Data-Driven, Physics-Informed Descriptors of Cation Ordering in Multicomponent Perovskite Oxides

The chemical flexibility and versatility of multicomponent perovskite oxides have enabled their various key properties, including ion conductivity for batteries and surface reactivity for heterogeneous catalysis and electrocatalysis. Notably, the ordering and distribution of different cations in multicomponent perovskite oxides have a profound influence on their optical absorption, magnetic behaviors, ion mobility, catalytic activity, and electromechanical properties for diverse applications.<br/><br/>Unfortunately, the design and optimization of multicomponent perovskite oxides have been severely hampered by the lack of physical rules that universally rationalize and accurately predict their cation ordering. Essentially, experimentalists have only established empirical principles correlating the cation ordering in such oxides with the atomic parameters of their constituent ions, which often fail at accurately classifying perovskite oxides into cation-ordered and disordered structures.<br/><br/>High-throughput first-principles calculations could provide new, accurate design principles of cation ordering, but this exploration is constrained by the lack of a systematic benchmark between ab initio calculations and experimentally quantified ordering across a broad oxide space. This lack of co-validation between simulations and experiments can further impede the data-driven discovery of multicomponent perovskite oxides by high-throughput virtual screening, machine learning, and experimental validation.<br/><br/>To address these knowledge gaps, we established new data-driven, physically interpretable descriptors that can accurately classify ~90% multicomponent perovskite oxides as cation-ordered or disordered for an experimental dataset of 190 multicomponent perovskites (https://arxiv.org/abs/2305.01806). Notably, these descriptors can be obtained from the low-cost density functional theory (DFT) calculations of a few simple prototype structures and rationalized as physics-informed material parameters, such as thermodynamic probability and configurational entropy. Such new cation-ordering design principles outperform their conventional counterparts, accelerating high-throughput virtual screening by balancing the breadth and cost in the DFT sampling of multicomponent oxides with different cation configurations. Overall, this work offers a rigorous benchmark between theory and experiments and demonstrates an effective paradigm for understanding and predicting chemical ordering in multicomponent materials across a vast chemical space.