)‖|[σ]s|‖s≤‖σ‖ and that ‖σ‖=‖|σ^|‖.

  3. (3)

    If T is not completely regular or if it is not T1, the Stone Čech construction can be made, but T fails to be a subspace of β⁢(T) and (E,p,T) can not be considered a sub-bundle of (E^,p^,β⁢(T)), even if Γb⁢(p) is full. In fact, if σ,τ∈Γb⁢(p) and s∈β⁢(T), then σ⁢Rs⁢τ if and only if σ⁢(t)=τ⁢(t) for each t∈e−1⁢(s).

Proposition 4.2.

The operations

  1. (1)

    ([σ]s,[τ]s)⟼[σ]s+[τ]s=[σ+τ]s:E^∨E^⟶E^,

  2. (2)

    ([σ]s,[τ]s)⟼[σ]s⁢[τ]s=[σ⁢τ]s:E^∨E^⟶E^, and

  3. (3)

    [σ]s⟼[σ]s∗=[σ∗]s:E^⟶E^

are well-defined and continuous.

Proof 4.3.

The well-definedness properties are straightforward and their verifications are omitted.

  1. (1)

    Consider an open basic neighborhood of [σ+τ]s in E^, of the form 𝒯ε⁢((σ+τ)^Q), where ε>0 and Q is an open neighborhood of s in β⁢(T).

    Suppose that r∈Q,δ>0, [γ1]r∈𝒯δ⁢(σ^) and [γ2]r∈𝒯δ⁢(τ^). Then

    infP∈𝒱⁢(r)supt∈e−1⁢(P)‖γ1⁢(t)−σ⁢(t)‖<δ⁢ and
    infP∈𝒱⁢(r)supt∈e−1⁢(P)‖γ2⁢(t)−τ⁢(t)‖<δ.

    There exists P∈𝒱⁢(r) such that ‖γ1⁢(t)−σ⁢(t)‖<δ and ‖γ2⁢(t)−τ⁢(t)‖<δ, for each t∈e−1⁢(P). Then ‖(σ⁢(t)+τ⁢(t))−(γ1⁢(t)+γ2⁢(t))‖<2⁢δ, for each t∈e−1⁢(P). Therefore supt∈e−1⁢(P)‖(σ⁢(t)+τ⁢(t))−(γ1⁢(t)+γ2⁢(t))‖≤2⁢δ, thus infV∈𝒱⁢(r)supt∈e−1⁢(V)‖(σ⁢(t)+τ⁢(t))−(γ1⁢(t)+γ2⁢(t))‖≤2⁢δ. By choosing δ<ε/2, one concludes that [γ1]r+[γ2]r=[γ1+γ2]r∈𝒯ε⁢((σ+τ)^Q).

  2. (2)

    To prove that ([σ]s,[τ]s)⟼[σ]s⁢[τ]s=[σ⁢τ]s:E^∨E^⟶E^ is a continuous function, we argue as before.

    Consider an open basic neighborhood, 𝒯ε⁢((σ⁢τ)^Q), of [σ⁢τ]s in E^, where ε>0 and Q is an open neighborhood of s in β⁢(T).

    Suppose that r∈Q,δ>0, [γ1]r∈𝒯δ⁢(σ) and [γ2]r∈𝒯δ⁢(τ). Then

    infP∈𝒱⁢(r)supt∈e−1⁢(P)‖γ1⁢(t)−σ⁢(t)‖<δ⁢ and
    infP∈𝒱⁢(r)supt∈e−1⁢(P)‖γ2⁢(t)−τ⁢(t)‖<δ.

    There exists P∈𝒱⁢(r) such that ‖γ1⁢(t)−σ⁢(t)‖<δ and ‖γ2⁢(t)−τ⁢(t)‖<δ, for each t∈e−1⁢(P). One has

    ‖γ1⁢(t)⁢γ2⁢(t)−σ⁢(t)⁢τ⁢(t)‖ =‖γ1⁢(t)⁢γ2⁢(t)−σ⁢(t)⁢γ2⁢(t)+σ⁢(t)⁢γ2⁢(t)−σ⁢(t)⁢τ⁢(t)‖
    =‖γ2⁢(t)⁢(γ1⁢(t)−σ⁢(t))+σ⁢(t)⁢(γ2⁢(t)−τ⁢(t))‖
    ≤‖γ2⁢(t)‖⁢‖γ1⁢(t)−σ⁢(t)‖+‖σ⁢(t)‖⁢‖γ2⁢(t)−τ⁢(t)‖,

    for each t∈e−1⁢(P).

