Molecular Dynamics Simulation of Titanium Hydride Stability using Machine-Learned Potentials

Abstract

Studies into of titanium hydrides are important for understanding hydrogen embrittlement processes in titanium alloys and developing promising metallic hydrogen storage systems. The Ti-H system has a phase diagram characterized by a phase transformation from a FCC to a tetragonal FCT crystal lattice. This transition was studied using machine learning of the interatomic potential based on the Moment Tensor Potential (MTP) model. To construct a potential able to accurately reproduce the energies of the system under tetragonal deformation, we created a training dataset containing both periodic structures and distorted configurations obtained at finite temperatures. To obtain a potential able to reproduce energy characteristics under tetragonal lattice distortion, we used an active learning technique, which controlled the degree of potential extrapolation compared to the training dataset and sequentially retrained the MTP. Molecular dynamics simulations have revealed that the FCT-FCC transformation temperature for TiH2 dihydride is 487 K, which is higher than experimentally observed values. We also found that at 300 K, a tetragonal lattice of TiHx hybrids is formed at a hydrogen content greater than 1,86.