[gpaw-users] Ni in a box doesn't convergence and spin-polarized set-up

[Ask Hjorth Larsen]

Hi Nick

On Mon, 12 Dec 2011, Nichols A. Romero wrote:

> Hi,
>
> Two related questions:
> 1. We tried to converge a Ni atom in a box and tried every combination
>   of parameters we could think of, but we never got it to converge.
>   Can someone send me an input file that works for an isolated Ni atom?

I don't know whether this is the same issue, but for an isolated Fe atom, 
consider the following:

* With the electron configuration 4s^1 3d^7, the s eigenvalue is lower
   than the d eigenvalue, so we have to move an electron from d to s

* And yet, with 4s^2 3d^6, the s eigenvalue is greater than the d
   eigenvalue.  So we have to move the electron back.

These two considerations make it impossible to converge the calculation 
unless artificially fixing the occupations, or alternatively by increasing 
the Fermi temperature to allow partial occupations.  The latter solution 
will result in a minimum energy being obtained with s and d channels being 
degenerate.  This appears to be one of the weird properties of Kohn-Sham 
DFT with LDA/GGAs.  I observed this using the AtomPAW class (from 
gpaw.atom.atompaw.  Not the DFT code called AtomPAW).


>
> 2. Can the set-up generator created a set-up based on a spin-polarized
>   configuration (instead of spin-paired)? I modified the default value
>   of spinpol in aeatom.py (AllElectronAtom constructor), but it still
>   didn't seem to work.
>

That hasn't been implemented.

Best regards
Ask