Gelfand spectra in Grothendieck toposes using geometric mathematics
RR. Duncan and P. Panangaden (Eds.)Quantum Physics and Logic 2012 (QPL2012)EPTCS 158, 2014, pp. 77–107, doi:10.4204/EPTCS.158.7
Gelfand spectra in Grothendieck toposesusing geometric mathematics
Bas Spitters
VALS-LRI, Universit´e Paris-Sud/INRIA Saclay [email protected]
Steven Vickers ∗ School of Computer Science, University of Birmingham,Birmingham, B15 2TT, UK. [email protected]
Sander Wolters †Radboud University Nijmegen, IMAPP [email protected]
In the (covariant) topos approach to quantum theory by Heunen, Landsman and Spitters, one asso-ciates to each unital C*-algebra A a topos T ( A ) of sheaves on a locale and a commutative C*-algebra A within that topos. The Gelfand spectrum of A is a locale Σ in this topos, which is equivalent to abundle over the base locale. We further develop this external presentation of the locale Σ , by notingthat the construction of the Gelfand spectrum in a general topos can be described using geometriclogic. As a consequence, the spectrum, seen as a bundle, is computed fibrewise.As a by-product of the geometricity of Gelfand spectra, we find an explicit external descriptionof the spectrum whenever the topos is a functor category. As an intermediate result we show thatlocally perfect maps compose, so that the externalization of a locally compact locale in a topos ofsheaves over a locally compact locale is locally compact, too. The main subject of this paper is the interplay between geometric logic and topos-theoretic approachesto C*-algebras (motivated by quantum theory). In particular, we consider the approach of Heunen,Landsman and Spitters [8, 21, 23, 22], although some of the ideas and techniques in this paper may turnout to be of interest to the related approach by Butterfield, Isham and D¨oring [25, 6, 7, 20, 15] as well;see [50] for a comparison. We are mainly interested in the spectral object of the topos approach and itsconstruction using geometric logic.We assume that the reader is familiar with the basics of topos and locale theory. All toposes areassumed to be Grothendieck toposes, and, in particular, every topos has a natural numbers object (NNO).In its general form, the theory of Grothendieck toposes is that of bounded toposes over some base topos S that embodies the ambient logic. We shall rarely need to be explicit about S , but our techniques willbe valid for S an arbitrary elementary topos with NNO and thus have wide constructive applicability. ∗ Supported by the UK Engineering and Physical Sciences Research Council, on the project EP/G046298/1 “Applications ofgeometric logic to topos approaches to quantum theory”. † Supported by N.W.O. through project 613.000.811. The standard reference [33] contains much of the needed material. In particular, Chapter II of this book demonstrates howsheaves on a topological space can be seen as bundles. Chapter III gives an introduction to Grothendieck toposes. Section VII.1contains background information on geometric morphisms. Finally Chapter IX gives all background information on locales.At certain points in this paper, in particular in Subsection 3.4, sheaf semantics is used. The relevant background material canbe found in Chapter VI of [33]. Another standard reference is the massive work [30, 31]. Gelfand spectra in Grothendieck toposes
The paper is split into two sections. Section 2 gives background information on geometric logic and,more importantly, the practical impact of the geometric mathematics that develops from it. For this wecan give a first definition as topos-valid constructions on sets (understood as objects of a topos) whichare preserved by inverse image functors of geometric morphisms. This is the geometric mathematics of sets , as objects of a topos.More profoundly, we can also consider geometricity of topos-valid constructions on locales, andmuch of our ability to do this follows from two results in [32]. The first provides a localic version of awell known result from point-set topology, that sheaves over X are equivalent to local homeomorphismswith codomain X . On the one hand, the sheaves are the “sets” in the topos Sh ( X ) of sheaves, or thediscrete locales – the frames are the powerobjects. On the other hand, the local homeomorphisms can beunderstood as the fibrewise discrete bundles over X , where we understand “bundle” in the very generalsense of locale map with codomain X . Applying an inverse image functor f ∗ to the sheaf correspondsto pulling back the bundle along f . Hence geometricity of a construction on the sheaves corresponds topreservation under pullback of the corresponding construction on the bundles.The second result in [32] is that internal frames in Sh ( X ) are dual to localic bundles over X . Thisimmediately allows us to extend our definition of geometricity to constructions on locales, namely aspreservation under pullback of the bundle constructions. Since the fibres are got as pullbacks alongpoints, the geometric mathematics works fibrewise and provides a fibrewise topology of bundles. Thisidea has already been explored in an ad hoc fashion in point-set topology – see, e.g., [26] –, and the notionof geometricity makes it much more systematic when one combines point-free topology with toposes.It should be noted that pullback of a bundle along a map f is not achieved by applying f ∗ to theinternal frame. This is already clear in the fibrewise discrete (local homeomorphism) case, since f ∗ does not preserve powerobjects, nor frame structure in general. [32] define a different functor f thattransforms frames to frames, but in practice it is often convenient to bypass the frames altogether andinstead use presentations of them by generators and relations. [41] shows that applying f ∗ to the presen-tation corresponds to pullback of the bundle, even though the middle step of presenting a frame is notgeometric.The practical effect of switching to generators and relations is that a locale is described by means ofa geometric theory whose models are the (generalized) points of the locale. Hence we explicitly describethe points rather than the opens, though the nature of geometric theories ensures that the topology isdescribed implicitly at the same time, by presenting a frame. Geometricity has the effect of restoringthe points to point-free topology, and allows us to define maps f : Y → X in two very intuitive waysas geometric transformations: as a map, f transforms points of Y to points of X , while as a bundle ittransforms points of X to locales, the fibres, and defines the bundle locale Y at the same time. This isexplained in more detail in [48].This switch of emphasis can be disconcerting if one thinks that a locale is its frame. However,calculating an internal frame is error prone – experience shows this even in the simpler case of presheaftoposes –, and one of the central messages of this paper is that the geometric methods often allow usto avoid the frame. On the other hand, sometimes the frame is still needed. Later on we show howto exploit local compactness to give a geometric calculation of it. Specifically, we prove a result aboutexponentiability of objects in a topos, which entails that if X is a locally compact locale and Y is a locallycompact locale in Sh ( X ) , with bundle locale Y over X , then Y is locally compact.Section 3 applies the previous discussion about geometricity to the HLS topos approach of quantumtheory. Since the topos is that of sheaves over a locale, the spectrum can be understood as a bundle[16], in which each fibre is the spectrum of a commutative C*-algebra, and the geometric approach willemphasize this. In particular, we concentrate on the external description of the bundle locale, which is as Spitters, Steven Vickers & Sander Wolters This section is divided into four parts. In Subsection 2.1 we briefly discuss geometric logic and associ-ated mathematical constructions. An important point is that the constructions that can be expressed bygeometric mathematics coincide with the constructions that are preserved when pulled back along theinverse image functor of a geometric morphism. Subsection 2.2 shifts attention from geometric logicto geometric mathematics (in particular by using geometric type constructs). We discuss geometric-ity of topos-valid constructions on sets (in the generalized sense of objects in a topos) and on locales.Subsequently, in Subsection 2.3 the geometricity of such constructions is further analysed when theseconstructions are carried out in toposes that are functor categories and in toposes of sheaves on a topo-logical space. Finally, in Subsection 2.4 we discuss exponentiability of locales.The results of this section are important in two ways. First, they allow us to talk about the opens,the elements of a frame (which is itself not a geometric concept), in a geometric way. Second, it entailsthat the external description of the Gelfand spectrum investigated in the next section is a locally compactlocale.
We briefly describe geometric theories and their interpretations in toposes. The discussion is brief be-cause the work in Section 3 concentrates on geometric mathematics, as described in Subsection 2.2,rather than geometric logic as presented in this subsection. The reader is advised to browse rather thancarefully read this subsection, and to consult it at a later stage, if needed.In this subsection we follow the discussion in [30, Section D1] where a much more detailed pre-sentation is given. Section 3 of [45] also treats this material, with more emphasis on geometric logic.Another good source is [33, Chapter X], but note that what is called geometric there is called coherent in [30, 31] and [45]: the difference is that “coherent” forbids infinite disjunctions. Finally, [48] providesan alternative introduction to some of the techniques presented in this section.The language of geometric logic is that of an infinitary, first-order, many-sorted predicate logic withequality . We start with a first-order signature Σ (where the notation Σ has nothing to do with thespectrum Σ in the following sections). The signature consists of a set S of sorts , a set F of functionsymbols and a set R of predicate symbols . Each function symbol f ∈ F has a type, which is a non-empty We shall not explicitly consider the geometric types of e.g. [45, Subsection 3.4]. Gelfand spectra in Grothendieck toposes finite list of sorts A , ..., A n , B . We write this as f : A × ... × A n → B , in anticipation of the categoricalinterpretation of the function symbols. The number n is called the arity of the function symbol. If thearity of a function symbol f is 0, i.e., when its type is a single sort B (we will write f : 1 → B ), f iscalled a constant . Each predicate P ∈ R also has a type A , ..., A n , where we also allow the empty list( n =
0) as a type. We will write P ⊆ A × ... × A n , again in anticipation of the categorical representation.A predicate with arity equal to 0, written as P ⊆
1, will be called a propositional symbol .Given the signature Σ , a context is a finite set of variables, each associated with some sort. It iscustomary to write the set as a vector (cid:126) x = ( x , ..., x n ) of distinct variables. Using Σ and the variables wecan define terms and formulae over Σ in that context. Terms , which all have a sort assigned to them, areinductively defined as follows: • Each variable x of sort A is a term of sort A . • If f : A × ... × A n → B is a function symbol, and t , ..., t n are terms of sort A , ..., A n , then f ( t , ..., t n ) is a term of sort B . In particular, every constant is a term.We can now construct geometric formulae in that context inductively as follows. They form thesmallest class which is closed under the following clauses:1. If R ⊆ A × ... × A n is a predicate, and t , ..., t n are terms of sort A , ..., A n , then R ( t , ... t n ) is aformula. For the particular case n =
0, even in the empty context every propositional symbol is aformula.2. If s and t are terms of the same sort, then s = t is a formula.3. Truth (cid:62) is a formula. If φ and ψ are formulae, then so is the conjunction φ ∧ ψ .4. Let I be any (index) set and for every i ∈ I , let φ i be a formula. Then (cid:87) i ∈ I φ i is a formula.5. If the variable u is not in the context (cid:126) x , and φ is a formula in context (cid:126) x ∪ { u } , then ( ∃ u ) φ is aformula in context (cid:126) x .Note that implication, negation and the universal quantifier are not allowed as connectives in the con-struction of geometric formulae.To show the context explicitly we shall denote a formula or term in context by (cid:126) x . φ or (cid:126) x . t . Note that aformula or term does not have to use all the variables in its context – some may be unused.We will also consider sequents φ (cid:96) (cid:126) x ψ , where φ and ψ are geometric formulae in context (cid:126) x . We willthink of the sequent as expressing that ψ is a logical consequence of φ in context (cid:126) x . A geometric theory T over Σ is simply a set of such sequents φ (cid:96) (cid:126) x ψ .The next step is to consider interpretations of geometric theories in toposes. The discussion will bevery brief, but the details can be found in the references stated at the beginning of this subsection. Let E be a topos (actually any category that has finite products would suffice for defining the structures,and also assuming pullbacks makes the definition of a homomorphism of Σ -structures nicer). Given asignature Σ , a Σ -structure M in E is defined as follows. For every sort A in Σ there is an associated object MA in E . For every function symbol f : A × ... A n → B there is an arrow M f : MA × ... × MA n → MB in E . A constant c : 1 → B is interpreted as an arrow Mc : 1 → MB , where 1 denotes the terminal objectof E . A predicate R of type A × ... × A n is interpreted as a monic arrow MR (cid:26) MA × ... × MA n .If M and N are Σ -structures in E , then a homomorphism of Σ -structures h is defined as follows.For each sort A in Σ there is an arrow h A : MA → NA . For each function symbol f : A × ... × A n → B ,we demand h B ◦ M f = N f ◦ ( h A × ... × h A n ) . If R ⊆ A × ... × A n is a predicate, then we demand that This should not be confused with the non-logical notion of context used in the topos approaches to quantum theory. as Spitters, Steven Vickers & Sander Wolters MR ⊆ ( h A × ... × h A n ) ∗ ( NR ) holds as subobjects of MA × ... × MA n , where the right hand side meanspulling the monic arrow NR (cid:26) NA × ... × NA n back along h A × ... × h A n .The Σ -structures in a