Learn-and-Match Molecular Cations for Perovskites

Forecasting the structural stability of hybrid organic/inorganic compounds, where polyatomic molecules replace atoms, is a challenging task; the composition space is vast, and the reference structure for the organic molecules is ambiguously defined. In this work, we use a range of machine-learning algorithms, constructed from state-of-the-art density functional theory data, to conduct a systematic analysis on the likelihood of a given cation to be housed in the perovskite structure. In particular, we consider both ABC(3) chalcogenide (I-V-VI3) and halide (I-II-VII3) perovskites. We find that the effective atomic radius and the number of lone pairs residing on the A-site cation are sufficient features to describe the perovskite phase stability. Thus, the presented machine learning approach provides an efficient way to map the phase stability of the vast class of compounds, including situations where a cation mixture replaces a single A-site cation. This work demonstrates that advanced electronic structure theory learning analysis can provide an efficient strategy superior to the conventional trial-and-error approach in combined with machine materials design.