Abstract:

BCC transition metals (TMs) have a variety of complex plastic deformation and fracture behaviours under different temperature and strain-rate conditions, which is found to be mainly governed by individual lattice defects behaviours like the kinetic movement of <111>/2 dislocations. Traditional empirical and semi-empirical force-field potentials generally have limitations on precise descriptions of those defects. This leads to difficulties in understanding those critical phenomena through large-scale molecular dynamic (MD) modelling. Recent machine learning (ML) potentials of molybdenum, trained by first-principle based datasets, are shown to be promising in describing most dislocation core properties accurately. However, some structures of mixed dislocation and Peierls barrier in BCC Mo still not be adequately reproduced. Therefore, in this presentation, a new deep learning interatomic potential of molybdenum developed via a hybrid descriptor in the ML Deep Potential framework (DP-HYB-Mo) will be introduced together with a benchmark discussion of its performance with respect to basic properties and defects behaviours.