Tags:artificial intelligence, computer science, Lemma learning, Proof generation, Satisfiability modulo assignment, Theory combination and theory inference

Abstract:

Search-based satisfiability procedures try to construct a model of the input formula by simultaneously proposing candidate models and deriving new formulae implied by the input. When the formulae are satisfiable, these procedures generate a model as a witness. Dually, it is desirable to have a proof when the formulae are unsatisfiable. Conflict-driven procedures perform nontrivial inferences only when resolving conflicts between the formulae and assignments representing the candidate model. CDSAT (Conflict-Driven SATisfiability) is a method for conflict-driven reasoning in combinations of theories. It combines solvers for individual theories as theory modules within a solver for the union of the theories. In this paper we endow CDSAT with lemma learning and proof generation. For the latter, we present two techniques. The first one produces proof objects in memory: it assumes that all theory modules produce proof objects and it accommodates multiple proof formats. The second technique adapts the LCF approach to proofs from interactive theorem proving to conflict-driven SMT-solving and theory combination, by defining a small kernel of reasoning primitives that guarantees that CDSAT proofs are correct by construction.