[Pw_forum] RE : RE : A question on atomic magnetic moments

Nicola Marzari nicola.marzari at materials.ox.ac.uk
Tue Feb 15 19:03:32 CET 2011



Dear Giovanni, Cyrille,


the iron dimer has a very complex electronic structure (see e.g. our
PRL 2006 Kulik et al) - different multiplets, and for each multiplet,
different symmetries.

In principle, there is no way to guess to which of the many
self-consistent states you'll converge to, and I suppose that what
you have here is that you have converged to different states.
You'll need to look at the total energy, kohn-sham eigenstates, and the 
symmetry of those to understand what you are looking at
(also, there is no point in having more bands than efermi + 3*degauss -
empty bands above that threshold are useless).

There is no easy way to force the code into one state or the other -
you could try using GGA+U - some U's will stabilize reliably one state, 
and then by reducing smoothly the U (or increasing it) you might be able
to have (as a function of U) the energy of your preferred electronic 
configuration.

Last - for a cluster or a molecule smearing is used, somewhat
inappropriately, to help a calculation converge when using
iterative approaches - it keeps you company during the voyage,
but shouldn't share your bed (the real reason d'etre for
smearing is to improve k-point integrations in metals).

				nicola


On 2/15/11 5:42 PM, Giovani Faccin wrote:
> Dear Cyrille,
>
> I recalculated the dimer with different values of nbnd. Now some
> interesting changes appeared:
>
> Value of nbnd    |   Magnetization
>
> 10     4.00
> 12     6.88
> 14     6.49
> 16     6.49
> 18     6.49
> 20     6.49
> 22     6.88
> 24     6.88
> 26     6.49
> 28     6.49
>


-- 
----------------------------------------------------------------------
Prof Nicola Marzari    Department of Materials    University of Oxford
Chair of Materials Modelling  Director, Materials Modelling Laboratory
nicola.marzari at materials.ox.ac.uk     http://mml.materials.ox.ac.uk/NM


More information about the Pw_forum mailing list