機械学習ポテンシャルを用いたBCC鉄へき開の大規模原子シミュレーション

抄録

It is well known that body-centered cubic (bcc) transition metals such as α-Fe become brittle below the ductile to brittle transition temperature (DBTT). Although the fracture occurs on a macroscopic scale, it consists of a series of interatomic bond breaking events; therefore, accurate atomistic modeling is critical to understanding this phenomenon. In this work, atomistic simulations of curved crack fronts of α-Fe were performed using an interatomic potential generated by a machine learning technique. For this purpose, large simulation boxes consisting of ∼26 million atoms were used. We obtained evidence that the cleavage plane is {100}, and the cleavages are accompanied by crack tip plasticity at elevated temperature.