Monday 28 September 2015, 15:00, Inria, 23 avenue d'Italie, 75013 Paris, room Orange: A framework for defining computational higher-order logics, PhD defense of Ali Assaf.


Link to manuscript:


The main aim of this thesis is to make formal proofs more universal by expressing them in a common logical framework. More specifically, we use the lambda-Pi-calculus modulo rewriting, a lambda calculus equipped with dependent types and term rewriting, as a language for defining logics and expressing proofs in those logics. By representing propositions as types and proofs as programs in this language, we design translations of various systems in a way that is efficient and that preserves their meaning. These translations can then be used for independent proof checking and proof interoperability. In this work, we focus on the translation of logics based on type theory that allow both computation and higher-order quantification as steps of reasoning.

Pure type systems are a well-known example of such computational higher-order systems, and form the basis of many modern proof assistants. We design a translation of functional pure type systems to the lambda-Pi-calculus modulo rewriting based on previous work by Cousineau and Dowek. The translation preserves typing, and in particular it therefore also preserves computation. We show that the translation is adequate by proving that it is conservative with respect to the original systems.

We also adapt the translation to support universe cumulativity, a feature that is present in modern systems such as intuitionistic type theory and the calculus of inductive constructions. We propose to use explicit coercions to handle the implicit subtyping that is present in cumulativity, bridging the gap between pure type systems and type theory with universes à la Tarski. We also show how to preserve the expressivity of the original systems by adding equations to guarantee that types have a unique term representation, thus maintaining the completeness of the translation.

The results of this thesis have been applied in automated proof translation tools. We implemented programs that automatically translate the proofs of HOL, Coq, and Matita, to Dedukti, a type-checker for the lambda-Pi-calculus modulo rewriting. These proofs can then be re-checked and combined together to form new theories in Dedukti, which thus serves as an independent proof checker and a platform for proof interoperability. We tested our tools on a variety of libraries. Experimental results confirm that our translations are correct and that they are efficient compared to the state of the art.