text
stringlengths
128
2.05k
(2) The embedding [MATH] clearly preserves existing limits. (3) Monomorphisms of [MATH] are precisely the morphisms [MATH] such that [MATH] is monic in [MATH] and each [MATH] is monic in [MATH] . We always represent them by collections with [MATH] where [MATH] is the inclusion map.
(4) Analogously, epimorphisms in [MATH] are precisely the morphisms [MATH] with [MATH] epic in [MATH] and each [MATH] epic in [MATH]
(5) [MATH] is wellpowered (or cowellpowered) iff [MATH] is. 2.3 Remark If [MATH] has pullbacks, then pullbacks of morphisms [EQUATION]
in [MATH] are computed as follows: form a pullback of [MATH] and [MATH] in [MATH] [EQUATION] and for each [MATH] form a pullback in [MATH]
[EQUATION] This defines an object [MATH] and morphisms [MATH] [MATH] of [MATH] (by the given functions [MATH] [MATH] and the given components [MATH] [MATH] ). It is easy to see that the following square
[EQUATION] is a pullback in [MATH] 2.4 Remark Recall that a finitely complete category with coproducts is infinitary extensive if coproducts are universal (i.e., preserved by pulling back along a morphism) and disjoint (i.e., coproduct injections are monic and their pairwise intersections are formed by the initial obje...
It is extensive if finite coproducts are universal and disjoint. 2.5 Proposition For every category [MATH] the following conditions are equivalent:
(i) [MATH] is extensive, (ii) [MATH] is infinitary extensive and (iii) [MATH] is finitely multi-complete. Proof. [MATH] iii is clear: extensivity includes finite completeness, so use Remark 2.2 (1).
iii [MATH] ii. We prove that coproducts in [MATH] are disjoint and universal. A coproduct of two objects [MATH] and [MATH] is obviously the collection [MATH] with [MATH] and [MATH] . The description of pullbacks in Remark 2.5 makes it clear that the pullback of the coproduct injections [MATH] and [MATH] is the empty co...
For the verification that coproducts are universal in [MATH] we also use just a coproduct [MATH] of a pair of objects, the general proof is again completely analogous. Let
[EQUATION] be an arbitrary morphism. Then [MATH] for [MATH] and [MATH] . Thus [EQUATION] where [MATH] and [MATH] . And the components of [MATH] define obvious morphisms [MATH] and [MATH] . It is easy to see that we get the following pullbacks
[EQUATION] where [MATH] [MATH] are the coproduct injections of [MATH] ii [MATH] i is trivial. 2.6 Lemma Let [MATH] be objects of [MATH] for [MATH] . A product of this family exists in [MATH] iff for every function [MATH] a multi-product of [MATH] for [MATH] exists in [MATH]
Proof. (1) We prove the necessity: Let [MATH] be a product [MATH] in [MATH] with projections [EQUATION] for [MATH] . Given [MATH] denote by [MATH] the set of all [MATH] with [MATH] for all [MATH] This defines cones in [MATH] as follows: for every [MATH] we have the [MATH] -component of [MATH]
[EQUATION] The set of these cones (indexed by [MATH] ) is a multi-product of [MATH] for [MATH] Indeed, every cone in [MATH] [EQUATION]
yields a cone [MATH] in [MATH] with [MATH] choosing [MATH] and the unique component of [MATH] being [MATH] . The fact that [MATH] has a unique factorization through [MATH] in [MATH] clearly implies that there exists a unique [MATH] and a unique factorization of [MATH] through [MATH] in [MATH]
(2) Conversely, if all the above multi-products exist in [MATH] , then [MATH] is given in [MATH] as follows. For every [MATH] choose a multi-product, indexed by a set [MATH] , as follows:
[EQUATION] This defines a morphism of [MATH] for every [MATH] : put [MATH] and let [EQUATION] be defined by [EQUATION] for all [MATH] with components
[EQUATION] as above. For every object [MATH] of [MATH] and every cone [MATH] [MATH] , in [MATH] we prove that there is a unique factorization through [MATH] . It then follows immediately that the same holds for families of more that one object as well.
The given cone chooses for every [MATH] a unique [MATH] , let [MATH] denote the resulting function. And [MATH] has a component [MATH] in [MATH] [MATH] . There exists a unique [MATH] and a unique factorization of [MATH] through [MATH] in [MATH] . This clearly implies that [MATH] factorizes uniquely through [MATH] in [MA...
