Bandgap prediction and design of halide double perovskites via ensemble machine learning

Dongwen Yang, Shulin Luo, Fahai Zhong, Lingrui Wang, Sanjun Wang, Wenyuan Zhang, Chunyao Niu, Chong Li, Fei Wang,
Bandgap prediction and design of halide double perovskites via ensemble machine learning,
Chemical Engineering Journal,
Volume 525,
2025,
169979,
ISSN 1385-8947,
https://doi.org/10.1016/j.cej.2025.169979.
(https://www.sciencedirect.com/science/article/pii/S138589472510822X)
Abstract: Owing to the compositional and structural versatility with tunable bandgaps, halide double perovskites (HDPs) have opened up a broad range of optoelectronic applications. Accurately measuring their bandgap values traditionally relies on experimental characterization or density functional theory (DFT) calculations. However, experimental methods are expensive. While DFT-based predicting methods are usually with long calculating time and high computational cost. The DFT results are mainly depended by the picked exchange correlation functional, which weaken the generalization ability of these methods. In this study, we utilize a general ensemble machine learning (EML) technique to accurately predict the bandgap values of HDPs with a low-cost flying way. A comprehensive HDPs dataset with 1320 A2BX6, 8348 A2B1+B3+X6, and 6120 A2B2+B′2+X6 were generated for EML. Importantly, we found that incorporating the DFT-PBE bandgap in the feature space can significantly improve the prediction results, achieving R2 of 0.95, 0.96, and 0.91, respectively. Moreover, based on the feature importance determined by EML, the relationship between the physical microstructure and the bandgap values were further analyzed. Setting the predicted bandgap values as screening criteria, we identified 58 compounds suitable for solar cell applications and 96 compounds for UV materials.
Keywords: Bandgap; Halide double perovskites; Ensemble machine learning; First-principles calculations; Optoelectronic materials design