Centrality Heuristics for Exact Model Counting

Model counting is the archetypical #P-complete problem consisting of determining the number of satisfying truth assignments of a given propositional formula. In this short paper, we empirically investigate the potential of employing graph centrality measures as a basis of search heuristics in the context of exact model counting. In particular, we integrate centrality-based heuristics into the search-based exact model counter sharpSAT. Our experiments show that employing centrality information significantly improves the empirical performance of sharpSAT, and also allows for simplifying the search heuristics compared to the current default heuristics of the model counter. In particular, we show that the VSIDS heuristic, which is an integral search heuristic employed in essentially all state-of-the-art conflict-driven clause learning Boolean satisfiability solvers, appears to be of very limited use in the context of model counting.