Artificial neural network interatomic potential for dislocation and fracture properties of Molybdenum

Abstract

A high dimensional artificial neural network interatomic potential for Mo is developed. To train and validate the potential density functional theory calculations on structures and properties that correlate to fracture, such as elastic constants, surface energies, generalized stacking fault energies, and surface decohesion energies, have been employed. The potential provides total energies with a root mean square error less than 5\;meV per atom both in the training and validation data sets. The potential was applied to investigate screw dislocation core properties as well as to conduct large scale fracture simulations. These calculations revealed that the 1/2$\langle111\rangle$ screw dislocation core is non-degenerate and symmetric and mode I fracture is brittle. It is anticipated that the thus constructed potential is well suited to be applied in large scale atomistic calculations of plasticity and fracture.