Machine-Learning Driven Molecular Dynamics Simulations of Hydrogen Binding and Diffusion in UO₂ Grain Boundaries

Hydrogen induced corrosion of uranium, which leads to the formation of toxic and pyrophoric UH₃, raises significant safety concerns for long-term storage of nuclear materials. Previous work suggests that hydrogen diffuses through the grain boundaries (GBs) of the passivating oxide layer to initiate hydriding reactions. However, the atomistic mechanisms underlying this phenomenon and the structural factors that control its initiation are not well understood. To address this knowledge gap, we use a high-throughput Density Function Theory (DFT) workflow to develop a machine-learned interatomic potential (MLIAP) based on the ChIMES formalism, which leverages linear combinations of many-body Chebyshev polynomials. We then investigate the adsorption and diffusion of H and H₂ within three different UO₂ coincidence site lattice (CSL) GBs: Σ3, Σ5, and Σ9. Our simulations cover temperatures of 100 – 1000 K and durations of ~10 ns, which are enabled only through use of our ChIMES MLIAP. Our results indicate that these GBs can act as hydrogen trapping sites, where diffusion is relatively slow compared to within the crystalline grain itself. We then use our computed diffusion rates as direct inputs for kinetic Monte Carlo simulations that can access millisecond timescales and help place bounds on hydriding initiation times. Our efforts thus provide fundamental atomistic insights that could guide the development of future corrosion mitigation strategies for the storage of nuclear materials.