Accurately predicting the crack behavior of magnesium-lithium (Mg-Li) alloys through atomic-scale simulations is essential for understanding crack-defect interactions and designing high-plasticity alloys. However, existing molecular dynamics potentials for Mg-Li alloys are inadequate for reliably modeling crack-related phenomena, such as stacking fault energies and dislocation core structures. In this work, by combining 1722 density functional theory (DFT) calculated data points with machine learning method, we develop a broadly applicable machine learning potential (MLP) for Mg-Li alloys. This MLP accurately reproduces lattice constants, elastic constants, surface energies, and generalized stacking fault energy curves, showing reasonable agreement with DFT data. Furthermore, MLP simulations reveal that either the continuous formation of basal ${\mathrm{I}}_2$ stacking faults or the formation of basal ${\mathrm{I}}_1$ stacking faults trigged by the cross-slip of pyramidal dislocation effectively relieves stress concentration at the crack tip, resulting in crack blunting. These findings help clarify the experimental debate regarding the dual role of basal slip: basal slip bands can also delay crack propagation, while discontinuous basal slip can initiate fatigue cracks.

Close

PrevNext

View figure in article

• Ductility
• Fracture
• Interatomic & molecular potentials
• Stacking faults
• Density functional theory
• Machine learning
• Molecular dynamics