Phase Transitions in the BaZrS₃ Chalcogenide Perovskite Through Machine Learning Potentials

Chalcogenide perovskites, particularly BaZrS3, has gained popularity in the last few years due to its great potential as an alternative lead-free photovoltaic absorber material. This is due to promising optoelectronic properties such as defect tolerance, strong dielectric screening, and light absorption [1]. However, some fundamental material physics, in particular polymorphic phase transitions, have not been explored in detail. Experimental studies have given conflicting results with Raman spectroscopy showing no signs of a phase transition[2], whilst XRD studies show an orthorhombic-to-tetragonal phase transition at 800K [3].<br/><br/>In this talk, we will introduce our machine learning potential model trained on perovskite structures with the neuroevolution potential method [4]. Through molecular dynamics calculations, we heat the experimentally reported orthorhombic <i>Pnma</i> phase and observe a first-order phase transition to a tetragonal <i>I4/mcm</i> phase at 610K. Upon further heating, we observe a second-order phase transition from the tetragonal phase to the cubic <i>Pm-3m</i> phase at 880K. We explain the order of these phase transitions through group-subgroup relationships and Landau theory.<br/><br/>Further analysis shows that the phase transitions are mediated through the M and R phonon modes associated with octahedral tilting, as is typically found in perovskite structures [5]. We analyze all possible Glazer tiltings to show that for the BaZrS₃ perovskite only the <i>Pnma --&gt;</i><i> I4/mcm --&gt;</i><i> Pm-3m</i> phase transition route is accessible through heating. We will also show a temperature-dependent XRD pattern to support further experimental studies.<br/><br/>To end, we will outline how various levels of anharmonicity (harmonic, quasi-harmonic, and fully anharmonic) and exchange-correlation functional (PBEsol and HSE06) impact on our predictions, highlighting some of the subtleties and limitations of each method.<br/><br/><br/>References:<br/><br/>[1] Sopiha et al, Adv. Opt. Mater. 2022 10 2101704.<br/>[2] Ye et al, arXiv: 2402.18957.<br/>[3] Jaykhedkar et al, Inorg. Chem. (2023) 62 12480.<br/>[4] Fan et al, J Chem. Phys. (2022) 157 114801.<br/>[5] Fransson et al, Chem. Mater. (2023) 35 8229.