Discovering novel halide perovskite alloys using multi-fidelity machine learning and genetic algorithm.

Area of Science:

• Materials Science
• Computational Chemistry
• Renewable Energy

Background:

• Stable halide perovskites are key for advanced photovoltaic (PV) absorbers.
• Current limitations in perovskite performance necessitate expanding the material pool.

Purpose of the Study:

• To develop accurate surrogate models for predicting perovskite properties.
• To perform inverse design for novel halide perovskite compositions using genetic algorithms (GA).

Main Methods:

• High-throughput density functional theory (DFT) calculations for ~800 perovskite alloys.
• Training multi-fidelity random forest regression models on DFT and experimental data.
• Employing GA with an objective function for inverse design.

Main Results:

• Screened over 150,000 hypothetical compounds, identifying thousands with desired properties (band gap 1-2 eV, efficiency >15%).
• Discovered hundreds of optimal compositions and phases through inverse design.
• Generated predictive ternary phase diagrams for material selection.

Conclusions:

• Machine learning effectively accelerates the discovery of stable, high-efficiency halide perovskites.
• The combined DFT, surrogate modeling, and GA approach is powerful for materials design.
• Identified promising candidates warrant further experimental investigation.