2.7 Remark Recall that a functor [MATH] is a left multi-adjoint if, for every object [MATH] , the category [MATH] has a multi-terminal set of objects. This is a set [MATH] [MATH] such that for every [MATH] there is a unique [MATH] and a morphism
[MATH] such that [MATH] , and moreover [MATH] is also unique. 2.8 Definition A category [MATH] is called cartesian multi-closed if it has finite products and each endofunctor [MATH] is a left multi-adjoint.
Explicitly, this means that for every pair of objects [MATH] and [MATH] there exists a set of objects [MATH] [MATH] and morphisms
[EQUATION] multi-universal in the expected sense: For every morphism [MATH] there exists a unique [MATH] for which some morphism [MATH] makes the following triangle
[EQUATION] commutative, and moreover, [MATH] is also unique. 2.9 Examples The following categories are cartesian multi-closed: (a) Every cartesian closed category.
(b) Every category with binary products satisfying [MATH] (e.g. every additive category). Indeed, given [MATH] , the family [MATH] indexed by all [MATH] is obviously multi-universal: every morphism [MATH] has the form [MATH] for a unique pair [MATH] and [MATH]
(c) Every dual of an extensive category (e.g. [MATH] ). Indeed, first observe that every object [MATH] has only a set of decompositions [MATH] in [MATH] (up to isomorphism). In fact, every such decomposition yields a morphism in [MATH] taking [MATH] to the left-hand summand and [MATH] to the right-hand one. And differe...
Now, given [MATH] , the family [MATH] in [MATH] indexed by all decompositions [MATH] and all [MATH] is obviously multi-universal: every morphism [MATH] in [MATH] has the form [MATH] in [MATH] for unique [MATH] and
[MATH] 2.10 Remark Let [MATH] have a strict initial object [MATH] , i.e., if a morphism [MATH] exists, then [MATH] is initial. Then [MATH] is cartesian closed iff it is cartesian multi-closed. Indeed, consider the above triangle for [MATH] : since [MATH] , we have a unique [MATH] , but for every [MATH] the unique [MATH...
This implies that for example the category [MATH] of topological spaces is not cartesian multi-closed. 2.11 Theorem Let [MATH] have finite products. Then [MATH] is cartesian closed iff [MATH] is cartesian multi-closed and has multi-products.
Proof. (1) Let [MATH] be cartesian closed. Given objects [MATH] [MATH] of [MATH] we denote the corresponding exponential object in [MATH] by
[EQUATION] together with the counit having components [EQUATION] (1a) We prove that [MATH] is cartesian multi-closed. It is our task to prove that for every morphism [MATH] in [MATH] there exists a unique [MATH] with a commutative triangle
[EQUATION] for some [MATH] , and moreover [MATH] is unique. Indeed, [MATH] is a morphism of [MATH] , thus we have a unique [MATH] with
[EQUATION] in [MATH] . This morphism [MATH] is given by choosing [MATH] and a morphism [MATH] in [MATH] with [MATH] Conversely, suppose the triangle above commutes, then we must prove [MATH] and [MATH] is as above. The choice of [MATH] and [MATH] in that triangle defines a morphism [MATH] with [MATH] , and since [MATH]...
(1b) [MATH] has multi-products. Indeed, for every collection [MATH] [MATH] in [MATH] let [MATH] be the object of [MATH] with components [MATH] , terminal in [MATH] . Since in [MATH] we have [MATH] and [MATH] has the exponential object [MATH] , it has the following product
[EQUATION] By Lemma 2.6 for every [MATH] a multi-product of [MATH] [MATH] exists in [MATH] . This proves our claim: put [MATH] (2) Let [MATH] be cartesian multi-closed and have multi-products. Then [MATH] has for every pair [MATH] [MATH] of objects a multi-universal collection of morphism [MATH] [MATH] w.r.t [MATH] . W...
