On Encoding LF in a Predicate Logic over Simply-Typed Lambda Terms

Abstract

Felty and Miller have described what they claim to be a faithful encoding of the dependently typed lambda calculus LF in the logic of hereditary Harrop formulas, a sublogic of an intuitionistic variant of Church's Simple Theory of Types. Their encoding is based roughly on translating object expressions in LF into terms in a simply typed lambda calculus by erasing dependencies in typing and then recapturing the erased dependencies through the use of predicates. Unfortunately, this idea does not quite work. In particular, we provide a counterexample to the claim that the described encoding is faithful. The underlying reason for the falsity of the claim is that the mapping from dependently typed lambda terms to simply typed ones is not one-to-one and hence the inverse transformation is ambiguous. This observation has a broad implication for other related encodings.