    Taking into account that ‖γ2⁢(t)−τ⁢(t)‖<δ and by choosing δ<1, one has that ‖γ2⁢(t)‖<‖τ⁢(t)‖+1 for each t∈e−1⁢(P).

    Then

    ‖γ1⁢(t)⁢γ2⁢(t)−σ⁢(t)⁢τ⁢(t)‖ ≤(‖τ⁢(t)‖+1)⁢‖γ1⁢(t)−σ⁢(t)‖+‖σ⁢(t)‖⁢‖γ2⁢(t)−τ⁢(t)‖
    ≤(‖τ⁢(t)‖+1)⁢δ+‖σ⁢(t)‖⁢δ
    =(‖τ⁢(t)‖+1+‖σ⁢(t)‖)⁢δ
    ≤(‖τ‖+‖σ‖+1)⁢δ.

    Consider δ=12⁢min⁡{1,ε‖τ‖+‖σ‖+1}. Then ‖γ1⁢(t)⁢γ2⁢(t)−σ⁢(t)⁢τ⁢(t)‖<12⁢ε, for each t∈e−1⁢(P). Furthermore supt∈e−1⁢(P)‖(σ⁢(t)⁢τ⁢(t))−γ1⁢(t)⁢γ2⁢(t)‖<ε, and infV∈𝒱⁢(r)supt∈e−1⁢(V)‖σ⁢(t)⁢τ⁢(t)−γ1⁢(t)⁢γ2⁢(t)‖<ε. It follows that [γ1]r⁢[γ2]r=[γ1⁢γ2]r∈𝒯ε⁢((σ⁢τ)^Q).

  3. (3)

    Consider an open basic neighborhood, 𝒯ε⁢((σ∗)^Q), of [σ∗]s in E^, where ε>0 and Q is an open neighborhood of s in β⁢(T). Let δ=ε/2 and suppose that [γ]r∈𝒯δ⁢(σ^Q). Then infP∈𝒱⁢(r)supt∈e−1⁢(P)‖γ⁢(t)−σ⁢(t)‖<δ, thus there exists P∈𝒱⁢(r), such that ‖γ⁢(t)−σ⁢(t)‖<δ, for each t∈e−1⁢(P). Since

    ‖γ∗⁢(t)−σ∗⁢(t)‖= ‖(γ⁢(t)−σ⁢(t))∗‖
    = ‖γ⁢(t)−σ⁢(t)‖<δ,

    for every t∈e−1⁢(P), then supt∈e−1⁢(P)‖γ∗⁢(t)−σ∗⁢(t)‖≤δ<ε and infP∈𝒱⁢(r)supt∈e−1⁢(P)‖γ∗⁢(t)−σ∗⁢(t)‖<ε.

    This proves that [γ∗]r∈𝒯ε⁢((σ∗)^Q).

Remark 4.4.

Let ξ∈Γb⁢(p^), σ=ξ↾T∈Γb(p), and s∈β⁢(T)∖T. Suppose that σ^⁢(s)≠ξ⁢(s) and that τ∈Γb⁢(p) is such that ξ⁢(s)=τ^⁢(s)=[τ]s. Note that the continuity of ξ and τ implies that for each ε>0, there exists V∈𝒱⁢(s), such that ‖|ξ⁢(r)−τ^⁢(r)|‖<ε, for each r∈V. However, since σ^⁢(s)≠τ^⁢(s), there exists ε>0 such that ε<infV∈𝒱⁢(s)supt∈e−1⁢(V)‖σ⁢(t)−τ⁢(t)‖, then for every V∈𝒱⁢(s), ε<supt∈e−1⁢(V)‖σ⁢(t)−τ⁢(t)‖. Therefore, for each V∈𝒱⁢(s), there exists t∈e−1⁢(V) such that ε<‖σ⁢(t)−τ⁢(t)‖. This is contradictory, then σ^⁢(s)=ξ⁢(s) it follows that σ^=ξ.

