W1-S_N2-BH: A Large-Scale CCSD(T)/CBS Kinetic Database

We present the W1-S_N2-BH database, a large-scale kinetic benchmark set of all-electron CCSD(T)/CBS reference barrier heights for 1881 nucleophilic substitution (S_N2) reactions calculated via the high-level W1w theory. The data set covers the chemical space spanned by five nucleophiles/leaving groups (NH₂^-, OH^-, F^-, PH₂^-, SH^-) and ten substituent R-groups (H, Me, C≡CH, NH₂, CN, OH, F, SH, Cl, Br), generating a comprehensive database that encompasses a wide range of structural and electronic configurations. The resulting barrier heights span a broad energetic range, providing a rigorous stress test for approximate methods across rapid, intermediate, and hindered kinetic regimes. We use this database to benchmark 40 density functional theory (DFT) and double-hybrid DFT methods, as well as 19 D4-corrected methods. The range-separated double-hybrid ωB97M-2 delivers exceptional overall performance (MAD = 1.19 kcal mol^-1), followed by the range-separated hybrid-meta-GGA ωB97M-V (MAD = 1.84 kcal mol^-1) and the range-separated hybrid-GGA CAM-B3LYP-D4 (MAD = 2.03 kcal mol^-1), all of which exhibit well-behaved error distributions. The deep-learning functional Skala (MAD = 4.05 kcal mol^-1) outperforms all conventional local functionals, demonstrating the potential of machine-learned DFT without exact exchange. Finally, we show that empirical dispersion corrections systematically deteriorate performance for methods that already underestimate barrier heights, as dispersion forces stabilize the compact transition structures more than the free reactants. The W1-S_N2-BH database establishes a rigorous, highly diverse, and accurate benchmark set of reaction barrier heights for the development and validation of next-generation density functionals and machine-learning potentials.