text
stringlengths
128
2.05k
3.15 Examples We give examples of cartesian pre-closed categories which are not cartesian multi-closed. (a) The category [MATH] is not cartesian multi-closed (by Remark 2.10 ). It is cartesian pre-closed because it is infinitary extensive, thus Theorems 1 and 2 of
apply. (b) [MATH] , the full subcategory of [MATH] spaces. The argument is the same. (c) The full subcategory [MATH] of [MATH] on all sets of cardinality other than continuum is cartesian pre-closed but not multi-closed. Indeed, this category has finite products. To prove that [MATH] is small, it is by Remark 3.4 suffi...
[MATH] whose left-hand component is the evaluation provides a weakly initial element of [MATH] . In fact, given [MATH] in [MATH] there exists a (not necessarily unique) [MATH] with [MATH]
[MATH] is not cartesian multi-closed because, having a strictly initial object, it would be cartesian closed, see Remark 2.10 . However, the sets [MATH] and [MATH] obviously fail to have an exponential object in [MATH] . (Since given [MATH] in [MATH] , its elements would bijectively correspond to maps [MATH] , hence [M...
We are going to improve the above Theorem by moving to local cartesian closedness. For that we use the characterization of locally cartesian closed exact completions [MATH] due to Carboni and Rosolini
, which we now recall. In that paper [MATH] was supposed to have weak finite limits. Recently, Emmenegger showed that the result is only correct if finite completeness is assumed, see
. The following Definition and Theorem are from 3.16 Definition By a weak simple product of a morphism [MATH] over [MATH] is meant an object [MATH] and morphisms
[MATH] and [MATH] forming a commutative triangle as follows [EQUATION] They are also requested to have the following weak universal property: given morphisms [MATH] and [MATH]
with [MATH] , there exists a morphism [MATH] such that [MATH] and [MATH] 3.17 Theorem A finitely complete category has a locally cartesian closed exact completion if and only if it has weak simple products.
3.18 Remark Weak simple products are also callled weak dependent products. The ”pre” version was called approximate dependent product in
it is a set of objects [MATH] [MATH] and morphisms [MATH] [MATH] [MATH] such that [MATH] . The weak universal property states that given [MATH] and [MATH]
with [MATH] , there exists [MATH] and a morphism [MATH] with [MATH] and [MATH] 3.19 Theorem For a complete category [MATH] the following conditions are equivalent:
(i) [MATH] is locally cartesian closed, (ii) [MATH] is cartesian closed, (iii) [MATH] is cartesian pre-closed, (iv) [MATH] has weak simple products, and
(v) [MATH] is cartesian pre-closed. Proof. iii [MATH] iv. In Lemma A1 of it is proved that a complete, cartesian pre-closed category [MATH] has approximate dependent products. We apply this to [MATH] to conclude that [MATH] has weak simple products. Indeed, given a morphism [MATH] , let
[MATH] and [MATH] [MATH] form an approximate dependent product in [MATH] . Since [MATH] preserves coproducts (by Example 2.1 (e) and Remark 2.2 (2)), the following morphisms [MATH]
and [MATH] are easily seen to form a weak simple product of [MATH] over [MATH] iv [MATH] i. [MATH] is locally cartesian closed by Theorem 3.17
[MATH] ii is clear. ii [MATH] iii. In Theorem 1 of every complete, infinitary extensive category [MATH] with [MATH] cartesian closed is proved to be cartesian pre-closed. Apply this to [MATH] , using Theorem 3.12
[MATH] ii follows from Theorem 3.12 3.20 Corollary Let [MATH] be a complete category which is a. cartesian closed, or b. additive, or
c. dual to an extensive category. Then [MATH] is locally cartesian closed. Thus categories [MATH] [MATH] [MATH] and [MATH] (= accessible set functors) are locally cartesian closed.
4. [MATH] is a pretopos but seldom a topos Recall that a pretopos is an exact and extensive category. That is, a category which is regular, has effective equivalence relations, and has finite coproducts which are disjoint and universal.
4.1 Lemma Let [MATH] be a finitely pre-complete category. Then [MATH] is a pretopos. Proof. By Remark 3.7 and Corollary 3.8 [MATH] is closed under existing limits in [MATH] , and it is obviously closed under colimits. Thus [MATH] shares with [MATH] all the exactness properties of colimits and finite limits. In particul...
