Li Research Group Computational Chemistry Lab

We are excited to announce the publication of our latest research titled "Surface Characterization of Cerium Oxide Catalysts Using Deep Learning with Infrared Spectroscopy of CO," in Materials Today Sustainability. This study pioneers the use of deep learning and density functional theory (DFT) to investigate the surface properties of cerium oxide (CeO₂) catalysts, widely used in gas-phase redox reactions and environmental applications.

 

In this study, we combined experimental infrared (IR) spectroscopy and DFT to develop a deep learning model capable of predicting surface structures of CeO₂ based on IR spectra. This approach provides a faster and more efficient method for analyzing surface characteristics compared to traditional spectroscopic techniques, offering valuable insights for catalyst design and optimization.

 

 

 

Key Findings:

• DFT-Enhanced Data for Machine Learning Models: By leveraging DFT calculations, we created a comprehensive dataset of vibrational frequencies, adsorption energies, and IR spectra of CO adsorbed on CeO₂ facets. This dataset was used to train deep learning models, significantly enhancing the accuracy of surface structure predictions.
• Deep Learning for Surface Characterization: The deep learning models accurately predicted CeO₂ facet distributions, CO-derived surface species, and adsorption energies from IR spectra. The predictions were validated against both synthetic and experimental IR spectra, demonstrating the effectiveness of combining machine learning with DFT for surface analysis.
• Applications in Catalyst Design: The approach provides new insights into how different CeO₂ facets interact with probe molecules like CO, allowing for better control over catalyst properties. This method also offers a valuable tool for optimizing CeO₂-based catalysts used in environmental applications such as CO₂ reduction and pollutant removal.

 

This collaboration with Prof. Wen-Yueh Yu emphasizes the transformative potential of combining machine learning with quantum mechanical simulations like DFT, accelerating research in catalyst development and surface science.

 

For more details, please access the full publication here.