# HSE06 - dos + band structure calculation using VASP

I'm a little confused with the order and flags needed to start the calculation with HSE06. Below is a step-by-step guide to what I've done so far.

Obs: I use GGA

For DOS

• Step (I) - Structure relaxation (my, INCAR without HSE flags):
Lattice Relaxation
NSW    =  300          (number of ionic steps)
ISMEAR =  0            (gaussian smearing method )
SIGMA  =  0.2         (please check the width of the smearing)
IBRION =  2            (Algorithm: 0-MD, 1-Quasi-New, 2-CG)
ISIF   =  3            (optimize atomic coordinates and lattice parameters)
EDIFFG = -1.5E-02      (Ionic convergence, eV/A)
PREC   =  Accurate     (Precision level)


### Question: GGA = PE Is it necessary in this step (I) and also in the next step (II)?

• Step (II) - Run the SCF calculation to obtain the WAVECAR

• Step (III) - Run SCF for DOS + HSE:

• +1 but please see this: mattermodeling.meta.stackexchange.com/q/421/5. Commented Nov 7, 2023 at 19:53
• According to VASP documentation: It is strongly recommended to start from a converged PBE calculation (ISTART = 1) before beginning with a DFT+HF method. You will need to have well converged WAVECAR for your third step. Regarding GGA tag, whether you specifically define that or not, VASP will have a default value which in older versions of VASP (prior to version 5.4.4), the default GGA tag was PE. I think in the newer version, the functional is in accordance with the POTCAR file you are using. Commented Nov 8, 2023 at 1:44

Perhaps only the first of the following is actually relevant to your question, but the second part gives some explanation on why the steps make sense in that order.

1. Is the GGA = PE INCAR flag necessary for a PBE calculation to be performed?

Like the comment said - looking at the default of GGA on the VASP Wiki, it says

Default: GGA = exchange-correlation functional in accordance with the POTCAR file

So if you use PBE POTCARs, VASP should default to PBE. Though, it might be useful to just leave the GGA = PE if you ever need to recall what you've done in the future.

1. Explanation of the order of steps

HSE06 (a hybrid functional) is a lot more expensive than PBE but can give better bandgaps (PBE often underestimates the gap). PBE generally gives good structures. With this in mind,

Step (I) : It makes sense to first relax the geometry with PBE (which is a lot cheaper than HSE06).

Step (II): While it is possible that a PBE converged WAVECAR has already been generated at the end of the geometry relaxation, it is likely to be a bad starting point for the HSE06 calculation as the basis used for the relaxation (and output WAVECAR) corresponds to the unrelaxed initial structure of the geometry optimization, which could be very different from the basis of the relaxed structure if the cell vectors have changed significantly. Hence, it is advisable to redo a static PBE SCF before using the converged PBE orbitals for the HSE06 calculation (make sure to set LWAVE to true to get the WAVECAR). [edited to reflect E.I. Joe's comments]

The reason for getting a PBE converged WAVECAR as a starting point for the HSE06 SCF is that convergence for the HSE06 run should be faster and more stable from PBE converged orbitals. Of course, it is possible to do the HSE06 calculation from scratch, but it would likely take much longer and more prone to convergence issues than starting with PBE converged orbitals.

• "It is possible that a PBE converged WAVECAR has already been generated at the end of the geometry relaxation if you set LWAVE to true (see LWAVE)" I think this advice needs to be adjusted slightly. Especially if the cell lattice parameters have changed, the basis used for the relaxation (and output WAVECAR) will be different from the one used in the self-consistent calculation, including HSE. This just leads to a worse starting point for a minimal cost of one more self-consistent loop.
– user8097
Commented Dec 20, 2023 at 16:01
• @E.I.Joe Good point - thanks. The answer was adjusted accordingly. Commented Dec 20, 2023 at 22:05