Reference Energies for Intramolecular Charge-Transfer Excitations

With the aim of completing our previous efforts devoted to local and Rydberg transitions in organic compounds, we provide a series of highly accurate vertical transition energies for intramolecular charge-transfer transitions occurring in (π-conjugated) molecular compounds. To this end, we apply a composite protocol consisting of linear-response CCSDT excitation energies determined with Dunning's double-ζ basis set corrected by CC3/CCSDT-3 energies obtained with the corresponding triple-ζ basis. Further basis set corrections (up to aug-cc-pVQZ) are obtained at the CCSD and CC2 levels. We report 30 transitions obtained in 17 compounds (aminobenzonitrile, aniline, azulene, benzonitrile, benzothiadiazole, dimethylaminobenzonitrile, dimethylaniline, dipeptide, β-dipeptide, hydrogen chloride, nitroaniline, nitrobenzene, nitrodimethylaniline, nitropyridine N-oxide, N-phenylpyrrole, phthalazine, and quinoxaline]. These reference values are then used to benchmark a series of wave functions [CIS(D), SOPPA, RPA(D), EOM-MP2, CC2, CCSD, CCSD(T)(a)*, CCSDR(3), CCSDT-3, CC3, ADC(2), ADC(3), and ADC(2.5)], the Green's function-based Bethe-Salpeter equation (BSE) formalism performed on top of the partially self-consistent evGW scheme considering two different starting points (BSE/evGW@HF and BSE/evGW@PBE0), and time-dependent density-functional theory (TD-DFT) combined with several exchange-correlation functionals (B3LYP, PBE0, M06-2X, CAM-B3LYP, LC-ωHPBE, ωB97X, ωB97X-D, and M11). It turns out that the CC methods including triples, namely, CCSD(T)(a)*, CCSDR(3), CCSDT-3, and CC3, provide rather small average deviations (≤0.10 eV), with CC3 emerging as the only chemically accurate approach. ADC(2.5) also performs nicely with a mean absolute error of 0.11 eV for a O(N⁶) formal scaling, whereas CC2 and BSE/evGW@PBE0 also deliver very satisfying results given their respective O(N⁵) and O(N⁴) computational scalings. In the TD-DFT context, the best performing functional is ωB97X-D, closely followed by CAM-B3LYP and M06-2X, all providing mean absolute errors around 0.15 eV relative to the theoretical best estimates.