Convergence of a simulated annealing algorithm for continuous global optimization. (Q5928424)

The author considers a simulated annealing algorithm for minimizing a continuous function \(f\) over a convex, compact, full-dimensional set \(X\). It is assumed that \(f\) has only a finite number of global minima over \(X\). The cooling schedule is based on the distance between the function value in the current point and an estimate of the global optimum. Under certain assumptions on the cooling schedule and on the distribution of the next candidate point it is shown that the considered simulated annealing algorithm converges in probability to a global optimum.