Reducing the Computational Complexity of Electron Correlation Techniques: FinalPhase I SBIR Report to AFOSR

摘要

Quantitative prediction of the structure, properties, and chemical reactivity ofmolecules by computer programs using first principles electronic structure methods is a valuable complement to experimental studies, because the latter are often difficult or impossible for transient reactive species. Unfortunately, the electronic structure methods which achieve quantitative chemical accuracy, such as those based on coupled cluster theory, are currently limited in applicability to small molecules of less than about ten first row atoms. This is because of the unphysical increase in computational cost with molecular size in standard implementations. This Phase I SBIR research has developed a new from the ground up approach to reformulating these proven electron correlation methods in terms of atom centered spatially localized quantities. The feasibility and formal advantages of the new formulation has been demonstrated in preliminary work based on the simplest electron correlation method, second order perturbation theory. Trial calculations on medium sized molecules indicated that recovery of both absolute electron correlation energy, and also recovery of relative energies across potential surfaces is excellent. The results have shown the potential of these new local correlation methods to allow the study of much larger molecules than can be treated conventionally, with roughly an order of magnitude improvement being possible.