Benchmarking Core-Level X-ray Absorption with MRSF-TDDFT, RASPT2, and Stochastic GAS Using the XABOOM Set

Rapid advances in X-ray free-electron laser facilities are producing X-ray absorption spectra (XAS) with unprecedented detail, creating a pressing need for theoretical methods that are both accurate and computationally efficient to interpret subtle spectral features. In this study, we benchmark two cost-effective electronic-structure approaches for an accurate description of XAS spectra: mixed-reference spin-flip time-dependent density-functional theory (MRSF-TDDFT) and single- and (extended) multistate restricted active-space perturbation theory (SS/(X)MS-RASPT2). We apply these methods to the near-edge X-ray absorption fine structure (NEXAFS) of small and medium-sized organic molecules from the XABOOM test set [J. Chem. Theory Comput. 17, 1618-1637 (2021)]. We show that MRSF-TDDFT and SS/(X)MSRASPT2 yield average errors below 0.5 eV for the lowest bright transitions compared with available gas-phase experimental K-edge spectra and reproduce both excitation energies and relative intensities across the full preionization region with excellent agreement. Furthermore, we introduce the Stochastic Generalized Active Space (S-GAS) method as a high-level tool to simulate XAS spectra and apply it to a subset of the XABOOM data set. In addition to providing accurate spectra, the S-GAS method serves as a benchmark for assessing the performance of the more approximate electronic-structure approaches. Taken together, these methods offer a robust framework for simulating XAS spectra, allowing us to systematically analyze the influence of orbital relaxation, core-valence separation, basis-set size, configuration space restrictions, relativistic effects, and vibrational zero-point energy. These findings provide practical guidelines for achieving high accuracy at reduced computational cost and for enabling more reliable interpretation of experimental NEXAFS features.