Following Andree Ehresmann's posting, and again in partial reply to Jean Benabou, I would like to add some considerations on the quotient of a category modulo a subcategory. With best regards Marco Grandis ------ 1. Generalised quotients of categories. A very general notion of generalised congruence in a category - also involving objects - can be found in a paper by Bednarczyk, Borzyszkowski and Pawlowski [BBP]. Here we will only consider a particular case, determined by the maps which we want to become identities. More precisely, given a category X and a set A of its arrows, X/A will denote the quotient of X modulo the generalised congruence generated by declaring every arrow in A to be equivalent to the identity of its domain. (It exists, because the generalised congruences of a category form a complete lattice, see [BBP].) The quotient p: X -> X/A is determined by the obvious universal property: - for every functor f: X -> Y which takes all the maps of A to identities, there is a unique functor f': X/A -> Y such that f = f'p. It is interesting to note that p automatically satisfies a 2- dimensional universal property, as one can easily deduce from the fact that natural transformations can be viewed as functors X -> Y^2, with values in the category of morphisms of Y. 2. Kernels and normal quotients of categories. This particular case can be made clearer when viewed at the light of general considerations on kernels and cokernels with respect to an *assigned ideal* of "null" arrows, studied in [Gr] - independently of the existence of a zero object. (For kernels with respect to an ideal, see also Ehresmann [Eh] and Lavendhomme [La].) Take, in Cat, the ideal of *discrete* functors, i.e. those functors which send every map to an identity; or, equivalently, consider as *null* objects in Cat the discrete categories and say that a functor is *null* if it factors through such a category (we have thus a *closed* ideal, according to an obvious Galois connection between set of maps and set of objects, see [Gr]). This ideal produces - by the usual universal properties formulated *with respect to null functors* - a notion of kernels and cokernels in Cat. Precisely, given a functor f: X -> Y, its kernel is the wide subcategory of all morphisms of X which f sends to identities of Y (V(f), in Benabou's notation), while its cokernel is the quotient Y -> Y/B, produced by the set-theoretical "arrow-image" B of f. A normal subcategory X' of X, by definition, is a kernel of some functor starting at X, or, equivalently, the kernel of the cokernel of its embedding. It is necessarily a wide subcategory; but, of course, there are wide subcategories which are not normal. Dually, a normal quotient p: X -> X' is the cokernel of some functor with values in X (or, equivalently, the cokernel of its kernel). A normal quotient is always surjective on objects (as it follows easily using its factorisation through its full image), but - of course - need not be surjective on maps. Now, the normal quotients of X are precisely those we have considered in point 1. Indeed, given a set A of arrows of X, the quotient X -> X/A is necessarily the cokernel of some functor f with values in X (eg, take the free category A' on the graph A and the resulting functor f: A' -> X). The normal quotients of a category X form a *lattice*, anti- isomorphic to the lattice of normal subcategories of X, via kernels and cokernels. (More generally, this holds replacing Cat with any category equipped with a closed ideal, and having kernels and cokernels wrt it; see [Gr].) 3. References [BBP] M.A. Bednarczyk - A.M. Borzyszkowski - W. Pawlowski, Generalized congruences-epimorphisms in Cat, Theory Appl. Categ. 5 (1999), No. 11, 266-280. [Eh] C. Ehresmann, Cohomologie a valeurs dans une categorie dominee, Extraits du Colloque de Topologie, Bruxelles 1964, in: C. Ehresmann, Oeuvres completes et commentees, Partie III-2, 531-590, Amiens 1980. (See also the Comments in the same volume, p. 845-847.) [Gr] M. Grandis, On the categorical foundations of homological and homotopical algebra, Cah. Topol. Geom. Diff. Categ. 33 (1992), 135-175. [La] R. Lavendhomme, Un plongement pleinement fidele de la categorie des groupes, Bull. Soc. Math. Belgique, 17 (1965), 153-185.
[Note from moderator: reference was mentioned before, but is more complete below] Don't know if this got through already as myu mail has been acting up: There is also Drinfeld's *math.KT/0210114* [abs <http://arxiv.org/abs/math.KT/0210114>, ps <http://arxiv.org/ps/math.KT/0210114>, pdf <http://arxiv.org/pdf/math.KT/0210114>, other <http://arxiv.org/format/math.KT/0210114>] : Title: *DG quotients of DG categories* Authors: * Vladimir Drinfeld <http://arxiv.org/find/math/1/au:+Drinfeld_V/0/1/0/all/0/1>* Comments: 50 pages, Latex; some typographical errors corrected, some references added Subj-class: K-Theory and Homology; Algebraic Geometry; Algebraic Topology; Category Theory jim Marco Grandis wrote:
Following Andree Ehresmann's posting, and again in partial reply to Jean Benabou, I would like to add some considerations on the quotient of a category modulo a subcategory.
With best regards
Marco Grandis
Hello, let me add some remarks to Marco Grandis' posting.
1. Generalised quotients of categories.
A very general notion of generalised congruence in a category - also involving objects - can be found in a paper by Bednarczyk, Borzyszkowski and Pawlowski [BBP].
I have not yet looked at that paper, but I think the "natural" thing is to consider equivalence relations R on a category C, which are subcategories of CxC (i.e. closed under composition, not necessary full; identities are in R by relexivity of R). In my diplomarbeit "Kongruenzrelationen auf Kategorien" from 1977, I considered that, but I was not the first one. Some years earlier there was a paper by Jacques Mersch from Liege (Belgium), which unfortunately appeared only in an internal publication of the university of Liege. Moreover, I think I remember that John Isbell did something on this subject.
The quotient p: X -> X/A is determined by the obvious universal property:
The universal property is also obvious in the general situation. For small categories, a functor with this property always exists, let's call it a quotient functor. For large categories it my happen that the hom-sets of the quotient become large, even if the hom-sets of the original category are small. In general, a congruence as above is not a kernel of a functor; the quotient functor may identify more morphisms. In my diplomarbeit, I rediscovered an example, which had already been found by Mersch. The quotient functors are exactly the regular epis in CAT. But unfortunately, they are not closed under compositon, so a quotient of a quotient of C need not be a quotient of C.
It is interesting to note that p automatically satisfies a 2- dimensional universal property, as one can easily deduce from the fact that natural transformations can be viewed as functors X -> Y^2, with values in the category of morphisms of Y.
Of course, this argument also works in the general situation. Greetings Reinhard Boerger
On 18 Jan 2006, at 09:36, V. Schmitt wrote:
Marco I am not very awake this morning but i think that this construction of formally inverted some arrow is well known for long (cf for instance Borceux's handbooks on localizations). Am i wrong? Cheers, Vincent
Categories of fractions are indeed very well-known, but satisfy a different universal property: to make *invertible* the assigned arrows (instead of making them *identities*). But you can view categories of fractions at the light of what I was saying. Take in Cat the (closed) ideal of functors which send every map to an isomorphism, or equivalently of those functors which factor through a groupoid. With respect to this ideal, the kernel of a functor f: X -> Y is the (wide and replete) subcategory of maps which f turns into isomorphisms, while the cokernel is the category of fractions of Y which inverts all arrows reached by f. Best regards Marco G. PS. And - thinking of Jean Pradine's message - yes, of course, quotient of groupoids are important, but have special features of their own; as he is pointing out.
participants (3)
-
jim stasheff -
Marco Grandis -
Reinhard Boerger