New computationally efficient quantum chemical models that capture static and dynamic correlation separately (2020)
Type of ContentTheses / Dissertations
Degree NameDoctor of Philosophy
PublisherUniversity of Canterbury
AuthorsStinson, Chrisshow all
This thesis presents a new family of computationally-efficient quantum chemical methods designed to capture static and dynamic correlation energies separately without double counting. Statically correlated systems require more than one Slater determinant to qualitatively describe the electronic wavefunction. A major challenge for modelling statically correlated systems is finding molecular orbitals that are simultaneously suitable to describe all relevant electronic configurations. Conventional multi-reference SCF methods overcome this problem by simultaneously optimizing molecular orbital and configuration interaction coefficients, but this procedure is computationally intensive and selecting the relevant configurations and orbitals is far from straightforward. Spin-flip non-orthogonal configuration interaction methods allow these processes to be decoupled by generating semi-optimized orbitals for excited-state configurations using high-spin reference Hartree-Fock calculations. The key insight of this thesis is that these expansions can be severely truncated in a physically-motivated manner to provide minimal determinant models that capture only static correlation, and to which a very simple second-order perturbation theory correction can be applied to recover the remaining dynamic correlation energy. Our minimal-determinant SF-NOCI and SF-NOCI-PT2 methods are applied to two simple model problems in which static correlation effects are important - dissociating LiH and twisting ethylene. SF-NOCI gives qualitatively-correct wavefunctions while SF-NOCI-PT2 energies show close agreement with experiment for ground-state dissociation energies and torsional barrier heights, and limited agreement for excited- state transition energies. This demonstrates the importance of optimizing orbitals for excited determinants. Even approximately optimized orbitals can substantially improve model accuracy while reducing complexity and computational cost. Accounting for dynamic and static correlation effects separately allows efficient models to be developed or used for both.