Spectra-Based Machine Learning for Predicting the Statistical Interaction Properties of CO Adsorbates on Surface.

Area of Science:

• Catalysis
• Surface Science
• Computational Chemistry

Background:

• Theoretical analyses of small-molecule adsorption often use simplified models.
• Real-world experiments involve multiple interacting molecules, necessitating advanced models.
• A gap exists between theoretical predictions and experimental observations in catalysis.

Purpose of the Study:

• To develop a comprehensive multimolecule adsorption model.
• To bridge the gap between theoretical and experimental catalysis research.
• To extract key interaction properties from spectra in real chemical environments.

Main Methods:

• Utilizing machine learning to predict adsorption properties.
• Employing conformationally averaged infrared and Raman spectra.
• Comparing machine learning predictions with theoretical derivations from ensembles.

Main Results:

• Machine learning accurately predicts average adsorption properties.
• The model performs well with large and indeterminate numbers of surface molecules.
• Quantitative spectra-averaged property relationships were established.

Conclusions:

• The developed model offers a robust theoretical framework for analyzing complex adsorption systems.
• This approach enhances the interpretation of experimental spectra in heterogeneous catalysis.
• It provides a pathway for more accurate predictions in real chemical environments.