Post-analysis-based clustering dramatically improves the Fiduccia-Mattheyses algorithm

This paper describes a new partitioning algorithm, BISECT, which is an extension of the Fiduccia-Mattheyses (FM) algorithm that recursively combines clustering and iterative improvement. A post analysis of sequences of moves in one pass generates disjoint subsets of nodes for clustering. After clustering BISECT is applied again on the compacted circuit. BISECT is stabler, achieves results that can be up to 73 times better than FM, and runs in linear time under suitably mild assumptions. BISECT also performs well in comparison with the Kernighan-Lin algorithm and simulated annealing. The empirical results show that BISECT is stable and is not very sensitive to the initial partition. For many random sparse graphs, BISECT achieves 0-cut bisections (balanced partitions)