Simulated annealing with adaptive cooling rates

Simulated annealing is known to be among the most robust optimization methods and many variations of it have attempted to improve the original algorithm. Here we present the molecular dynamics-based adaptive-cooling simulated annealing (ACSA) method, where the cooling rates are adaptively adjusted during the optimization based on the instantaneous values of the heat capacity. At the temperatures where more states are thermally accessible and the heat capacity is above a certain threshold, the algorithm chooses a slow cooling rate to ensure that the system is well optimized and doesn’t get trapped in one of the undesirable minima, and a faster cooling rate is chosen at all other times to decrease the optimization time. The efficiency of the ACSA algorithm is calculated with respect to the original simulated annealing algorithm, when applied to the Lennard-Jones cluster optimization. The ACSA is about twice as efficient as the original algorithm for the Lennard-Jones clusters with 25-40 atoms and its performance improves for lager clusters.