11
$\begingroup$

I tried to relax a 4x4x1 supercell of ferromagnetic monolayer material using the default settings (ALGO = Normal) but it didn't converge. So, I switched to ALGO = Fast and the results are converging normally now. Is this setting safe? Will this affect the accuracy of the results? The used INCAR file is below :

ENCUT = 600 eV
PREC = Accurate
LREAL = Auto
EDIFFG = -0.001
EDIFF = 1E-8 
LCHARG = .FALSE.
LWAVE = .FALSE.
ISMEAR = 0
ADDGRID = .TRUE.
SIGMA = 0.03
NSW = 299
IBRION = 2
ISIF = 3
ISPIN = 2
MAGMOM = 16*2.0 32*0.0
ALGO = Fast # This was Normal before editing
#Mixer
AMIX     = 0.2
BMIX     = 0.00001
AMIX_MAG = 0.8
BMIX_MAG = 0.00001

LASPH      = .TRUE.

NCORE = 2
$\endgroup$
11
$\begingroup$

Changing ALGO should make no difference in an ideal world. However, when you invoke spin polarization, you may find a different magnetic state from both algorithms. The best practice would be to ensure that you converge to the right solution.

That being said, the NORMAL algo is normally more robust than Fast. This in general might be a bad sign for your system. You can also try the ALL algo and see what that gives.

I see you have also added an incar, here is some general advice that might influence convergence.

  • ADDGRID is a spooky keyword, I would say never use it but sometimes it helps. I suggest leaving it off, convergence issues tend not to be fixed by it in general in these magnetic systems
  • You are using a small sigma value, I would suggest using 0.2 and reduce it to your desired value after convergence. It will not influence geometry much but will make convergence much easier.
  • Leave the mixing settings at their defaults most of the time. You can try this approach as a first attempt, but if it doesn't fix the problem do not keep it.
  • Consider running a spin paired calculation first as a single point calculation, save the WAVECAR/CHGCAR, then add magnetization. This often helps as well.
  • EDIFF = 1e-8 is insanely accurate, use something more like 1e-4 or 1e-5 for geometry optimization. If you find that you cannot converge the geometry you can raise it or switch to a VTST geometry optimizer which uses forces which are less sensitive to this.
$\endgroup$
4
  • $\begingroup$ Good point about the magnetic moments. That's a real subtlety. $\endgroup$ – Andrew Rosen Jan 15 at 18:57
  • 1
    $\begingroup$ It has given me many angry noises at my desk in the past few days $\endgroup$ – Tristan Maxson Jan 15 at 19:02
  • $\begingroup$ I have added my iNCAR file could you please have a look at it? Don't you think that the mixing tags are the source of the problem? $\endgroup$ – Chi Kou Jan 15 at 19:04
  • 1
    $\begingroup$ @ChiKou I have added some additional advice based on prior experience. Maybe some can be applied. $\endgroup$ – Tristan Maxson Jan 15 at 19:14

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.