topos E and their homomorphisms define a category Str Σ ( E ) . Let F : E → F be a functor between toposes. F need not come from a geometric morphism, but we do assume it to beleft exact. Any such functor induces a functor Str Σ ( F ) : Str Σ ( E ) → Str Σ ( F ) in a straightforward way.The next step is to introduce models of a geometric theory T over Σ . In order to do this we need tointerpret terms and formulae-in-context for a Σ -structure in E . This can be done inductively, in muchthe same way as in using the internal language of the topos. Details can be found in [30, SubsectionD1.2]. For a given Σ -structure M , the end result is that a formula-in-context (cid:126) x . φ , where (cid:126) x = ( x , ..., x n ) are variables with associated sorts A , ..., A n , is interpreted as a subobject M ( (cid:126) x . φ ) of MA × ... × MA n .A Σ -structure M in a topos E is called a model for a geometric theory T if for every sequent φ (cid:96) (cid:126) x ψ in T we have M ( (cid:126) x . φ ) ⊆ M ( (cid:126) x . ψ ) , where we view the interpretation of the formulae as subobjects of MA × ... × MA n . We write Mod T ( E ) for the full subcategory of Str Σ ( E ) whose objects are the modelsof T .Although any left-exact functor F : E → F induces a functor Str Σ ( F ) between the associated cat-egories of Σ -structures, in general F does not preserve those “geometric constructions” used to inter-pret formulae, and consequently Str Σ ( F ) does not restrict to model categories. On the other hand, if F = f ∗ , the inverse image of a geometric morphism f , then Str Σ ( F ) does restrict to a functor Mod T ( F ) : Mod T ( E ) → Mod T ( F ) . A nice proof of the restriction is given in [33, Section X.3]. Although only fi-nite joins are considered there, this proof is in particular interesting because it also treats (non-geometric)formulae that use implication and the universal quantifier, and shows that models of a theory using thisadditional structure are only preserved by the inverse image functor of a geometric morphism when thegeometric morphism is open. In general, we shall describe a topos-valid construction as geometric if itis preserved (up to isomorphism) by inverse image functors.One important kind of mathematical structure that can be expressed by geometric logic is finitaryalgebraic theories, such as monoids, (Abelian) groups, rings and lattices. Their axioms all take the shape (cid:62) (cid:96) (cid:126) x s = t where s and t are terms. These theories are clearly geometric, as are many-sorted algebraictheories such as pairs of rings and modules. The theory of partially ordered sets, but also ordered groupsor ordered rings, is geometric. More general examples are local rings, finite sets and (small) categories.See [30, Example D1.1.7] for more details.Sometimes it can be hard to see whether some given structure is geometric. Just because it is pre-sented by a theory that uses ∀ or ⇒ in formulae, does not mean that there is no equivalent geometrictheory that describes the structure.One example of structure that is definitely not geometric is that of frames: complete lattices withbinary meet distributing over arbitrary joins, and homomorphisms preserving joins and finite meets.This will be important in Section 3, where we will look at spectra of commutative unital C*-algebrasin toposes: the spectrum is described by a frame. The non-geometricity of frames is shown by the factthat the inverse image functor of a geometric morphism does not necessarily map frames to frames.C*-algebras provide another example: inverse image functors need not map C*-algebras to C*-algebras.Still, we can say a lot about the spectrum of a commutative C*-algebra by paying attention to geometricconstructions, as we will see in Subsection 3.2.Although there is not a single geometric theory whose models are the frames, they are closely relatedto the important class of propositional geometric theories – those with no declared sorts so that the signa-ture consists entirely of propositional symbols. Each such theory can be straightforwardly translated intoa presentation of a frame by generators and relations, with propositional symbols and axioms becomingthe generators and relations, and then the points of the corresponding locale are equivalent to the models2 Gelfand spectra in Grothendieck toposes of the theory.Defining the notion of geometric theory and their models in toposes is just the start of an introductionto geometric logic. The next step would be to discuss the rules of inference. However, for the purposesof this paper we have treated enough and the reader interested in the rules of inference is directed to [30,Section D1.3].More important for us will be the way that geometric theories are used to describe the generalizedpoints of a Grothendieck topos F , the geometric morphisms E → F to F from another topos. When-ever we refer to points we mean generalized points in the above sense, and not the smaller class of global points , i.e., points where E = Set . A key fact is that for each F there is a geometric theory T such that the points of F are equivalent to the models of T in E . F is said to classify the theory T .Conversely, every geometric theory has a classifying topos.The same techniques can also be applied to localic toposes, i.e., those of the form Sh X , where X is alocale. They classify propositional geometric theories.The geometric approach will continually ask – What are the points? What theory does it classify?This will come as a shock to those used to thinking of the points as insufficient – as, indeed, the globalpoints are – and calculating concretely with the topos or the frame. In Subsection 2.1 we gave formal definitions of geometric theories and their interpretation in toposes.In what follows, however, we shall make little use of formal geometric theories. This is not just for theusual reason, that in practical mathematics it is tedious to work formally. We shall also be extensivelyusing type-theoretic features of the logic that have not been fully formalized. Our informal treatmentwill be what we describe as “geometric mathematics”.It has two levels. A geometric mathematics of “sets” (understood as objects of a topos) consistsof those structures, constructions and theorems that are preserved by inverse image parts of geometricmorphisms. The next level is a geometric mathematics of locales, and we shall explain geometricity herein terms of bundles.A key property of geometric logic, distinguishing it from finitary first-order logics, is that the in-finitary disjunctions allow us to characterize some constructions (necessarily geometric) uniquely up toisomorphism by using geometric structure and axioms. These constructions include N , Z and Q (thoughnot R and C ) , free algebras more generally, colimits and finite limits. Using these constructions infor-mally, but in the knowledge that they could be formalized within geometric theories, allows us to dealwith geometric theories and their models in a way that is closer to a “geometric mathematics” than tologic. We shall do this extensively in the rest of the paper.To explain how this extends to constructions on locales, we shall need to recall the equivalencebetween internal locales and bundles. For simplicity we assume we are working in a topos E = Sh ( X ) ,where X is a locale in Set , though the idea works more generally. Then the category Loc Sh ( X ) oflocales in Sh ( X ) (defined as the dual of the category of internal frames) is equivalent to the slice category Loc / X , of locales in Set over X ([32] or [30, Section C1.6]). A locale Y in the topos E = Sh ( X ) can berepresented by a locale map p : Y → X . We will call a map a bundle when we view it in this way, andthink of it as a family of locales, the fibres, parametrized by (generalized) points of X . Let x : X (cid:48) → X be – or an arbitrary base topos S , elementary with NNO. We do not assume classical sets. as Spitters, Steven Vickers & Sander Wolters
83a point of X , and consider the pullback diagram X (cid:48) × X Y (cid:15) (cid:15) (cid:47) (cid:47) Y p (cid:15) (cid:15) X (cid:48) x (cid:47) (cid:47) X If X (cid:48) =
1, so x is an ordinary (global) point of X , then the pullback X (cid:48) × X Y is the fibre of p over x . Whenwe generalize to arbitrary X (cid:48) then we may think of the pullback as the generalized fibre of p over thegeneralized point x .If we define toposes E = Sh ( X ) and F = Sh ( X (cid:48) ) , then x becomes a geometric morphism x : F → E . As we have noted earlier, the inverse image part x ∗ : E → F does not preserve frame structure.Nonetheless, there is a functor x : Frm E → Frm F whose action on the corresponding bundles is bypullback along x . Moreover it is a right adjoint to x ∗ : Frm F → Frm E , the direct image part of x , which,unlike x ∗ , does preserve frames.For each object U in a topos E there is a corresponding discrete locale whose frame is the power-object P U . The corresponding bundles are the ´etale bundles, or local homeomorphisms ([33, ChapterII]), whose fibres are normally called stalks. If F is a geometric morphism with codomain E then f ( P U ) ∼ = P f ∗ ( U ) , and it follows that applying f ∗ corresponds to pulling back the bundle of the dis-crete locale. Our notion of geometricity with regard to constructions on objects of toposes consisted ofpreservation under inverse image functors. If we translate the construction into one on the bundles, wesee that geometricity comes to mean preservation under pullback.In this form we have a notion of geometricity that can also be understood with regard to locales,once they are interpreted as bundles. Another way to say the same thing is that the mathematics worksfibrewise (for generalized fibres), since fibres are just pullbacks of bundles.We often make use of the following observation. Let p : Y → X be an arbitrary localic bundle: wetake it that there is an internal locale in Sh X corresponding to it. If y : W → Y is a (generalized) point of Y , then we obtain a point x = py of X and a map y (cid:48) : W → x ∗ Y to the pullback x ∗ Y : W y (cid:36) (cid:36) y (cid:48) (cid:33) (cid:33) x ∗ Y (cid:15) (cid:15) (cid:47) (cid:47) Y p (cid:15) (cid:15) W x (cid:47) (cid:47) X Thus we may describe the points of Y as the pairs ( x , y (cid:48) ) , where x is a point of X and y (cid:48) is a point of thefibre x ∗ Y .This description is geometric, because the structure and properties of the diagram are preserved byany change of base (pullback along maps into W ). It follows that, for a geometric theory for Y , we needone whose models are the points ( x , y (cid:48) ) .We shall also need to consider the specialization order (cid:118) on Y , and for this the analysis needs to bemore refined. The straightforward part is that within a single fibre x ∗ Y we have y (cid:48) (cid:118) y (cid:48) iff ( x , y (cid:48) ) (cid:118) ( x , y (cid:48) ) .The ⇒ direction follows immediately from the map x ∗ Y → Y , while the converse follows from the factthat in Loc the pullback property determines the hom- po sets into a fibre, as well as the homsets. Also, if ( x , y (cid:48) ) (cid:118) ( x , y (cid:48) ) then x (cid:118) x by the map p .4 Gelfand spectra in Grothendieck toposes
For a more precise analysis of the specialization between fibres we shall assume that p is an opfi-bration in the 2-categorical sense. (A similar analysis applies when p is a fibration.) For a definition,see [30, B4.4.1]. However, easier to understand in our situation is to define the opfibrational structuregenerically, as in [16], which also discusses other ways in which fibrations and opfibrations relate to thetopos approaches to quantum theory. Consider the two maps X i : X S → X where i is either of the twoprincipal points ⊥ , (cid:62) of S . We have obvious maps Y S → ( X i ) ∗ Y , defined on ( x ⊥ , y (cid:48)⊥ , x (cid:62) , y (cid:48)(cid:62) ) by forgettingone of the y (cid:48) i s. Then p is an opfibration if the map Y S → ( X ⊥ ) ∗ Y has a left adjoint over ( X ⊥ ) ∗ Y . We canwrite it in the form ( x ⊥ , x (cid:62) , y (cid:48)⊥ ) (cid:55)→ ( x ⊥ , y (cid:48)⊥ , x (cid:62) , r x ⊥ x (cid:62) ( y (cid:48)⊥ )) ,thus focusing attention on the fibre map r x ⊥ x (cid:62) : x ∗⊥ Y → x ∗(cid:62) Y . Proposition 1.
Let p : Y → X be a locale map that is a 2-categorical opfibration. Then ( x , y (cid:48) ) (cid:118) ( x , y (cid:48) ) in Y iff x (cid:118) x in X and r x x ( y (cid:48) ) (cid:118) y (cid:48) in x ∗ Y .Proof. ⇒ : We have already noted that x (cid:118) x . Using the adjunction we find ( x , y (cid:48) , x , r x x ( y (cid:48) )) (cid:118) ( x , y (cid:48) , x , y (cid:48) ) and the conclusion follows. ⇐ : Using the fact that the adjoint maps into Y S , we know ( x , y (cid:48) ) (cid:118) ( x , r x x ( y (cid:48) )) (cid:118) ( x , y (cid:48) ) .There is a dual result if p is a fibration, giving fibre maps contravariantly r x (cid:62) x ⊥ : x ∗(cid:62) Y → x ∗⊥ Y . Then ( x , y (cid:48) ) (cid:118) ( x , y (cid:48) ) in Y iff x (cid:118) x in X and y (cid:48) (cid:118) r x x ( y (cid:48) ) in x ∗ Y .If a locale property or construction is understood in terms of frames, then its geometricity itself islikely to be non-obvious, because frames are non-geometric. One potential way round this is to checkpreservation by the functors f , but this can be difficult without concrete knowledge of f . A morepractical approach is usually to work with frame presentations by generators and relations. A frame O X that is presented by a set of generators G and a set of relations R will be denoted as O X ∼ = Fr (cid:104) G | R (cid:105) . Forthe reader unfamiliar with frame presentations, see e.g. [39, Chapter 4] for an introduction. A general butprimitive class of frame presentations is that of GRD-systems [41, Section 5], but there are various morerestricted kinds that are better adapted to particular problems. Of whatever kind, the central property isthat the relation between the presentation and the corresponding bundle is geometric – that is, preservedby bundle pullback. For the topos approach(es) to quantum theory we are in particular interested in toposes that are eitherfunctor categories or sheaves on a locale. In this subsection we consider geometricity for constructionson sets and locales in such toposes. As before, we are considering topos-valid constructions, so sets andtheir elements are understood in their generalized sense.For functor categories, the fact that geometric constructions are preserved under the inverse imagefunctor of any geometric morphism entails the following lemma.
