Mapping XANES spectra on structural descriptors of copper oxide clusters using supervised machine learning

Area of Science:

• Materials Science
• Catalysis
• Computational Chemistry

Background:

• Supported subnanometer metal oxide clusters exhibit enhanced reactivity, but their atomic-level structural characterization remains challenging.
• X-ray absorption near edge structure (XANES) spectroscopy is sensitive to local geometry, and theoretical modeling of XANES spectra is reliable.
• Supervised machine learning offers a powerful approach to extract structural information from experimental spectra.

Purpose of the Study:

• To apply a supervised machine learning method to analyze grazing incidence XANES spectra of size-selective copper oxide clusters.
• To correlate experimental XANES data with atomic-level structural descriptors under operando methanation conditions.

Main Methods:

• Utilized a supervised machine learning approach, specifically a convolution neural network, trained on theoretical XANES spectra.
• Applied the trained model to "invert" experimental grazing incidence XANES data from copper oxide clusters.
• Measured spectra in operando conditions during the methanation reaction.

Main Results:

• Successfully extracted Cu-Cu coordination numbers, serving as structural descriptors.
• Distinguished between different structural motifs, specifically Cu2O-like and CuO-like, of copper oxide clusters.
• Reliably evaluated average cluster sizes, demonstrating dynamic structural transformations during the reaction.

Conclusions:

• The machine learning approach effectively bridges experimental XANES data and atomic-level structural information for supported subnanometer clusters.
• Understanding the dynamic structural changes of copper oxide clusters under reaction conditions is crucial for catalysis.
• This work has significant implications for establishing structure-composition-function relationships in heterogeneous catalysis.