Accurate prediction of electronic properties of open-shell transition metal complexes is essential for materials design and catalysis. Nevertheless, the properties that make these materials and molecules so compelling also make them extremely challenging to study accurately with any computational model. Although density functional theory (DFT) remains the method of choice for its balance of speed and accuracy in computational screening, semi-local approximations in DFT, such as the generalized gradient approximation (GGA), suffer from many electron self-interaction errors that causes them to predict erroneous spin states and geometries, barrier heights and dissociation energies, and orbital energies, to name a few. I will outline our recent efforts to both understand and correct these errors with a focus on predictive modeling of transition metal chemistry: i) We describe how common approximations to recover the derivative discontinuity (i.e., DFT+U and global or range separated hybrids) affect the density properties of transition metal complexes and correlated solids in a counterintuitive manner with respect to exact references, ii) We demonstrate recovery of the flat-plane condition that is a union of the requirement of piecewise linearity with electron removal or addition as well as unchanged energy when changing the spin of an electron in isoenergetic orbitals. We accomplish this at no computational cost over semi-local DFT by building from scratch our judiciously-modified DFT (jmDFT) functionals designed to oppose errors inherent in semi-local functionals. We show the connection to but divergence of these functional forms from hybrid expressions and standard DFT+U. This helps to explain why both common approximations increase static correlation errors. We also present fundamental expressions for determining these parameters, mitigating any empiricism in the method. iii) Finally, time permitting, I will describe our efforts to overcome DFT inaccuracies through the development of data-driven structure-method relationships, including in artificial neural networks that can predict sensitivity of spin-state ordering in transition metal complexes to changes in the exchange-correlation functional.

Recovering exact conditions at semi-local DFT cost to mitigate energy and density errors for transition metal chemistry

256th National Meeting (2018)

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