A Multilevel k–way Partitioning Algorithm for Finite Element Meshes using Competing Ant Colonies

The self–organizing properties of ant colonies are employed to tackle the classical combinatorial optimization problem of graph partitioning. Structural information from the graph is mapped onto an environment upon which a number of colonies compete for resources. Using Genetic Programming, a Foraging Strategy is evolved which when executed by the ants in each colony leads to a restructuring of the global environment corresponding to a good partition. Multiple colonies allows for simultaneous k–way partitioning which can provide better partitions than current algorithms which are based on recursive bisection.