Demon algorithms and their application to optimization problems

We introduce four new general optimization algorithms based on the 'demon' algorithm from statistical physics and the simulated annealing (SA) optimization method. These algorithms reduce the computation time per trial without significant effect on the quality of solutions found. Any SA annealing schedule or move generation function can be used. The algorithms are tested on traveling salesman problems including Grotschel's 442-city problem (1984) with results comparable to SA. Applications to the Boltzmann machine are considered.