On Calculations of Transition States in Rare Event

In  the study of rare event,  transition state is one of central concepts and  in mathematics, it can regarded as a saddle point (SP) search problem.

This talk will review the works  in my group about the methods of finding these saddle points, including the gentlest ascent dynamics and the iterative minimization algorithms.

The main focus is then on the new development of the methods for computationally intensive function such as the  free energy surface by adopting the recent development of machine learning techniques.

This method combines a statistical method, Gaussian process regression (GPR) and the SP dynamics,  Gentlest Ascent Dynamics. We sequentially detect  the SP by GAD applied to the surrogate model and update the surrogate GPR by an active learning strategy based on the design of experiment so  that the uncertainty of the trained surrogate model is minimized and the efficiency gets a significant improvement.  This is joint work with Dr Hongqiao Wang and Dr Shuting Gu.