Trouble with hybrid functional calculations using Ferwe=-2

[Kousei]

Dear All,

I am currently trying to perform a structural optimization of the excited state of CaF2:Eu2+ using the FERWE tag.
Here is the INCAR of the calculation.

Code: Select all

System = CaF2_Eu2+
ENCUT = 500
LHFCALC = .TRUE.
HFSCREEN = 0.2
ALGO = All
TIME = 0.4
FERWE = 390*1.0 1*0.0 1*1.0 58*0.0 390*1.0 1*0.0 1*1.0 58*0.0 390*1.0 1*0.0 1*1.0 58*0.0 390*1.0 1*0.0 1*1.0 58*0.0
FERDO = 384*1.0 66*0.0 384*1.0 66*0.0 384*1.0 66*0.0 384*1.0 66*0.0
IBRION = 2
ICHARG = 2
ISMEAR = -2
ISTART = 1
ISPIN = 2
ISIF = 2
LORBIT = 11
MAGMOM = 1*7.0 323*0
NUPDOWN = 7
NBANDS = 450
NSW = 50
NPAR = 1
SIGMA = 0.1

The calculation completes without any error messages, but I have identified a problem where the occupancy that should have been specified in the FERWE tag does not hold until the end of the calculation. I would like to know if there is a possible cause for this.

Many thanks in advance,
Kousei

[martin.schlipf]

Did you try whether this problem is specific to your system or holds in general? FERWE should not be updated differently for normal DFT calculations, as far as I know, so it seems you should be able to reproduce this with a simpler system. It is generally a good idea to try the feature that you want in a small system to avoid wasting computational resources if something goes wrong.

For the future, please also remember that we need a complete set of input files to reproduce the issue.

[Kousei]

Thank you for your comment.
I followed your suggestion and did the calculation in the unit cell, but still occured the same problem.
I have attached the files and would appreciate your advice.

[martin.schlipf]

Thank you for this input. I could reproduce the issue. It seems that there is a workaround by using ALGO = Normal instead, if that works for your system.

[Kousei]

Thank you very much.
By changing the ALGO to Normal, the calculation converged well.

[martin.schlipf]

We have analyzed this in a bit more detail and updated the wiki entries for ISMEAR, LDIAG, and ALGO.

The bottom line is that you should use ALGO = VeryFast or ALGO = Damped to avoid bands changing the order. In either case you should set LDIAG = F to avoid a diagonalization step that can also change the order of the bands.