Universal Rapid Machine Learning Models for Predicting Unconvoluted and Convoluted X-ray Absorption Spectra

X-ray absorption near-edge structure (XANES) serves as a powerful analytical technique for probing the local three-dimensional (3D) atomic structure in materials and molecules. Rapid computation of XANES based on 3D structures of materials or molecules constitutes the cornerstone and a vital element of quantitative XANES analysis. Here, we introduce an XANES prediction model that accepts 3D structures as input and generates either XANES or unconvoluted XANES as output, demonstrating excellent generalizability across diverse broadening. The model's predictive accuracy is validated for both hard X-ray XAS (exemplified by K-edge spectroscopy of 3d/4d transition metals) and soft X-ray XAS (demonstrated via sulfur K-edge spectroscopy). By adopting the model, researchers can predict XANES of multiple elements using a single unified model, eliminating the need for element-specific models. Remarkably, even with sparse 3D structural data for target elements, the model reliably predicts the corresponding XANES spectra. This model empowers researchers to perform real-time validation of hypothesized 3D structures during XAS beamline experiments.