Machine-Learning Interatomic Potentials for Radiation Damage Analysis in High-Temperature Superconductors

The superconducting properties of rare-earth barium copper oxides (REBCOs) materials such as YBa2Cu3O7−δ (YBCO) are subject to profound changes when crystal defects are present. Neutron irradiation in compact fusion reactors, for example, leads to structural damage and degradation of such properties. Atomistic simulations can thus provide

insight into the microscopic features of radiation damage.

We develop and compare Machine-Learning Potentials (MLPs) like ACE and MACE, trained on extensive DFT data from CP2K, including a vast range of YBCO structures across oxygen stoichiometries, along with its sub-phases. The models reproduce equations of state, elastic constants, defect formation energies, and the orthorhombic–tetragonal transition in YBCO7, outperforming classical potentials. We also benchmark Threshold Displacement Energies (TDEs), analyze potential-energy-surface curvature under displacement, and assess liquid-phase accuracy relevant to heat spikes and recrystallization.

Finally, a statistically significant set of collision cascades simulations is performed across PKA energies and species at 20 K and 300 K. This comparison highlights accuracy and trade-offs among MLP frameworks and provides insight into how oxygen content influences defect dynamics in REBCO superconductors.