Intuitionistic model constructions and normalization proofs

The traditional notions of strong and weak normalization refer to properties of a binary reduction relation. In this paper we explore an alternative approach to normalization, in which we bypass the reduction relation and instead focus on the normalization function, that is, the function that maps a term to its normal form. We work in an intuitionistic metalanguage, and characterize a normalization function as an algorithm that picks a canonical representative from the equivalence class of convertible terms. This means that we also get a decision algorithm for convertibility. Such a normalization function can be constructed by building an appropriate model and a function quote, which inverts the interpretation function. The normalization function is then obtained by composing the quote function with the interpretation function. We also discuss how to get a simple proof of the property that constructors are one-to-one, which is usually obtained as a corollary of Church–Rosser and normalization in the traditional sense. We illustrate this approach by showing how a glueing model (closely related to the glueing construction used in category theory) gives rise to a normalization algorithm for a combinatory formulation of Godel System T. We then show how the method extends in a straightforward way when we add cartesian products and disjoint unions (full intuitionistic propositional logic under a Curry–Howard interpretation) and transfinite inductive types such as the Brouwer ordinals.