Lemma 2. ([30, Corollary D1.2.14(i)]) Let T be a geometric theory over a signature Σ and let C beany small category. A Σ -structure M in the topos [ C , Set ] is a T -model iff for every object C ∈ C the Σ -structure ev C ( M ) is a T -model in Set . Here ev C : [ C , Set ] → Set denotes the functor that evaluates atthe object C. There is an isomorphism
Mod T ([ C , Set ]) ∼ = [ C , Mod T ( Set )] . There are also topological notions of fibration, and they are different from the one we use. as Spitters, Steven Vickers & Sander Wolters ev C is the inverse image part of a geometricmorphism. The observation that we have an isomorphism of categories of models uses the fact that ahomomorphism of Σ -structures in [ C , Set ] can be identified with a natural transformation between the Σ -structures, viewed as functors C → Str Σ ( Set ) .[30, Corollary D1.2.14(ii)] shows that for a spatial locale X , it suffices to check the fibrewise naturefor fibres over the global points. Let T be a geometric theory over a signature Σ . Then a Σ -structure M in the topos Sh ( X ) is a T -model iff, for each global point x , the fibre x ∗ ( M ) is a T -model in Set .Unfortunately, from a geometric point of view, spatiality is an uncommon property that often dependson classical reasoning principles. Lemma 2 is geometrically justified, because the Yoneda embedding Y already provides enough points of the form Y ( C ) .If the geometric theory T in Lemma 2 is classified by the topos E , then the lemma tells us thatgeometric morphisms from [ C , Set ] to E are equivalent to C -indexed diagrams of points of E . In fact,this can be used to show that [ C , Set ] is exponentiable as a topos, with E [ C , Set ] classifying the theory of C -indexed diagrams of points of E (see e.g. [28]).We should also ask what are the geometric morphisms to [ C , Set ] , i.e., its points. Definition 3.
Let C be a small category, and F : C op → Set a presheaf. F is flat if it has the followingproperties1. For some C ∈ C , F ( C ) is inhabited.2. If x i ∈ F ( C i ) (i = , ), then there are morhisms f i : C i → D in C , and y ∈ F ( D ) , such that x i = F ( f i )( y ) .3. If f , g : C → D in C and y ∈ F ( D ) with F ( f )( y ) = F ( g )( y ) , there is is some morphism h : D → Esuch that h f = hg, and some z ∈ F ( E ) such that y = F ( h )( z ) . The theory of flat presheaves over C is geometric, and its models are equivalent to the points of [ C , Set ] (see e.g. [45]).The role of the flatness conditions becomes clearer if one considers the Grothendieck construction ,which turns a presheaf F into a category (cid:82) F , the so-called category of elements of F . Its objectsare pairs ( C , u ) with u ∈ F ( C ) , and a morphism from ( C , u ) to ( D , v ) is an arrow f : C → D such that u = F ( f )( v ) . Then F is flat iff (cid:82) F is filtered, and in that case F can be thought of as a filtered diagramof representable presheaves Y ( C ) (which are themselves flat). In fact, as point of [ C , Set ] , F is a filteredcolimit of representables.We shall mostly use this in the case where C is a poset P . Lemma 4.
For a poset P, flat presheaves F : P op → Set correspond up to isomorphism with ideals of P.Proof.
Let F be a flat presheaf, and p ∈ P . By Definition 3(2) F ( p ) contains at most one element. Bythis observation, we identify the set of objects of (cid:82) F with a subset I ⊆ P , where p ∈ I iff there exists an(automatically unique) element x ∈ F ( p ) , i.e., ( p , x ) ∈ (cid:82) F . Under the identification of (cid:82) F with I , thearrows of (cid:82) F translate to the order relation of P , restricted to I . By conditions (1) and (2) of Definition 3, I is non-empty and upward directed. Condition (3) automatically holds for posets. The set I is downwardclosed as F is a presheaf. For a flat functor F , we have identified the category (cid:82) F with an ideal I of P . Conversely, to each ideal I of P , we can associate a flat functor F , which has I as its category ofelements.Then the flat presheaves are simply the ideals of P , and we see that the functor topos [ P , Set ] is thetopos of sheaves over the locale Idl P whose points are the ideals of P and whose opens are the Scottopens of the ideal completion, or, equivalently, the Alexandrov opens (up-closed subsets) of P .6 Gelfand spectra in Grothendieck toposes
A technique that we shall find useful in various places is a simple form of iterated forcing . Suppose D is an internal category in [ C , Set ] , hence a functor from C to Cat . A topos [ D , [ C , Set ]] of internaldiagrams can be defined, together with a geometric morphism to [ C , Set ] . A point of it is a pair ( F , G ) where F is flat presheaf over C and G is a flat presheaf over D ( F ) , i.e. colim ( C , x ) C ( C ) , the filteredcolimit of the diagram corresponding to F .It turns out (see e.g. [30, C2.5]) that [ D , [ C , Set ]] too is a presheaf topos, over the category C (cid:110) D defined as follows. Its objects are pairs ( C , D ) with C an object of C and D ∈ D ( C ) , where D : C → Set is the object of objects of D . An arrow ( f , g ) : ( C , D ) → ( C (cid:48) , D (cid:48) ) in C (cid:110) D is given by an arrow f : C → C (cid:48) in C and an arrow g ∈ D ( C (cid:48) ) , g : D ( f )( D ) → D (cid:48) .As an example, consider the case where D is discrete, corresponding to an object U : C → Set in [ C , Set ] . Then the geometric morphism [ D , [ C , Set ]] → [ C , Set ] is the local homeomorphism correspond-ing to U . We thus see that the external view of the internal discrete locale U is also given by a presheaftopos, on C (cid:110) U . Its objects are pairs ( C , u ) ( C an object of C , u ∈ U ( C ) ), and a morphism from ( C , u ) to ( D , v ) is a morphsm f : C → D such that v = U ( f )( u ) . This subsection presents a general result (Theorem 7) on exponentiability. It is not specifically aboutgeometricity, but is related because geometricity would like to replace the non-geometric construction ofthe frame O X by the locale exponential S X , which has the same points ( O X is the set of points of S X )and is geometric. However, the exponential exists only if X is locally compact, and so we are going tobe interested in local compactness in order to define the elements of the frame geometrically.In fact, the following known theorem [24] holds for locales in any elementary topos. Theorem 5.
Let X be a locale, then the following are equivalent:1. X is locally compact.2. The functor ( − ) × X : Loc → Loc has a right adjoint ( − ) X .3. The exponential S X exists, where S denotes the Sierpi´nski locale. We are going to build up to Theorem 6, stating that ‘locally perfect maps compose’, where p : Y → X is locally perfect if the corresponding internal locale Y in Sh ( X ) is locally compact. The precise formof our result is that if X is a locally compact locale, and p : Y → X is locally perfect, then Y is locallycompact. However, this can be relativized. If also q : Z → Y is locally perfect, then by working in Sh ( X ) we find that pq is locally perfect.We shall apply this to the spectrum of an internal C*-algebra to show that its external description isa locally compact locale.Consider p : Y → X as above, and assume for a moment that the exponential S Y exists. A point of S Y is equivalent to a map Y → S , which corresponds to an open U ∈ O Y . By assumption Y is locallycompact in Sh ( X ) , so the exponential S Y exists in Sh ( X ) , where S denotes the internal Sierpi´nski locale.Using the fact that Loc Sh ( X ) is equivalent to Loc / X , the locale S Y has an external description by a localemap q : S YX → X , for some locale S YX . The external description of S is the projection π : S × X → X . As The terminology is debatable, and will depend on whether one calls p perfect or proper for the situation where Y iscompact. In the present paper we are following Johnstone’s usage [27], where a map f : X → Y between topological spaces iscalled locally perfect if the following condition holds. If U ∈ O X and x ∈ U , then there exists an open neighborhood V of x ,an open neighborhood N of f ( x ) , and a set K ⊆ X , such that K ⊆ U ∩ f − ( N ) and for each b ∈ N , the fibre K b is compact. Heshows that if f is locally perfect, then f ∗ ( X ) is a locally compact locale in Sh ( Y ) , and the converse holds if Y satisfies the T D separation property. as Spitters, Steven Vickers & Sander Wolters q ∗ ( { x } ) over a point x in X is given by S Y x ,where we write Y x for the fibre x ∗ Y = p − ( { x } ) .An open U ∈ O Y and a point x ∈ X give an open U x in the fibre Y x , as in the figure below. Thisin turn is equivalent to a map Y x → S , which is a point of q ∗ ( { x } ) . This suggests that the global pointsof S Y correspond exactly to the global sections of the bundle q : S YX → X . That is, global points of S Y correspond to maps σ : X → S YX such that q ◦ σ = id X .The informal reasoning above can be made precise in order to prove that for a locally compact X ,and Y a locally compact in Sh ( X ) , with external description p : Y → X , the locale Y is locally compact.We define locales by providing their geometric theories (of generalized points) and we define continuousmaps as constructively described transformations of points of one locale to points of another locale;see [45, Sec. 4.5]. Theorem 6 (locally perfect maps compose) . If X is a locally compact locale, and Y a locally compactlocale in Sh ( X ) , with external description p : Y → X , then Y is locally compact.Proof.
The locale Y is the locale with (generalized) points the pairs ( x , t ) such that x ∈ X , t ∈ Y x . Thatis, we view the map p as a bundle; see [48] and the discussion of the bundle picture in Subsection 2.2.As before, S YX denotes the external description of the internal locale S Y in Sh ( X ) . Since the exponentialis geometric [41, Sec. 10.3], the generalized points of S YX are the pairs ( x , w ) such that x ∈ X , w ∈ S Y x .The internal evaluation map ev : S Y × Y → S , part of the geometric structure of S Y , must then correspondto a map ev : S YX × X Y → S given by (( x , w ) , y ) (cid:55)→ w ( y ) . We define E as the locale with (generalized)points those σ : X → S YX such that q ◦ σ = id X . Here local compactness of X allows us to define theexponential ( S YX ) X , and an equalizer captures the section condition to give a sublocale. To define the mapev : E × Y → S , we first define the map E × Y → S YX × X Y by ( σ , y ) (cid:55)→ ( σ ( py ) , y ) and then compose itwith the map ev : S YX × X Y → S .To complete the proof, we require that if g : Z × Y → S , then there is a unique ˜ g : Z → E such that g = ev ◦ ( ˜ g × id Y ) . This condition amounts to saying that for all z , y , g ( z , y ) = w ( y ) , where ˜ g ( z )( py ) = ( py , w ) .Assuming this condition, consider ˜ g ( z )( x ) = ( x , w ) , for some w in S Y x . For all y in Y x we must have˜ g ( z )( py ) = ˜ g ( z )( x ) = ( py , w ) , and by the condition w ( y ) = g ( z , y ) . Hence w is uniquely determined foreach x , and so ˜ g is unique. Reversing the argument, and relying heavily of geometricity, we see that itleads to a definition of ˜ g , given g .8 Gelfand spectra in Grothendieck toposes
Theorem 6 holds in a much more abstract categorical form. This can be applied when the base X isa non-localic topos that is exponentiable in the category of toposes – our prime example will be a toposof the form [ C , Set ] where C is not a poset. Theorem 7.
