Exploring Structure-Sensitive Relations for Small Species Adsorption Using Machine Learning.

Area of Science:

• Catalysis
• Materials Science
• Computational Chemistry

Background:

• Accurate prediction of adsorption energies on heterogeneous catalyst surfaces is vital for predicting reactivity and screening materials.
• Existing adsorption linear scaling relations often lack accuracy and are limited to single adsorbate/site types, hindering predictions of structure sensitivity and optimal catalyst design.

Purpose of the Study:

• To develop a machine learning model for accurate prediction of adsorption energies on heterogeneous catalyst surfaces.
• To identify critical factors influencing adsorption energies and preferred binding sites.
• To outperform traditional linear scaling relations in accuracy and scope.

Main Methods:

• Utilized machine learning on nearly 300 density functional theory calculations.
• Developed generalized coordination number scaling relations.
• Incorporated species valency, electronic coupling, site type, and coordination environment into the model.

Main Results:

• Generalized coordination number scaling relations showed accuracy for oxygen and high-valency carbon species but failed for others.
• The machine learning model accurately predicts adsorption energy and preferred site, outperforming linear scalings.
• Identified valency, electronic coupling, site type, and coordination environment as critical factors for small species adsorption.

Conclusions:

• Machine learning models offer a more accurate and general approach to predicting adsorption energies compared to linear scaling relations.
• The developed model can reveal the structure sensitivity of chemical reactions and guide catalyst design for enhanced activity.
• The methodology is validated for transition metals and applicable to single-atom alloys.