Mechanical properties of V-4Ti-4Cr alloy from molecular dynamics with a neural-network potential

Abstract

Machine-learning interatomic potentials enable large-scale atomistic simulations of vanadium alloys relevant to fusion applications, but reliable training and validation remain challenging in multicomponent systems. Here, we develop a descriptor-based DeepMD-DPA1 potential for the V-Ti-Cr system using a two-stage workflow: broad configuration sampling driven by the MatterSim foundation model followed by fine-tuning to density-functional-theory data computed with VASP. The model achieves root-mean-square errors of 9.2 meV/atom for energies and 0.23 eV/angstrom for force components. Using this potential in large-cell LAMMPS simulations, we compute Young's modulus, bulk modulus, and Poisson's ratio for V-4Ti-xCr and V-xTi-4Cr alloys at T = 300 K and T = 1073 K. We find that increasing Cr fraction increases the elastic moduli of the alloy, while increasing Ti fraction decreases them; all compositions are softer at 1073 K than at 300 K. Two-phase simulations give a melting temperature for pure V of 1950 K, in good agreement with experimental data.