We study leading-edge local search heuristics for balanced hypergraph partitioning and Boolean satisfiability, intending the generalization of such heuristics beyond their original domains. We adapt the Fiduccia Mattheyses (FM) hypergraph partitioning heuristic to Boolean Satisfiability (SAT), and the WalkSAT SAT solver to hypergraph partitioning. Focusing on balanced hypergraph partitioning, we propose a combination of the classical FM heuristic and our "cross-over" heuristic WalkPart, and empirically show that it is more powerful than each component alone. Empirically, we show a 15% improvement in net cut and a 7% improvement in runtime over a leading-edge implementation of the FM heuristic.
Arathi Ramani, Igor L. Markov