Variational versus perturbational treatment of spin-orbit coupling in relativistic density functional calculations of electronic g factors: Effects from spin-polarization and exact exchange

Different approaches are compared for relativistic calculations of electronic g factors of molecules with light atoms, transition metal complexes, and selected complexes with actinides, using density functional theory (DFT) and Hartree-Fock (HF) theory. The comparison includes functionals with range-separated exchange. Within the variationally stable zeroth-order regular approximation (ZORA) relativistic framework, g factors are obtained with a linear response (LR) method where spin-orbit (SO) coupling is treated as a linear perturbation, a spin-polarized approach based on magnetic anisotropy (MA) that includes SO coupling variationally, and a quasi-restricted variational SO method previously devised by van Lenthe, van der Avoird, and Wormer (LWA). The MA and LWA approaches were implemented in the open-source NWChem quantum chemistry package. We address the importance of electron correlation (DFT vs HF), the importance of including spin polarization in the g tensor methodology, the question of whether the use of nonrelativistic spin density functionals is adequate for such calculations, and the importance of treating spin-orbit coupling beyond first-order. For selected systems, the extent of the DFT delocalization error is explicitly investigated via calculations of the energy as a function of fractional electron numbers. For a test set of small molecules with light main group atoms, all levels of calculation perform adequately as long as there is no energetic near-degeneracy among occupied and unoccupied orbitals. The interplay between different factors determining the accuracy of calculated g factors becomes more complex for systems with heavy elements such as third row transition metals and actinides. The MA approach is shown to perform acceptably well for a wide range of scenarios.