Fast Blocked Clause Decomposition with High Quality

Abstract

Any CNF formula can be decomposed two blocked subsets such that both can be solved by BCE (Blocked Clause Elimination). To make the decomposition more useful, one hopes to have the decomposition as unbalanced as possible. It is often time consuming to achieve this goal. So far there have been several decomposition and post-processing algorithms such as PureDecompose, QuickDecompose, EagerMover etc. We found that these existing algorithms are often either inefficient or low-quality decomposition. This paper aims at improving the decomposition quality, while keeping the runtime of algorithms under control. To achieve this goal, we improve the existing BCE, and present two new variants of PureDecompose, a new heuristic decomposition called LessInterfereDecompose, and a new post-processing algorithm called RsetGuidedDecompose. Combining these new techniques results in a new algorithm called MixDecompose. In our experiments, there is no application formula where the quality of PureDecompose+EagerMover is better than MixDecompose. In terms of speed, MixDecompose is also very fast. Our average runtime is a little longer, but the worst-case runtime is shorter. In theory, our two variants of PureDecompose requires linear time in the number of clauses. By limiting the size of the touch list used by BCE, we can guarantee always that MixDecompose runs in linear time.