Adsorbate chemical environment-based machine learning framework for heterogeneous catalysis.

Area of Science:

• Computational chemistry
• Materials science
• Catalysis

Background:

• Heterogeneous catalysis is governed by atomic-scale factors like morphology and adsorbate coverage.
• Modeling these phenomena requires analyzing vast numbers of atomic configurations.
• Existing methods struggle with the complexity of diverse adsorbate binding and substrate structures.

Purpose of the Study:

• To introduce a novel computational workflow, ACE-GCN, for screening catalyst surface configurations.
• To develop accurate catalyst surface models for complex systems under reaction conditions.
• To accelerate the discovery and design of efficient heterogeneous catalysts.

Main Methods:

• Developed Adsorbate Chemical Environment-based Graph Convolution Neural Network (ACE-GCN) workflow.
• Applied ACE-GCN to model NO on Pt3Sn(111) for nitrate electroreduction.
• Applied ACE-GCN to model OH* on Pt(221) for oxygen reduction reaction.
• Trained ACE-GCN on ~10% of DFT-relaxed configurations.

Main Results:

• ACE-GCN successfully models atomistic configurations with diverse adsorbates and morphologies.
• The model accurately describes relative stabilities of configurations for complex catalytic systems.
• Achieved high accuracy in predicting surface behavior with a fraction of computational cost.

Conclusions:

• ACE-GCN significantly accelerates the analysis of complex catalyst surface configurations.
• This approach enables more rigorous descriptions of catalysts under in-situ conditions.
• Expected to advance the development of novel catalysts for critical chemical transformations.