[MATH] and [MATH] . The exponential object [MATH] in [MATH] is formed as follows: let for every [MATH] a multi-universal collection be given
[EQUATION] This defines a morphism [EQUATION] whose index function [MATH] is [MATH] and whose components are [MATH] . We prove that in [MATH] we have
[EQUATION] with the above counit [MATH] . Indeed, to give a morphism [MATH] means to choose for every [MATH] (a) indices [MATH] and [MATH] , and
(b) a morphism [MATH] And to give a morphism [MATH] means to choose for every [MATH] an index [MATH] and a morphism [MATH] , where the latter is equivalent to choosing [MATH] and a morphism from [MATH] to [MATH] , due to the universal property of [MATH]
Turning to a general pair of objects [MATH] and [MATH] , the existence of exponential objects [MATH] and the fact that [MATH] is a coproduct in [MATH] of [MATH] imply
[EQUATION] the above product exists because [MATH] has multi-products (see Lemma 2.6 ). 2.12 Corollary [MATH] is cartesian closed for every complete, additive category.
2.13 Definition A category [MATH] is called locally cartesian multi-closed if every slice [MATH] is cartesian multi-closed. 2.14 Theorem
Let [MATH] be a complete category. Then [MATH] is locally cartesian closed iff [MATH] is locally cartesian multi-closed. Proof. Let [MATH] be locally cartesian multi-closed. For every object [MATH] of [MATH] we verify that [MATH] is cartesian closed. This category is isomorphic to the product of categories [MATH] [MATH...
Since [MATH] has products, from its cartesian multi-closedness follows that [MATH] is cartesian closed (see Theorem 2.11 ). And a product of cartesian closed categories is cartesian closed (with exponential objects defined componentwise).
Conversely, let [MATH] be locally cartesian closed. For every object [MATH] of [MATH] the category [MATH] is isomorphic to [MATH] and is thus cartesian closed. Therefore [MATH] is cartesian multi-closed (see Theorem 2.11 ), proving that
[MATH] is locally cartesian multi-closed. 2.15 Corollary Let [MATH] be a complete category with a strict initial object. Then [MATH] is (locally) cartesian closed if and only if [MATH] is (locally) cartesian closed.
This follows from Remark 2.10 , Theorem 2.11 and Theorem 2.14 2.16 Examples (i) For the topos [MATH] of finite sets, [MATH] is not cartesian closed: [MATH] does not have multi-products (see Remark 2.2 (1)). Analogously, [MATH] is not cartesian closed where [MATH] is the category of finite abelian groups.
(ii) In contrast, [MATH] is locally cartesian closed, indeed a topos, see Example 2.1 (c). (iii) [MATH] is cartesian closed by Corollary 2.12
(iv) The category [MATH] of small categories is cartesian closed but not locally so. Since it is complete and has a strict initial object, we conclude that [MATH] is cartesian closed but not locally so.
2.17 Open problem Is [MATH] locally cartesian closed? 2.18 Definition By a multi-topos is meant a finitely complete, cartesian multi-closed category with a subobject classifier.
2.19 Theorem For a complete category [MATH] the completion [MATH] is a topos iff [MATH] is a multi-topos. Proof. (1) Let [MATH] be a multi-topos with a subobject classifier [MATH] . Then the morphism
[EQUATION] corresponding to [MATH] is a subobject classifier of [MATH] , which by Theorem 2.11 proves that [MATH] is a topos. Indeed, given a subobject [MATH] with [MATH] (see Remark 2.2 (2)), the corresponding pullbacks in [MATH]
[EQUATION] yield the following morphism [EQUATION] The function [MATH] assign to every [MATH] the left-hand object of [MATH] , (with the component [MATH] above) and to every [MATH] the right-hand one. The square below
[EQUATION] is a pullback in [MATH] . This follows easily from Remark 2.3 It remains to prove that the pulback ( 2.1 ) determines [MATH] uniquely. Suppose [MATH] makes the corresponding square a pullback in [MATH] , too. For every [MATH] the commutativity of that square proves that
[MATH] is the left-hand object, [MATH] , iff [MATH] . Thus, [MATH] . For every [MATH] we are to prove that the following square [EQUATION]
is a pullback in [MATH] – then [MATH] and the proof is complete. Indeed, given a commutative square in [MATH] [EQUATION] define [MATH] by [MATH] [MATH] and [MATH] for all [MATH] Then the last square and the pullback ( 2.1 ) with [MATH] in place of [MATH] imply that the following square
[EQUATION] commutes in [MATH] . Thus, it factorizes uniquely through the modified square ( 2.1 ), which proves that [MATH] factorizes uniquely through ( 2.2 ), as required.
