Embedding Deduction Modulo into a Prover

Deduction modulo consists in presenting a theory through rewrite rules to support automatic and interactive proof search. It induces proof search methods based on narrowing, such as the polarized resolution modulo. We show how to combine this method with more traditional ordering restrictions. Interestingly, no compatibility between the rewriting and the ordering is requested to ensure completeness. We also show that some simplification rules, such as strict subsumption eliminations and demodulations, preserve completeness. For this purpose, we use a new framework based on a proof ordering. These results show that polarized resolution modulo can be integrated into existing provers, where these restrictions and simplifications are present. We also discuss how this integration can actually be done by diverting the main algorithm of state-of-the-art provers. Whatever their applications, proofs are rarely searched for without context: mathematical proofs rely on set theory, or Euclidean geometry, or arithmetic, etc.; proofs of program correctness are done using e.g. pointer arithmetic and/or theories defining data structures (chained lists, trees, . . . ); concerning security, theories are used for instance to model properties of encryption algorithms. It is therefore essential to have theoretical foundations and practical methods that handle theories conveniently and efficiently. For this purpose, there are two directions: to develop methods that are really specific to a particular theory; or to develop a generic framework that can handle all theories. The first option is appealing for efficiency reasons: for instance, combining a SAT solver with the Simplex method leads to very powerful SMT solvers for linear arithmetic. However, developing methods for new theories is hard. Even the combination of such specific methods is not trivial, although there have been a lot of interesting results in that direction in the recent years. In this paper, we are more interested in the second option: having a generic way to handle theories efficiently. A naive way to do so would be to use an axiomatization of the theory, but in general, this approach would be really inefficient for automated proving. Somehow, we need to present the theory so as to take advantage of its properties. A first idea is to use the consistency of the theory. When proving a goal in a consistent theory by refutation, resolving the clauses of the theory is useless, since it will not bring out a contradiction. This idea defines the set-of-support