This establishes that □^:Γb⁢(p)⟶Γb⁢(p^),σ⟼σ^ is a surjective map.

On the other hand, if σ^=τ^, then by restricting to T one obtains that σ=τ, namely □^ is an injective map, hence □^ is an isometric star isomorphism between Γb⁢(p) and Γb⁢(p^).

In the remainder of this section we verify that E^s is a C*-algebra, for each s∈β⁢(T).

Let (E,p,T) be a bundle of C*-algebras, where T is a completely regular and T1 space. The set Γb⁢(p) of all bounded sections for p, with pointwise operations and the norm given by

‖σ‖=supt∈T‖σ⁢(t)‖t,

is known to be a C*-algebra [5].

Consider the Stone-Čech compactification β⁢(T) of T.

For each s∈β⁢(T), let

Is={σ∈Γb⁢(p):infV∈𝒱⁢(s)supt∈e−1⁢(V)‖σ⁢(t)‖t=0}.

The collection Is just defined is a closed ideal of Γb⁢(p), as proven below.

To begin with, observe that the section σ0, where σ0⁢(t) is the zero element of the C*-algebra Et, belongs to Is.

  1. (1)

    Let σ and τ be elements of Is and ε>0. Let V∈𝒱⁢(s) such that

    supt∈e−1⁢(V)‖σ⁢(t)‖t⁢<ε2andsupt∈e−1⁢(V)∥⁢τ⁢(t)∥t<ε2.

    Then ‖σ⁢(t)+τ⁢(t)‖t≤‖σ⁢(t)‖t+‖τ⁢(t)‖t<ε for each t∈e−1⁢(V), which implies that σ+τ∈Is.

  2. (2)

    If σ⁢τ∈Is and ε>0, let V∈𝒱⁢(s) such that

    supt∈e−1⁢(V)‖σ⁢(t)‖t⁢<εandsupt∈e−1⁢(V)∥⁢τ⁢(t)∥t<ε.

    Then ‖σ⁢(t)⁢τ⁢(t)‖t≤‖σ⁢(t)‖t⁢‖τ⁢(t)‖t<ε for each t∈e−1⁢(V), so σ⁢τ∈Is.

  3. (3)

    Suppose σ∈Is, k∈ℂ, k≠0, and ε>0. Let V∈𝒱⁢(s) such that

    supt∈e−1⁢(V)‖σ⁢(t)‖t<ε|k|.

    Then ‖k⁢σ⁢(t)‖t=|k|⁢‖σ⁢(t)‖t<ε for each t∈e−1⁢(V). Therefore, k⁢σ∈Is.

  4. (4)

    Since for each σ∈Γb⁢(p) and t∈T, ‖σ⁢(t)‖t=‖σ∗⁢(t)‖t, then σ∗∈Is whenever σ∈Is.

  5. (5)

    If σ∈Is, τ∈Γb⁢(p), τ≠σ0, and ε>0, there exists V∈𝒱⁢(s) such that

    supt∈e−1⁢(V)‖σ⁢(t)‖t<ε‖τ‖.

    Then, for each t∈e−1⁢(V),

    ‖σ⁢τ⁢(t)‖t=‖σ⁢(t)⁢τ⁢(t)‖t≤‖σ⁢(t)‖t⁢‖τ‖<ε.

    Hence σ⁢τ∈Is. Similarly, τ⁢σ∈Is.

  6. (6)

    To see that Is is closed, take τ∈Is¯. If ε>0 and σ∈Is is such that ‖σ−τ‖<ε2, there exists V∈𝒱⁢(s) such that

    supt∈e−1⁢(V)‖σ⁢(t)‖t<ε2.

    Note that

    ‖σ−τ‖=supt∈T‖σ⁢(t)−τ⁢(t)‖t<ε2.

    In particular,

    ‖σ⁢(t)−τ⁢(t)‖t<ε2,for each ⁢t∈e−1⁢(V).