4.2 Remark By the same argument we derive that if [MATH] is finitely pre-complete, then (a) [MATH] is infinitary extensive, see Remark 2.4
(b) every monomorphism in [MATH] is regular, and (c) every epimorphism in [MATH] is regular. For small categories [MATH] , we know that [MATH] is a topos. The converse ”almost” holds:
4.3 Theorem Given a category [MATH] with copowers and finite intersections, then [MATH] is a topos iff [MATH] is essentially small.
Proof. (1) We consider first the case that [MATH] is a preordered class. We are to prove that if [MATH] is a topos, then [MATH] is essentially small. Since [MATH] has a terminal object, [MATH] has a pre-terminal set of objects, and since [MATH] is wellpowered, so is [MATH] . It is easy to see that a wellpowered preorde...
(2) Next consider [MATH] having an object [MATH] which is the domain of a parallel pair of distinct morphisms. Since for every cardinal [MATH] the coproduct injections of [MATH] are then pairwise distinct, [MATH] is not essentially small. We are going to prove that [MATH] is not topos.
Assuming that [MATH] has a subobject classifier [MATH] , we derive a contradiction. Since [MATH] contains all hom-functors, Yoneda lemma implies that (up to natural isomorphism) [MATH] is given as follows: on objects [MATH] we have
[EQUATION] and on morphisms [MATH] the map [MATH] forms preimages of subfunctors under [MATH] . We are going to prove that this presheaf [MATH] is not petty, which is the desired contradiction.
For every set of elements of [MATH] [EQUATION] we are going to find an object [MATH] of [MATH] and an element of [MATH] [EQUATION]
such that [MATH] for all [MATH] and all [MATH] . Our choice of [MATH] is a copower of the above object [MATH] [EQUATION] where [MATH] is a cardinal with
[EQUATION] Denote by [MATH] [MATH] ) the coproduct injections. Choose arbitrary two elements [MATH] in [MATH] and form the intersection [MATH]
[EQUATION] Given any distinct elements [MATH] in [MATH] , they yield the same intersection: [EQUATION] This follows from the (obvious) fact that there is an isomorphism [MATH] with [MATH] and [MATH]
Our choice of the subfunctor [MATH] of [MATH] uses the above morphism [MATH] : put [MATH] and [MATH] for all [MATH] and define a small presheaf [MATH] as the wide pushout of the following morphisms
[EQUATION] Up to natural isomorphism, [MATH] can be described as follows: on objects [MATH] put [EQUATION] where [MATH] is the least equivalence relation merging, for every morphism [MATH] , all [MATH] with [MATH] to one element (denoted by [MATH] ). The equivalence class of a morphism [MATH] not factorizing through [M...
This small presheaf [MATH] is a subobject of [MATH] since the cocone [MATH] fulfills [MATH] (independent of [MATH] ). This yields a natural transformation
[EQUATION] It assigns to [MATH] for [MATH] the value [EQUATION] and to [MATH] not factorizing through [MATH] the value [EQUATION]
This morphism [MATH] is monic: (a) Given distinct [MATH] [MATH] , then [MATH] since [MATH] is monic: indeed, each [MATH] is a split monomorphism (the codiagonal splitting it).
(b) Given distinct [MATH] and [MATH] not factorizing through [MATH] , then in case [MATH] we have [MATH] since [MATH] is a (split) monomorphism, and in case [MATH] we have [MATH] Otherwise [MATH] would factorize through [MATH] (and [MATH] through [MATH] ) due to the above pullback [MATH]
(c) Given [MATH] and [MATH] not factorizing through [MATH] , then [MATH] : since [MATH] is a split monic and [MATH] , the equality [MATH] would yield a factorization [MATH] or [MATH] through [MATH]
It remains to prove [EQUATION] Assuming to the contrary [MATH] , we prove that the morphisms [EQUATION] are pairwise distinct, contradicting [MATH]
For elements [MATH] [MATH] of [MATH] with [MATH] we are going to verify that [MATH] . The equality [MATH] means that in [MATH] we have a pullback as follows:
[EQUATION] Apply this to [MATH] in the component of [MATH] indexed by [MATH] : we get [EQUATION] with [EQUATION] since [MATH] . (Observe that [MATH] clearly does not factorize through [MATH] .) The codiagonal [MATH] yields
[EQUATION] with [EQUATION] Indeed, naturality of [MATH] gives, since [MATH] [EQUATION] Thus, the above pullback applied to [MATH] provides
[EQUATION] Since [MATH] does not factorize through [MATH] , neither does [MATH] , therefore [MATH] lies in some component [MATH] of [MATH] and [MATH] . Thus [MATH] factorizes through [MATH] , proving [MATH] , and [MATH] also factorizes through [MATH] , hence [MATH] , as desired.
