Implementation change of the TB09 functional

[F. Fuchs]

Hello modelling experts,

I made a couple of DFT calculations using meta-GGA functional by Tran and Blaha. I used ATK 12.8 before and now switched to ATK 15. I realized that the results changed. Two example observations:
1) In order to achieve the same band gap in bulk silicon, I now need to use a different value for the c-parameter.
2) I did some two-probe simulations using version 12.8 and looked at the local density of states. Near the electrodes I observed some deviation from bulk properties in the electrodes (it was already bulk-like further away from the electrodes, so it was not a convergence problem with respect to the device length) . This problem however,  disappeared when using ATK 15 (which is very nice!).

My question is: What changed between ATK 12.8 and ATK 15 related to the Tran and Blaha functional? In the release of ATK 14 I read "New MGGA library and implementation - now also with TPSS functional" and I suppose that this causes the deviations. I would like to know some more details. How meaningful are the results calculated with the older implementation in ATK 12.8?

Thank you!

[Nordland]

Hey Fuchs.

There is a set of minor changes to make it more precise that has been implemented. The old result should still be valid, but an old c-parameter would have to be reevaluate to give new consistent results. There are one important changes that is the most likely cause of the different that you see.

• The MGGA contains the kinetic energy density which is straight forward to calculate for the  valence density (since we know the wave function), but for the nonlinear core correction which comes from the pseudo-potential we only have a density and hence there is no direct way to calculate the kinetic energy density for this. We tested various approximation for kinetic energy density for this nonlinear core-correction and picked the approximation that gave the best results.

[F. Fuchs]

Thank you very much for your helpful reply! Indeed, I achieve almost identical results with the 2015 version when using a different value for the c-parameter.
(Btw.: I observed a slight shift of the band structure with respect to the Fermi energy when looking at SiNW structures. This shift is however rather small; approx. 0.01 eV.)

Is there some more documentation of the used approximations, which you have mentioned, available? I couldn't find any, yet.