Polarization of core orbitals and computation of nuclear quadrupole coupling constants using Gaussian basis sets

Most standard Gaussian basis sets for first row atoms, even large sets designed to converge on a 'complete basis set' limit, systematically overestimate the electric field gradient at nuclear sites for first row atoms, resulting in errors of up to 15% in the computation of nuclear quadrupole coupling constants. This error results from a failure to include tight d functions, which permit the core 1s orbitals to distort under the influence of the field of the nuclear quadrupole. Augmentation of standard basis sets with a single set of single-exponent d functions, matched to the reciprocal square of the nominal 1s radius, reduces these errors by up to 90%.