Quick screening stable double perovskite oxides for photovoltaic applications by machine learning

Rapid discovery of functional materials remains a public challenge because traditional trial and error methods are general inefficient, especially when thousands of candidates are treated. Machine learning (ML) is essential to deal with a large number of data sets, predict unknown material properties and reveal the relationship between structures and properties. Herein, in order to find double perovskite oxide (DPO) materials for solar cells, we design a framework and develop a robust ML model to predict band gaps of DPOs based on a dataset containing band gap values of 236 experimentally studied perovskite oxides. Successfully, 236 promising stable ferroelectric photovoltaic DPOs with suitable band gaps are screened out from 4,058,905 candidate compositions. The developed ML model provides an excellent predictive performance (R2 : 0.932, RMSE : 0.196 eV) based on only three component features. Moreover, our statistical graph confirms the previous studies that tuning the electronegativity difference between oxygen and B site cation via doping foreign cations could change the band gaps of perovskite oxides. These findings show that ML is very promising not only for predicting the properties, but also for investigation on the physical law.

Automated summary

Traditional trial and error methods are inefficient when dealing with a large number of candidates, but machine learning can rapidly discover functional materials and reveal the relationship between structures and properties. In this study, a machine learning model was developed to predict the band gaps of double perovskite oxide (DPO) materials for solar cells, resulting in the screening of 236 promising DPOs with suitable band gaps. The developed model showed excellent predictive performance and confirmed previous research findings through statistical analysis.