Critical analysis of fragment-orbital DFT schemes for the calculation of electronic coupling values

TL;DR

This paper critically evaluates different formulations of fragment-orbital DFT schemes for calculating electronic couplings, introducing a new charged-fragment variant that improves accuracy on test sets.

Contribution

The authors present a new FO-DFT variant using charged fragments and systematically compare its performance with existing schemes for electronic coupling calculations.

Findings

New charged-fragment FO-DFT improves accuracy for HAB11

Hybrid functionals enhance electronic coupling estimates

Guidelines for selecting optimal FO-DFT variants based on task

Abstract

We present a critical analysis of the popular fragment-orbital density-functional theory (FO-DFT) scheme for the calculation of electronic coupling values. We discuss the characteristics of different possible formulations or 'flavors' of the scheme which differ by the number of electrons in the calculation of the fragments and the construction of the Hamiltonian. In addition to two previously described variants based on neutral fragments, we present a third version taking a different route to the approximate diabatic state by explicitly considering charged fragments. In applying these FO-DFT flavors to the two molecular test sets HAB7 (electron transfer) and HAB11 (hole transfer) we find that our new scheme gives improved electronic couplings for HAB7 (-6.2% decrease in mean relative signed error) and greatly improved electronic couplings for HAB11 (-15.3% decrease in mean relative signed error). A systematic investigation of the influence of exact exchange on the electronic coupling values shows that the use of hybrid functionals in FO-DFT calculations improves the electronic couplings, giving values close to or even better than more sophisticated constrained DFT calculations. Comparing the accuracy and computational cost of each variant we devise simple rules to choose the best possible flavor depending on the task. For accuracy, our new scheme with charged-fragment calculations performs best, while numerically more efficient at reasonable accuracy is the variant with neutral fragments.