Different results in DOSCAR file when NUPDOWN flag is used

[ledssiul]

I was doing maybe the most simple test. One isolated Rh atom using a simple INCAR, a big box, GGA, and just Gamma point.

ISPIN = 2
VOSKOWN = 1
ISMEAR = 0 (Gaussian)
SIGMA = 0.01
LORBIT = 10

As expected, everything seems to be all right and well converged....
BUT when I set up

NUPDOWN = 3

flag in the INCAR file I got clear discrepancies in the DOSCAR file. The integrated DOS(up) and integrated DOS(dwn) seem to be wrong. In the DOSCAR file I am getting 16 electrons and of course also the height of the discrete spectrum (DOS) change. Well, the location of the picks are correct. Taking a look in the OUTCAR file, in both cases the number of electrons is equal to 9 and the magnetization equal to 3 as expected, also, the occupation for each band is 1.

Here an small portion of the wrong DOSCAR file around the fermi energy,

E_Fermi = -5.20709742
...
-5.320 0.0000E+00 0.0000E+00 0.1000E+02 0.0000E+00
-5.293 0.0000E+00 0.3518E-04 0.1000E+02 0.9333E-06
-5.267 0.0000E+00 0.1949E+02 0.1000E+02 0.5172E+00
-5.240 0.0000E+00 0.2047E+03 0.1000E+02 0.5949E+01
-5.214 0.0000E+00 0.1918E+01 0.1000E+02 0.6000E+01
-5.187 0.0000E+00 0.9359E-07 0.1000E+02 0.6000E+01
-5.161 0.0000E+00 0.0000E+00 0.1000E+02 0.6000E+01
...

Of course, if I remove the NUPDOWN variable I recover the correct number of electrons, intDOS(up) = 6, intDOS(dwn)=3. Maybe it is important to say that I also try an additional "static" run using NSW = 0 and ISTART = 1 to calculate DOSCAR, even when I think it is not necessary to do it because the FFT-grid has not changed. I also try Rh_1, Rh_2, Co_1, and Co_2 with the same wrong results.

Here my question is, is there any dependence of the DOSCAR file with respect to NUPDOWN? Maybe something missing?

I am using version vasp.5.2.2

Thank you very much in advance for any clarification at this respect.

Luis

PS. I try to do an intensive search in the forum but I did not find any answer. I apologize in case some one else answer this question before.

[admin]

if NUPDOWN is set,
1) the Fermi level is calculated from the electrons which are set to spin down explicitely (and it is not adjusted to the spin-up window again)
2) the weight factors in the integrated DOS are multiplied by 2.
However, for a free Rh atom (as you report)
-- the results are basically thes same, independent of whether NUPDOWN is set (proveided that you converge to the electronic
ground state of the free atom, following Hund's rune)
-- free atoms have a discrete line sprectrum, not a DOS. Therefore it is more appropriate to check the KS-levels.

--

[ledssiul]

Perfect, clear!

Thanks a lot!
<span class='smallblacktext'>[ Edited Thu May 27 2010, 09:10PM ]</span>