    It follows that

    ‖τ⁢(t)‖t=‖τ⁢(t)−σ⁢(t)+σ⁢(t)‖t≤‖τ⁢(t)−σ⁢(t)‖t+‖σ⁢(t)‖t<ε,

    for each t∈e−1⁢(V).

    Therefore supt∈e−1⁢(V)‖τ⁢(t)‖t≤ε and infV∈𝒱⁢(s)supt∈e−1⁢(V)‖τ⁢(t)‖t=0. Consequently τ∈Is.

Being Is a closed ideal of Γb⁢(p) secures that Γb⁢(p)/Is is a C*-algebra.

Now, consider the equivalence relation Rs defined on Γb⁢(p) by σ⁢Rs⁢τ if and only if

infV∈𝒱⁢(s)supt∈e−1⁢(V)‖σ⁢(t)−τ⁢(t)‖t=0,

and let [σ]s be the equivalence class of σ modulo Rs. Then [σ]s=σ+Is, indeed

τ∈[σ]s, if and only if, ⁢τ⁢Rs⁢σ,
if and only if, ⁢τ−σ∈Is,
if and only if, ⁢τ=σ+(τ−σ)∈σ+Is.

Thus E^s=Γb⁢(p)/Rs=Γb⁢(p)/Is.

This proves that the fibers of the metric bundle (E^,p^,β⁢(T)) obtained by localization from the data provided by the bundle of C*-algebras (E,p,T) also have a C*-algebra structure.

Remark 4.5.

For the bundle of C∗-algebras (E,p,T) under consideration the entire process of constructing a bundle of C*-algebras can be carried out over any compactification S of the space T. Again, the fibers over a point t∈T remain unchanged, provided that the set Γb⁢(p) is full, while the fibers over the elements of the complement of T in S are new C*-algebras.

Example 4.6.

Let T be a completely regular Hausdorff space. The algebra Cb⁢(T) of all continuous complex valued bounded functions can be viewed as a bundle of C∗–algebras. In fact, consider E=T×ℂ with the product topology and let p=π1:E⟶T be the first projection. The triplet (E,p,T) is a bundle of C∗–algebras where each bounded section σ, can be identify with the continuous bounded function π2∘σ:T⟶ℂ, also if f:T⟶ℂ is a continuous bounded function and IdT is the identity map of T, then IdT×f is a bounded section for p and each one of its fibers can be identify with the set of complex numbers.

Let (E^,p^,β⁢(T)) be the bundle of C∗–algebras obtained by localization from (E,p,T). For each σ∈Cb⁢(T), denote by σ~ the unique continuous extension of σ to β⁢(T), such extention exists because of the boundness of σ, that is, if T⁢⟶𝑒⁢β⁢(T) is the canonical injection of T into β⁢(T), then σ~∘e=σ. We claim that the map

μs: E^s⟶ℂ
[σ]s= σ^⁢(s)⟼σ~⁢(s)

is an isometric star isomorphism between E^s and ℂ, for each s∈β⁢(T).

Indeed, let s∈β⁢(T).

  1. (1)

    In this item we will prove that ‖σ^⁢(s)‖s=|σ~⁢(s)|, for each σ∈Cb⁢(T), from which it will follow the well definiteness property of μs, indeed, if σ⁢Rs⁢τ, then (σ−τ)⁢Rs⁢ 0, so σ−τ^⁢(s)=0^⁢(s), then ‖σ−τ^⁢(s)‖s=0 and thus |σ~⁢(s)−τ~⁢(s)|=|σ−τ~⁢(s)|=0. This also will settle the injectivity of μs.

