Breaking Symmetries in SAT Matrix Models

Abstract

Symmetry occurs naturally in many computational problems. The use of symmetry breaking techniques for solving search problems reduces the search space and therefore is expected to reduce the search time. Recent advances in breaking symmetries in SAT models are mainly focused on the identification of permutable variables via graph automorphism. These symmetries are denoted as instance-dependent, and although shown to be effective for different problem instances, the advantages of their generalised use in SAT are far from clear. Indeed, in many cases symmetry breaking predicates can introduce significant computational overhead, rendering ineffective the use of symmetry breaking. In contrast, in other domains, symmetry breaking is usually achieved by identifying instance-independent symmetries, often with promising experimental results. This paper studies the use of instance-independent symmetry breaking predicates in SAT. A concrete application is considered, and techniques for symmetry breaking in matrix models from CP are used. Our results indicate that instance-independent symmetry breaking predicates for matrix models can be significantly more effective than instance-dependent symmetry breaking predicates.