Key points are not available for this paper at this time.
Accurate and efficient prediction of nuclear magnetic resonance (NMR) parameters remains a central challenge in electronic-structure theory, particularly across chemically diverse nuclei and bonding environments. Using our analytic gauge-including atomic orbital (GIAO) implementation for XYG3-type doubly hybrid density functionals (xDHs), we benchmarked eight xDH@B3LYP variants and xDH-PBE0 for NMR shielding constants and chemical shifts against the curated NS372 data set of main-group species, taking CCSD(T)/pcSseg-3 as the reference. We compared the results with conventional density functional approximations and their current-dependent variants, MP2-based methods, as well as previously identified high-performing B2PLYP-type double hybrids. The results show that the xDH family yields the lowest overall mean absolute deviations (MADs). In particular, XYGJ-OS provides the best overall balance for both shielding constants and chemical shifts, with MADs of 2.40 and 1.77 ppm, respectively, and exhibits consistently strong performance across a range of nuclei. Across the xDH series, the balance between opposite-spin and same-spin PT2 contributions helps rationalize the systematic shift from deshielding to overshielding behavior. Overall, this study demonstrates that xDHs are highly promising for quantitative prediction of NMR parameters and provides useful guidance for the future development of density functionals targeting magnetic-response properties.
Lai et al. (Sat,) studied this question.