A comparison of models for calculating nuclear magnetic resonance shielding tensors

The direct ~recomputation of two-electron integrals! implementation of the gauge-including atomic orbital ~GIAO! and the CSGT ~continuous set of gauge transformations! methods for calculating nuclear magnetic shielding tensors at both the Hartree-Fock and density functional levels of theory are presented. Isotropic C, N, and O magnetic shielding constants for several molecules, including taxol ~C47H51NO14 using 1032 basis functions! are reported. Shielding tensor components determined using the GIAO and CSGT methods are found to converge to the same value at sufficiently large basis sets; however, GIAO shielding tensor components for atoms other than carbon are found to converge faster with respect to basis set size than those determined using the CSGT method for both Hartree-Fock and DFT. For molecules where electron correlation effects are significant, shielding constants determined using ~gradient-corrected! pure DFT or hybrid methods ~including a mixture of Hartree-Fock exchange and DFT exchange-correlation! are closer to experiment than those determined at the Hartree-Fock level of theory. For the series of molecules studied here, the RMS error for C chemical shifts relative to TMS determined using the B3LYP hybrid functional with the 6-3111G(2d ,p) basis is nearly three times smaller than the RMS error for shifts determined using Hartree-Fock at this same basis. Hartree-Fock C chemical shifts calculated using the 6-31G* basis set give nearly the same RMS error as compared to experiment as chemical shifts obtained using Hartree-Fock with the bigger 6-3111G(2d ,p) basis set for the range of molecules studied here. The RMS error for chemical shifts relative to TMS calculated at the Hartree-Fock 6-31G* level of theory for taxol ~C47H51NO14) is 6.4 ppm, indicating that for large systems, this level of theory is sufficient to determine accurate C chemical shifts. © 1996 American Institute of Physics. @S0021-9606~96!01914-X#