On the invariant measure of non-reversible simulated annealing

Abstract

We give a criterion to ensure convergence of non-reversible simulated annealing algorithms to the set of global minima of the target function U. We show, that such conditions only have to take into account the structure of the local minima of U. Moreover we give an example showing that in general (i.e. without any further condition) non-reversible simulated annealing may converge to suboptimal points.