Accurate bandgaps of photovoltaic kesterites from first-principles DFT+U

TL;DR

This paper demonstrates that a parameter-free DFT+U approach accurately predicts the bandgaps of photovoltaic kesterites, outperforming some advanced methods and providing insights into defect effects with low computational cost.

Contribution

It introduces a DFT+U method with linear response evaluated parameters that reliably predicts bandgaps of kesterite materials without empirical tuning.

Findings

DFT+U marginally outperforms GW in bandgap prediction.

Correcting all atomic subspaces improves results over traditional methods.

Hund's J corrections via DFT+U+J worsen predictions, but BLOR functional restores accuracy.

Abstract

Streamlined prediction of the electronic properties of photoactive materials warrants a Density Functional Theory (DFT) based approach that (i) yields reliable bandgaps, (ii) is free of empirically tuned parameters, and (iii) exhibits low computational overhead. Here we show that for Cu2ZnSnS4 and Cu2ZnGeS4 kesterite photovoltaic materials, all three of these demands are met by the DFT plus Hubbard U technique (DFT+U) with corrective parameters evaluated via minimum-tracking linear response. The predicted bandgaps are found to even marginally outperform those from the self-consistent GW approach. Key to this method's success is the application of Hubbard U corrections to all atomic subspaces that dominate the conduction and valence band edges, as opposed to the conventional approach of correcting 3d and 4f atomic states. Intriguingly, the inclusion of Hund's J corrections via the extended DFT+U+J functional significantly worsens these results. This under performance can be ameliorated through the use of the Burgess-Linscott-O'Regan (BLOR) flat-plane based Hubbard U plus Hund's J functional, with bandgap predictions in close agreement with the conventional DFT+U method. The DFT+U method is also used to predict defect-induced changes to the bandgap and associated formation energies, in 1,728-atom supercells.