4.4 Example The categories [MATH] and [MATH] are locally cartesian closed (by Theorem 3.19 ) but not toposes. 4.5 Open problem Is there a category [MATH] which is not essentially small with [MATH] a topos?
4.6 Remark (1) Menni characterized categories whose exact completion is a topos, see . He defined a generic proof as a map [MATH] such that for every map [MATH] there exists a map [MATH] for which [MATH] factorizes through [MATH] , and [MATH] factorizes through [MATH] . And he proved (in Theorem 1.2) that a category ha...
(2) Consequently, the theorem just proved implies for complete, cartesian pre-preclosed categories [MATH] that the free colimit completion [MATH] is a topos iff the free coproduct completion [MATH] has a generic proof.
4.7 Corollary If a complete, cartesian pre-closed category with copowers is not essentially small, then its free coproduct completion does not have a generic proof.
See Theorem 3.19 The following examples are proved by Menni in a somewhat more technical manner (see , Propositions 5.7 and Lemma 5.5):
4.8 Example (1) The category [MATH] does not have a generic proof. Indeed, it is equivalent to [MATH] (see Example 2.1 (c)). (2) For every small category [MATH] the free coproduct completion has a generic proof. The exact completion, which is the presheaf category for [MATH] , is namely a topos.
5. [MATH] is sometimes wellpowered but not often This last section is devoted to the question whether [MATH] is wellpowered or cowellpowered. We first observe that the two problems are more or less equivalent.
5.1 Proposition Every cowellpowered category [MATH] is also wellpowered. The converse holds for finitely pre-complete categories [MATH]
Proof. (1) Let [MATH] be cowellpowered. Following Remark 4.2 , every monomorphism [MATH] in [MATH] is monic in [MATH] , too. Form the pushout in [MATH]
[EQUATION] Since [MATH] and [MATH] are small, so are [MATH] and [MATH] . Thus we obtain a quotient [MATH] in [MATH] Distinct subobjects of [MATH] in [MATH] yield distinct quotients in [MATH] . Indeed, [MATH] certainly has the corresponding property, therefore, so does [MATH] . Since [MATH] has only a set of quotients, ...
(2) Let [MATH] be finitely pre-complete, and [MATH] be wellpowered. Following Remark 4.2 , every epimorphism [MATH] is regular in [MATH] . Hence it is determined by its kernel pair [MATH] . Since [MATH] is a subobject of [MATH]
[MATH] is cowellpowered. 5.2 Proposition If a wellpowered category [MATH] has all morphisms monic, then [MATH] is cowellpowered.
Proof. (1) For every functor [MATH] denote by [MATH] a class of objects representing all [MATH] with [MATH] up to isomorphism. Then we verify that [MATH] is small iff [MATH] is a set.
Indeed, let [MATH] be small. If [MATH] , then [MATH] is a set because [MATH] has only a set of subobjects. If [MATH] is a colimit of a diagram with objects [MATH] [MATH] , then clearly [MATH] . This is a set if [MATH] is small.
Conversely, let [MATH] be a set. Recall that [MATH] is a colimit of the diagram of its elements. More precisely, we form the category [MATH] of all pairs [MATH] where [MATH] is an object of [MATH] and [MATH] . Morphisms [MATH] are those morphisms [MATH] of [MATH] with [MATH] The diagram
[EQUATION] has canonical colimit [MATH] . Since [MATH] is a set, [MATH] is essentially a small category, thus, [MATH] is small. (2) The category [MATH] clearly has non-empty pre-limits. (Indeed, choose an object [MATH] of the given diagram [MATH] . Then every cone of [MATH] has a subobject of [MATH] as its domain.) Thu...
