The invariance of a wavefunction with respect to unitary transformation of inactive, active or virtual canonical molecular orbitals (CMOs) makes it possible to produce localized molecular orbitals (LMOs) that are more compact than the CMOs. The LMOs are utilized to improve the performance and scaling of many quantum chemistry methods, from effective fragment potentials (EFP) to multi-reference singles and doubles configuration interaction (MRSDCI). While the localization of the occupied MOs usually does not present any difficulty, in the space of the virtual MOs traditional localization functionals exhibit multiple irrelevant extrema, hindering the localization. In this work, we present the implementation of two virtual MO localization methods: (i) the projection method based on the Subotnik and Head-Gordon approach [1]; (ii) the direct minimization of orbital variance, based on the approach of Jansik et. al.[2]. The analysis of the computational efficiency of the methods on different computer architectures is performed, the degree of LMO localization is compared, and the suitability of the generated LMOs for MRSDCI and EFP methods is investigated.[1] Subotnik et.al., JCP 123, 114108 (2005)[2] Jansik et.al., JCP 134, 194104 (2011)

Localization of virtual molecular orbitals: Projection vs. direct optimization

247th National Meeting (2014)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.