A Graph Neural Network-Based Approach to XANES Data Analysis.

Area of Science:

• Materials Science
• Computational Chemistry
• Spectroscopy

Background:

• Three-dimensional (3D) structure determination is vital for linking material properties to performance.
• X-ray absorption near edge structure (XANES) spectroscopy provides atomic-scale local 3D structural information.
• Accurate 3D structure analysis is essential for materials design and discovery.

Purpose of the Study:

• To present a novel approach for simulating XANES spectra using a 3D graph neural network (3DGNN).
• To develop a faster and more accurate method for 3D structure determination from XANES data.
• To integrate advanced computational tools into materials analysis frameworks.

Main Methods:

• Development of a customized 3D graph neural network (3DGNN) model named XAS3Dabs.
• Direct input of 3D system structures into the XAS3Dabs model.
• Incorporation of geometric features within the model's message passing blocks.
• Graph extraction focusing on edges related to the absorbing atom to reduce redundancy.

Main Results:

• XAS3Dabs achieves faster simulation times compared to traditional XANES fitting methods when combined with optimization algorithms.
• The model demonstrates superior accuracy in XANES prediction over existing machine learning models.
• The approach shows enhanced model performance and robustness across various hyperparameters.
• The method is generalizable for simulating spectra across different absorption edges and systems.

Conclusions:

• XAS3Dabs offers a powerful and efficient tool for analyzing local 3D structures using XANES data.
• The model's accuracy and speed make it suitable for online data processing and real-time analysis.
• This approach is expected to be a key component of future online 3D structure analysis frameworks, particularly for facilities like the High-Energy Photon Source (HEPS).