Predicting and understanding photocatalytic CO₂ reduction reaction with IR spectroscopy-based interpretable machine...

Area of Science:

• Catalysis
• Materials Science
• Computational Chemistry

Background:

• Selective conversion of carbon dioxide (CO2) into valuable products is crucial.
• Correlating microscopic spectral data with macroscopic catalytic performance remains challenging.

Purpose of the Study:

• To establish an accurate and interpretable relationship between vibrational spectral signals and catalytic performance using machine learning (ML).
• To uncover hidden physical insights for photocatalytic CO2 reduction.

Main Methods:

• Advanced machine learning approaches were employed to analyze infrared (IR) spectral signals.
• The model's generalizability was tested on a Bi5O7I photocatalytic system.

Main Results:

• The ML model accurately predicted CO production activity and selectivity from IR spectral data.
• A novel strategy to steer CO selectivity was identified and validated through reaction mechanism analysis.

Conclusions:

• Machine-learned spectroscopy offers a powerful tool for identifying reaction control factors in CO2 conversion.
• This approach lays the foundation for targeted optimization and reverse design of catalysts.