Application of machine-learning for construction of bias potential: a case study of add-atom hyperdynamics and straight screw dislocation migration

Abstract

In this report we describe a bias potential for add-atom global hyperdynamics on the basis of machine-learning (ML) interatomic potential (namely, Moment Tensor Potential, MTP). We compare the results obtained using the ML-bias potential with the ones obtained with conventional bond-boost bias potential. We also discuss possibilities for construction of a ML-bias potential for acceleration a migration of straight screw dislocation.