Machine Learning-Assisted Density Functional Theory Screening of Double Perovskite Materials for Red-Light Emission

Our study integrates machine learning (ML) with density functional theory (DFT) calculations to construct a hybrid database comprising 1168 inorganic single and double perovskite materials, enabling the screening and prediction of red-light-emitting materials based on bandgap selection. A comparative evaluation of five regression models, Support Vector Regression (SVR), Decision Tree (DT), Extreme Gradient Boosting (XGB), Gradient Boosting Regression (GBR), and Random Forest Regression (RFR), reveals that the XGB model achieves the highest predictive accuracy for bandgaps. We use Mean Absolute Error (MAE) and coefficient of determination ( R 2 ) to assess the model screening performance ( R 2 ≈ 0.96 in the test set, MAE ≈ 0.227 eV in XGB). Using this model, Cs 2 InAgCl 6 and Cs 2 AgAsCl 6 were identified from 30 candidate materials as satisfying the red-light emission bandgap range (0.5–2.0 eV). Additionally, Cs 2 LaAgCl 6 and Cs 2 ScAgCl 6 were selected as controls for subsequent DFT calculations to systematically evaluate their mechanical stability, electronic structure, and optical properties. The results indicate that Cs 2 AgAsCl 6 is a potential candidate for high-efficiency red-light LEDs. Furthermore, Cs 2 ScAgCl 6 warrants further exploration for its potential application in blue-light LEDs.