Recall that a set of objects in category is said to be a sieve if the implication
is true for every arrow . Dually, a set of objects is said to be a cosieve if the implication
is true for every arrow . Notice that a subset is a sieve iff its complement is a cosieve. We shall often identify a sieve with the full subcategory of spanned by the objects in . If is a sieve (resp. cosieve) then there exists a unique functor such that (resp. ). This defines a bijection between the set of sieves (resp. cosieves) in and the set of functors .
We shall say that an object of the category is a barrel. The bottom of a barrel is the sieve and its top is the cosieve .
The opposite of a barrel is a barrel . Notice that and that . The opposition functor is an automorphism of the category .
The category is cartesian closed.
We have to show that every barrel is exponentiable. For this, it suffices to verify that the functor is a Conduché fibration?. But the Conduché condition? is trivially satisfied, since the arrow in has no nontrivial factorisation. Hence the category is cartesian closed.
The internal hom between two barrels and can be described as follows. We have and . Thus, an object of the category is either a functor or a functor . The morphisms between two functors in is a natural transformation, and similarly for a morphism between two functors in . If and , then morphism in the category is a barrel map which is extending the functor . We leave to the reader the description of the composition law between the different kind of morphisms.
Consider the functor
which associates to a barrel the pair of categories . The functor has a left adjoint and a right adjoint. The left adjoint associates to a pair of categories the category equipped with the canonical functor . The right adjoint associates to a pair of categories their join? equipped with the canonical functor .
Recall that a distributor between two categories and is defined to be a functor We shall regard the set as functor of two variables, contravariant in and covariant in . For example, the functor is defining the unit distributor . The distributors form a category
The opposite of a distributor is the distributor obtained by putting . We shall denote by the element corresponding to an alement , so that
To every functor we can associate two distributors and by putting
for every object and every object . Notice that . An adjunction between two functors and is exactly an isomorphism of distributors .
it is often convenient to represent an element of a distributor by a dotted arrow or a wavy arrow and more simply by a plain arrow if the context is clear. If and , we shall write the element as a composite , Dually, if , we shall write the element as a composite ,
To every distributor we can associate a category called the collage of constructed as follows: and for , we put
The composition of arrows in is defined as above. The associativity of composition is equivalent to the functorialty of .
There is a unique functor such that and . This shows that the collage category carries the structure of a barrel. Notice the isomorphism . We would like to say that collage operation is a functor
but we have not yet defined the category of distributors . A map between two distributors and is defined to be triple , where and are two functors and is a natural transformation,
Maps of distributors can be composed and this defines the category . It is obvious from this construction that the functor
which associates to a distributor the pair of categories is a Grothendieck fibration. A map of distributors induces a functor between the collage categories,
This defines the collage functor
The collage functor is an equivalence of categories.
Let us define the inverse functor in the other direction
It associates to a barrel the distributor defined by putting for every pair of objects . It is easy to see that there is a natural isomorphism for every barrel and a natural isomorphism for every distributor .
The first of the following two triangles commutes strictly, whilst the second commutes only up to a natural isomorphism.
We shall say that a distributor is locally representable if the functor is representable for every object and say that is is representable if it is isomorphic to a distributor for a functor . Dually, we shall say that is locally corepresentable if the functor is representable for every object and say that is corepresentable if it is isomorphic to a distributor for a functor .
A distributor is representable iff it is locally representable iff the full subcategory is reflexive. Dually, is corepresentable iff it is locally corepresentable iff the full subcategory is coreflexive.
(12) Let us suppose that is represented by a functor together with a natural isomorphism . Then the map is a natural isomorphism
for every object and this shows that is locally representable. (23) If is locally representable, then for each object , there exists an object together with an element which represents the functor . By definition, for every object and every element , there exists a unique morphism such that . This means that the morphism of the category is reflecting the object into the subcategory . Thus, is a reflective subcategory of . (31) If is a reflective subcategory of , let us show that the distributor is representable. For each object , let us choose an object together with a morphism which reflects the object into (we are using the axiom of choice here). Then for every morphism in , there exists a unique morphism such that , Let us put . This defines a functor which represents the distributor .
The proof that a locally representable distributor is representable depends on the axiom of choice. A locally representable distributor is called an anafunctor by Makkai here.