5.3 Proposition [MATH] is cowellpowered for every finitely complete, wellpowered category [MATH] such that (1) all hom-sets of objects that are neither terminal nor initial are non-empty, and
(2) morphisms with non-terminal codomains have (split epi, mono)-factorizations. Proof. (a) We first prove for every coproduct of representables
[EQUATION] that it has only a set of subobjects in [MATH] . Since [MATH] is closed under coproducts in [MATH] and [MATH] is infinitary extensive, it is sufficient to prove that every representable functor has only a set of subobjects.
To give a subobject [MATH] of [MATH] means to give, for every object [MATH] of [MATH] , a subset [MATH] such that for all morphisms
[MATH] we have [EQUATION] In case [MATH] is a terminal object, then [MATH] is determined by a class [MATH] such that no morphism with domain in [MATH] has codomain in the complement [MATH] . Due to (1) either [MATH] or its complement consists of objects isomorphic to [MATH] (initial) or [MATH] (terminal), thus, the num...
Suppose [MATH] is non-terminal. We prove that [MATH] is determined by its values at (i) 1 and (ii) subobjects of [MATH] . Since [MATH] is wellpowered, this proves that there is only a set of possibilities.
For that, consider an object [MATH] . Given [MATH] factorize it as a split epic [MATH] followed by a monic [MATH] If [MATH] splits [MATH] then [MATH] and, since [MATH] , we have [MATH] . Conversely, since [MATH] [MATH] whenever
[MATH] . This proves our claim that [MATH] is determined by [MATH] and all [MATH] (b) The full subcategory of [MATH] formed by all petty functors is cowellpowered. Indeeed, we only need to prove that coproducts of hom-functors [MATH] have sets of quotients in [MATH] . Observe that [MATH] is also a coproduct of hom-func...
(c) Following Remark 3.4 , it is sufficient to prove that every subfunctor of a hom-functor is petty. 1. Consider first [MATH] for the terminal object [MATH] . To give a subfunctor [MATH] means to specify a class [MATH] of objects such that
[MATH] and [MATH] implies [MATH] ; then the subfunctor assigns [MATH] to objects of [MATH] and [MATH] else. Due to (1), the only possibilities are [MATH] or [MATH] consisting of initial objects. In both cases every element [MATH]
yields a weakly initial object of [MATH] , thus [MATH] is petty. 2. Next let [MATH] be a subfunctor of [MATH] where [MATH] is non-terminal. Without loss of generality assume that [MATH] for all [MATH] and that on morphisms [MATH] we have [MATH] . Since [MATH] is wellpowered, we have a set [MATH]
[MATH] of subobjects representing the images of all morphisms [MATH] [MATH] , lying in [MATH] . As above, each [MATH] lies in [MATH] The elements [MATH] [MATH] , form a weakly initial set in [MATH] : for every [MATH] as above we have [MATH]
5.4 Example [MATH] is wellpowered and cowellpowered for all of the following categories and their duals: (1) [MATH] (2) Vector spaces over every field.
(3) Sets and partial functions. Indeed, [MATH] as well as [MATH] satisfy the assumptions of the above Proposition. ( [MATH] was the reason for the complicated condition (2).)
Recall that a category [MATH] is concrete it there exists a faithful functor [MATH] 5.5 Remark (1) For a finitely pre-complete category [MATH] , the category [MATH] is concrete if and only if it is wellpowered.
Indeed, [MATH] is finitely complete (Corollary 3.8 ) and monomorphisms are regular (Remark ( 4.2 ). Thus the result follows from
Theorem 4.1(iii). (2) The limit completion of a category [MATH] is dual to [MATH] . Thus the limit completion of the categories in the above example are wellpowered and cowellpowered.
5.6 Remark (1) The category [EQUATION] of all accessible set functors is wellpowered and cowellpowered. Indeed, recall from Remark 3.1 (e) that this is precisely the category [MATH] . Recall also that this category is a locally cartesian closed pretopos (see Corollary 3.20 and Lemma 4.1 ).
