Maintaining Arc-Consistency in Stochastic Local Search

Combining advantages of different paradigms such as stochastic local search and constraint propagation is a research area of current interest, but implementing such hybrids is a non-trivial task. This paper describes a straightforward combination of existing search and modeling techniques, yielding a local search algorithm that maintains arc-consistency for binary CSPs. On a set of bandwidth graph colouring benchmarks it finds colourings better than those found by an equivalent forward checking algorithm, and almost as good as those found by a specialized colouring algorithm. It is embedded within a more general local search algorithm for integer linear programs, allowing further novel hybrids.