Unraveling phase-dependent hydrogen diffusion in MgH₂ using a deep potential framework

Hydrogen diffusion plays an important role in the sorption kinetics of magnesium hydrides. In this work, a machine-learning-based Deep Potential (DP) model for the Mg-H system was developed using the DP-GEN active learning framework and density functional theory (DFT) calculations. The accuracy of the DP model was validated through comparisons of energies, forces, phonon dispersions, energy-volume relationships, and elastic properties, showing excellent agreement with DFT results. Using the developed DP potential, large-scale molecular dynamics simulations were performed to investigate hydrogen diffusion in tetragonal MgH 2 (mp-23710) and orthorhombic MgH 2 (mp-23712). The results show that hydrogen diffusivity increases with temperature in both phases, while orthorhombic MgH 2 exhibits significantly higher diffusion coefficients than the tetragonal phase. CI-NEB calculations reveal that the enhanced diffusion originates from lower hydrogen migration barriers in the orthorhombic structure. For both polymorphs, vacancy-assisted diffusion is energetically more favorable than interstitial diffusion, and the introduction of hydrogen vacancies markedly promotes hydrogen transport. Radial distribution function analysis further indicates that the weaker local ordering and more flexible lattice environment of orthorhombic MgH 2 facilitate hydrogen migration. This work provides an accurate DP potential for Mg-H systems and offers atomistic insights into the phase-dependent hydrogen diffusion behavior of MgH 2 .