Hi All, Since we have a semantic characterisation of containers, we can look at Thorstens question
Alternatively for containers: Every endofunctor which has an initial algebra is a container. Obviously, this will only hold in the syntax of Type Theory, i.e. the classifying predicative pretopos with W-types.
Surely, it must be the case that all syntactically definable endofunctors are accessible. And, being syntactic constructions, I'd not be too surprised if they preserved connected limits ... Unless I've got the wrong end of the stick which is entirely possible. All the best Neil On 28 Feb 2008, at 15:32, Thorsten Altenkirch wrote:
Hi Sam,
the issue you address has to do with the question of predicative vs. impredicative, I think.
As you say in a topos any monotone operator on the lattice of subsets of a given type has a least (and greatest fixpoint). In a predicative theory like Martin-Loef's Type Theory these fixpoints only exist if the monotone operator is defined in a *strictly* positive way.
There seems to be no simple example of such a fixpoint because all simple examples turn out to be strictly positive. Here is the simplest one I can think of in the moment: Let (A,@) be a partial combinatory algebra, e.g. the natural number with an encoding of Turing machines. Then choose a subset R \subseteq A to define an antitone operator F on subsets of A by
F P = { x:A | all y:A.P y -> R (x @ y) }
Now clearly FF (do F twice) is monotone and hence will have a least fixpoint in a topos. However, by a proof-theoretic analysis (I think) we can show that the fixpoint cannot be constructed in Martin-Loef Type Theory. (It would be nicer to have a semantical argument).
Btw this is just an encoding of Reynold's construction to show that there are no set-theoretic models of System F. In System F you get a fixpoint of (X -> 2) -> 2 which corresponds to choosing R to be an encoding of 2, e.g. R = {\xy.x,\xy.y}.
There is a semantic characterisation of strictly positive operators on subsets which is similar to the concept of containers, I have used this in joint work with Andreas Abel a while ago (our JFP 2002 paper and TYPES 99). It would be nice, if one could show that all operators which have fixpoints are one of those. Alternatively for containers: Every endofunctor which has an initial algebra is a container. Obviously, this will only hold in the syntax of Type Theory, i.e. the classifying predicative pretopos with W-types.
I am not sure this mismatch is a problem: Do you ever need fixpoints of non-strict positive operators?
Cheers, Thorsten
On 27 Feb 2008, at 19:07, Sam Staton wrote:
Various people (including Moerdijk & Palmgren, Hyland & Gambino, Abbott & Altenkirch & Ghani) have considered locally cartesian closed pretoposes with W-types, ie initial algebras for generalized polynomial functors, aka 'containers'. (These W-types are an appropriate analysis of the fixed points of intensional constructions and so these categories have been motivated as "constructive toposes", as models of a particular version of Martin-Lof's type theory).
Lcc pretoposes are Heyting pretoposes. For any object X, consider those monotone endofunctions on the class Sub(X) that can be constructed from the pullback functors and their left and right adjoints. By my calculations, in a topos, these monotone endofunctions have least and greatest fixed points (by internalizing them and then using Tarski's fixed point theorem).
It seems that, in general, these monotone endofunctions will not have least or greatest fixed points in lcc pretoposes with W-types. (I haven't proved this, though, and perhaps I have missed a neat trick.)
If I'm not mistaken, this is a mismatch. Lcc pretoposes let you work both intensionally and extensionally, yet when it comes to W- types, we only consider the intensional constructions. Is there a reason for this?-- are these fixed points "non-constructive" or "impredicative" in some sense?
Has anybody thought about these things before? Have I missed something? All the best, Sam
[My example is bisimilarity for transition systems, as studied in concurrency theory. This is conventionally calculated as a maximum fixed point for a monotone endofunction on the lattice of binary relations. I can define a 'constructive' version of bisimilarity using W-types, and then take the image. This differs from the usual notion, unless choice is assumed.]
This message has been checked for viruses but the contents of an attachment may still contain software viruses, which could damage your computer system: you are advised to perform your own checks. Email communications with the University of Nottingham may be monitored as permitted by UK legislation.