Unsupervised machine learning reveals eigen reactivity of metal surfaces

Area of Science:

• Surface Chemistry
• Materials Science
• Computational Chemistry

Background:

• Metal surface reactivity is crucial for catalysis but lacks a precise definition.
• Current methods for comparing metal surface activity rely on specific probe adsorbates, leading to variable results.

Purpose of the Study:

• To define and characterize the intrinsic reactivity of metal surfaces.
• To establish a framework for understanding metal surface behavior independent of probe molecules.

Main Methods:

• Utilized unsupervised machine learning, specifically principal component analysis (PCA).
• Analyzed a dataset of binding strengths for 10 probe adsorbates across 48 metal surfaces.

Main Results:

• Identified two dominant eigenvectors representing intrinsic metal surface reactivity.
• Defined these eigenvectors as covalent reactivity and ionic reactivity.
• Ionic reactivity correlates with work function; covalent reactivity relates to electronic density of states at the Fermi level.

Conclusions:

• Metal surface reactivity is a two-dimensional vector, not a scalar value.
• This finding provides a new perspective on surface interactions and catalyst design.