A high accuracy machine-learning potential model for Mo-Re binary alloy

Molybdenum is a promising candidate material for advanced nuclear reactors. However, its application in nuclear energy facilities is limited by its intrinsic brittleness, a common characteristic of body-centered cubic transition metals, which often exhibit poor plasticity and workability. The addition of Re to Mo can exploit the “Re softening effect” to enhance plasticity. To better understand the physical origin of this effect and explore the nanoscale atomistic mechanisms in Mo-Re alloys under service conditions, atomic-scale simulation methods, such as molecular dynamics (MD), are widely used as a complementary theoretical tool to experimental studies. However, the reliability of MD simulations is constrained by the limitations of existing empirical interatomic potentials. To address this challenge, this study employs state-of-the-art deep-potential methods to develop a machine learning-based interatomic potential for Mo-Re alloys. This advanced potential model achieves first-principles accuracy across a wide range of material properties, including elastic constants, surface energies, point defects, dislocations, and melting points, within a single potential. It enables high-accuracy atomic-scale simulations and investigations into the microstructural evolution of Mo-Re alloys under complex multi-field coupling conditions (irradiation, heat, and stress), which will establish the theoretical foundation for understanding the Re softening effect.