Chen H.-ATang P.-HChen G.-JChang C.-CPao C.-W.CHIEN-CHENG CHANG2021-08-052021-08-05202119487185https://www.scopus.com/inward/record.uri?eid=2-s2.0-85104369462&doi=10.1021%2facs.jpclett.1c00410&partnerID=40&md5=bc6d1d92f21fe065d6c92aa9906d4b3dhttps://scholars.lib.ntu.edu.tw/handle/123456789/577055Revealing the process-structure-property (PSP) relationships of chemically complex mixed-ion perovskite requires comprehensive insights into correlations between microstructures and chemical compositions. However, experimentally determining the microstructural information about complex perovskites over the composition space is a challenging task. In this study, a machine learning enabled energy model was trained for MAyFA1-yPb(BrxI1-x)3 mixed-ion perovskite for fast and extensive sampling over the compositional/permutational spaces to map the ion-mixing energies, chemical ordering, and atomic strains. Correlation analysis indicated the strong lattice distortion in the high-MA/Br concentration regime is the primary reason for poor device performance - strong lattice distortion induces high mixing energy, resulting in phase segregation and defect formation. Hence, mitigating lattice distortion to retain the single-phase solid solution is one necessary condition of the optimal composition of mixed-ion perovskites. The present study therefore provides insights into the microstructures as well as the guidelines for determining the optimal composition of mixed-ion perovskite materials. ? 2021 American Chemical Society.Ions; Machine learning; Microstructure; Mixing; Monte Carlo methods; Turing machines; Chemical compositions; Complex perovskite materials; Complex perovskites; Correlation analysis; Lattice distortions; Microstructural information; Monte Carlo sampling; Optimal composition; Perovskite; article; correlation analysis; machine learning; practice guideline[SDGs]SDG7journal article10.1021/acs.jpclett.1c00410338226322-s2.0-85104369462