Direct global optimization of Onsager-Machlup action using Action-CSA

Abstract

A computational approach to find multiple transition paths via the direct global optimization of Onsager-Machlup (OM) action using the recently proposed Action-CSA approach is presented. The Action-CSA approach performs a global search on a trajectory space based on the conformational space annealing (CSA) algorithm. The approach uses analytic minimization routines by using the analytic gradients of the symmetric second-order temporal discretization of the OM action. For benchmark purposes, the lowest OM action transition paths of alanine dipeptide are compared with those from Langevin dynamics simulations. The new approach finds multiple low OM action transition paths, and they are in reasonable agreement with the LD simulation results. However, it is also identified that the low OM action paths obtained with Action-CSA significantly suffer from discretization error, which may result in unphysical transition paths. This suggests that finding transition paths via the global optimization of OM action with temporal discretization should not be performed, and more sophisticated numerical methods or alternative action integrals are necessary to obtain smooth and accurate paths.