An analytically derived cooling schedule for simulated annealing (Q2460137)

The authors consider two global optimization problems, one with a continuous domain and the other with a discrete domain. They solve both problems by an adaptive search algorithm. The adaptive search algorithm is designed to model an idealized version of simulated annealing by assuming that points can be sampled exactly according to a sequence of the Boltzmann distributions. The authors analytically derive a cooling schedule, i.e. the function generating the sequence of temperature parameters which is necessary for an adaptive search. Their choice of cooling schedule ensures linearity in the expected number of sample points needed to be arbitrary close to a global optimum. The authors give a computational study comparing their adaptive cooling schedule with several other cooling schedules on continuous and discrete test problems from the literature.