General simulated annealing

Simulated annealing is a new kind of random search methods developed in recent years. It can also be considered as an extension to the classical hill-climbing method in AI—probabilistic hill climbing. One of its most important features is its global convergence. The convergence of simulated annealing algorithm is determined by state generating probability, state accepting probability, and temperature decreasing rate. This paper gives a generalized simulated annealing algorithm with dynamic generating and accepting probabilities. The paper also shows that the generating and accepting probabilities can adopt many different kinds of distributions while the global convergence is guaranteed.