(2) Let [MATH] be a topos. By Theorem 2.11 [MATH] is cartesian multi-closed. Denote by [MATH] the subobject classifier of [MATH] Let [MATH] be the element given by the indexing function of [MATH] , and let
[EQUATION] be the unique component of [MATH] . For every subobject [MATH] in [MATH] , since [MATH] is also monic in [MATH] , we have a morphism [MATH] forming a pullback as follows:
[EQUATION] Thus [MATH] chooses [MATH] (since [MATH] does) and the unique component [MATH] defines a pullback in [MATH] as follows:
[EQUATION] Conversely, given [MATH] making the last square a pullback in [MATH] , then the corresponding morphism [MATH] makes the above square a pullback. Therefore, [MATH] is unique. This proves that [MATH] is a subobject classifier in [MATH]
2.20 Corollary For a complete category [MATH] with a strict initial object [MATH] is a topos if and only if [MATH] is a topos. This follows from the above Theorem and Remark 2.10
However, we have seen in Example 2.16 (i) that for the topos [MATH] the category [MATH] is not even cartesian closed. 3. [MATH] is often locally cartesian closed
We now turn to the free completion [MATH] of a category [MATH] under colimits. In the present section we concentrate on the question whether [MATH] is (locally) cartesian closed, in the next ones we ask whether it is a topos and study the (co)wellpoweredness of it.
Recall the concept of a free completion [EQUATION] under colimits: the category [MATH] is cocomplete, and there is a full embedding [MATH] such that every functor [MATH] with [MATH] cocomplete has an extension to a colimit-preserving functor [MATH] , unique up to natural isomorphism.
3.1 Example (a) [MATH] is the free colimit completion of the terminal category. (b) [MATH] is the free colimit completion of the two-element chain.
(c) For small categories [MATH] is the presheaf category [MATH] and [MATH] is the Yoneda embeddding. (d) For general categories [MATH] can be described as the full subcategory of [MATH] on all small functors , i.e., small colimits of hom-functors (see
2.29). And [MATH] is the codomain restriction of the Yoneda embedding, we denote it by [MATH] (e) [MATH] is the category of all accessible set functors. Indeed, for every locally presentable category [MATH] a functor into [MATH] is accessible iff it is a small presheaf on [MATH] (see
). Thus [MATH] is the category of all accessible set-valued functors on [MATH] The completions [MATH] and [MATH] are closely related: the former one is an exact completion of the latter. Recall that a category is exact if it has finite limits, regular factorizations with regular epimorphisms stable under pullbacks, and...
, is an exact category [EQUATION] with a full embedding [MATH] such that every functor [MATH] with [MATH] exact has an extension to an exact functor [MATH] unique up to natural isomorphism. The following is Lemma 3 in
3.2 Lemma For every finitely complete category [MATH] we have [EQUATION] Freyd introduced the following concepts in 3.3 Definition
(1) By a pre-limit of a diagram [MATH] is meant a set of cones such that every cone of [MATH] factorizes through some of them (not uniquely in general). A category is pre-complete if every diagram has a pre-limit. Dual concept: pre-colimit and pre-cocomplete category.
(2) A functor [MATH] is petty if it is a quotient of a coproduct of representable functors. It is lucid if, moreover, for every pair [MATH] of natural transformations the equalizer of [MATH] has a petty domain.
3.4 Remark (a) In the definition of a lucid functor we can restrict [MATH] to representable functors. Thus, in categories where all subfunctors of representable functors are petty, any petty functor is lucid.
(b) A functor [MATH] is petty iff its category [MATH] of elements has a weakly initial set. That is, we have a set of objects [MATH]
[MATH] of [MATH] and elements [MATH] such that every element [MATH] has, for some [MATH] and some morphism [MATH] of [MATH] , the form
[MATH] 3.5 Examples (a) Complete categories are pre-complete. (b) Small categories are pre-complete. (c) Accessible categories are pre-cocomplete. Indeed, if [MATH] is a diagram in a [MATH] -accessible category and if [MATH] is chosen so that
(i) every object of [MATH] is [MATH] -presentable, and (ii) [MATH] has less than [MATH] morphisms, then the (essentially small) set of all cocones with [MATH] -presentable codomains is a pre-colimit of [MATH]
The following is essentially Lemma 1 in 3.6 Lemma Every lucid functor is small. If [MATH] has finite pre-limits, then every small functor [MATH] is lucid.
