Generalizing Core-Guided Max-SAT

Recent work has shown the value of using unsatisfiable cores to guide maximum satisfiability algorithms (Max-SAT) running on industrial instances [5,9,10,11]. We take this concept and generalize it, applying it to the problem of finding minimal correction sets (MCSes), which themselves are generalizations of Max-SAT solutions. With the technique's success in Max-SAT for industrial instances, our development of a generalized approach is motivated by the industrial applications of MCSes in circuit debugging [12] and as a precursor to computing minimal unsatisfiable subsets (MUSes) in a hardware verification system [1]. Though the application of the technique to finding MCSes is straightforward, its correctness and completeness are riot, and we prove both for our algorithm. Our empirical results show that our modified MCSes algorithm performs better, often by orders of magnitude, than the existing algorithm, even in cases where the use of cores has a negative impact on the performance of Max-SAT algorithms.