Mechanism- and Data-Driven Exploration of a Global Descriptor for CO₂ Reduction

Area of Science:

• Catalysis
• Materials Science
• Computational Chemistry

Background:

• Atomic-scale structure-performance relationships are crucial for catalyst design.
• Existing descriptors often use local information, limiting accuracy in predicting catalytic performance.
• A gap exists between computational predictions and experimental results for nanoparticle catalysts.

Purpose of the Study:

• To develop a global descriptor for copper nanoparticles (NPs) to understand size effects in CO2 reduction reaction (CO2RR).
• To establish a novel computational framework for predicting catalytic performance at experimental scales.
• To correlate microscopic structure with macroscopic catalytic performance.

Main Methods:

• A mechanism- and data-driven approach was employed.
• Surface oxidation (*OH coverage) was identified as a key global property.
• A multiscale neural network framework (ScaleNet) was developed to predict *OH coverage on NPs.

Main Results:

• ScaleNet accurately predicted *OH coverage on Cu NPs at experimental scales, overcoming limitations of density functional theory (DFT).
• The predicted *OH coverage strongly correlated with experimentally observed CO2RR activity and selectivity.
• The framework demonstrated excellent accuracy and extrapolation ability by integrating global and local information.

Conclusions:

• *OH coverage serves as a reliable global descriptor for Cu NP catalytic performance in CO2RR.
• The ScaleNet framework offers a novel learning paradigm for nanoscale research and catalyst optimization.
• This work provides valuable insights for designing and improving catalysts based on atomic-scale understanding.