[Wannier] Wannier90 underestimating the band gap compared to my DFT result

[Pascal Salzbrenner]

Dear all,

I am trying to generate a wannier90 tight-binding model for CdTe (using wannier90-1.2 for the initial post postprocessing step and wannier90-3.1 for the wannierisation). In general, it reproduces the behaviour of the bands well compared to my DFT (VASP) result, but the conduction band is, in many places, somewhat lower using the wannier90 model than it is when generating the band structure directly from DFT. In particular, at the Gamma point, the band gap is underestimated by around 0.1 eV.
I have tried using a variety of different projectors (even projector sets of different sizes, with between 16 and 24 projectors per unit cell). The combinations I have used are composed of atomic s- (with r=1 as well as r=2) and p-orbitals, as well as different hybridisations. I have also tried adding orbitals using the "random" keyword. The attached figure was generated by projecting on the atomic sp2 and pz orbitals on both Cd and Te. At the Gamma point, the conduction band is highly s-like in character, but adding s-orbitals to these projectors does not improve the fit around the Gamma point, and it worsens it in other regions of the BZ.
I have tried excluding the higher unoccupied bands during the projection using the "exclude_bands" tag, and I have tuned the frozen window. In all these calculations, the disagreement at the gamma point persists. Does anyone have any idea what I am doing wrong, or what I could try to get the two methods to agree at this point? The "wannier90.win“ file, as well as the output band structure for the projection of the sp2 and pz orbitals (which are representative of the calculations I tried) can be found here: https://drive.google.com/open?id=15uGvHXHVFY_iq-LDOtTeK5JHkrun6jJ6. I’d be grateful for any help.

Thanks in advance!

Pascal Salzbrenner
MPhil Student
Monserrat Group
Theory of Condensed Matter, University of Cambridge