Towards a geometry of recursion

Any mathematical theory of algorithms striving to offer a foundation for programming needs to provide a rigorous definition for an abstract algorithm. The works reported by Girard (1988) in [10] and by Moschovakis (1989, 1995) in [29], [30], [31] are among the best examples of such attempts. They both try to offer a mathematically precise and rigorous formulation of an abstract algorithm, intend to keep the algorithmic flavour present and take the notion of recursion as primary and central. The present work is motivated by Girard's GoI 2 paper (Girard (1988) [10], which offers a treatment of recursion in terms of fixed points of linear functions. It is situated in the context of the geometry of interaction (GoI) program and is carried out in the concrete setting of the space of bounded linear maps on a Hilbert space. In this paper, we extend the work in Girard (1988) [10] to the context of traced unique decomposition categories, once again emphasizing the role of abstract trace in the theory of computing. We show that traced unique decomposition categories enriched over partially additive monoids or their variations suffice to axiomatize and hence extend the work in Girard's GoI 2 paper. The theory developed here allows us to formulate an abstract notion of an algorithm as a pair of morphisms in a traced unique decomposition category, an abstract notion of computation as the execution formula (defined using the trace operator) applied to an algorithm, and finally a notion of deadlock-freeness for algorithms. In addition, we can treat recursive definitions, fixed points and fixed point operators in a uniform way in terms of traced unique decomposition categories.