The orthorhombic to tetragonal transition in YBa₂Cu₃O₇ with a machine learning potential

The superconducting properties of YBa₂Cu₃O_7-δ (YBCO) are strongly affected by defects in the O sublattice, which commonly occur in the synthesis process or by irradiation with energetic particles. Such defects are also spontaneously formed with temperature and are believed to be responsible for the orthorhombic to tetragonal phase transition occurring at ~1000 K. In this presentation, I will show the first machine-learning (ML) interatomic potential trained on density functional theory (DFT) calculations which is able to reproduce this transition in stoichiometric YBCO with great accuracy.

The transformation occurs by formation of O Frenkel pairs, where O atoms from the CuO chain (O1 site) migrate to the interstitial positions in between the chains (O5 site). We observe also involvement of O atoms from the apical site (O4) in this rearrangement. Moreover, we show with free energy calculations that the ordered orthorhombic structure is the thermodynamically stable structure at low temperatures, whereas the disordered tetragonal structure is stable at high temperatures. The present work proves that ML potentials trained on DFT calculations can accurately describe atomistic processes in YBCO in the normal state, and paves the way for the investigation of defect diffusion in this material at any temperature.