A Curve Swarm Algorithm for Global Search of State Transition Paths

Abstract

Atomistic scale simulation can predict state transition processes such as adsorption, diffusion, and reaction. The challenge of the accurate prediction is to obtain a global view of many local minima and saddle points on the potential energy surface (PES) of a material system. The transition conditions are determined by the saddle points on the PES with the minimum energy barrier between local minima. In this paper, a new algorithm is developed to exhaustively search local minima and saddle points within a region of PES in order to provide a global view of the energy landscape. Unlike the existing saddle point search methods, the algorithm represents a transition path by a parametric Bezier curve with control points. It uses multiple groups of such curves, each of which represents a multistage transition path. During the searching process, each group of curves communicates with others to maintain cohesion and avoid collision based on a collective potential model. The algorithm is integrated with density functional theory calculation and demonstrated by diffusion of hydrogen atoms in the FeTiH system.