Let C be a category with finite limits, and let X be an exponentiable object in C . Letp : Y → X be an object Y of C / X , let Z be an object of C , and suppose the exponential Z Y exists in C / X .We shall write it as q : Z YX → X . Then Z Y exists in C .Proof. By the considerations above Theorem 6, we arrive at the following candidate for the exponential Z Y . Take the equalizer E (cid:31) (cid:127) eq (cid:47) (cid:47) ( Z YX ) X q X (cid:43) (cid:43) (cid:112) X (cid:113) ◦ ! (cid:51) (cid:51) X X , where (cid:112) X (cid:113) : 1 → X X denotes the transpose of the identity arrow of X . Note that the exponentials ( Z YX ) X and X X exist in C by exponentiability of X . Also note that the global points of E are exactly the globalsections of q . Next, we need to find a suitable evaluation map ev : E × Y → Z . For the definition of ev we will make use of the internal evaluation arrow Z Y × Y → Z . Externally this gives the followingcommuting triangle: Z YX × X Y ev (cid:47) (cid:47) (cid:35) (cid:35) Z × X π (cid:125) (cid:125) X With some abuse of notation, we denote the map π ◦ ev : Z YX × X Y → Z again by ev . For the next step indefining the evaluation map, the diagram given below is commutative by the definition of E . E × Y π (cid:15) (cid:15) E × p (cid:47) (cid:47) E × X π (cid:47) (cid:47) eq × X (cid:47) (cid:47) ( Z YX ) X × X ev (cid:47) (cid:47) ! × X (cid:121) (cid:121) q X × X (cid:29) (cid:29) Z YXq (cid:15) (cid:15) × X (cid:112) X (cid:113) × X (cid:42) (cid:42) X X × X ev (cid:35) (cid:35) Y p (cid:47) (cid:47) X The evaluation maps and exponentials in this diagram exist because of the exponentiability of X .By the universal property of pullbacks this diagrams yields an arrow E × Y → Z YX × X Y . Taking thecomposition with ev : Z YX × X Y → Z coming from the internal exponential Z Y gives the desired evaluationmap ev : E × Y → Z .It remains to check that this map satisfies the desired universal property. Maps Z × Y → Z correspondbijectively with maps ( Z × X ) × X Y → Z × X over X . By the existence of the internal exponent Z Y thelatter maps correspond bijectively with maps Z × X → Z YX over X . Using exponentiability of X maps Z × X → Z YX correspond bijectively with maps Z → ( Z YX ) X . The maps Z × X → Z YX that are maps over X precisely correspond to the maps Z → ( Z YX ) X that factor through E . This proves that E is indeed anexponential Z Y . as Spitters, Steven Vickers & Sander Wolters Y is exponentiable in C / X , then Y is exponentiablein C . Theorem 6 is the special case where C is the category of locales. The topos approaches to quantum theory were inspired by the work of Butterfield and Isham [25, 20, 6, 7].The original “BI” approach, or presheaf approach (as it is formulated using a topos of presheaves) wassubsequently developed by D¨oring and Isham [15]. In the present paper we are particularly interestedin the different formulation of [21], which we call the copresheaf approach. What these two approachesshare is their use, for a given quantum system described by a C*-algebra, of a topos whose internal logicembodies the idea of fixing a generic context , or classical perspective on that system. Thus in the internallogic, the system has a classical phase space. Our aim here is to show how this internal phase space canbe represented externally by a bundle, whose fibres are the phase spaces for individual contexts.In the approaches cited, a unital C*-algebra A is studied through the poset C ( A ) of commutativeunital C*-subalgebras of A , ordered by inclusion, and using a topos of functors from C ( A ) to Set .At this point the presheaf and copresheaf approaches diverge, using contravariant and covariant func-tors respectively. We shall focus now on the copresheaf approach with its topos [ C ( A ) , Set ] . To studythe operator algebra A from the perspectives of its commutative C*-subalgebras, it is replaced by thecovariant functor A : C ( A ) → Set , A ( C ) = C , where the arrows are mapped to inclusions. In the topos [ C ( A ) , Set ] , A becomes an internal unital commutative C*-algebra.From the internal perspective of the topos, the quantum-mechanical observables, now in the form ofthe commutative C*-algebra A , look more like the observables of a classical physical theory. Crucial forthis is the topos-valid version of Gelfand duality, described in the work of Banaschewski and Mulvey [3,2, 4] as a duality between the category of unital commutative C*-algebras and the category of compact,completely regular locales. A more explicit and fully constructive description of this Gelfand dualityis given in [9, 10]. By this duality, A is isomorphic to the C ∗ -algebra of continuous, complex-valuedfunctions on a certain compact, completely regular locale Σ , the spectrum of A , which we think of as aphase space. For further discussion of the ideas of the topos approach, such as the treatment of states, see[21, 23, 22, 8, 16]. We also mention exciting recent work which reconstructs the Jordan algebra structurefrom the spectral object [18, 19, 14, 13, 12].In this section we apply the geometricity ideas of the previous section to the topos approach describedin [21]. In particular, we concentrate on the description of the spectrum of internal commutative unitalC*-algebras in toposes. The first subsection examines the geometricity of parts of the definition of C*-algebra. This is a central issue, since the C*-algebra structure itself is not geometric, so it must bereplaced by a geometric structure that, while not being the whole C*-algebra, nonetheless supports ageometric construction of the spectrum.The second subsection examines the geometricity of the construction of the spectrum. The thirdsubsection is concerned with finding an explicit description of the spectrum of the algebra A in theparticular topos [ C ( A ) , Set ] , which is central to the topos approach. The fourth subsection generalizesthis description to spectra of commutative unital C*-algebras in functor categories in general. In the fifthsubsection we consider extending the copresheaf approach to algebraic quantum field theory. Throughoutthe emphasis will be on the role of geometric logic. In the presheaf approach the smaller class of von Neumann algebras are typically considered, rather than arbitrary unitalC*-algebras. Gelfand spectra in Grothendieck toposes
In a topos E , a C*-algebra A (always assumed unital here) –1. is an associative, unital algebra over C with an involution ∗ , which is anti-multiplicative ( ( ab ) ∗ = b ∗ a ∗ ) and conjugate linear ( ( za + b ) ∗ = ¯ za ∗ + b ∗ , where z ∈ C ),2. has a norm (cid:107)−(cid:107) , which is submultiplicative and satisfies the axiom (cid:107) aa ∗ (cid:107) = (cid:107) a (cid:107) , and3. is complete with respect to the norm.However, the definition is not geometric, for three reasons.First, with regard to the algebraic structure (1) above, already C – as set of complex numbers ratherthan a locale – is not a geometric construction. For a truly geometric account, a C*-algebra would haveto be a locale rather than a set. In [4], a C*-algebra is defined as an algebra over the Gaussian numbers Q [ i ] , upon which completeness then allows the action of Q [ i ] to be extended to C . Apart from this, thealgebraic part is straightforward. It consists of arrows + , · : A × A → A for addition and multiplication,an arrow ∗ : A → A for the involution, an arrow Q [ i ] × A → A for scalar multiplication, and constants0 , → A for the unit and zero element. These arrows are to render all desired diagrams commutative.Note that we made use of the geometricity of Q [ i ] . In the language of Subsection 2.1, in a formalgeometric theory we could declare a sort k , add structure and axioms to force it to be isomorphic to Q × Q , and then define the appropriate operations as ring with involution.Second, the norm is not geometric. We shall develop a geometric theory of “commutative G*-algebra”, more general than unital commutative C*-algebras, but expressing enough of the structure todefine the spectrum. For the norm, we shall ignore the condition (cid:107) a (cid:107) = → a =
0, as it is not geometric :so we have only a semi-norm. As explained in [43], the semi-norm as described in [4] with a binaryrelation N ⊆ A × Q + can be understood as a map (cid:107)·(cid:107) : A → ←−−− [ , ∞ ] taking its values in the upper reals .Then ( a , q ) ∈ N if (cid:107) a (cid:107) < q . The third reason, connected to the first one, is that the completeness of A with respect to the norm is not geometric.What we seek, therefore, is a geometric notion that generalizes commutative C*-algebras, and onwhich we can still, and geometrically, calculate the spectrum.There is a geometric core to the definition of C*-algebra, in the notion of semi-normed pre-C*-algebra – that is, a *-algebra over Q [ i ] , but dropping completeness and weakening the norm to a semi-norm. However, that runs into problems because at a certain point in constructing the spectrum we needto know the the order on the self-adjoints A sa , or, alternatively, its positive cone. In a C*-algebra a self-adjoint is (non-strictly) positive iff it is a square, but without completeness we cannot guarantee the vitalproperty that the sum of squares is still a square – for example, 2 might not be a square.Our way round this is to use the preordered archimedean rings of [9]. Definition 8.
A commutative Q -algebra R is called preordered if it has a positive cone , i.e., a subset Pthat contains all squares and is closed under addition and multiplication by Q + . The preorder is thengiven by a ≤ b if b − a ∈ P.The preorder R is archimedean if, in addition, for each a ∈ R there is some r ∈ Q such that a ≤ r.A commutative G*-algebra is a commutative Q [ i ] -*-algebra A for which the self-adjoint part A sa isa preordered archimedean ring. When the norm is expressed as a subobject of A × Q + , the potential non-geometricity of the axiom is visible in (cid:107) a (cid:107) = Q + . In this ad hoc naming, we write ‘G’ for ‘geometric’. We have not attempted to define non-commutative G*-algebras, sinceit is not so easy to order the self-adjoints when they don’t commute. as Spitters, Steven Vickers & Sander Wolters If A is a commutative C*-algebra in a topos, then its spectrum Σ A is the space of the continuous *-algebrahomomorphisms x : A → C . The classical theory then says that A is isomorphic to the complex algebra ofcontinuous maps from Σ A to C , and its self-adjoint part A sa is isomorphic to the real algebra of continuousmaps from Σ A to R . The topology on Σ A is the weak-* topology, and another way to say this is that asubbasis of opens is provided by the sets of the form { x | x ( a ) > } for a ∈ A self-adjoint.This suggests that if we want a geometric description of the points of the spectrum, we should usethe elements a ∈ A sa to form propositional symbols (let us say D ( a ) ), and add axioms to say that in thissystem D ( a ) behaves like { x | x ( a ) > } . This was done in [4].Coquand [9] defines, for any preordered archimedean ring, the spectrum (which he calls the maximalspectrum) in a point-free way.If A is a commutative C*-algebra, A is a commutative G*-algebra by restricting C to Q [ i ] and definingits positive cone to be the set of all squares. This is closed under addition and multiplication by positiverationals. Also, the archimedean property follows from the existence of the norm. Then [10] its point-freeGelfand spectrum is isomorphic to the spectrum constructed in [9].As mentioned before, a point-free approach will give axioms characterizing the behaviour of formalsymbols D ( a ) ( a ∈ A sa ) with intended meaning { x | x ( a ) > } . The axioms in [9] are: D ( a ) ∧ D ( − a ) (cid:96) ⊥ (1a) D ( a ) (cid:96) ⊥ if a ≤ D ( a + b ) (cid:96) D ( a ) ∨ D ( b ) (1c) (cid:62) (cid:96) D ( ) (1d) D ( ab ) (cid:97)(cid:96) ( D ( a ) ∧ D ( b )) ∨ ( D ( − a ) ∧ D ( − b )) (1e) D ( a ) (cid:96) (cid:95) < r ∈ Q D ( a − r ) . (2)These are all intuitively clear in terms of the intended meaning. For example, (1a) says that for no x can we have both x ( a ) > x ( − a ) = − x ( a ) >
0, while (2) says that if x ( a ) >
0, then x ( a ) > r forsome r > x of the spectrum can be described geometrically by theset of those elements a ∈ A sa for which x ( a ) >
0. In fact, we could take it (but we shall modify this view)that that set is the point x . Such a set must conform with the axioms. For example, it cannot contain both a and − a .We do not wish to recap those substantial calculations, but there is an important structural part of thedevelopment of which we shall make considerable use. The six axioms above (1,2) are a propositionalgeometric theory, with propositional symbols D ( a ) indexed by elements a of A sa . As mentioned inSection 2.1, it can be used to present a frame F A , with the D ( a ) as generators and the axioms as relations.Then the points of the spectrum correspond to completely prime filters of the frame. However, the firstgroup of axioms (1) do not use infinitary disjunctions, and so could be taken as presenting a finitarydistributive lattice.2 Gelfand spectra in Grothendieck toposes
Definition 9.
Let A be a commutative G*-algebra. We write L A for the distributive lattice presented bygenerators D ( a ) (a ∈ A sa ) subject to the above relations (1). (In [9], L A is referred to as Spec r ( A sa ) .) Unlike the frame F A , the lattice L A is constructed geometrically from A . This is not noted explicitlyin [9], but is a consequence of the way it is presented by generators and relations because geometricconstructions include free algebras, and generating and factoring out congruences. The techniques areas described in [36]. It is, in fact, possible to give a description that is more concrete than that of thegeneral universal algebra, though we shall not need it here. [42] shows how to construct, geometrically,a free distributive lattice as a quotient of a double finite powerset, in this case F F A sa . [9] then gives anexplicit concrete description of when, in L A , we have D ( a ) ∧ · · · ∧ ( a n ) ≤ D ( b ) ∨ · · · ∨ D ( b m ) .It follows that points x of the spectrum can also be represented as subsets of x ⊆ L A . Each subsetcorresponds to a map from L A to the subobject classifier Ω , and respecting the axioms (1) then amountsto saying that this map preserves meets and joins (all finitary), in other words that x is a prime filter of L A . It is a filter if it is up-closed and closed under finite meets, and it is prime if whenever it contains afinite join, then it also contains one of the elements joined.At this point we can thus describe the points of the spectrum of A geometrically as prime filters x of L A , not arbitrary ones, but those respecting axiom (2), in other words such that if D ( a ) ∈ x then D ( a − r ) ∈ x for some 0 < r ∈ Q . This is slightly awkward because, although we have reduced muchto the lattice, we still have to refer explicitly to the elements of A . The next stage will remove thisawkwardness, and at the same time give access to a general geometric treatment of compact regularlocales. Definition 10.
