Adaptation in stochastic tunneling global optimization of complex potential energy landscapes

Global optimization remains one of the great challenges in scientific computing. One particular successful approach is the usage of tunneling functions to cross barriers and transition states more easily thus allowing for a fast scan of the potential energy surface under investigation. In this paper we develop for the first time a performance measurement procedure for stochastic tunneling approaches and derive an adaptive algorithm that is steered by this performance measure. The proposed algorithm is based on a scale-free measure and thus applicable to general stochastic optimization schemes. We found for a very hard optimization problem the computational effort to be some order of magnitude lower while at the same time increasing the accuracy by a factor of three.