Quantitative Structural Description of Zeolites by Machine Learning Analysis of Infrared Spectra

Area of Science:

• Materials Science
• Computational Chemistry
• Crystallography

Background:

• Machine learning (ML) algorithms can uncover complex correlations in spectroscopic data.
• Infrared (IR) spectroscopy is a key technique for analyzing materials, including zeolites.
• Understanding structure-spectrum relationships in zeolites is crucial for their design and application.

Purpose of the Study:

• To apply ML algorithms to simulated IR spectra of zeolites.
• To establish correlations between zeolite structural information and spectral features.
• To predict zeolite structural characteristics using IR spectra.

Main Methods:

• Utilized ML algorithms on theoretically simulated IR spectra of 230 zeolite frameworks.
• Trained ML models for classification to predict tilings and secondary building units (SBUs).
• Employed the ExtraTrees algorithm for regression on an expanded dataset of 470 spectra, including modified structures.

Main Results:

• Achieved >89% accuracy in predicting natural zeolite tilings and SBUs.
• Obtained >90% prediction accuracy for average Si-O distances, Si-O-Si angles, and TO4 tetrahedra volumes.
• Demonstrated the potential of IR spectra as a quantitative tool for zeolite characterization.

Conclusions:

• ML analysis of IR spectra effectively establishes structure-spectrum correlations in zeolites.
• The study successfully predicted key structural features of zeolites with high accuracy.
• This approach offers novel avenues for utilizing IR spectroscopy in quantitative zeolite analysis.