A distributive lattice L is said to be normal if whenever a ∨ b = (cid:62) then there are x andy such that a ∨ y = x ∨ b = (cid:62) and x ∧ y = ⊥ . Defining a (cid:48) (cid:48) a (a (cid:48) well inside a) if there is y such thata ∨ y = (cid:62) and a (cid:48) ∧ y = ⊥ , then another way to express normality is that if a ∨ b = (cid:62) , then there is somea (cid:48) (cid:48) a with a (cid:48) ∨ b = (cid:62) . We also write (cid:16) a for the set { a (cid:48) | a (cid:48) (cid:48) a } .If L is a normal distributive lattice, then a prime filter x ⊆ L is regular if whenever a ∈ x, then a (cid:48) ∈ xfor some a (cid:48) (cid:48) a. The theory of regular ideals of a normal distributive lattice L is geometric. We could describe thisby a propositional theory along the lines of that used earlier for Σ A , but to illustrate the use of geometricmathematics we present a predicate theory that describes the regular ideals directly. It has a predicatesymbol x ⊆ L , and axioms as follows. (cid:62) (cid:96) x ( (cid:62) ) (3a) x ( a ) ∧ x ( b ) (cid:96) ab x ( a ∧ b ) (3b) x ( ⊥ ) (cid:96) ⊥ (3c) x ( a ∨ b ) (cid:96) ab x ( a ) ∨ x ( b ) (3d) x ( a ) (cid:96) a ( ∃ a (cid:48) )( a (cid:48) (cid:48) a ∧ x ( a (cid:48) )) (3e)(Note that the logical symbols (cid:62) , ∧ , ⊥ , ∨ are overloaded here, denoting both operators in the lattice L and logical connectives. The two usages are syntactically quite different, though clearly the axioms setup a semantic connection between them.) Axioms (3a,3b) say that x is a filter, (3c,3d) that it is prime,and (3e) that it is regular.This is presented as a predicate theory, and so, in principle, has a classifying topos that might notbe localic. However, it is localic, and this is evident from the fact that no new sorts are declared in the as Spitters, Steven Vickers & Sander Wolters L (the regular spectrum of L ) for the locale just defined, whose points are the regularprime filters of L .Coquand [9] proved (i) L A is normal, and (ii) axiom (2) is equivalent to the regularity axiom a (cid:96) (cid:87) a (cid:48) (cid:48) a a (cid:48) . This is important as it implies that the spectrum Σ A is isomorphic to RSpec L A , the regularspectrum of L A .This completes our geometric construction of the spectrum from the C*-algebra. We replaced thenon-geometric commutative C*-algebra by the more general and geometric commutative G*-algebra.The spectrum of this G*-algebra A is constructed geometrically by first constructing the normal distribu-tive lattice L A and subsequently constructing its regular spectrum.If f : A → B is a homomorphism of commutative G*-algebras, then geometricity ensures that it givesa lattice homomorphism L f : L A → L B , and (contravariantly) a map Σ f : Σ B → Σ A . If y is a regular primefilter of L B , then Σ f ( y ) is its inverse image L − f ( y ) . If f is an inclusion, we shall generally write L AB for L f and ρ BA for Σ f .In the next two lemmas we shall consider formal expressions φ ( x , . . . , x n ) = φ ( x i ) n built using finitemeets and finite joins from generators D ( x i ) , where x i is a variable ranging over A sa . From [9] we knowthat φ ( a i − r ) n (cid:48) φ ( a i ) n if 0 < r ∈ Q . [9, Corollary 1.7] also shows that if φ is a finite join of generators,and φ ( a i ) n = L A for some elements a i ∈ A sa , then φ ( a i − r ) n = < r ∈ Q . This clearlyextends to the case where φ is a finite meet of finite joins of generators, and hence to arbitrary distributivelattice expressions. Lemma 11.
Let A be a commutative G*-algebra, and suppose v (cid:48) u = φ ( a i ) n in L A . Then there is some < r ∈ Q such that v ≤ φ ( a i − r ) n .Proof. We have u ∨ w = v ∧ w = w , and – extending the list of a i s as necessary – we canwrite w = ψ ( a i ) n . By the previous discussion we can find 0 < r ∈ Q such that φ ( a i − r ) n ∨ ψ ( a i − r ) n = ψ ( a i − r ) n ≤ w , we conclude that v ≤ φ ( a i − r ) n .The following lemma will be important when we come to describe the external spectrum. Lemma 12.
Let f : A → B be a homomorphism of G*-algebras, and suppose u ∈ L A and v (cid:48) L f ( u ) inL B . Then there is some u (cid:48) (cid:48) u such that v ≤ L f ( u (cid:48) ) .Proof. Writing u = φ ( a i ) n , we find that v (cid:48) φ ( f ( a i )) n . Applying Lemma 11, we can find 0 < r ∈ Q suchthat v ≤ φ ( f ( a i ) − r ) n = L f ( u (cid:48) ) where u (cid:48) = φ ( a i − r ) n (cid:48) u .The opens of RSpec L are described in [21] as the regular ideals of L , i.e. those ideals I such that if (cid:16) a ⊆ I then a ∈ I . This follows from the coverage theorem, a general result of topos-valid locale theory,but unfortunately it is not geometric. The problem lies in the regularity condition, which amounts to ( ∀ a (cid:48) )( a (cid:48) (cid:48) a → I ( a (cid:48) )) (cid:96) a I ( a ) where the left-hand side is not a geometric formula.In order to gain access to geometric methods, we replace the regular ideals by the “rounded (cid:48) -ideals”. They differ in the way they use elements of L to represent opens: a regular ideal comprises thoseelements of L that are less than the open, whereas the (cid:48) -ideal comprises those that are well inside.4 Gelfand spectra in Grothendieck toposes
The concept of rounded (cid:48) -ideal is geometric, and so there is a locale RIdl ( L , (cid:48) ) (or, for short,RIdl L ) whose points are the rounded (cid:48) -ideals. This much follows [40] from the simple fact that ( L , (cid:48) ) is a continuous information system — that is, (cid:48) is an idempotent relation: transitive ( (cid:48) ◦ (cid:48) ⊆ (cid:48) ) andinterpolative ( (cid:48) ◦ (cid:48) ⊇ (cid:48) ).What is important in our situation is that the points of the locale RIdl L form a frame, and indeedthe frame of opens for RSpec L . This is expressed succinctly in Theorem 15, that RIdl L is the localeexponential S RSpec L , which allows us to use Theorem 5.Given any idempotent relation < on a set X , [40] in effect defines a rounded < -ideal to be a modelof the geometric theory with one predicate symbol I ⊆ X and axioms a (cid:48) < a ∧ I ( a ) (cid:96) a (cid:48) a I ( a (cid:48) ) (cid:62) (cid:96) ( ∃ a ) I ( a ) I ( a (cid:48) ) ∧ I ( a (cid:48)(cid:48) ) (cid:96) a (cid:48) a (cid:48)(cid:48) ( ∃ a )( a (cid:48) < a ∧ a (cid:48)(cid:48) < a ∧ I ( a )) The corresponding locale is RIdl ( X , < ) , and its opens are the rounded upsets of X under < .The proof of the following theorem is instructive. Although the theorem is stated entirely for locales,the proof is fairly simple if one digresses into non-localic toposes. Note that the discussion in Section 2.2of opfibrations holds equally for maps (geometric morphisms) between toposes. A map from S to a topos E is a pair of points of E , together with a specialization morphism between them. Theorem 13.
Let X be a locale, and ( P , < ) a continuous information system in Sh ( X ) . Let p : Y → X bethe external description of the internal locale
RIdl ( P , < ) . Then p is an opfibration.Proof. Let E be the classifying topos for the theory of continuous information systems ( Q , < ) , and let q : F → E be the bundle in which the fibre over ( Q , < ) is the rounded ideal completion of ( Q , < ) . Hence F classifies triples ( Q , <, J ) where ( Q , < ) is a continuous information system and J is a rounded idealfor it. We first show that q is an opfibration.Let f : ( Q ⊥ , < ) → ( Q (cid:62) , < ) be a homomorphism of continuous information systems, in other words,a function f : Q ⊥ → Q (cid:62) that preserves < . By [46, Proposition 2.10], we obtain a map RIdl ( f ) :RIdl ( Q ⊥ , < ) → RIdl ( Q (cid:62) , < ) , mapping J (cid:55)→ (cid:16) f ( J ) . We have RIdl ( f )( J ) (cid:118) J (cid:48) iff f extends to a ho-momorphism (necessarily unique) from ( Q ⊥ , <, J ) to ( Q (cid:62) , <, J (cid:48) ) , and from this it follows that the map (( Q ⊥ , < ) , ( Q (cid:62) , < ) , J ) (cid:55)→ ( Q ⊥ , <, J , Q (cid:62) , <, RIdl ( f )( J )) provides the left adjoint required for an opfibra-tion; see Proposition 1.Returning to the situation as in the statement, by the definition of classifying topos the continuousinformation system in Sh ( X ) gives a map ( P , < ) : X → E , and p is the pullback of q along it. From thegeometricity of the definition of opfibration, we see that the property is preserved under pullback.To see that (cid:48) is an idempotent relation (which is well known), first note that in any distributive lattice L we have that if b (cid:48) ≤ a (cid:48) (cid:48) a ≤ b then b (cid:48) (cid:48) b , and if a (cid:48) (cid:48) a then a (cid:48) ≤ a . From this is follows that (cid:48) istransitive. If moreover L is normal, then (cid:48) is interpolative. For if a (cid:48) (cid:48) a with y as in the definition, thenby normality a (cid:48)(cid:48) ∨ y = (cid:62) for some a (cid:48)(cid:48) (cid:48) a , and then also a (cid:48) (cid:48) a (cid:48)(cid:48) . Another useful fact is that if a i (cid:48) a ( i = , a ∨ a (cid:48) a . We find also that the rounded (cid:48) -ideals of L are exactly the ordinary ideals I (with respect to ≤ ) that are rounded in the sense that I = (cid:16) I . Proposition 14.
Let L be a normal distributive lattice. Then there is a bijection between its regularideals and its (cid:48) -ideals. Note that the ‘R’ in RIdl stands for rounded , not regular . [40] calls it Idl ( X ,< ) , but we want to stress the roundedness. as Spitters, Steven Vickers & Sander Wolters Proof.
First, if J is a regular ideal then (cid:16) J is a (cid:48) -ideal. The only mildly non-obvious part of this is thatif a (cid:48) i (cid:48) a ( i = , y i as in the definition, then a (cid:48) ∨ a (cid:48) (cid:48) a using y ∧ y .Next, if I is a (cid:48) -ideal then we write r (cid:104) I (cid:105) = { a ∈ L | (cid:16) a ⊆ I } .To show this is an ideal, suppose (cid:16) a i ⊆ I and b (cid:48) a ∨ a with y as in the definition. Then we canfind a (cid:48) i (cid:48) a i with a (cid:48) ∨ a (cid:48) ∨ y = (cid:62) and it follows that b (cid:48) a (cid:48) ∨ a (cid:48) ∈ I . It is regular, because if (cid:16) a ⊆ r (cid:104) I (cid:105) then (cid:16)(cid:16) a ⊆ I ; but (cid:16)(cid:16) a = (cid:16) a by interpolativity of (cid:48) , so a ∈ r (cid:104) I (cid:105) . Indeed, it is the smallest regular idealcontaining I .It is now easy to show this gives a bijection, with r (cid:104) (cid:16) J (cid:105) = J and (cid:16) r (cid:104) I (cid:105) = I (using (cid:16) I = I ).We now prove our central result in this section, which gives a geometric account of the opens ofRSpec L . One can compare it with Stone’s representation theorem for Boolean algebras B (see [29]),where the frame of opens for the spectrum of B is the ideal completion of B . In fact, our result generalizesthat, since every Boolean algebra is normal, with (cid:48) coinciding with ≤ . Theorem 15.
If L is a normal distributive lattice, then
RIdl L ∼ = S RSpec L .The evaluation map RIdl L × RSpec L → S takes ( I , x ) to the top point (cid:62) whenever the ideal I meetsthe filter x.Proof. We use [41, Theorem 12.7], which states that if X and W are locales for which S X exists and ishomeomorphic to the double powerlocale P W , then S W exists and is homeomorphic to X . In our case wetake X to be RIdl L , which is exponentiable with its opens the rounded upsets of L , and W to be RSpec L ,so it remains to calculate the double powerlocale P W and show that its points are again the roundedupsets of L .We use the Double Coverage Theorem [38, Theorem 7] to calculate its points. As explained in [44],the calculation is an analogue of one for the lower powerlocale that is directly derived from the usualCoverage Theorem [29]. Given a site in the form of a meet-semilattice S with a meet-stable coverage (cid:67) , the corresponding locale has for its points the filters F of S that “split” (cid:67) in the sense that if a (cid:67) U and a ∈ F then F meets U . Then its lower powerlocale has for its points the upsets of S (thus we dropthe requirement that meets should be respected) that split (cid:67) . For the double powerlocale we require a DL-site , in which S is a distributive lattice and the coverage is stable for both meets and joins: the pointsfor the locale are the prime filters that split (cid:67) , and for the double powerlocale they are the upsets thatsplit (cid:67) .To apply this theorem we must show that the covers a (cid:67) (cid:16) a in L are meet- and join-stable. Meet-stability means that a ∧ b is covered by { a (cid:48) ∧ b | a (cid:48) (cid:48) a } . It is not true that if a (cid:48) (cid:48) a then a (cid:48) ∧ b (cid:48) a ∧ b .However, knowing that a ∧ b (cid:67) (cid:16) ( a ∧ b ) , we argue that if c (cid:48) a ∧ b then c (cid:48) a (cid:48) (cid:48) a for some a (cid:48) , and so c ≤ a (cid:48) ∧ b . Join-stability is similar, but with joins.It is clear that an upset splits (cid:67) iff it is rounded with respect to (cid:48) , and the result follows. The finalpart derives from the definition of the evaluation map for RIdl L .Now that the notion of commutative G*-algebra has been settled, and the geometricity of the con-struction of its spectrum has been demonstrated, the following theorem is immediate. Theorem 16.
Let A be a unital commutative C*-algebra in a topos E and let Σ A be the Gelfand spectrumof A. Let Sh E ( Σ A ) → E be the unique localic geometric morphism corresponding to the locale Σ A , and Gelfand spectra in Grothendieck toposes let f : F → E be any geometric morphism. Consider the commutative G*-algebra f ∗ ( A ) , and let Σ f ∗ ( A ) be its spectrum in F . Then we have a pullback square Sh F ( Σ f ∗ ( A ) ) (cid:15) (cid:15) (cid:47) (cid:47) Sh E ( Σ A ) (cid:15) (cid:15) F f (cid:47) (cid:47) E We finish the section by observing that Theorem 15 and the results in [11, 47] provide entirelyanaloguous results for the geometric theory of integrals on a G*-algebra. Following [22] these internalintegrals correspond to quasi-states externally.