    Suppose that |σ~⁢(s)|=m. Obviously, for each V∈𝒱⁢(s), one has supr∈V|σ~⁢(r)|≥m and taking into account the density of e⁢(T) in β⁢(T), it follows that supt∈e−1⁢(V)|σ~((e(t))|=supt∈e−1⁢(V)|σ(t)|≥m and ‖σ^⁢(s)‖s=infV∈𝒱⁢(s)supt∈e−1⁢(V)|σ⁢(t)|≥m.
    On the other hand, by continuity of σ~, for each ε>0, there exists a neighborhood V of s such that |σ~⁢(r)|<m+ε, for each r∈V. In particular, |σ~⁢(e⁢(t))|=|σ⁢(t)|<m+ε, for each t∈e−1⁢(V), thus supt∈e−1⁢(V)|σ⁢(t)|≤m+ε and
    ‖σ^⁢(s)‖s=infV∈𝒱⁢(s)supt∈e−1⁢(V)|σ⁢(t)|≤m+ε. It follows that

    ‖σ^⁢(s)‖s=infV∈𝒱⁢(s)supt∈e−1⁢(V)|σ⁢(t)|=m=|σ~⁢(s)|.

  2. (2)

    Let λ∈ℂ. Define the section σ:T⟶E by σ⁢(t)=(t,λ). Then μs⁢([σ]s)=λ. It follows that μs is surjective.

  3. (3)

    Let σ∈Cb⁢(T) and σ~ its continuous extension to β⁢(T), then ‖σ‖≤‖σ~‖ and assuming that ‖σ‖<‖σ~‖ and taking to∈β⁢(T)∖T such that |σ~⁢(to)|=‖σ~‖, there exists an open neighborhoo Q of to and t∈Q∩e⁢(T) such that |σ⁢(t)|>ε where ε=12⁢(‖σ~‖+‖σ‖), hence |σ⁢(t)|>‖σ‖, then, by contradiction, it follows that ‖σ‖=‖σ~‖. This means that the map
    □~:Cb⁢(T)⟶C⁢(β⁢(T)),σ⟼σ~ is an isometric star isomorphisms, in particular the spectrum of Cb⁢(T) is β⁢(T).

    Example 4.7.

    Let T be a locally compact Hausdorff space and S=T∪{ω} its Alexandroff compactification. As in the previous example, let’s consider the bundle of C∗–algebras (E,p,T), where E=T×ℂ and p=π1. Let (E^,p^,S) be the bundle of C∗–algebras obtained by localization from (E,p,T). To examine the fiber E^ω, consider the equivalence relationship Rω in Cb⁢(T) defined by f⁢Rω⁢g, if and only if, infKsupt∈T∖K|f⁢(t)−g⁢(t)|=0, where K runs throughout the collection of all compact subsets of T. Let f,g∈Cb⁢(T) and suppose that limt→∞h⁢(t)=1, that is, if h⁢Rω⁢h1, where h1 is the constant function one, then f⁢Rω⁢(f⁢h). Consequentely,

    [f]ω =[f⁢h]ω=[f]ω⁢[h]ω
    =[f]ω⁢[h1]ω=[h1]ω⁢[f]ω.

    Hence [h1]ω turns out to be the multiplicative identity of E^ω.

    The class [h1]ω generates a subalgebra of E^ω, which is isometrically isomorphic to ℂ. However, if for instance T=ℝ, there are plenty of functions in Cb⁢(T), like the function defined by f⁢(x)=sin⁡x, whose class is outside of the generated subalgebra.

    Then the fiber above the point ω, contrary to example above, could be far away from being the set of complex numbers ℂ.

5. Some Categorical Considerations

We choose the Stone-Čech compactification in this work in view of the richness provided by its universal property, which extends to the category of bundles of C*-algebras, as will be seen in this section.

Let (E,p,T)⟶(F,q,S) bundles of C∗-algebras, ℓ:T⟶S be a continuous function and consider the set Fℓ={(t,m)∈T×F:ℓ⁢(t)=q⁢(m)}=⋃{{t}×Fℓ⁢(t):t∈T}.

The diagram

FℓFTSπ2π1qℓ

is a pullback in the category of topological spaces and continuous functions.

Definition 5.1.

A morphism (ℓ,Λ):(E,p,T)⟶(F,q,S) in the category of bundles of C∗-algebras is a pair of continuous functions ℓ:T⟶S and Λ:Fℓ⟶E such that

FℓETΛπ1p

is a commutative diagram and Λ↾{t}×Fℓ⁢(t) is a homomorphism of C∗-algebras, for each t∈T.

