Integrating CNF and BDD based SAT solvers

Abstract

This paper presents an integrated infrastructure of CNF and BDD based tools to solve the Boolean Satisfiability problem. We use both CNF and BDDs not only as a means of representation, but also to efficiently analyze, prune and guide the search. We describe a method to successfully re-orient the decision making strategies of contemporary CNF tools in a manner that enables an efficient integration with BDDs. Keeping in mind that BDDs suffer from memory explosion problems, we describe learning-based search space pruning techniques that augment the already employed conflict analysis procedures of CNF tools. Our infrastructure is targeted towards solving those hard-to-solve instances where contemporary CNF tools invest significant search times. Experiments conducted over a wide range of benchmarks demonstrate the promise of our approach.