A Machine Learning Potential Model for Simulating Dynamic Mechanical Response of Pb-Sn Alloy

Lead is a low-melting-point metal with a complex temperature-pressure phase diagram. Alloying with tin further reduces its melting temperature, making lead-tin alloys an important model material for studying dynamic mechanical responses and failure behavior. However, experimental characterization of atomic-scale dynamic failure mechanisms in PbSn alloys remains challenging due to current technical limitations. Non-equilibrium molecular dynamics (NEMD) simulations can track atom trajectories and reveal key dynamic processes under dynamic loading-unloading. It thus serves as a critical alternative tool. Yet, the reliability of molecular dynamics relies on the accuracy of interatomic potentials, and currently, no high-accuracy potential exists for PbSn alloys under dynamic conditions. In this work, we develop a machine-learning interatomic potential (DP-PbSn) for PbSn alloys using a concurrent learning scheme. This potential achieves first-principles accuracy across a wide thermodynamic range (0–100 GPa, 0–5000 K), reliably predicting fundamental properties (e.g., lattice constants, elastic constants), defect energetics (e.g., surface energy, stacking fault energy, vacancy formation energy), as well as melting curves and shock Hugoniot curves, demonstrating its suitability for dynamic simulations. Leveraging this potential, we conduct preliminary NEMD simulations to investigate the dynamic mechanical responses of pure Pb and PbSn alloys, elucidating the influence of Sn on phase transitions and plastic deformation under dynamic loading. The DP-PbSn serves as a robust theoretical tool for high-accuracy non-equilibrium molecular dynamics, providing essential insights for experimental studies on the dynamic damage behavior of PbSn alloys.