Proof. (1) Every lucid functor [MATH] is small. Indeed, since [MATH] is petty, we have a quotient [MATH] [MATH] [MATH] a set. Thus [MATH] is the coequalizer of its kernel pair
[EQUATION] which yields a petty functor [MATH] (since [MATH] is the equalizer of [MATH] and [MATH] and [MATH] is petty). Therefore, [MATH] is also a quotient [MATH] [MATH] [MATH] a set. Consequently, [MATH] is the coequalizer of [MATH] and [MATH] , hence [MATH] is small.
(2) If [MATH] has finite pre-limits, every small functor [MATH] is lucid. Indeed, by the usual reduction of colimits to coproducts and coequalizers we have a coequalizer in [MATH] as follows
[EQUATION] where [MATH] and [MATH] are sets. By 1.4 and 1.2 in , both of the coproducts above are lucid, and by 1.9 in , applied to [MATH] , it follows that [MATH] is lucid.
3.7 Remark Whenever a limit of a diagram in [MATH] exists, it is formed objectwise. That is, [MATH] is closed under existing limits in [MATH] . This follows easily from Yoneda Lemma.
3.8 Corollary [MATH] is (finitely) complete iff [MATH] is (finitely) pre-complete. Proof. Indeed, from (finite) pre-completeness of [MATH] it follows that (finite) limits of lucid functors are lucid, see 10 , (1.7) 1.12] . Thus 3.6 implies that [MATH] is closed under (finite) limits in [MATH]
Conversely, let [MATH] be (finitely) complete. For every (finite) diagram [MATH] in [MATH] its composite with the Yoneda embedding
[MATH] has, by 3.7 , a petty pointwise limit in [MATH] . This is clearly equivalent to [MATH] having a pre-limit in [MATH] 3.9 Remark
Analogously, [MATH] has finite products iff [MATH] has finite pre-products. 3.10 Lemma Given objects [MATH] [MATH] of [MATH] for which [MATH] has the following product
[EQUATION] it follows that a pre-product of [MATH] [MATH] exists in [MATH] More generally, given a diagram [MATH] for which [EQUATION]
exists in [MATH] , then [MATH] has a pre-limit of the diagram [MATH] Proof. We prove the first statement, the latter is analogous. The functor [MATH] assigns to every object [MATH] the collection of all [MATH] where [MATH] is a morphism of [MATH] for some [MATH] . This follows from [MATH] being closed under colimits an...
[EQUATION] for every [MATH] . Consequently, if [MATH] is the set of all [MATH] with [MATH] for every [MATH] , then the cones [EQUATION]
for [MATH] form the desired pre-product. 3.11 Definition (see A category [MATH] is called cartesian pre-closed if it has finite products and for every pair of objects [MATH] and [MATH]
the functor [MATH] is small. The second author proved in Proposition 1 that for pre-complete categories [MATH] cartesian closedness of [MATH] is equivalent to cartesian pre-closedness of [MATH] . We will show that this also implies pre-completeness of [MATH]
3.12 Theorem Let [MATH] have finite products. Then [MATH] is cartesian closed if and only if [MATH] is pre-complete and cartesian pre-closed.
Proof. Let [MATH] be cartesian closed. Then [MATH] is pre-complete. Indeed, for every diagram [MATH] we form the constant diagram [MATH] with value [MATH] , a terminal object, and use the exponential object [MATH] where [MATH] . The existence of [MATH] implies (due to colimits preserved by functor [MATH] in [MATH] ) th...
[EQUATION] exists in [MATH] . Apply Lemma 3.10 to [EQUATION] We deduce that a pre-limit of [MATH] exists. The rest follows from Proposition 1.
3.13 Remark In a cartesian pre-closed category every pair [MATH] and [MATH] of objects posseses a set of morphisms [EQUATION] with the universal property of Definition 2.8 except that neither [MATH] nor [MATH] are required to be unique. Indeed, by Remark
3.4 (b) this is precisely the fact that the functor [MATH] is petty. It is an open question whether (in analogy to the free coproduct completion) the pettiness of the above functors implies that they are small, i.e., that our category is cartesian pre-closed.
3.14 Example Every cartesian multi-closed category is cartesian pre-closed. Indeed, given the multi-universal morphism [MATH] [MATH] of Definition 2.8 , then in [MATH] we have
[EQUATION] This is clear since to give a morphism [MATH] means precisely to give a (unique) [MATH] and a (unique) moprhism [MATH] . Thus, [MATH] is small.