Mining definitions in Kissat with Kittens.

Abstract

Bounded variable elimination is one of the most important preprocessing techniques in SAT solving. It benefits from discovering functional dependencies in the form of definitions encoded in the CNF. While the common approach pioneered in SatELite relies on syntactic pattern matching, our new approach uses cores produced by an embedded SAT solver, Kitten. In contrast to a similar semantic technique implemented in Lingeling based on BDD algorithms to generate irredundant CNFs, our new approach is able to generate DRAT proofs. We further discuss design choices for our embedded SAT solver Kitten. Experiments with Kissat show the effectiveness of this approach.