EUROPEAN CONGRESS AND EXHIBITION ON ADVANCED MATERIALS AND PROCESSES - EUROMAT 2021

Studying fundamental mechanisms of deformation in metals and alloys requires dependable interatomic potentials because dislocations and cracks are above scales accessible with first-principles calculations. While classical potential forms like the modified embedded atom method (MEAM) have been successfully employed in many cases, non-fcc metals and almost all alloys are not modeled sufficiently quantitative. As a first step towards Mg-alloys we present a broadly applicable machine learned potential for pure Magnesium in the Behler-Parrinello neural-network framework trained on first-principles density-functional theory. We show that the potential predicts dislocation and crack structures very well and subsequently apply it to cross-slip of prismatic screw dislocations, which is not accessible to first-principles approaches. Prismatic slip is achieved by double-cross-slip of stable basal dislocations in steps of c/2 driven by a shear stress on the prismatic plane. The geometry of the observed process compares very well with the process deduced from experiments, the enthalpy barrier not. This mimics the stress-driven double-kink nucleation in bcc elements: the geometry of the mechanism is predicted well, but stress and activation barriers are overestimated by first-principles predictions.