Charged supercell calculations

[Martin]

Dear ABINITers,

I tried to perform charged (super)cell calculations, but I end up with the following error msg

m_ebands.F90:1460:ERROR
Too large difference in no. of electrons:,
Expected= 761.00 Calculated= 0.00

I have attached the input and log files. Guess I used the nband/oocopt/occ/charge parameters in an inconsistent way.

Thanks in advance!

Best regards,

Martin Haeufel
TU Munich, WSI (T33)
http://www.wsi.tum.de

[Robin]

Dear Martin,

Does this error exist in a neutral system?

Sincerely,
Guangfu Luo

[Martin]

Dear Guangfu Luo,

thanks for the immediate reply.

No, I do not get any errors for the neutral cell. In the neutral case I only set nband; there is no need to set occopt, occ, or charge in that case.

Question: What additional input variables do I have to set, when switching from a neutral cell to a (single positive) charged cell?

Best,
Martin

[Robin]

Dear Martin,

I did not find any inconsistent parameters in your input file. Actually, the success of the GS and KSS calculations indicate that they work fine. To figure out the cause, maybe you can try to calculate a system with 2 positive charges.

Some features of Abinit can not be combined together yet. Maybe this is another one.

Sincerely,
Guangfu Luo

[Martin]

I did not find any inconsistent parameters in your input file. Actually, the success of the GS and KSS calculations indicate that they work fine. To figure out the cause, maybe you can try to calculate a system with 2 positive charges.

I just ran the calculation with charge +2. Same error.

Any other ideas? Thanks!

Best, Martin

[mverstra]

occopt 0
occ 380*2 1 *0

whereas in your log for ds2:
Occupation numbers for ikpt= 1
1 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

looks like there is only 1 kpoint with all occ=0. Try with a normal occopt for metallic systems (4 or 7), or else iscf -3. DS3 may be ignoring the occ you provide, and reading the 0s in from the DS2 wavefunctions.

matthieu

[Martin]

I tried a spin-polarized calculation (nsppol=2), using occopt=1 to get a semiconductor, for charged supercells of odd charge, e.g. +1, and it worked.

Best,
M.

[mverstra]

Ok great - did you get a spin polarized result?

matthieu

[Martin]

Yes, which is consistent with the published results. Is it necessary to use nsppol=2 every time one has an odd number of electrons in the cell?

Martin

[mverstra]

It depends on the system:

* Al has 3 electrons, but it is fine in non-sppol because it's a real metal. One band is half filled and everything is fine.

* If you have a semiconductor + a localized level or an atom, of course, you need spin-polarization, as you can't "double-fill" part of a band.

matthieu