The functor which associates to a functor the distributor induces an equivalence between the category and the full subcategory of spanned by the representable distributors.
A functor has a right adjoint iff the distributor is corepresentable, and it has a left adjoint iff the distributor is representable.
The distributors also form a bicategory? whose objects are the small categories. The composite of a distributor with a distributor is the distributor defined by putting
where the tensor product? between the contravariant functor and the covariant functor is used.
The composition law
is coherently associative and the distributor is a unit for this composition.
This follows from the properties of the tensor product for a right -“module” and a left -“modules”. By definition, the tensor product is a quotient of the matrix-product
and the canonical map
is universal among the maps satisfying the compatibility condition for . If is a -“bimodule” and is a right -“module”, then the associativity isomorphism
can be obtained by showing that the two sides are actually isomorphic to the triple-tensor product . By definition, the triple-tensor product of is a quotient of the triple matrix-product , and the canonical map
is universal among the maps satisfying the “trilinearity” conditions and . The unit isomorphism is obtained by showing that the right action is a universal “bilinear” map.
The composition functor is divisible? on both sides. Hence the functor has a right adjoint for every distributor . By construction, we have
where the hom set of maps in the category is used. Dually the functor has a right adjoint for every distributor . By construction, we have
where the hom set of maps in the category is used.
Let us denote by the category whose objects are the cocomplete locally small categories and whose morphisms are the cocontinuous functors. For any small category , we have
For a fixed , the composition functor
is cocontinuous, since it has a right adjoint. This defines a functor ,
Dually, the composition functor
is cocontinuous, since it has a right adjoint. This defines a functor ,
Notice the canonical isomorphisms and .
(Morita-Watts-Lawvere-Benabou) The functors and defined above are equivalence of categories.
To every functor in is associated a pair of adjoint cocontinuous functors
where for a functor , and where is the left Kan extension? along of a functor . The two functors can represented by distributors. For every we have
and this means that we have a canonical isomorphism , where is the distributor defined by putting For every we have
and this means that we have a canonical isomorphism , where is the distributor obtained by putting From the adjunction , we obtain an adjunction between distributors (beware that ). The unit of this adjunction is a map in and the counit is a map in . The counit is the obvious map
defined by composing the pairs of arrows . The composite is the distributor
The unit is the map induced by the functor .
The bicategory of distributors is symmetric monoidal. The tensor product of a distributor with a distributor is the distributor defined by putting
The tensor product functor
is really a cartesian product in the category . More precisely, we have a canonical isomorphism
in the category of cylinders , where is the collage barrel of a distributor .
If , and are small categories, then the categories and are equivalent, since they are both isomorphic to the the category
The equivalence
is actually natural when and are varying in the bicategory of distributors. In other words, the endo-functor of the bicategory is left adjoint to the endo-functor . It follows that the objects and are mutually dual in the symmetric monoidal bicategory . Hence the bicategory is compact closed. Let us examine this duality explicitly. If and , then is a distributor . If and , then is a distributor The distributor is given by the functor and the distributor by the same functor . Notice that is a covariant functor of , whilst is a contravariant functor of . Notice also that .
A monoidal bicategory is a tri-category. In this context, the adjunction identities are taking the form of a pair of isomorphisms between distributors,
The domain of is the composite of the distributors, Thus, is a coend,
Notice that the product in the integrant is covariant in , whilst the product is contravariant. The isomorphism is then induced by the map
which takes a quadruple to their composite in , Dually,the domain of is defined to be the composite This domain is isomorphic to , since and . The isomorphism is induced by the isomorphism .
In a compact monoidal category, every map has a transpose . In the monoidal bicategory of distributors, show that the transpose of a distributor is its opposite .
Historical notes.
The notion of distributor was first introduced by Lawvere in a talk that he gave in 1966 at a meeting in Oberwolfach. They were used to represent cocontinuous functors between presheaf categories. The theory of distributors was later developed extensively by Bénabou who introduced also the notion of bicategory. The collage category is also due to him, but the terminology “collage category” was introduced by Street in his Rendiconti paper.
