Screening for lead-free inorganic double perovskites with suitable band gaps and high stability using combined machine learning and DFT calculation

Abstract

To accelerate the application of perovskite materials in photovoltaic solar cells, developing novel lead-free perovskite materials with suitable band gaps and high stability is vital. However, laborious experiment and density functional theory (DFT) calculation are time-consuming and incapable to screen promising perovskites rapidly and efficiently. Here, we proposed a novel search strategy combining machine learning and DFT calculation to screen 5,796 inorganic double perovskites. The eXtreme Gradient Boosting Regression (XGBR) algorithm was first applied to build a robust and predictive machine learning (ML) model for perovskite materials. XGBR algorithm yielded a lower mean square error (MSE) than both Artificial Neural Network (ANN) algorithm and Support Vector Regression (SVR) algorithm. From the ML model, two novel lead-free inorganic double perovskites: Na2MgMnI6, K2NaInI6, were obtained, suitable direct bandgaps of 1.46 eV for K2NaInI6 and 1.89 eV for Na2MgMnI6, which are similar to the organic–inorganic perovskite (MAPI3) CH3NH3PbI3 (Eg = 1.6 eV), high thermal stability and good optical properties were also confirmed by DFT calculation. The method of combining ML and DFT exhibits high accuracy and significantly speeds up the discovery of promising perovskite materials.