Heterogeneous dynamics in yttrium dihydride with deep learning potential

Our recent study based on ab initio molecular dynamics (AIMD) simulations [Y. Huang et al., Sci. Rep. 15, 18144 (2025)] reveals that hydrogen diffusion in YH2 at elevated temperatures proceeds through a correlated mechanism. In the current work, we develop a deep-learning interatomic potential for YH2 trained on AIMD data, enabling large-scale classical molecular dynamics simulations to elucidate the microscopic details of this cooperative mechanism. Our simulations reproduce and extend the AIMD predictions, capturing an order-disorder transition at a characteristic temperature of Tα ∼800 K accompanied by quasi-one-dimensional, string-like hydrogen displacements. A second crossover at Tλ ∼ 950 K denotes the superionic-like transition, which coincides with a discernible peak in the specific heat capacity. We show that above Tα, the hydrogen atoms, which are constrained within the yttrium lattice, exhibit spatially heterogeneous dynamics, manifested as distinct regions of slow- and fast-moving atoms. Further analysis establishes a buildup of dynamic heterogeneity that peaks above Tα and gradually diminishes beyond Tλ. The progression of dynamic heterogeneity in YH2 shows a remarkable similarity to that of fluorites with a close-packed structure and establishes YH2 as a Type II superionic-like conductor at high temperatures. Our work opens new avenues for exploring superionic-like transport in metal hydrides, a class of materials in which such behavior has remained largely unexplored.