Accurate Crystal Structure\nPrediction of New 2D Hybrid\nOrganic–Inorganic Perovskites

Low-dimensional hybrid organic–inorganic perovskites\n(HOIPs)\nare promising electronically active materials for light absorption\nand emission. The design space of HOIPs is extremely large, as a variety\nof organic cations can be combined with different inorganic frameworks.\nThis not only allows for tunable electronic and mechanical properties\nbut also necessitates the development of new tools for in silico high\nthroughput analysis of candidate materials. In this work, we present\nan accurate, efficient, and widely applicable machine learning interatomic\npotential (MLIP) trained on 86 diverse experimentally reported HOIP\nmaterials. This MLIP was tested on 73 experimentally reported perovskite\ncompositions and achieves a high accuracy, relative to density functional\ntheory (DFT). We also introduce\na novel random structure search algorithm designed for the crystal\nstructure prediction of 2D HOIPs. The combination of MLIP and the\nstructure search algorithm reliably recovers the crystal structure\nof 14 known 2D perovskites by specifying only the organic molecule\nand inorganic cation/halide. Performing this crystal structure search\nwith ab initio methods would be computationally prohibitive but is\nrelatively inexpensive with the MLIP. Finally, the developed procedure\nis used to predict the structure of a totally new HOIP with cation\n(<i>cis</i>-1,3-cyclohexanediamine). Subsequently, the new\ncompound was synthesized and characterized, which matches the predicted\nstructure, confirming the accuracy of our method. This capability\nwill enable the efficient and accurate screening of thousands of combinations\nof organic cations and inorganic layers for further investigation.