| Résumé | In this article, we examine the ability of the exchange-hole dipole moment (XDM) model of dispersion to treat large supramolecular systems. We benchmark several XDM-corrected functionals on the S12L set proposed by Grimme, which comprises large dispersion-bound host-guest systems, for which back-corrected experimental and Quantum Monte Carlo (QMC) reference data are available. PBE-XDM coupled with the relatively economical and efficient pc-2-spd basis set gives excellent statistics (mean absolute error (MAE) = 1.5 kcal/mol), below the deviation between experimental and QMC data. When compared only to the (more accurate) QMC results, PBE-XDM/pc-2-spd (MAE = 1.2 kcal/mol) outperforms all other dispersion-corrected DFT results in the literature, including PBE-dDsC/QZ4P (6.2 kcal/mol), PBE-NL/def2-QZVP (4.7 kcal/mol), PBE-D2/def2-QZVP′ (3.5 kcal/mol), PBE-D3/def2-QZVP′(2.3 kcal/mol), M06-L/def2-QZVP (1.9 kcal/mol), and PBE-MBD (1.8 kcal/mol), with no significant bias (mean error (ME) = 0.04 kcal/mol). PBE-XDM/pc-2-spd gives binding energies relatively close to the complete basis-set limit and does not necessitate the use of counterpoise corrections, which facilitates its use. The dipole-quadrupole and quadrupole-quadrupole pairwise dispersion terms (C<inf>8</inf> and C<inf>10</inf>) are critical for the correct description of the dimers. XDM-corrected functionals different from PBE that work well for small dimers do not yield good accuracy for the large supramolecular systems in the S12L, presenting errors that scale linearly with the dispersion contribution to the binding energy. |
|---|
