Dear Jean, Thank you for your reply, but I'm afraid I do not understand your first paragraph. The condition I'm interested in is self-dual, but the notion of locally fully faithful functors is not. For instance, the projection from a coslice category to its base is not locally fully faithful. It is an interesting suggestion though. This seems to be the correct version: * U : C -> D is a functor such that, for every object A in C, the induced functor A/C -> UA/D is locally fully faithful. Thus one might say "colocally locally fully faithful". Expanding the definitions a bit, this reduces to the condition on "categories of factorisations" that Steve Lack mentioned in his reply. If I am not mistaken, locally fully faithful functors are in particular colocally locally fully faithful. Thus, by what you said in the following paragraphs, discrete fibrations (resp. discrete opfibrations) are examples of colocally locally fully faithful functors generalising the projection from a slice (resp. coslice) category to its base. This may be useful for transporting (weakly) orthogonal factorisation systems. Best wishes, -- Zhen Lin On 6 August 2014 04:08, Jean Bénabou <jean.benabou@wanadoo.fr> wrote:
Dear Zhen Lin,
Before I answer your question let me give names to arbitrary maps of C: f: X --> Y, g: Z --> W, x: Z -->X, y: W --> Y and h: W --> X . Your condition reads: for every pair (f,g) and every triple (x,y,h') where: yg = fx and h': UW --> UX satisfies : h' Ug = Ux and Uf h' = Uy, there exits a unique h: W -->X such that: Uh = h', hg = x and fh = y . One verifies first that this condition is satisfied for all pairs (f,g) iff it is satisfied for f arbitrary and g is an identity. In that case that means that f is what I call hypercartesian (which in the anglo-american literature is called cartesian). Since f is arbitrary , the condition becomes: (i) Every map of C is hyper cartesian. Let me call U locally full and faithful (lff) iff for every object X of C the obvious functor C/X --> D/UX is full and fathull. In my mail to Joyal and the catgory list, dated July 28 I already mentioned that (i) is equivalent to (ii) U is lff. I also said, it is obvious, that U full and faithful => U is lff . I mentioned also the case of groupoids, with a sharper result than the one you stated, namely: If C is a groupoid, every functor U: C --> D, where D is arbitrary, is lff.
Let me add a remark which was not in my mail to Joyal, namely, the previous property characterizes groupoids. More precisely we have:
PROPOSITION 1. Let C be a category. The following are equivalent: (i) C is a groupoid (ii) Every functor with domain C is lff (iii) The unique functor C --> 1 is lff.
There are MANY MORE properties of lff functors which would be too long to give here. Let me mention a few which are not in my mail to Joyal. The following theorem generalizes greatly the previous proposition.
THEOREM. Let U: C --> D be a fibration. The following are equivalent: (i) U is lff (ii) All the fibers of U are groupoids. (iii) U is conservative (i.e. reflects isomorphisms).
Such fibrations are very important. Because of (ii) they have sometimes been called groupoid fibrations. In particular, it follows from (iii), that for such a fibration if D is a groupoid so is C.
You said that the functors satisfying your condition are stable by composition. This result can be strengthened since we have:
PROPOSITION 2. Let U and V be functors such that the composite UV is defined. If U is lff, then UV is lff iff V is.
The following result is easy to prove but nevertheless important for many theoretical reasons.
THEOREM 2. lff functors are stable by pull back along any functor.
I could add many significant results, in particular about cartesian functors, or orthogonality but this mail is already a bit long, and I apologize for this length. Thus there is no need to give a name to the property you mentioned, locally full and faithful describes precisely this property.
Best wishes, Jean
[For admin and other information see: http://www.mta.ca/~cat-dist/ ]