In this subsection we fix a unital C*-algebra A and calculate, in various forms, the spectrum as it arisesin the copresheaf approach.To summarize the notation, C ( A ) is the poset of unital commutative C*-subalgebras of A , partiallyordered by inclusion, and the topos [ C ( A ) , Set ] is the category of copresheaves on C ( A ) , equivalent to thesheaf topos Sh ( Idl C ( A )) . A is the tautological copresheaf mapping each context C to itself, and mappingeach arrow D → C in C ( A ) to the inclusion D (cid:44) → C . It is a unital commutative C*-algebra, internal tothe topos [ C ( A ) , Set ] and leads – internally – to a normal distributive lattice L A and a Gelfand spectrum Σ A ∼ = RIdl ( L A ) . This compact regular locale is of interest to the copresheaf model, as it internally playsthe role of a phase space.We seek the external representions of the Gelfand spectrum and other internal locales as bundles,which we shall typically denote by removing underlinings – sometimes in an ad hoc way. The externalrepresentation of Σ A will be p : Σ A → Idl C ( A ) .First we calculate the locale Σ A . We characterize both its points (Theorem 17), in line with thegeometric approach, and its opens (Theorem 19). We find it convenient to use local compactness todescribe the opens geometrically as points of the exponential S Σ A , and this is easily translated into adescription of the frame O Σ A .The explicit description of the opens was previously given Subsection 2.2 of [50]. Although theproof given there has the advantage of not using any advanced topos-theoretic methods, it has somedisadvantages too. The proof hides the role of geometric reasoning. As we have seen in the previoussubsection, it is because of geometricity that the spectrum is so closely related to the spectra of thecommutative C*-subalgebras (or, for reader familiar with the work of Butterfield and Isham, why thespectrum is so closely related to the spectral presheaf of their approach). Another disadvantage of theproof in [50] is that it is not clear how it can be generalized when the topos [ C ( A ) , Set ] is replaced by adifferent topos. This point is also related to the geometricity being hidden. Theorem 17.
The points of Σ A can be geometrically described as the pairs ( I , x ) where I is an ideal of C ( A ) and x is a subset of ∏ C ∈ C ( A ) L C satisfying the following properties:1. If ( C , a ) ∈ x then C ∈ I.2. If C ⊆ D ∈ I then ( C , a ) ∈ x iff ( D , L CD ( a )) ∈ x.3. If C ∈ I then { a ∈ L C | ( C , a ) ∈ x } is a regular prime filter in L C . as Spitters, Steven Vickers & Sander Wolters Proof.
Recall from Subsection 2.2 that a point of Σ A is equivalent to a pair ( I , x ) where I is an ideal of C ( A ) (i.e. a point of the locale Idl C ( A ) ), and x is a point of I ∗ Σ A , in other words, a regular prime filterof L I . Now I is the filtered colimit (actually here a directed join) of the principal ideals ↓ C for C ∈ I , andgeometric constructions preserve filtered colimits. It follows that L I ∼ = colim C ∈ I L C .Each element of L I is the image of some ( C , a ) , so to specify the filter x of L I it suffices to specify x ⊆ ∏ C ∈ C ( A ) L C . Then condition (1) says that x is in the colimit over C ∈ I , and (2) expresses the factthat equality in the filtered colimit derives from ( C , a ) = ( D , L CD ( a )) . The final condition expresses theregular prime filter property. However, some care is needed with regularity. Actually, regularity saysthat if ( C , a ) ∈ x then there is some D ∈ I with C ⊆ D and some b (cid:48) L CD ( a ) in L D such that ( D , b ) ∈ x .Lemma 12 says that in this case there is some a (cid:48) ∈ L C with a (cid:48) (cid:48) a and b ≤ L CD ( a (cid:48) ) , and we then havethat ( C , a (cid:48) ) ∈ x .Note how this geometric description provides a subbase for the topology (a set of generators for theframe) as the pairs ( C , a ) , where C ∈ C ( A ) and a ∈ L C . The point ( I , x ) is in this open iff C ∈ I and ( C , a ) ∈ x .We now use local compactness of Σ A to give an explicit description in geometric form of the internalframe, as S Σ A . We shall write S Σ A for its external description. Theorem 18.
The points of S Σ A are pairs ( I , U ) where I is an ideal of C ( A ) and U is a subset of ∏ C ∈ C ( A ) L C satisfying the following properties.1. If ( C , a ) ∈ U then C ∈ I.2. If C ⊆ D ∈ I then ( C , a ) ∈ U iff ( D , L CD ( a )) ∈ U .3. If C ∈ I then { a ∈ L C | ( C , a ) ∈ U } is an ideal in L C .4. If ( C , a ) ∈ U then there is some ( D , b ) ∈ U with C ⊆ D and L CD ( a ) (cid:48) b.Proof. The first part is an application of Theorem 15. The rest is done in exactly the same way asTheorem 17, except we have to take care in expressing the fact that L I is rounded under (cid:48) , becauseLemma 12 does not apply. Note that in the case where I is a principal ideal ↓ D , U is equivalent to a (cid:48) -ideal of L D and hence to an open of Σ D .We can now give an explicit description of the frame O Σ A . Theorem 19. Σ A is locally compact. The points of S Σ A are the C ( A ) -indexed families U , where each U C is a (cid:48) -ideal of L C , and if C ⊆ D, then L CD ( U C ) ⊆ U D .Proof. By Theorem 7 we know that Σ A is locally compact, and that S Σ A has the sections of S Σ A for itspoints. Since a section is a map from Idl C ( A ) , we can use Lemma 2 to see that the sections are asdescribed in the statement.In terms of frames, it is now immediate that O Σ A is isomorphic to the subframe of ∏ C ∈ C ( A ) O Σ C comprising those elements U such that for any C ⊆ D , ρ ∗ DC ( U C ) ⊆ U D .Unlike the case for the external S Σ A , it is a non-trivial calculation to calculate the internal frame O ( Σ A ) = pt ( S Σ A ) We should like to emphasize that this non-trivial calculation is usually unnecessary.However much one might like to think that the internal locale “is” the internal frame, it is usually betterto identify the locale with its external description.8
Gelfand spectra in Grothendieck toposes
Theorem 20.
The internal frame O ( Σ A ) is given as a copresheaf by O ( Σ A ) = { U ∈ ∏ C ⊆ D O Σ D | if C ⊆ D ⊆ D then ρ ∗ D D ( U D ) ⊆ U D } . Proof.
By Yoneda’s Lemma, O ( Σ A )( C ) ∼ = [ C ( A ) , Set ]( Y ( C ) , O ( Σ A ) , where Y : C ( A ) op → [ C ( A ) , Set ] is the Yoneda embedding. Now, internally, O ( Σ A ) is the set of pointsof S Σ A , in other words its discrete coreflection. It follows that morphisms Y ( C ) → O ( Σ A ) are equivalentto locale maps Y ( C ) → S Σ A ∼ = RIdl ( L A ) ,and we can analyse these by their external representation. Externally, the local homeomorphism for Y ( C ) has bundle space Idl ( ↑ C ) . This can be proved by iterated forcing – see Section 2.3. Hence weseek maps Idl ( ↑ C ) → S Σ A over Idl C ( A ) and, by Lemma 2, the required maps are the same as monotonefamilies U D ( C ⊆ D ), with each U D in the fibre of S Σ A over ↓ D , in other words a rounded ideal of L D .“Monotone” is with respect to the given order on C ( A ) and the specialization order on S Σ A . UsingProposition 1 and Theorem 13, we see that if C ⊆ D ⊆ D , then we require RIdl ( L D D )( U D ) ⊆ U D .Now suppose I ⊆ L D is a (cid:48) -ideal, and y ⊆ L D a regular prime filter. Then the point ρ ∗ D D ( y ) = L − D D ( y ) is in the open I iff, as subsets of L D , they meet – they have inhabited intersection. Clearly this isequivalent to L D D ( I ) meeting y , which in turn is equivalent to y being in RIdl ( L D D )( I ) = (cid:16) L D D ( I ) .It follows that RIdl ( L D D ) = ρ ∗ D D , and we have the description in the statement. In this subsection we generalize Theorem 19 to unital commutative C*-algebras in toposes that are func-tor categories. Subsequently, in Subsection 3.5, we use this result when we explore examples of functorcategories (other than [ C ( A ) , Set ] , or [ C ( A ) op , Set ] ) which may be of interest to the topos approachesto quantum theory. As a first step, we use presheaf semantics to identify all C*-algebras in functorcategories. The reader unfamiliar with presheaf semantics may want to consult [33, Chapter VI].Let C be any small category. Below we prove the following: Proposition 21.
The object A (with additional structure + , · , ∗ , ) is a C*-algebra in the topos [ C , Set ] iffit is given by a functor A : C → CStar , where
CStar is the category of C*-algebras and ∗ -homomorphismsin Set . The C*-algebra A is commutative iff each A ( C ) is commutative. The algebra A is unital iff everyA ( C ) is unital and for each f : C → D, the ∗ -homomorphism A ( f ) : A ( C ) → A ( D ) preserves the unit.Proof. It follows from Lemma 2 and the discussion in Subsection 3.2 that a semi-normed ∗ -algebra over C in [ C , Set ] is equivalent to a functor A : C → Set , such that each A ( C ) is a semi-normed ∗ -algebra over C , and, for each arrow f : D → C in C , the function A ( f ) : A ( D ) → A ( C ) is a ∗ -homomorphism suchthat (cid:107) A ( f )( a ) (cid:107) D ≤ (cid:107) a (cid:107) C . We used (cid:107) · (cid:107) C to denote the semi-norm on A ( C ) . The internal semi-norm N issubmultiplicative and satisfies the C*-property iff each semi-norm (cid:107) · (cid:107) C is submultiplicative and satisfiesthe C*-property (cid:107) a ∗ a (cid:107) C = (cid:107) a (cid:107) C .Recall that the semi-norm N of A is defined as a subobject of A × Q + . The internal semi-norm isconnected to the external semi-norms by the identities N ( C ) = { ( a , q ) ∈ A ( C ) × Q + | (cid:107) a (cid:107) C < q } , (4) as Spitters, Steven Vickers & Sander Wolters (cid:107) · (cid:107) C : A ( C ) → R + , (cid:107) a (cid:107) C = inf { q ∈ Q + | ( a , q ) ∈ N ( C ) } . (5)Note that the fact that ∗ -homomorphisms are contractions, in the sense that (cid:107) A ( f )( a ) (cid:107) C ≤ (cid:107) a (cid:107) C , preciselystates that N defined by (4) is a well-defined subobject of A × Q + . The semi-norm N is a norm iff itsatisfies the axiom ( ∀ q ∈ Q + ( a , q ) ∈ N ) ⇒ ( a = ) . (6)By the rules of presheaf semantics, externally, this axiom translates to: for each C ∈ C the semi-norm (cid:107) · (cid:107) C is a norm.Completeness can be checked in the same way as in [21], because the axiom of dependent choiceworks in any presheaf topos. For completeness, we thus need to check the axiom ∀ f ∈ A N ( ( ∀ n ∈ N ∀ m ∈ N ( f ( n ) − f ( m ) , − n + − m ) ∈ N ) ⇒ ( ∃ a ∈ A ∀ n ∈ N ( a − f ( n ) , − n ) ∈ N ) ) . Note that for any object C ∈ C , the elements of A N ( C ) correspond exactly to sequences ( a n ) n ∈ N in A ( C ) .By presheaf semantics, the axiom for completeness holds iff for every object C ∈ C and any sequence ( a n ) n ∈ N in A ( C ) , if C (cid:13) ( ∀ n ∈ N ∀ m ∈ N ( a n − a m , − n + − m ) ∈ N ) , then C (cid:13) ( ∃ a ∈ A ∀ n ∈ N ( a − a n , − n ) ∈ N ) . This can be simplified by repeated use of presheaf semantics, and the identity (cid:107) A ( f )( a n ) − A ( f )( a m ) (cid:107) D = (cid:107) A ( f )( a n − a m ) (cid:107) D ≤ (cid:107) a n − a m (cid:107) C , where f : C → D is any arrow. In the end, the axiom of completeness simplifies to the statement thatgiven an object C ∈ C and any sequence ( a n ) n ∈ N in A ( C ) such that for any pair n , m ∈ N we have ( a n − a m , − n + − m ) ∈ N ( C ) , there exists an element a ∈ A ( C ) such that for every n ∈ N , ( a − a n , − n ) ∈ N ( C ) . By definition of N this simply states that each A ( C ) is complete with respect to the norm (cid:107) · (cid:107) C .This completes the proof that C*-algebras in [ C , Set ] are equivalent to functors C → CStar .Proposition 21 makes it easy to calculate the spectrum Σ A of any commutative unital C*-algebra A in any functor category, using the methods of Subsection 3.3. The first step is the construction ofthe distributive lattice L A , which we will simply denote as L . This is done as in Subsection 3.2. Theconstruction is geometric, based on the G*-algebra structure of A , and so from the discussion in Section 2we can derive L : L : C → Set , L ( C ) = L A ( C ) , (7) L ( f ) = L A ( f ) : L A ( C ) → L A ( D ) .As in Subsection 3.3, we can now describe the spectrum Σ in various ways. To start with, it isdescribed externally as a localic geometric morphism Σ → [ C , Set ] . We must keep in mind that if C isnot a poset, then neither Σ A nor [ C , Set ] is a localic topos: they classify non-propositional theories andare generalized spaces in the sense of Grothendieck. However, the bundle Σ A → [ C , Set ] is still a localicbundle , because it is a localic geometric morphism [30, A4.6.1]. Its fibres are locales. This all followsbecause it arises from an internal locale in [ C , Set ] .00 Gelfand spectra in Grothendieck toposes