(2) This category has also been studied by Barto who has also proved that it is concrete, and universal . That is, every concrete category can be fully embedded into [MATH]
Recall that a clique on a set [MATH] is a graph whose arrows lead from every node to all distinct nodes. Graph homomorphisms between cliques are precisely the monic maps. Therefore, if [MATH] is a clique on
[MATH] nodes, then (a) for all ordinals [MATH] infinitely many morphisms exist from [MATH] to [MATH] , and (b) no morphism from [MATH] to [MATH] exists if [MATH]
Recall that a class [MATH] of morphisms of a category [MATH] is right-cancellative if [MATH] implies that [MATH] 5.7 Definition A class of objects [MATH] for [MATH] of a category is called clique-like provided that there exists a right-cancellative
class [MATH] of morphisms such that (a) for every ordinal [MATH] more than one morphism from [MATH] to [MATH] exists in [MATH] , and
(b) no morphism from [MATH] to [MATH] exists in [MATH] either (b1) for all pairs of ordinals [MATH] or (b2) for all pairs of ordinals [MATH]
5.8 Proposition If a category [MATH] has a clique-like class of objects, then [MATH] is not cowellpowered. Proof. For every ordinal [MATH] denote by
[EQUATION] the multiple coequalizer of the following morphisms of [MATH] (a full subcategory of [MATH] ): [EQUATION] for all morphisms [MATH] of [MATH] . These small functors [MATH] are quotients of
[MATH] in [MATH] It remains to prove that they are pairwise non-isomorphic. Indeed, consider [MATH] Assume first that (b1) holds. Choose distinct morphisms [MATH] in [MATH] . They are merged by [MATH] because the element [MATH] of [MATH] is sent to [MATH] by [MATH] and to [MATH] by [MATH] . But [MATH] does not merge [M...
[MATH] cannot be a composition [MATH] for any [MATH] in [MATH] and any [MATH] . Thus [MATH] cannot be merged with any distinct morphism from [MATH] to [MATH] by [MATH]
In the case that (b2) holds the proof is the same, we just calculate the components [MATH] and [MATH] 5.9 Example In the category [MATH] of sets and monomorphisms we have a clique-like collection: choose a set [MATH]
of power [MATH] and [MATH] all morphisms. Thus [MATH] is not cowellpowered. Compare this with the fact that by Proposition 5.2 [MATH] is.
5.10 Example For the following categories [MATH] the colimit completions [MATH] and [MATH] are not cowellpowered: we present a clique-like class of objects in [MATH] as well as in its dual. Here [MATH] is always chosen to be the class of all morphisms of [MATH]
(1) The category of graphs: let [MATH] be a clique on [MATH] vertices. This is a clique-like class in [MATH] . Now take the same class except changing [MATH]
to the complete graph on 2 vertices (so that more than one homomorphism exists from every nonempty graph into [MATH] ). The result is a clique-like class in [MATH]
(2) Any alg-universal category [MATH] , i.e, one admitting a full embedding [MATH] The monograph presents a number of examples of such categories, e.g., semigroups or rings with unit.
If [MATH] is a clique-like collection in [MATH] or [MATH] , then [MATH] is one in [MATH] or [MATH] , resp. (3) The category [MATH] of accessible set functors is, as we have seen above, cowellpowered. By Remark 5.6 (2), this category is alg-universal, therefore [MATH] is not cowellpowered.
(4) The category [MATH] of algebras on one unary operation. To present a clique-like class, we first construct unary algebras [MATH] for [MATH] with no homomorphisms from [MATH] to [MATH] whenever
[MATH] Recall from the concept of the rank [MATH] of an element [MATH] of an algebra [MATH] which is either an ordinal or [MATH] , an element larger than all ordinals. We work with [MATH] given by a set [MATH] and its endomap [MATH] . The rank is defined to be [MATH] iff there exist elements [MATH] with [MATH]
for all [MATH] and [MATH] . For all other elements [MATH] of [MATH] rank is defined by [EQUATION] Every homomorphism [MATH] is nondecreasing on ranks:
[EQUATION] for all [MATH] . See , Proposition 5.2. Let us construct algebras [MATH] for [MATH] such that [MATH] contains an element
[MATH] of rank [MATH] , but no element of rank at least [MATH] . We proceed by transfinite induction: For [MATH] put [MATH] [MATH] and [MATH]