The diffusion-driven orthorhombic to tetragonal transition in YBa$_2$Cu$_3$O$_7$ derived with a machine learning interatomic potential

Abstract

Defects in high temperature superconductors such as YBa$_2$Cu$_3$O$_7$ (YBCO) critically influence their superconducting behavior, as they substantially degrade or even suppress superconductivity. With the renewed interest in cuprates for next-generation superconducting magnets operating in radiation-harsh environments such as fusion reactors and particle accelerators, accurate atomistic modeling of defects and their dynamics has become essential. Here, we present a general-purpose machine-learning interatomic potential for YBCO, based on the Atomic Cluster Expansion (ACE) method and trained on Density Functional Theory (DFT) data, with particular emphasis on defects and their diffusion mechanisms. The potential is validated against DFT calculations of ground-state properties, defect formation energies of oxygen Frenkel pairs and diffusion barriers for their formation. Remarkably, the potential captures the diffusion-driven orthorhombic to tetragonal transition at elevated temperatures, a transformation that is difficult to describe with empirical potentials, elucidating how the formation of oxygen Frenkel pairs in the basal plane governs this order-disorder transition. The ACE potential introduced here enables large-scale, predictive atomistic simulations of defect dynamics and transport processes in YBCO, providing a powerful tool to explore its stability, performance, and functionality under realistic operating conditions. Moreover, this work proves that machine learning interatomic potentials are suitable for studies of quaternary oxides with complex chemistry.