Screening Perovskites from ABO₃ Combinations Generated by Constraint Satisfaction Techniques Using Machine Learning

Area of Science:

• Materials Science
• Computational Chemistry
• Machine Learning

Background:

• Perovskite oxides exhibit desirable properties for diverse applications.
• Traditional screening methods (experimental and DFT) are costly and time-consuming.
• Efficient methods are needed to identify promising perovskite candidates.

Purpose of the Study:

• To develop a machine learning (ML) approach for identifying perovskite oxides.
• To predict perovskite formability and stability from ABO3 combinations.
• To accelerate the discovery of novel perovskite materials.

Main Methods:

• Formulated perovskite identification as a constraint satisfaction problem.
• Utilized charge neutrality and Goldschmidt tolerance factor criteria.
• Engineered 16 features from 21 to train ML models on 343 known ABO3 compounds.
• Developed separate ML models for formability and stability prediction.

Main Results:

• Achieved high precision (0.983, 0.971) and recall (1.00, 0.943) for formability and stability models, respectively.
• Identified 1373 formable perovskites and 430 stable perovskites from 2229 ABO3 combinations.
• Highlighted key features for predicting formability (A-O bond length, tolerance, octahedral factors) and stability (B-site elemental and structural features).

Conclusions:

• The ML approach effectively predicts perovskite formability and stability.
• Identified 338 promising perovskite candidates for future research.
• Demonstrated the potential of ML to significantly expedite materials discovery in perovskite science.