To make clear the analogy with the localic case, we shall use the same notation for a locale as for itstopos of sheaves. For example, 1 will often denote the topos of sets, and S the topos Sh S , whose objectsare functions. In practice this does not cause confusion, but one should refrain from asking whether thelocale actually “is” the frame or the topos. Rather, one should think that a locale is described equally wellby its frame of opens or its topos of sheaves. Note that geometric morphisms between localic toposes areequivalent to maps between the locales, and we shall frequently refer to geometric morphisms in generalas maps between the toposes.For the rest of this subsection we shall let C be a small category, and let A be a C*-algebra in [ C , Set ] .We write p : Σ A → [ C , Set ] for the localic bundle corresponding to the internal spectrum Σ A . Theorem 22. (Analogue of Theorem 17.) The points of Σ A can be geometrically described as the pairs ( F , x ) where F is a flat presheaf over C and x is a subset of ∏ ( C , u ) ∈ ( (cid:82) F ) L C satisfying the followingproperties:1. If f : ( C , u ) → ( D , v ) in (cid:82) F then ( C , u , a ) ∈ x iff ( D , v , L f ( a )) ∈ x.2. If ( C , u ) is an object in (cid:82) F then { a ∈ L C | ( C , u , a ) ∈ x } is a regular prime filter in L C .Proof. The proof is essentially the same as that of Theorem 17, once one has taken on the fact (Subsec-tion 2.3, just below Definition 3) that the points of [ C , Set ] are the flat presheaves over C and that eachis a filtered colimit of representables Y ( C ) . Theorem 23. (Analogue of Theorem 18.) The frame of opens of Σ A has internal localic form S Σ A = RIdl ( L A ) . Externally, the points of the bundle locale S Σ A are pairs ( F , U ) , where F is a flat presheaf of C and U is a subset of ∏ ( C , u ) ∈ ( (cid:82) F ) L C satisfying the following properties:1. If f : ( C , u ) → ( D , v ) in (cid:82) F, then ( C , u , a ) ∈ U iff ( D , v , L f ( a )) ∈ U .2. If ( C , u ) is an object in (cid:82) F, then { a ∈ L C | ( C , u , a ) ∈ U } is an ideal in L C .3. If ( C , u , a ) ∈ U , then there is some f : ( C , u ) → ( D , v ) in (cid:82) F and some ( D , v , b ) ∈ U with L f ( a ) (cid:48) b.Proof. Essentially the same as for Theorem 18.At this point in Subsection 3.3 we moved on to giving an explicit description of the frame O Σ A .Here we must be more careful, since in general Σ A is a non-localic topos. Any topos map (geometricmorphism) has a hyperconnected-localic factorization [30, A4.6.1], and by applying this to a map E → E has a localic reflection L ( E ) . Its opens are the maps E → S . In other words, since S classifies subsingletons, the opens of L ( E ) are the subobjects of 1 in E . If E is exponentiable as atopos (as is the case for [ C , Set ] ), then the opens of L ( E ) are the points of S E and can be calculatedimmediately by the method of Theorem 19. Theorem 24. (Analogue of Theorem 19.) The opens of Σ A (equivalently: the opens of the localic reflec-tion L ( Σ A ) ) can be geometrically described as the elements U ∈ ∏ C ∈ C Σ A ( C ) such that if f : C → D is amorphism in C , then Σ ∗ A ( f ) ( U C ) ≤ U D .Proof. Using exponentiability as in Theorem 19, we can calculate the opens as the points of S Σ A andthence as the sections of S Σ A → [ C , Set ] . Sections are described using Lemma 2. For each object C of C we require a (cid:48) -ideal of L A ( C ) , hence an element of O Σ A ( C ) . For each morphism f : C → D we require ahomomorphism of points of S Σ A over f , and the homomorphism property comes down to the condition Σ ∗ A ( f ) ( U C ) ≤ U D . as Spitters, Steven Vickers & Sander Wolters ( Σ A ) ), i.e. the sheaves of Σ A rather than just its opens (which are not enough, and only describe the localic reflection). Again onewould use exponentiability, but this time to calculate [ set ] Σ instead of S Σ . Here, [ set ] denotes the objectclassifier , classifying the geometric theory with one sort and no other ingredients. Its models are justsets (carrying the single sort), so the maps Σ → [ set ] are the sheaves of Σ (by which we just mean theobjects of the topos, even in this non-localic case). [49] shows how to give a geometric descriptionof the sheaves over any stably compact locale when presented by a strong proximity lattice, and as aspecial case works for a compact regular locale when presented by a normal distributive lattice L . Thesheaves are, first of all, finitary sheaves over the distributive lattice L , i.e. those presheaves F that satisfyfinite instances of the sheaf pasting condition. In addition, they must satisfy a continuity condition that F ( a ) ∼ = colim a (cid:48) a (cid:48) F ( a (cid:48) ) .We now proceed by analogy with Theorem 23, but replacing S by [ set ] . Internally in [ C , Set ] wehave a geometric theory describing the sheaves over the internal locale Σ A . This corresponds externallyto a bundle ( not localic) [ set ] Σ A → [ C , Set ] . We can now describe the points of the external topos [ set ] Σ A as pairs ( F , G ) where F is a flat presheaf over C and G is a sheaf (presheaf with finitary pasting andcontinuity) over colim ( C , u ) ∈ ( (cid:82) F ) L A ( C ) . After this, the sheaves of Σ can be described as the sections of [ set ] Σ A → [ C , Set ] .Finally in this subsection, we calculate the internal frame of Σ A . Theorem 25. (Analogue of Theorem 20.) The internal frame O ( Σ A ) is given as an object of [ C , Set ] by O ( Σ A )( C ) = { U ∈ ∏ f : C → D O Σ D | if C f → D g → D then ρ ∗ g ( U f ) ≤ U g f } Proof.
Just as in Theorem 20, we find that O ( Σ A )( C ) is the set of maps (geometric morphisms) from Y ( C ) to RIdl L A over [ C , Set ] . We can calculate the external form of the discrete locale Y ( C ) usingiterated forcing (Subsection 2.3). Its topos is [ C (cid:110) Y ( C ) , Set ] , where now Y ( C ) denotes the discreteinternal category. From the definition of Y we see that the objects of C (cid:110) Y ( C ) are the pairs ( D , f ) with f : C → D in C , and a morphism from ( D , f ) to ( D , f ) is a morphism g : D → D in C such that f = g f .Now combining Lemma 2 with Theorem 23, we see that the maps we are looking for are functors U from C (cid:110) Y ( C ) to the models as described in Theorem 23. For each object f : C → D , its image U f mustbe such that the flat presheaf F as described in the Theorem is representable (for D ). Its representabilityallows us to combine conditions (2) and (3) to say that we have a (cid:48) -ideal in L A ( D ) , hence an open of Σ A ( D ) . Now for each morphism g : D → D (cid:48) , with f (cid:48) = g f , we need a homomorphism U f → U f (cid:48) over g . Inother words, if a ∈ U f (as (cid:48) -ideal of L A ( D ) ), then L A ( g ) ( a ) ∈ U f (cid:48) . When the (cid:48) -ideals of L A ( D ) are viewedas opens of Σ A ( D ) , this is equivalent to saying that U f ≤ ρ ∗ A ( g ) ( U f (cid:48) ) . We now see that the internal frame isas described. The copresheaf approach is based on algebraic quantum theory in the sense that in this approach quantumtheory is described using abstract C*-algebras. In this section we seek to extend the copresheaf approach,as already suggested in [21], to the Haag–Kastler formalism, which is an algebraic approach to quantumfield theory. Introductions to the Haag–Kastler formalism, or algebraic quantum field theory (AQFT), canbe found in [1, 17]. In this formalism (where, for the sake of simplicity we consider Minkowski spacetime M ) the physical content of a quantum field theory is described by a net of C*-algebras O → A ( O ) ,02 Gelfand spectra in Grothendieck toposes where O ranges over certain (open connected causally complete) regions of spacetime. This means thatwe associate to each region O of spacetime of interest, a C*-algebra A ( O ) . We think of the self-adjointelements of A ( O ) as the observables that can be measured in the region O . With this in mind, we canmake the assumption that if O ⊆ O , then A ( O ) ⊆ A ( O ) . If K ( M ) denotes the set of the spacetimeregions of interest, partially ordered by inclusion, then a net of C*-algebras defines a covariant functor A : K ( M ) → CStar . We assume that the algebras A ( O ) are unital for convenience.By Proposition 21 an AQFT is a C*-algebra A internal to [ K ( M ) , Set ] . Note that A is in generalnot commutative. As for the copresheaf approach we can Bohrify the C*-algebra A . This means thatwe make it commutative by considering it as a copresheaf over the poset of commutative subalgebras.The difference with the copresheaf approach is that the Bohrification takes place internal in the topos [ K ( M ) , Set ] , instead of the topos Set . We obtain a commutative C*-algebra internal to a topos (whichin turn is internal to a functor category), and, using the ideas of the preceding paragraphs, we describethe points of the Gelfand spectrum of this commutative C*-algebra.Instead of the Haag–Kastler formalism, we could have considered the more general and more recentlocally covariant quantum field theories [5]. This amounts to replacing the poset K ( M ) by a morecomplicated category of manifolds and embeddings (which is no longer a poset). Although Bohrificationof the locally covariant field theories can be described using the same ideas of Subsection 3.4, we stickwith the Haag–Kastler formalism, as this makes the presentation a bit easier.An internal unital commutative C*-subalgebra of A is simply a subobject C of A such that for each O ∈ K ( M ) , C ( O ) is a commutative unital C*-algebra in Set . These internally defined commutativeC*-subalgebras form a poset C ( A ) in [ K ( M ) , Set ] and we can consider the (internal) functor categoryover this poset. By using iterated forcing (see Section 2.3), we can describe this functor category withina functor category using a single Grothendieck topos over Set , given by the site K ( M ) (cid:110) C ( A ) . In this(composite) topos, the Bohrified net is given by the functor ( O , C ) (cid:55)→ C ( O ) , where O ∈ K ( M ) and C isa commutative unital C*-subalgebra of A | ↑ O . Before considering Gelfand Duality, we first simplify thetopos in which we are working. Instead of labeling the objects of the base category by subalgebras C of A | ↑ O , we only concentrate on the part C ( O ) . This motivates using the topos [ P , Set ] , where the poset P is defined as follows: an element ( O , C ) ∈ P , consists of an O ∈ K ( M ) and a C ∈ C O : = C ( A ( O )) ,and the order relation is given by ( O , C ) ≤ ( O , C ) iff O ⊆ O , C ⊆ C . We are interested in the unital commutative C*-algebra A : P → Set , ( O , C ) (cid:55)→ C in the topos [ P , Set ] .Note that A (( O , C ) ≤ ( O , C )) is the inclusion map C (cid:44) → C .Next, we want to compute the points of the locale Σ , which is the external description of the Gelfandspectrum of A . An easy way to do this is by using the reasoning in [37]. As a category, the topos [ P , Set ] is equivalent to the topos Sh ( P ↑ ) , where P ↑ is the set P , equipped with the Alexandrov upset topology.As in [34] we can find a site P (cid:110) Σ such that Sh Sh ( P ) ( Σ ) , the topos of sheaves over Σ , internal to Sh ( P ↑ ) is equivalent to Sh ( P (cid:110) Σ ) . The locale Σ is the locale generated by the posite P (cid:110) Σ . We use the positedescription P (cid:110) Σ in order to find the points. We start with the functor L , the distributive lattice objectin [ P , Set ] , given by L : P → Set , L ( O , C ) = L C , L (( O , C ) ≤ ( O , C )) : L C → L C , [ a ] C (cid:55)→ [ a ] C . The elements of P (cid:110) Σ are triples ( O , C , [ a ] C ) , where O ∈ K ( M ) , C ∈ C O and [ a ] C ∈ L C . The order ofthis poset is given by ( O , C , [ a ] C ) ≤ ( O , C , [ a ] C ) , iff O ⊆ O , C ⊆ C , [ a ] C ≤ [ a ] C . as Spitters, Steven Vickers & Sander Wolters P (cid:110) Σ is equipped with the following covering relation (cid:67) , which is inherited from the coveringrelation (cid:67) , exploiting the fact that we are working over P . We have a covering ( O , C , [ a ] C ) (cid:67) W iff for W = { [ b ] C ∈ L C | ( O , C , [ b ] C ) ∈ W } , the condition [ a ] C (cid:67) W holds in L C . Note that the covering relationon P (cid:110) Σ is completely described in terms of covering relations on the L C .A point σ of the external spectrum Σ corresponds to a completely prime filter of P (cid:110) Σ . Recall thata filter σ is a nonempty, upward closed and lower directed subset of P (cid:110) Σ , and that σ is completelyprime if it satisfies ( O , C , [ a ] C ) ∈ σ and ( O , C , [ a ] C ) (cid:67) W , implies U ∩ σ (cid:54) = /0 . Let σ be a point of Σ . It is straightforward to show that R = { O ∈ K ( M ) | ∃ C ∈ C O , ∃ [ a ] C ∈ L C , s.t. ( O , C , [ a ] C ) ∈ σ } is an ideal of K ( M ) . Fix any O ∈ R and consider the set I O = { C ∈ C O | ∃ [ a ] C ∈ L C s.t. ( O , C , [ a ] C ) ∈ σ } . For any O ∈ R , I O is an ideal of C O . For a pair O ∈ R and C ∈ I O , define σ O , C : = { [ a ] C ∈ L C | ( O , C , [ a ] C ) ∈ σ } . As in [37], it can be shown that σ O , C is a completely prime filter of L C . A completely prime filter σ O , C on L C corresponds to a unique point λ ( O , C ) of the Gelfand spectrum Σ C .Next, we show how for different O ∈ R , C ∈ I O , the λ ( O , C ) ∈ Σ C are related. Let, for some fixed O ∈ R , D ⊂ C in C O . Let a ∈ D + and assume that [ a ] C ∈ σ O , C . By the order on P (cid:110) Σ , ( O , C , [ a ] C ) ≤ ( O , D , [ a ] D ) ∈ σ , where we used that σ is a filter, and therefore it is upward closed. For any a ∈ D + , if [ a ] C ∈ σ O , C ,then [ a ] D ∈ σ O , D . The filter σ O , D can be viewed as a frame map σ O , D : O Σ C → X Da = { λ ∈ Σ D | (cid:104) λ , a (cid:105) > } to 1 iff λ ( O , D ) ∈ X Da , iff [ a ] D ∈ σ O , D . If ρ CD : Σ C → Σ D is the restrictionmap, then σ O , C ◦ ρ − CD : O Σ D → λ ( O , C ) | D . At the level of points of Σ D , theimplication ∀ a ∈ D + , [ a ] C ∈ σ O , C ⇒ [ a ] D ∈ σ O , D translates to: ∀ a ∈ D + , (cid:0) σ O , C ( X Ca ) = (cid:1) ⇒ (cid:0) σ O , D ( X Da ) = (cid:1) . As the X Da form a basis of the Hausdorff space Σ D , and ρ − ( X Da ) = X Ca , this can only mean that σ O , D = σ O , C ◦ ρ − DC . In other words, whenever D ⊆ C , one has λ ( O , D ) = λ ( O , C ) | D .Assume that O (cid:48) ⊂ O in K ( M ) and that C ∈ C O (cid:48) . In P (cid:110) Σ , ∀ [ a ] C ∈ L C , ( O , C , [ a ] C ) ≤ ( O (cid:48) , C , [ a ] C ) . If [ a ] C ∈ σ O , C , then by the filter property of σ , [ a ] C ∈ σ O (cid:48) , C . We conclude that if O (cid:48) ⊆ O in R and C ∈ C O (cid:48) ,then λ ( O (cid:48) , C ) = λ ( O , C ) . Hence: Theorem 26.
