Differnet NPAR gives different ground states

[mgverg]

Hi!

I am running vasp5.4.1 for the following POSCAR and PBE POTCAR

VOS
1.0
6.2610000000000001 0.0000000000000000 0.0000000000000000
0.0000000000000000 6.2610000000000001 0.0000000000000000
0.0000000000000000 0.0000000000000000 4.1010000000000000
V O S
2 10 2
Direct
0.0000000000000000 0.5000000000000000 0.6176000000000000
0.5000000000000000 0.0000000000000000 0.3824000000000000
0.0000000000000000 0.5000000000000000 0.2200000000000000
0.5000000000000000 0.0000000000000000 0.7800000000000000
0.1830000000000001 0.9620000000000000 0.2850000000000000
0.8169999999999999 0.0380000000000000 0.2850000000000000
0.6830000000000001 0.4620000000000000 0.7150000000000001
0.3169999999999999 0.5380000000000000 0.7150000000000001
0.9620000000000000 0.8169999999999999 0.7150000000000001
0.0380000000000000 0.1830000000000001 0.7150000000000001
0.4620000000000000 0.3169999999999999 0.2850000000000000
0.5380000000000000 0.6830000000000001 0.2850000000000000
0.5000000000000000 0.5000000000000000 0.5000000000000000
0.0000000000000000 0.0000000000000000 0.5000000000000000



And I get different ground states depending on the NPAR. If I use NPAR=4 I get a ferromagnetic phase, but if I use NPAR=6 I get non-magnetic results. The rest of the input files are the same. How can this be possible? And which ground state is the correct one?

[henrique_miranda]

Could you please also provide the INCAR + the identifiers of the POTCARS you used as well as the KPOINTS file?

Kind regards,
Henrique Miranda

[mgverg]

This is the INCAR file:

ENCUT = 550
PREC = High
EDIFF= 0.1E-03
ISTART = 0 ; ICHARG = 2
LSORBIT = .TRUE.
SAXIS= 0 0 1
MAGMOM= 900*0.0
#LREAL= Auto
LCHARG= .False.
NBANDS = 164
#
#LORBIT=11
LMAXMIX=4
#
NPAR=4

POTCARS:

PAW_PBE V 08Apr2002
5.00000000000000000
parameters from PSCTR are:
VRHFIN =V: p6 d4 s1
LEXCH = PE

PAW_PBE O 08Apr2002
6.00000000000000000
parameters from PSCTR are:
VRHFIN =O: s2p4
LEXCH = PE

PAW_PBE S 17Jan2003
6.00000000000000000
parameters from PSCTR are:
VRHFIN =S : s2p4
LEXCH = PE

KPOINTS:

KPOINTS
0
Gamma
11 11 11

[alex]

Hi,

please also check if you have a different number of bands (NBANDS). This might yield your different groundstates in the end, because the iteration becomes a slightly different drift. However, the one with the lowest energy wins. ;-)

Cheers,

Alex

[mgverg]

Hi,

I do have different number of bands. However even if I fix the same number of bands for all the different NPAR I get different ground states. Moreover, every time I run a job I need to test it with different NPARs to make sure I get the correct ground state? How can I be sure I have the correct ground state.

Thanks

[henrique_miranda]

Indeed, the effective number of bands in our calculation might be changed by VASP depending on the value of NPAR.
The number of bands should be divisible by NPAR.
So in your case you have:
NBANDS = 164
with NPAR = 4 you get 164/4=41 bands/core which is ok
with NPAR = 6 you get 164/6=27.33333 bands/core which means VASP internally will use NBANDS=168 from (28*6)

If you set NBANDS=168 and redo the two runs with NPAR=4 and NPAR=6 do you still get different magnetic ground-states?
Can you attach the two OUTCARS?

Kind regards,
Henrique Miranda

[mgverg]

If I use NBANDS = 168 I get 0 magnetic moment for all NPARs (4, 6 and 8 ). However, how can I know this in advance?
I am not running one compound, I am doing a massive run.

If I use NPAR=4 and NBANDS=100 I get a magnetic moment and the total energy is lower, so this is the real ground state.

So what do I do?

[henrique_miranda]

Ok, so the difference in magnetization you where having might have been due to a different number of bands being used in your calculation.

I find it strange that you get lower energy when using a lower number of bands (100 vs 168).
Can you attach the OUTCAR files of these two runs?