Last band in my calculation is occupied

[hipertrofia]

I am doing a simulation of a solvated molecule. The number of electrons is 1078 and the calculation is non spin polarized, so the number of occupied bands should be exactly 539 with 2 electrons in each since the system is closed shell. I have checked that the geometries are reasonable etc. I am seeing very strange behavior, from the list band occupancies:

Code: Select all

.
.
.
    535      -3.1926	  2.00000
    536      -3.1896	  2.00000
    537      -2.0266	  1.93484
    538      -1.9933	  1.81429
    539      -1.8666	  0.25087
    540       0.9179	  0.00000
    541       1.0167	  0.00000
.
.
.
    739       7.6158	  0.00000
    740       7.6457	  0.00000
    741       7.6730	  0.00000
    742       1.7514	  0.00000
    743       6.3634	  0.00000
    744      -5.5186	  2.00000

As you can see, the last 3 bands are out of order and this messes up the occupancies of the other bands as well. I am using the ALGO = fast tag. This is my INCAR:

Code: Select all

ISTART = 0
PREC = Normal
ENCUT = 600
ISMEAR = -1 ; SIGMA = 0.030
ISYM = 0
IBRION = 2
ISIF = 2
NSW = 1
EDIFF = 1.E-05
LCHARG = .FALSE.
LWAVE = .FALSE.
LPLANE = .FALSE.
NPAR = 8
LSCALU = .FALSE.
ALGO = fast
ISPIN = 1

Could someone please explain to me what I'm doing wrong to induce this strange behavior?

[Neutrino]

Hi hipertrofia,

This looks like the effect of smearing and is not unexpected. In a self-consistent calculation (no relaxation or MD), you can get rid of it by ISMEAR=-5.
For insulators and semiconductors (where water falls within), ISMEAR=0 (Gaussian) is recommended. In general smearing will disrupt the occupancy a bit but it is also needed to accelerate the convergence of the electronic structure.


Best,
M.

[hipertrofia]

Thanks for your reply. I do not think the smearing is the cause of the problem, see below.

My point is that there should not be fractional occupancies even with large smearing, because band 539 should be the highest occupied one and should be separated by a large gap from the next one. Band number 744 seems to be spurious and the mere fact that its energy is not in increasing order with respect to the previous band(s) is already suspicious.

I have done further testing and verified I only experience this problem for high basis set cutoff energies and when using the RMM-DIIS algorithm. With the Davidson solver this problem does not show. Also, with RMM-DIIS plus low cutoffs (e.g. 300 eV) this problem does not show.

In the description of the RMM-DIIS solver in the VASP wiki (http://cms.mpi.univie.ac.at/wiki/index. ... O#RMM-DIIS) there is a mention to the ENINI keyword and how it's defaulted to ENINI = ENCUT. I might try that, although for the time being changing to the Davidson iterator should do.