arXiv subjects
Publications and source records attributed to Marc Lasson.
We present a refinement of the Calculus of Inductive Constructions in which one can easily define a notion of relational parametricity. It provides a new way to automate proofs in an interactive theorem prover like Coq.
Reynold's abstraction theorem is now a well-established result for a large class of type systems. We propose here a definition of relational parametricity and a proof of the abstraction theorem in the Calculus of Inductive Constructions (CIC), the underlying formal language of Coq, in which parametricity relations' codomain is the impredicative sort of propositions. To proceed, we need to refine this calculus by splitting the sort hierarchy to separate informative terms from non-informative terms. This refinement is very close to CIC, but with the property that typing judgments can distinguish informative terms. Among many applications, this natural encoding of parametricity inside CIC serves both theoretical purposes (proving the independence of propositions with respect to the logical system) as well as practical aspirations (proving properties of finite algebraic structures). We finally discuss how we can simply build, on top of our calculus, a new reflexive Coq tactic that constructs proof terms by parametricity.
Let P be any pure type system, we are going to show how we can extend P into a PTS P' which will be used as a proof system whose formulas express properties about sets of terms of P. We will show that P' is strongly normalizable if and only if P is. Given a term t in P and a formula F in P', P' is expressive enough to construct a formula "t ||- F" that is interpreted as "t is a realizer of F". We then prove the following adequacy theorem: if F is provable then by projecting its proof back to a term t in P we obtain a proof that "t ||- F".
We present an adaptation, based on program extraction in elementary linear logic, of Krivine & Leivant's system FA_2. This system allows to write higher-order equations in order to specify the computational content of extracted programs. The user can then prove a generic formula, using these equations as axioms, whose proof can be extracted into programs that normalize in elementary time and satisfy the specifications. Finally, we show that every elementary recursive functions can be implemented in this system.