A machine learning potential of Mg-Y alloy and its application on pyramidal dislocation

Alloying magnesium (Mg) with yttrium (Y) has been reported to improve the plasticity of Mg at room temperature by activating cross-slip of screw dislocations. However, the cross-slip of screw dislocations in Mg-Y alloy remains limited understood in atomic simulations, due to the inadequate interatomic potentials. Here, we create a deep learning potential (DLP) suitable to investigate the dislocation behavior in Mg-Y alloy. The newly DLP reproduces basic physical properties, solute-solute interactions, and dislocation core structures in reasonable agreement with density functional theory (DFT) data. Using the DLP, we find that the dominant slip plane in pure Mg is the pyramidal II plane, with cross-slip following an II→I→II pathway. Alloying with a small concentration of Y (<1 at. %) facilitates cross-slip by lowering the energy barrier, which finally enhances the plasticity of Mg. At higher Y concentrations, the dominant slip plane shifts to the pyramidal I plane, and the cross-slip pathway transitions to I→II→I. The energy barrier increases progressively in this concentration which cause the decrease of plasticity. These findings provide the direct atomic scale evidence for the transition of cross-slip behavior, consistent with experimental observations reporting the activity of both pyramidal I and II planes and cross-slip between them.