This article has two related aims. The first is to study the categorical setting of Abramsky, Haghverdi, & Scott's untyped linear combinatory algebras [6], and the second is to relate this to the much more recent work by Abramsky & Heunen on Frobenius algebras in the infinitary setting [7]. The key to this is (extensional) reflexivity (i.e. the property of an object being isomorphic to its own internal hom[ → ]). We first characterise extensionally reflexive objects in compact closed categories, then consider when & how this property may be 'strictified' -how we may give a monoidally equivalent category where the isomorphisms exhibiting reflexivity are in fact identity arrows. This results in small two-object compact closed categories consisting of a unit object and a single (non-unit) strictly reflexive object. We then move on to studying the endomorphism monoids of such objects from an algebraic rather than logical or categorical viewpoint. We demonstrate that these necessarily contain an interesting inverse monoid that may be thought of as Richard Thompson's iconic group F together with the equally iconic bicyclic monoid B of semigroup theory, with non-trivial interactions between the two derived from the Frobenius algebra identity -and claim this as a particularly significant example of the (unitless) Frobenius algebras of [7]. We first develop the theory from a purely theoretical point of view, then move on to develop concrete examples, based on the algebra and category theory behind [21,22,6]. The concrete examples we give are based on the traced monoidal category of partial injections, and reflexive objects in the compact closed category that results from applying the Int or GoI construction. We then give compact closed categories, monoidally equivalent to compact closed subcategories of Int(pInj), where this reflexivity is exhibited by identity arrows, and show how the above algebraic structures (Thompson's F , the bicyclic monoid, and Frobenius algebras) arise in a fundamental manner.
This work is of course !( ) dedicated to Samson Abramsky.
Historical backgroundThe starting point for this chapter is Girard's Geometry of Interaction programin particular the first two parts. It is by now well-established that compact closed 1 categories model key aspects of this (see [6,25] for a good account), and this observation motivated the name of Abramsky's categorical GoI construction [1] (see Section 7).However, this immediately poses an interesting puzzle. In [21], Girard makes the rather cryptic comment that his system 'forgets types', even though the stated aim was to produce a model of the polymorphically typed System F , rather than a purely untyped system. The explanation seems to be that the GoI system moves from a rigidly typed system to an entirely untyped system, in order to build a more flexible (polymorphic) type system on top of this . The claim that there is an untyped logical system at the core of Girard's GOI was borne out in Abramsky, Haghverdi, and Scott's paper [6] that gave an untyped combinat...