Comprehensive Structural Descriptors Enable Machine Learning Prediction of Perovskite Band Characteristics.

Area of Science:

• Materials Science
• Condensed Matter Physics
• Computational Chemistry

Background:

• Perovskites are key materials for next-generation optoelectronics.
• Accurate prediction of electronic properties like bandgap is vital for device design.
• Conventional machine learning struggles with soft perovskite lattice distortions.

Purpose of the Study:

• Develop physically meaningful structural descriptors for halide perovskites.
• Improve machine learning model accuracy for predicting electronic properties.
• Investigate structure-property relationships in Br-alloyed perovskites.

Main Methods:

• Introduced novel descriptors: octahedral distortion, cation displacement, segregation index.
• Applied machine learning to predict bandgap, VBM, and CBM.
• Analyzed two Br-alloyed lead halide perovskite systems (Cs$_{0.25}$FA$_{0.75}$PbI$_{x}$Br$_{3-x}$ and CsPbI$_{x}$Br$_{3-x}$).

Main Results:

• Achieved high accuracy in predicting bandgap, VBM, and CBM positions.
• Demonstrated the importance of structural information in ML models.
• Discovered an inverse correlation between Pb-X-Pb bond angle variation and bandgap.

Conclusions:

• Physically meaningful structural descriptors enhance ML predictions for perovskites.
• Structural order influences electronic properties, specifically the VBM contribution of Bromine.
• This approach aids in rational design of high-performance perovskite optoelectronics.