ABINIT Discussion Forums

Hi!

I have been trying for days to get a converged solution of an isolated chromium atom, but no luck here. Maybe I'm doing something wrong with the input. My final approach was to use a spin-polarized calculation using the occupations below to get the 3d^5, 4s^1 configuration (see options below). When trying nsppol=1 for anti-ferromagnetic calculations, abinit keeps complaining about missing (a)symmetry to differentiate between spin-up / spin-down states. I'm using the FHI pseudopotentials together with LDA/LDS (ixc=7).

nsppol 2
occopt 2
nband 3 3
occ 5 1 0 0 0 0
spinat 0.0 0.0 3.0

The total energy jumps between -8.2 and -9.5 Ha, even after about 70 SCF iterations. What can I do about this?

Many thanks,
Oliver

Hi,

The code probably oscillates between two solutions, a magnetic and non-magnetic one with different energies. You should be able to confirm this by looking at your log.

You could probably solve this problem by imposing a density matrix as explained in the LDA+U tutorial (you can set U to zero).

If you are only calculating for an isolated Cr atom, then an anti-ferromagnetic calculation is impossible as you need at least two atoms to make it work (one spin up, one spin down).

Good luck
Simon

Dear Simon,

thank you for the fast reply! Of course, I will immediately counter this with another question (actually, shame on me, but I'm a very novice in DFT still...)

The LDA+U tutorial shows application together with the PAW technique. Is this mandatory? (PAW data for chromium is rather rare, while HGH and FHI pseudopotentials are readily available).

Best regards,
Oliver

Dear Oliver,

DFT+U is available only in conjunction with PAW atomic datasets and not with Norm Conserving Pseudopotentials.
As underlined my Simon, you could impose the density matrix to help the convergency, but in order to be efficient, you need to use a non zero value of U
(with U=0, the imposition will have no effects).

Alternatively, and firstly, you might retry with value of "occ" between 0 and 1, see for example test v5/14.

Good Luck
Bernard

Dear Bernard,

thanks for the info! As I would rather stay with the more tested PsPs instead of the rare PAW data for chromium, I will try it with the occ numbers as you suggested first!

Btw: is there any option in Abinit to keep the magnetization fixed such that only one of the solutions will be aimed at during SCF cycles?

Yours,
Oliver

Dear Oliver,

You might try fixmom
(read http://www.abinit.org/documentation/hel ... tml#fixmom)
or spinat (http://www.abinit.org/documentation/hel ... tml#spinat).

Best
Bernard