Incremental Encoding and Solving of Cardinality Constraints

Abstract

Traditional SAT-based MaxSAT solvers encode cardinality constraints directly as part of the CNF and solve the entire optimization problem by a sequence of iterative calls of the underlying SAT solver. The main drawback of such approaches is their dependence on the number of soft clauses: The more soft clauses the MaxSAT instance contains, the larger is the CNF part encoding the cardinality constraints. To counter this drawback, we introduce an innovative encoding of cardinality constraints: Instead of translating the entire and probably bloated constraint network into CNF, a divide-and-conquer approach is used to encode partial constraint networks successively. The resulting subproblems are solved and merged incrementally, reusing not only intermediate local optima, but also additional constraints which are derived from solving the individual subproblems by the back-end SAT solver. Extensive experimental results for the last MaxSAT evaluation benchmark suitew demonstrate that our encoding is in general smaller compared to existing methods using a monolithic encoding of the constraints and converges faster to the global optimum.

This work was partly supported by the German Research Council (DFG) as part of the Transregional Collaborative Research Center “Automatic Verification and Analysis of Complex Systems” (SFB/TR 14 AVACS).