In a letter adressed to me (April 11,2010), Anders Kock has expressed his view on the invention of distributors. I find his opinion worth to be made public (with his permission):
The notion of profunctor/bimodule was not a creation of 1966, but is a result of an evolution, which includes for instance the section (p.22-23) in Cartan-Eilenberg on (the bicategory of) bimodules, and their tensor product (= composition in the bicategory), and the tensor product of a covariant and a contravariant functor on a category, cf. Watts’ contribution in the LaJolla volume. The important thing about the evolution of these notions was that they led to the notion of bicategory - likewise an evolution having many stages and inputs, including 2-categories, and monoidal categories. In this evolution, Benabou is a main actor. The identification of profunctors with cocontinuous functors between categories of presheaves is likewise part of an evolution, where ancestors are Morita’s characterization of equivalences between module categories, and Watts’ characterization of cocontinuous functors between module categories (Proc.Amer.Math.Soc, 1960).
I can agree with that. Most mathematical ideas are the result of an evolution. Mutations are playing an important role in biological evolution. What about mutations of mathematical ideas? What could be the role of mutations in the evolution of mathematical ideas?
The notion of symmetric monoidal bicategory has a complicated history. Let me quote Street:
An important point here was the realization that there are really three kinds of commutativity for monoidal bicategories and then it was important to find the correct definition of braided monoidal bicategory. Kapranov and Voevodsky made a definition of braiding on (I think) a slightly restricted notion of monoidal bicategory: and they left out an axiom. The missing axiom was corrected by Larry Breen and Baez-Neuchl. In my paper with Day we made up the term “syllepsis” for the extra notion of commutativity between “braiding” and “symmetry”.
Let me quote Breen:
Actually, I did not “correct” the Kapranov -Voevodsky definition, in fact I was aware of the full definition at least as far back as 1988. I wrote about this in a letter to Deligne, dated 4 Feb. 1988, (now on the website). I did not write this up as a paper at the time, but returned to this topic (in a more general topos context) in the last chapter of my Asterisque monograph (vol. 225, 1994), where I discuss these questions and in particular explicitly mention on page 148 the necessary correction to Kapranov-Voevodsky. In my letter the associativities are strict, so that only the commutativity conditions are considered. I had played around with the more complete diagrams involving both associativity and commutativity but I hadn’t at the time worked out the lax definition of the 2-categorical braiding axioms as Kapranov-Voevodsky did in their paper.
Many people have discovered independently that the symmetric monoidal bicategory is compact closed. It was known to the Australian school for several decades. The general case of the symmetric monoidal bicategory of -distributors, for a closed symmetric monoidal category , is treated in the paper of Day and Street referred to below.
Books:
F. Borceux: Handbook of Categorical Algebra in 3 volumes, CUP, (1994)
Papers:
J. Baez, M. Neuchl Braided Monoidal 2-categories, Advances in Math, 121 (1996) 196-244
J. Bénabou: Introduction to bicategories, Springer Lecture Notes in Math 47, 1-17, (1967)
J. Bénabou: 2-Dimensional limits and colimits of distributors, Mathematisches Forschungsinstitut Oberwolfach, Tagungsbericht 30 (1972) 6–7.
J. Bénabou: Les Distributeurs, Univ. Catholique de Louvain, Séminaire de Mathématiques Pures, Rapport 33 (1973)
J. Bénabou: Distributors at work, (2000), pdf
B. Day, R. Street: Monoidal bicategories and Hopf algebroids, Advances in Math. 129 (1997) 99-157; MR99f:18013.
M. Kapranov and V.Voevodsky: 2-categories and Zamolodchikov tetrahedra equations, Proc. Symp. Pure Math. 56 (1994), part 2, p. 177-259.
M. Kapranov and V.Voevodsky: Braided monoidal 2-categories and Manin-Schechtman higher braid groups, J. Pure Appl. Algebra, 92 (1994), 241-267.
R. Street: Cauchy characterization of enriched categories, Rendiconti del Seminario Matematico e Fisico di Milano 51 (1981) 217-233; MR85e:18006. Reprints in Theory and Applications of Categories 4 (2004) 1-16, link
R. Street: Two constructions on lax functors, Cahiers topologie et géométrie différentielle 13 (1972) 217-264; MR50#436; numdam