Guiding blind signal separation in X-ray absorption spectroscopy using supervised machine learning

Area of Science:

• Materials Science
• Catalysis
• Spectroscopy

Background:

• Heterogeneous catalysts exhibit complex, time-dependent structures under working conditions, with active and passive species coexisting.
• Decoupling contributions from multiple species in spectroscopic data is challenging for sample-averaging techniques like X-ray absorption spectroscopy (XAS).

Purpose of the Study:

• To develop a data-driven, supervised machine learning (ML) approach to distinguish contributions of spectroscopically distinct species in complex materials.
• To enable accurate quantification of uncertainties in spectral decomposition.

Main Methods:

• Utilized a supervised ML approach trained on large datasets of ab initio calculated X-ray absorption spectra for diverse compounds.
• Applied the method to both model datasets and experimental time-resolved operando XAS data.

Main Results:

• Successfully demonstrated the ML method's ability to identify and separate spectral contributions from coexisting species.
• Applied to a copper oxide catalyst during electrocatalytic CO2 conversion, revealing a charge disproportionation process.
• Identified the formation of copper oxalate species and metallic Cu under open circuit conditions.

Conclusions:

• The developed ML approach effectively deconvolutes complex XAS data, aiding in the identification of active catalytic species.
• The method provides a robust tool for spectral decomposition and uncertainty quantification in operando studies.
• Revealed a novel charge disproportionation mechanism in copper oxide catalysts during CO2 electroreduction.