Performance of density functional theory in computing nonresonant vibrational (hyper)polarizabilities

A set of exchange-correlation functionals, including BLYP, PBE0, B3LYP, BHandHLYP, CAM-B3LYP, LC-BLYP, and HSE, has been used to determine static and dynamic nonresonant (nuclear relaxation) vibrational (hyper)polarizabilities for a series of all-trans polymethineimine (PMI) oligomers containing up to eight monomer units. These functionals are assessed against reference values obtained using the Møller-Plesset second-order perturbation theory (MP2) and CCSD methods. For the smallest oligomer, CCSD(T) calculations confirm the choice of MP2 and CCSD as appropriate for assessing the density functionals. By and large, CAM-B3LYP is the most successful, because it is best for the nuclear relaxation contribution to the static linear polarizability, intensity-dependent refractive index second hyperpolarizability, static second hyperpolarizability, and is close to the best for the electro-optical Pockels effect first hyperpolarizability. However, none of the functionals perform satisfactorily for all the vibrational (hyper)polarizabilities studied. In fact, in the case of electric field-induced second harmonic generation all of them, as well as the Hartree-Fock approximation, yield the wrong sign. We have also found that the Pople 6-31+G(d) basis set is unreliable for computing nuclear relaxation (hyper)polarizabilities of PMI oligomers due to the spurious prediction of a nonplanar equilibrium geometry.