Insights Into Radiation Damage in YBa2Cu3O7−δ From Machine Learned Interatomic Potentials

Accurate prediction of radiation damage in YBa 2 Cu 3 O 7 − δ (YBCO) is essential for assessing the performance of high-temperature superconducting (HTS) tapes in compact fusion reactors. Existing empirical interatomic potentials have been used to model radiation damage in stoichiometric YBCO, but fail to describe oxygen-deficient compositions, which are ubiquitous in industrial Rare-Earth Barium Copper Oxide conductors and strongly influence superconducting properties. In this work, we demonstrate that modern machine-learned interatomic potentials enable predictive modelling of radiation damage in YBCO across a broad range of oxygen stoichiometries, at a greater fidelity than previous empirical models. We employ two recently developed approaches: an Atomic Cluster Expansion (ACE) potential and a tabulated Gaussian Approximation Potential (tabGAP). Both machine-learned models are shown to accurately reproduce Density Functional Theory (DFT) energies, forces, and threshold displacement energy distributions, providing a quantitatively reliable description of atomic-scale collision processes. Molecular dynamics simulations of 5 keV cascades predict significantly enhanced peak defect production and recombination relative to a previous empirical potential, indicating qualitatively different cascade evolution. Moreover, these new machine learning models generate increased proportions of copper and oxygen vacancies compared to previous models, in direct agreement with experimental observations. By explicitly varying oxygen deficiency, we further show that total defect production exhibits only a weak dependence on stoichiometry, providing new insight into the robustness of radiation damage processes in oxygen-deficient YBCO. Finally, fusion-relevant 300 keV cascade simulations reveal amorphous regions with characteristic dimensions comparable to the superconducting coherence length, consistent with electron microscopy observations of neutron-irradiated HTS tapes. These results establish machine-learned interatomic potentials as powerful, computationally efficient tools for uncovering radiation-damage physics in YBCO, enabling predictive simulations across technologically relevant compositions and irradiation conditions.