Let (E,p,T) be a bundle of C∗-algebras, β⁢(T) be the Stone-Čech compactification of T and e:T⟶β⁢(T) be the canonical map. Consider the function Λ:E^e⟶E defined by Λ⁢(t,[σ]e⁢(t))=σ⁢(t).

The pair (e,Λ) is a morphism from (E,p,T) in (E^,p^,β⁢(T)).

Indeed, the diagram

E^eETΛπ1p

is commutative and for each t∈T, the function Λ↾{t}×E^e⁢(t) is a homomorphism of C∗-algebras between {t}×E^e⁢(t) and Et.

Suppose that S is compact Hausdorff space, that (F,q,S) is a bundle of C∗-algebras, that there exists a full set of global sections for q and that (ℓ,Δ):(E,p,T)⟶(F,q,S) is a morphism of bundles of C∗-algebras. One has that ℓ:T⟶S and Δ:Fℓ⟶E are continuous functions, that the diagram

FℓETΔπ1p

commutes and that Δ↾{t}×Fℓ⁢(t) is a homomorphism of C∗-algebras, for each t∈T.

Consider ℓ^ the extention of ℓ to β⁢(T) and the set

Fℓ^ ={(k,a)∈β⁢(T)×F:ℓ^⁢(k)=q⁢(a)}
=⋃{{k}×Fℓ^⁢(k):k∈β⁢(T)}.

For (k,a)∈Fℓ^, let αa be a global section of the bundle (F,q,S) such that αa⁢(q⁢(a))=a. Define

Ω:Fℓ^ ⟶E^
(k,a) ⟼[αΔa]k.

It is straightforward that Ω is a well defined function and that Ω↾{k}×Fℓ^⁢(k) is a homomorphism of C∗-algebras, for each k∈β⁢(T).

Therefore, there exists a unique morphism (ℓ^,Ω):(E^,p^,β⁢(T))⟶(F,q,S) such that the diagram

(E,p,T)(F,q,S)(E^,p^,β⁢(T))(ℓ,Δ)(e,Λ)(ℓ^,Ω)

commutes.

6. Some additional considerations

  1. (1)

    On Fell Bundles: We consider in this item the definition of Fell Bundle given by A. Kumjian [8].

    In order to make a first move in the direction of possible applications of the results of this work to the Fell bundles studied in [2] we consider a Fell bundle (E,p,G), where for now we assume the restrictive case of an infinite discrete groupoid G. Consider the Stone-Čech compactification of the groupoid G constructed in [1]. Here one has that the set of composable pairs β⁢(G)(2) of β⁢(G) is β⁢(G)(2)=⋃u∈G0Gu¯×Gu¯=⋃u∈G0β⁢(Gu)×β⁢(Gu), where the space of units G0 is the set of elements of the form g⁢g−1, with g∈G, the functions rang r:G⟶G0 and source d:G⟶G0 are defined by r⁢(g)=g⁢g−1 and d⁢(g)=g−1⁢g respectively, and for each u∈G0, Gu=d−1⁢({u}) and Gu=r−1⁢({u}). The theorems 3.1 and 3.2 in [1] guarantee that there is a unique operation ∗ in β⁢(G) extending in an appropriate manner the operation in G, characterized as follows: If (θ,η)∈β⁢(G)(2), there exists u∈G0 such that (θ,η)∈Gu¯×Gu¯. If {gi}i∈I is a net in Gu that converges to θ and {hj}j∈J is a net in Gu that converges to η, then θ∗η=limilimjgi⁢hj. The inversion in β⁢(G) is obtained extending the inversion defined in G.

