Improving the transferability of density functional theory predictions through correlation analysis: Structural and energetic properties of NiX alloys (X=C, Si, Ge, and Sn)

Abstract

Listen

Translate article

This work reports on the performance of density functional theory (DFT) for a series of single and binary systems, aiming for a quantitative description of $\mathrm{Ni}X$ ($X=\mathrm{C}$, Si, Ge, and Sn) alloys. Both semilocal GGA and a meta-GGA density functional, with and without dispersion corrections, are tested. We found in our study that no single functional simultaneously provides an accurate quantitative description of the investigated structural and energetic properties. However, the spread in computed DFT data could be rationalized in terms of the distribution of reduced density gradients and differences in the evolution of the exchange enhancement factors for different functionals. We demonstrate how to construct a regression model based on data from several density functionals that increases the predictivity of semilocal DFT. We foresee that the use of regression models (or extensions of it) can be valuable in the development of more accurate density functionals that in the future could provide a quantitative accuracy for complex multicomponent systems.