Machine-Learning-Augmented Chemisorption Model for CO2 Electroreduction Catalyst Screening.

Area of Science:

• Materials Science
• Computational Chemistry
• Catalysis

Background:

• Predicting surface reactivity of metal alloys is crucial for catalyst design.
• Accurate modeling of adsorbate interactions on multimetallic surfaces remains challenging.

Purpose of the Study:

• To develop a machine-learning model for fast and accurate prediction of metal alloy surface reactivity.
• To identify promising multimetallic alloys for CO2 electrochemical reduction.

Main Methods:

• Utilized artificial neural networks trained on ab initio adsorption energies and electronic fingerprints.
• Employed scaling relations for adsorption energies to enhance prediction capabilities.
• Analyzed network response to input feature perturbations for fundamental understanding.

Main Results:

• Achieved prediction of adsorbate interactions on multimetallics with approximately 0.1 eV error.
• Outperformed the two-level interaction model in predictive accuracy.
• Identified promising {100}-terminated multimetallic alloys for CO2 reduction to C2 species.

Conclusions:

• Machine-learning augmented chemisorption models enable efficient high-throughput catalyst screening.
• This approach enhances understanding of chemical bonding on metal surfaces.
• The model facilitates the design of advanced catalysts for sustainable chemical transformations.