AI-driven battery material analysis improves XANES predictions by 55%

Understanding the atomic and electronic structure of materials relies heavily on X-ray absorption near-edge structure (XANES) analysis, but accurately simulating these spectra is often time-consuming and computationally intensive, especially when studying dynamic processes. Zichang Lin, Wenjie Chen, Yitao Lin and colleagues at Tsinghua University have developed a new approach to overcome these limitations by employing artificial intelligence. To predict XANES spectra with remarkable accuracy, the research team first built a crystal graph neural network trained on extensive simulation data for 48 different elements. Importantly, this model was subsequently refined using a small amount of experimental data through a transfer learning strategy, greatly increasing its ability to match real-world observations and reducing prediction errors of key spectral features of elements such as sulfur, titanium, and iron by approximately 55%, paving the way for rapid and reliable materials analysis.

When large amounts of data need to be analyzed, such as in-situ characterization of battery materials, it is usually too complex to achieve the required accuracy and timeliness. To address these issues, an artificial intelligence (AI) model for XANES prediction was developed. However, existing models are trained using simulated data, resulting in large discrepancies between predicted and experimental spectra, and their universality across different elements is not well studied.

XAS characterization of minerals and materials

His research focuses on X-ray absorption spectroscopy (XAS) and its applications in characterizing the local structure and electronic properties of various materials, including minerals, catalysts, energetic materials, and nanomaterials. Computational techniques such as density functional theory are used to model XAS spectra, determine local structure, and aid materials discovery. Machine learning techniques such as t-SNE and PCA are employed to analyze XAS data and extract meaningful information, often leveraging material databases such as Material Projects and AFLOWLIB.

Crystal Graph network accurately predicts XANES spectra

Scientists have achieved a breakthrough in predicting X-ray absorption near-edge structure (XANES) spectra, a key technique for understanding the atomic and electronic structure of materials, by developing a crystal graph neural network model called CGXAS. Training the model on a comprehensive dataset of 341,405 simulated XANES spectra containing 48 different elements resulted in a significantly lower mean relative squared error of 0.020223 in predicting XANES features. This universal model, CGXAS Uni, demonstrated superior accuracy and versatility compared to element-specific models. Experiments demonstrate the power of transfer learning to improve model predictions using limited experimental data and address common mismatches between simulated and real-world spectra.

Calibrating CGXAS Uni using a small dataset containing only 48 spectra for sulfur, 40 spectra for titanium, and 45 spectra for iron significantly reduced the edge energy mismatch error of K-edge XANES for these elements by approximately 55%. Measurements confirm that the resulting CGXAS Exp model accurately predicts XANES features and bridges the gap between computationally efficient simulations and accurate experimental observations. This provides a new method for fast, universal, and experimentally calibrated XANES predictions, opening up possibilities for real-time analysis of materials, especially in complex systems such as battery materials, and enabling more efficient in-situ characterization.

AI predicts XANES spectra across elements

Scientists have developed a new approach to predicting X-ray absorption near-edge structure (XANES) spectra, which provides valuable insight into the atomic and electronic structure of materials. Recognizing the limitations of existing simulation methods in terms of both accuracy and speed when analyzing large datasets, the team created an artificial intelligence model capable of universal XANES predictions across 48 different factors. The model, initially trained on simulated data, achieves surprisingly low error rates and shows the potential for rapid analysis of complex materials. The researchers used transfer learning to further refine the model and calibrate it on a limited set of experimental XANES data, significantly reducing the differences between predicted and experimental spectra for elements such as sulfur, titanium, and iron. This calibration process reduced the prediction error of key spectral features by approximately 55%, demonstrating the model's ability to adapt to real-world data and improve its predictive power. The success of this approach stems from the ability to train the model on a diverse and universal dataset and learn the fundamental relationships between material structure and XANES spectra, regardless of the specific element being studied.