5
$\begingroup$

I have conducted HSEsol relaxation on a monolayer unit cell of 3 atoms using VASP, with the settings below :

NCORE = 2

# Self Consistent Calculation step:
ISTART = 0
ICHARG = 2

#Electronic Relaxation:
ENCUT = 600 eV
EDIFF = 1.0E-7
PREC = Accurate
LREAL = .FALSE.
ALGO = Normal
LCHARG = .FALSE.
LWAVE = .FALSE.

#Relaxation:
ISMEAR = 0
SIGMA = 0.05
IBRION = 2
ISIF = 3
NSW = 100
EDIFFG = -0.001
GGA COMPAT = .FALSE.
NBANDS = 24

# Magnetism:
ISPIN= 2
MAGMOM = 1*2.0 2*0.0
LASPH      = .TRUE.
LMAXMIX = 4
VOSKOWN = 1

#Mixer
AMIX     = 0.2
BMIX     = 0.00001
AMIX_MAG = 0.8
BMIX_MAG = 0.00001

#HSEsol:
GGA = PS
LHFCALC = .TRUE.
HFSCREEN = 0.2
PRECFOCK=Fast    

K-mesh
0
G
6 6 1

Do you think that these settings are accurate enough to continue with them ? Should I increase PRECFOCK to Normal or No ?

$\endgroup$
3
  • 1
    $\begingroup$ what do you wish to compute? Convergence tests are normally needed to be sure. That being said, ENCUT/KPOINTS are likely converged geometries. The rest is hard to say $\endgroup$ – Tristan Maxson Feb 2 at 8:06
  • $\begingroup$ I want to relax my structure for electronic and magnetic properties. Other than ENCUT and KPOINTS, do you that other tags are accurate enough ? $\endgroup$ – Chi Kou Feb 2 at 10:03
  • 1
    $\begingroup$ Most other tags look okay I think. I will provide a quick answer. $\endgroup$ – Tristan Maxson Feb 2 at 12:52
4
$\begingroup$

This is the common case of needing to do convergence tests, but lets estimate based on prior experience. Your ENCUT is probably high enough, 600 is almost surely greater than ENMAX depending on what elements you are using. Your KPOINTS are also sufficient for geometry relaxation at the HSEsol level in my experience, even for really small structures.

Obviously you will use a higher number of k points for the band structure since you need to trace the band path, so I won't address that as a lack of convergence. You will also need to increase k points for the density of states, probably. Without seeing a structure, I can say that there is a good chance it is worth the time to do the finest grid of k points you can afford to do. If this is doable for geometry relaxation, I strongly suspect that you can afford more time for the density of states.

Finally, EDIFF is probably overconverged. A value of 1.0E-5 is probably sufficient for a geometry relaxation, you can rerun with 1.0E-7 afterwards to make sure nothing significantly changes. I am also unsure that you need GGA_COMPAT, I think it defaults to false. The mixer settings should also probably not be changed unless you have an issue. Before trying those mixing settings, try ALGO=All first.

$\endgroup$
2
  • $\begingroup$ I totally agree with what you have said, but why ALGO=All should be tried first ? $\endgroup$ – Chi Kou Feb 2 at 13:05
  • 1
    $\begingroup$ It tends to be more stable for HSE06. It converges well for anything semiconductor/insulator like. I find it is competitive with Fast/Normal even for PBEsol calculations for semiconductors. $\endgroup$ – Tristan Maxson Feb 2 at 21:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.