Applying Active Learning to the Screening of Molecular Oxygen Evolution Catalysts

Area of Science:

• Catalysis
• Materials Science
• Computational Chemistry

Background:

• The oxygen evolution reaction (OER) is key for green hydrogen production.
• Current OER catalysts are expensive or inefficient due to high overpotentials.
• Computational screening is needed to discover novel, cost-effective OER catalysts.

Purpose of the Study:

• To develop machine learning algorithms for enhanced computational screening of molecular OER catalysts.
• To improve the efficiency of discovering new OER catalysts.
• To identify catalysts that circumvent limitations imposed by linear scaling relations.

Main Methods:

• Utilized Gaussian process regression (GPR) to predict binding energies.
• Applied active learning schemes to guide computational simulations.
• Derived an acquisition function to identify promising catalyst candidates.

Main Results:

• Demonstrated improved computational efficiency in catalyst screening.
• Showed the algorithm guides simulations towards candidates with desirable OER descriptor values.
• Identified a method to circumvent linear scaling relation constraints for OER overpotentials.

Conclusions:

• Machine learning significantly enhances the discovery of novel OER catalysts.
• The developed approach can accelerate the identification of efficient and affordable catalysts for green hydrogen.
• Provides a perspective on screening stable and active molecular OER catalysts.