A point σ of Σ is described by a triple ( R , I R , λ R , I ) , where: • R is an ideal in K ( M ) . Gelfand spectra in Grothendieck toposes • The function I R associates to each O ∈ R , an ideal I O of C O satisfying two conditions. Firstly,if O ⊆ O , then I O ∩ C O ⊆ I O . Secondly, if C i ∈ I O i , where i ∈ { , } , then there is an O ∈ R and a C ∈ I O such that O i ⊆ O and C i ⊆ C. • The function λ R , I associates to each O ∈ R and C ∈ I O , an element λ O , C ∈ Σ C , such that ifO ⊆ O and C ⊆ C , then λ O , C = λ O , C | C . The two conditions in the second bullet point are included to ensure that the set I = { ( O , C ) ∈ P | O ∈ R , C ∈ I O } is an ideal of P . Mathematically, the theorem would look more elegant if it were formulated in termsof ideals of P instead of using pairs ( R , I R ) , but that description would miss an important physicalpoint. Namely, a spacetime point x ∈ M corresponds to a specific filter of K ( M ) , consisting of all O ∈ K ( M ) containing x . However, a point σ of Σ is labelled by an ideal R of K ( M ) and not by afilter. If we want the points of the phase space to be indexed by points of the spacetime M , it mightbe interesting to look at the contravariant functor Σ : P op → Set , ( O , C ) (cid:55)→ Σ C . Remaining on the topicof applying topos approaches to quantum theory to algebraic quantum field theory, this functor Σ isalso interesting when we consider the work by Nuiten on this subject [35]. Nuiten investigates relationsbetween independence conditions on nets of operator algebras on the one hand, and sheaf conditions onthe corresponding Bohrified functors on the other. As argued in [51], the sheaf condition of Nuiten canbe viewed as a sheaf condition on the functor Σ : P op → Set (although strictly speaking the coveringrelation which is involved does not satisfy all conditions for the basis of a Grothendieck topology).
The authors would like to thank Klaas Landsman for his comments, which greatly improved this paper,and also an anonymous referee for his or her comments.
References [1] Huzihiro Araki (1999):
Mathematical theory of quantum fields . International Series of Monographs onPhysics
The spectral theory of commutative C*-algebras: the constructive Gelfand-Mazur theorem . Quaestiones Mathematicae
The spectral theory of commuta-tive C*-algebras: the constructive spectrum . Quaestiones Mathematicae
A globalisation of the Gelfand duality theorem . Annals of Pure and Applied Logic
The generally covariant locality principle—anew paradigm for local quantum field theory . Comm. Math. Phys.
A topos perspective on the Kochen-Specker theorem: II. Con-ceptual Aspects and Classical Analogues . International Journal of Theoretical Physics as Spitters, Steven Vickers & Sander Wolters [7] Jeremy Butterfield & Chris J. Isham (2002):
A topos perspective on the Kochen-Specker theorem: IV. IntervalValuations . International Journal of Theoretical Physics
Intuitionistic Quantum Logicof an n-level System . Foundations of Physics
39, pp. 731–759, doi:10.1007/s10701-009-9308-7.[9] Thierry Coquand (2005):
About Stone’s Notion of Spectrum . Journal of Pure and Applied Algebra
Constructive Gelfand duality for C*-algebras . Mathematical Pro-ceedings of the Cambridge Philosophical Society
Integrals and Valuations . Journal of Logic and Analysis
Flows on Generalised Gelfand Spectra of Nonabelian Unital C ∗ -Algebras and TimeEvolution of Quantum Systems . Available at http://arxiv.org/abs/1212.4882 .[13] Andreas D¨oring (2012): Generalised Gelfand Spectra of Nonabelian Unital C ∗ -Algebras . Available at http://arxiv.org/abs/1212.2613 .[14] Andreas D¨oring & John Harding (2010): Abelian subalgebras and the Jordan structure of a von Neumannalgebra . Available at http://arxiv.org/abs/1009.4945 .[15] Andreas D¨oring & Chris Isham (2011): ‘What is a Thing?’: Topos Theory in the Foundations of Physics .In Bob Coecke, editor:
New Structures in Physics , chapter 13,
Lecture Notes in Physics
The Born Rule as Structure of Spectral Bun-dles . In Bart Jacobs, Peter Selinger & Bas Spitters, editors:
Proceedings of the 8th International Workshopon Quantum Physics and Logic, Nijmegen 2011 , EPTCS
95, pp. 81–90, doi:10.4204/EPTCS.95.8.[17] Rudolf Haag (1996):
Local Quantum Physics: Fields, Particles, Algebras , second edition. Texts and Mono-graphs in Physics, Springer-Verlag, Berlin, doi:10.1007/978-3-642-61458-3.[18] Jan Hamhalter (2011):
Isomorphisms of ordered structures of abelian C ∗ -subalgebras of C ∗ -algebras . J.Math. Anal. Appl.
Structure of associative subalgebras of Jordan operator alge-bras . The Quarterly Journal of Mathematics
A topos perspective on the Kochen-Speckertheorem: III. Von Neumann Algebras as the Base Category . International Journal of Theoretical Physics
A topos for algebraic quantum theory . Communica-tions in mathematical physics
Bohrification . In Hans Halvorson, ed-itor:
Deep Beauty , Cambridge University Press, pp. 271–313, doi:10.1017/CBO9780511976971.008.ArXiv:0909.3468.[23] Chris Heunen, Nicolaas P. Landsman & Bas Spitters (2012):
Bohrification of operator algebras and quantumlogic . Synthese
Function Spaces in the Category of Locales . In B. Banaschewski & R.-E. Hoffmann,editors:
Continuous Lattices: Proceedings, Bremen, 1979 , Lecture Notes in Mathematics
A topos perspective on the Kochen-Specker theorem: I. Quan-tum States as Generalized Valuations . International Journal of Theoretical Physics
Fibrewise Topology . Cambridge Tracts in Mathematics
91, Cambridge University Press,Cambridge, doi:10.1017/CBO9780511896835. Gelfand spectra in Grothendieck toposes [27] Peter T. Johnstone (1984):
Open locales and exponentiation . In J.W. Gray, editor:
Mathematical Appli-cations of Category Theory , Contemporary Mathematics
30, American Mathematical Society, pp. 84–116,doi:10.1090/conm/030/749770.[28] Peter T. Johnstone & Andr´e Joyal (1982):
Continuous categories and exponentiable toposes . Journal of Pureand Applied Algebra
Stone Spaces . Cambridge Studies in Advanced Mathematics
3, Cambridge UniversityPress.[30] P.T. Johnstone (2002):
Sketches of an Elephant: A Topos Theory Compendium, vol. 1 . Oxford LogicGuides
44, Oxford University Press.[31] P.T. Johnstone (2002):
Sketches of an Elephant: A Topos Theory Compendium, vol. 2 . Oxford LogicGuides
44, Oxford University Press.[32] Andr´e Joyal & Miles Tierney (1984):
An extension of the Galois theory of Grothendieck . Memoirs of theAmerican Mathematical Society
Sheaves in Geometry and Logic . Springer, doi:10.1007/978-1-4612-0927-0.[34] I. Moerdijk (1986):
Continuous Fibrations and Inverse Limits of Toposes . Compositio Mathematica
58, pp.45–72. Available at .[35] Joost Nuiten (2011):
Bohrification of local nets of observables . Ph.D. thesis, Utrecht. Available at http://arxiv.org/abs/1109.1397 .[36] Erik Palmgren & Steven Vickers (2007):
Partial Horn Logic and Cartesian Categories . Annals of Pure andApplied Logic
The Space of Measurement Outcomes as a Spectral Invariant for Non-CommutativeAlgebras . Foundations of Physics
42, pp. 896–908, doi:10.1007/s10701-011-9619-3.[38] S.J. Vickers & C.F. Townsend (2004):
A Universal Characterization of the Double Powerlocale . TheoreticalComputer Science
Topology via Logic . Cambridge University Press.[40] Steven Vickers (1993):
Information Systems for Continuous Posets . Theoretical Computer Science
The Double Powerlocale and Exponentiation: A Case Study in Geometric Reasoning . Theory and Applications of Categories
12, pp. 372–422. Available at .[42] Steven Vickers (2004):
Entailment Systems for Stably Locally Compact Locales . Theoretical ComputerScience
Localic Completion of Generalized Metric Spaces I . Theory and Applications ofCategories
14, pp. 328–356. Available at .[44] Steven Vickers (2006):
Compactness in Locales and in Formal Topology . Annals of Pure and Applied Logic
Locales and Toposes as Spaces . In Marco Aiello, Ian E. Pratt-Hartmann &Johan F.A.K. van Benthem, editors:
Handbook of Spatial Logics , chapter 8, Springer, pp. 429–496,doi:10.1007/978-1-4020-5587-4 8.[46] Steven Vickers (2009):
Localic Completion of Generalized Metric Spaces II: Powerlocales . Journal of Logicand Analysis
A Monad of Valuation Locales . Available at .[48] Steven Vickers (2014):
Continuity and Geometric Logic . Journal of Applied Logic as Spitters, Steven Vickers & Sander Wolters [49] K. Viglas (2004):
Topos Aspects of the Extended Priestley Duality . Ph.D. thesis, Department of Computing,Imperial College, London.[50] Sander Wolters (2013):
A Comparison of Two Topos-Theoretic Approaches to Quantum Theory.
Communi-cations in Mathematical Physics
Independence Conditions for Nets of Local Algebras asSheaf Conditions . Available at http://arxiv.org/abs/1309.5639http://arxiv.org/abs/1309.5639