Data-Driven Investigation of the Synthesizability and Bandgap of Double Perovskite Halides

Abstract

Double perovskite halide materials have been widely used in batteries, light-emitting diodes, and solar cells. Thus, investigations of the fundamental properties of the double perovskite halide to search for an ideal structure are crucial. In this study, a surrogate model is developed to predict the formation energy, convex hull energy, and bandgap of A(2)BB'X-6 type double perovskite halide structures. The material properties of 13 542 candidate structures are predicted and validated through first-principles calculations. Without double perovskite halide information during training, the prediction accuracy for the formation energy is obtained as an R-squared value of 0.770 and Root Mean Square Error (RMSE) of 0.404 eV atom(-1). For the convex hull energy, an accuracy of 0.642 is obtained. For the bandgap, R-squared score of 0.427 and an RMSE of 1.235 eV are achieved. Furthermore, the optimization process confirms that adding only 850 (6%) double perovskite halide structures to the training set increases the R-squared value to 0.90 for the formation energy. In the bandgap, more data are needed; 3550 data (68.2%) are added to achieve an R-squared score of 0.9. The current study successfully predicts the fundamental properties of double perovskite halides for the accelerated discovery of ideal structures.