There's a whole genre of stochastic optimization algorithms in the same family - parallel tempering in particular has some nice theoretical properties (it stric

There's a whole genre of stochastic optimization algorithms in the same family - parallel tempering in particular has some nice theoretical properties (it strictly dominates simulated annealing, assuming you have computational power to run all the replicas, and has a light guarantee that "eventually" it will converge), and there are adaptive methods to optimize the step sizes, number of replicas, and temperature ranges.