    With the aim of extending the operation defined in E2={(e1,e2):(p⁢(e1),p⁢(e2))∈G(2)} to the set E^2={([σ]θ,[τ]η):(θ,η)∈β⁢(G)(2)}, proceed as follows: If (θ,η)∈G(2) pick μ∈Γb⁢(p) satisfying μ⁢(θ⁢η)=σ⁢(θ)⁢τ⁢(η). Otherwise, consider u∈G0 such that (θ,η)∈Gu¯×Gu¯ and nets {gi}i∈I in Gu and {hj}j∈J in Gu converging to θ and η, respectively. These nets can be constructed in such a manner that gi⁢hj≠gi1⁢hj1 if (i,j)≠(i1,j1). Choose a bounded section μ for p such that μ⁢(gi⁢hj)=σ⁢(gi)⁢τ⁢(hj) for each i∈I and j∈J and define the function ([σ]θ,[τ]η)⟼[μ]θ∗η:E^2⟶E^. The involution on E^ can be obtained in a similar way.

  2. (2)

    On Duality: duality is a widely discussed topic in categories of C*-algebras, bundles of C*-algebras and Fell Bundles. In [2], the authors present a non-commutative version of the Gelfand-Naimark duality between the categories of compact Hausdorff topological spaces and commutative unital C*-algebras. In [3] the author represents C*-algebras with some additional structure on Fell bundles over locally compact ètale groupoids what turns out to be a non-commutative extension of the Gelfand duality.

    In [10] a duality was established between the category of pairs of C*-algebras (A,B), such that B is a central subalgebra of a C∗-algebra A with a unit element and the category of bundles of C*-algebras with compact base space, the corrresponding bundle (E,p,T) to the pair (A,B) is constructed over the spectrum T of B and its fibers are quotiens of A module the ideals generated by the maximal ideals of B. This duality is an extension to the noncommutative case of the duality between the category of commutative C∗-algebras with a unit element and the category of compact spaces.

Acknowledgements.
The authors are indebted to the reviewer for the numerous and accurate suggestions that led to a considerable improvement of this paper.
Funding.
This paper contains results obtained by the first author in her investigation supported by Universidad Nacional Facultad de Ciencias (Bogotá).
Author contributions.
Conceptualization, C. N. and J V.; formal analysis, C. N.; investigation, C.N. and J. V.; writing—original draft preparation, C. N.; writing—review and editing, C.N. and J. V.. The two authors have read and agreed to the published version of the manuscript.

References

  • [1] F. Behrouzi, The Stone Čech compactification of groupoids, Ukrainian Mathematical Journal 67, no. 4 (2015), 515–527.
  • [2] P. Bertozzini, R. Conti, and N. Pitiwan, Discrete non-commutative Gelfand-Naimark duality, East-West J. Mathematics 21, no. 2 (2019), 103–143.
  • [3] T. Bice, Dauns-Hofmann-Kumjian-Renault duality for Fell bundles and structured C*-algebras, arxiv:2104.12415.
  • [4] J. Dauns and K. H. Hofmann, Representations of rings by sections, Mem. Amer. Math. Soc. 83 (1968).
  • [5] K. H. Hofmann, Representations of algebras by continuous sections, Bulletin of the American Mathematical Society 78, no. 3 (1972), 291–373.
  • [6] K. H. Hofmann, The Dauns–Hofmann theorem revisited, J. Algebra Appl. 10, no. 1 (2011), 29–37.
  • [7] K. H. Hofmann, Some bibliographical remarks on representations of algebras by continuous sections, in: K. H. Hofmann and J. R. Liukkonen (eds.), Recent Advances in the Representation Theory of Rings and C∗–Algebras by Continuous Sections, Memoires of the American Mathematical Society 148 (1974), 177–182.
  • [8] A. Kumjian, Fell Bundles over groupoids, Proceedings of American Mathematical Society 126, no. 4 (1998), 1115–1125.
  • [9] C. M. Neira and J. Varela, Localization with change of the base space in uniform bundles and sheaves, Revista Colombiana de Matemáticas 41, no. 2 (2007), 303-333.
  • [10] J. Varela, Duality of C∗-Algebras, in: K. H. Hofmann and J. R. Liukkonen (eds.), Recent Advances in the Representation Theory of Rings and C∗–Algebras by Continuous Sections, Memoires of the American Mathematical Society 148 (1974).
  • [11] J. Varela, Sectional representation of Banach modules, Math Z. 139 (1974), 55–61.
  • [12] J. Varela, On the existence of uniform bundles, Revista Colombiana de Matemáticas XVIII (1984), 1–8.