题目：Thermodynamic Stability Landscape of Halide Double Perovskites via High‐Throughput Computing and Machine Learning
作 者：Zhenzhu Lia, Qichen Xua, Qingde Suna, Zhufeng Houb, Wan‐Jian Yina*
单位：aCollege of Energy, Soochow Institute for Energy and Materials InnovationS (SIEMIS), and Key Laboratory of Advanced Carbon Materials and Wearable Energy Technologies of Jiangsu Province Soochow University, Suzhou 215006, China
bCenter for Materials Research by Information Integration National Institute for Materials Science, 1-2-1 Sengen, Tsukuba, Ibaraki 305-0047, Japan
摘 要：Formability and stability issues are of core importance and difficulty in current research and applications of perovskites. Nevertheless, over the past century, determination of the formability and stability of perovskites has relied on semiempirical models derived from physics intuition, such as the commonly used Goldschmidt tolerance factor, t. Here, through high‐throughput density functional theory (DFT) calculations, a database containing the decomposition energies, considered to be closely related to the thermodynamic stability of 354 halide perovskite candidates, is established. To map the underlying relationship between the structure and chemistry features and the decomposition energies, a well‐functioned machine learning (ML) model is trained over this theory‐based database and further validated by experimental observations of perovskite formability (F1 score, 95.9%) of 246 A2B(I)B(III)X6 compounds that are not present in the training database; the model performs a lot better than empirical descriptors such as tolerance factor t (F1 score, 77.5%). This work demonstrates that the experimental engineering of stable perovskites by ML could solely rely on training data derived from high‐throughput DFT computing, which is much more economical and efficient than experimental attempts at materials synthesis.