Thermodynamic stability prediction of organic-inorganic hybrid perovskites using machine learning

Hybrid organic-inorganic perovskites (HOIPs) exhibit exceptional photovoltaic conversion efficiency and have broad applications in solar energy applications. However, their structural stability remains a key factor limiting their development and application. Accurate prediction of HOIPs stability is therefore critical. In this study, the thermodynamic stability of HOIPs was quantified by their relative energies, with lower values indicating higher stability. Based on a dataset of 1346 perovskite samples, we employed two feature selection strategies, recursive feature elimination (RFE) and stepwise method, to build machine learning models. The gradient boosting regression model with RFE exhibited the best performance, achieving a highR²score of 0.993. Key features-including the average group number (N_g), average anionic radius (r_i), B-site lattice constantc(c_B), and the lowest energy of the atomic orbitals in the X-site (E_X)-were identified and found to correlate clearly with thermodynamic stability. Using the trained model, we predicted the relative energies of 1584 previously unreported perovskite candidates, among which 106 compounds with negative relative energy were predicted to be potentially stable. Further density functional theory calculations confirmed 26 stable candidates and their thermodynamic stability was further validated using theab initiomolecular dynamics. The proposed screening workflow in this study offers approximately a 15-fold improvement in efficiency compared to traditional computational methods, providing valuable guidance for the rapid identification of stable perovskites.