Data-driven discovery of electrocatalysts for CO₂ reduction using active motifs-based machine learning.

Area of Science:

• Electrochemistry
• Materials Science
• Computational Chemistry

Background:

• Electrochemical carbon dioxide reduction (CO2RR) offers a pathway for CO2 mitigation and valuable chemical synthesis.
• Machine learning (ML) accelerates catalyst discovery but is often limited to narrow chemical spaces and incomplete activity predictions.
• CO2RR can produce diverse chemical products, necessitating catalysts with high selectivity.

Purpose of the Study:

• To establish a high-throughput virtual screening strategy for identifying active and selective CO2RR catalysts.
• To overcome limitations of current ML approaches by integrating a CO2RR selectivity map.
• To guide researchers on catalyst stoichiometry and morphology for enhanced performance.

Main Methods:

• Developed a hybrid ML model combined with a CO2RR selectivity map for virtual screening.
• Predicted catalytic activity and selectivity for 465 metallic catalysts across four target products.
• Validated computational predictions through experimental methods.

Main Results:

• Identified promising catalytic activity and selectivity for CO2RR across a wide range of metallic catalysts.
• Discovered previously unreported, favorable catalytic behaviors in Cu-Ga and Cu-Pd alloys.
• The screening strategy successfully predicted catalyst performance beyond existing databases.

Conclusions:

• The integrated ML and selectivity map strategy enables efficient discovery of active and selective CO2RR catalysts.
• Cu-Ga and Cu-Pd alloys show significant potential for CO2RR applications.
• This approach accelerates the identification of novel catalysts and provides valuable design insights.