CHAMP: A multipass algorithm for Max Sat based on saver variables

In this paper, we introduce the concept of saver variables in Max Sat and demonstrate their contribution to the performance of solvers for this problem. We present two types of saver variables: high-rank savers and consensual savers. We show how to incorporate them in various ways into an iterated algorithm, CHAMP, for Max Sat. We conduct an extensive empirical evaluation on two collections of instances — instances from a past Max Sat competition and random instances. It turns out that, by using savers, the number of unsatisfied clauses may be reduced by more than 70% in some families. Moreover, a refined version CHAMP＋ of CHAMP improves the results even further. We show that by combining CHAMP＋ with CCLS, a state-of-the-art solver, we obtain better solutions for many Max Sat instances.