Machine learning meets X-ray absorption spectroscopy

Scientists at Lawrence Livermore National Laboratory (LLNL) have developed a new approach that can quickly predict the structure and chemical composition of dissimilar materials.

In a new study in materials chemistryLLNL scientists Wonseok Jeong and Tuan Anh Pham have developed an approach that combines machine learning and X-ray absorption spectroscopy (XANES) to resolve the chemical species of amorphous carbon nitride.

This study provides new deep insights into the local atomic structure of the system and, more broadly, is important in establishing an automated framework for the rapid characterization of heterogeneous materials with complex structures. This is a step.

Elucidating the atomic structure of heterogeneous materials, such as carbonaceous residues produced in the detonation of high explosives, poses a major challenge for materials scientists. This process is often labor intensive and often involves the use of empirical parameters.

To address this open challenge, the team's integrative approach begins by developing the potential of machine learning to efficiently explore the vast configuration space of amorphous carbon nitride as a representative system. This neural network-based model enables the identification of representative local structures within the material and provides insight into how these structures vary with chemical composition and density.

By combining these machine learning possibilities with high-fidelity atomic simulations, the researchers established correlations between local atomic structure and spectroscopic features. This correlation serves as a basis for interpreting XANES experimental data, allowing important chemical information to be extracted from complex spectra.

“In our study, we aimed to address the long-standing challenge of characterizing detonation products and disordered materials in general by integrating computational and experimental methods,” said Jeong, first author of the paper. says.

“Our approach not only improves our understanding of these materials, but also lays the foundation for similar studies across different material systems and characterization methods. It can be easily employed to predict elemental speciation and provide inputs that have contributed to improvements in the explosion model,'' said Pham, the project's principal investigator.

The findings of this study represent a major advance in the field of materials science and provide a powerful framework for understanding atomic speciation in disordered systems. Furthermore, the versatility of this approach means that it can be easily adapted to investigate other material classes and experimental characterization probes, paving the way for real-time interpretation of spectroscopic measurements.

This study involved collaboration between researchers from different backgrounds, highlighting the interdisciplinary nature of LLNL research. As scientists continue to explore the frontiers of materials design and characterization, such innovative approaches have the potential to open up new opportunities for innovation and scientific discovery, Jeong said. Ta.

Other co-authors on the paper include Wenyu Sun, Marcos Calegari Andrade, Liwen Wan, Trevor Willey, and Michael Nielsen.