[Wien] leaking core charge and ‘.lcore’

[Peter Blaha]

> I am working on a structure containing Osmium.  In init, I got a warning
> that Os 5s core charge (about 1%) leaked out of the sphere (RMT=2.08 set
> by setrmt).

I don't know the energy of Os 5s. If it is just a little below -6. Ry, 
just lower E-core, if it is at lower energies, .lcore should be fine.
The solution with   .lcore should be perfectly ok in your case.
.lcore directs the code to use the spherical core-density (also beyond RMT)
and do a superposition of these densities using dstart (see the dayfile).

.lcore would NOT be ok if Os-O distances become so small, that the Os-5s 
band gets a significant band-width due to the interaction with O.

RMTs of 2.5 and 1.3 violate the RMT criteria and together with a large 
RKmax ?) you get "numerical linear dependencies and ghostbands). Usually 
it is save to change the suggestions of setrmt by a small amount (eg. 
2.18 for Os), but not that much.

>
> I increased the Os RMT to 2.5 manually, decreasing the RMT of its O
> neighbor to 1.3.  Then there were no warnings in lstart, but the
> calculation crashed after a couple of iterations with QTL-B / “semicore
> band-ranges too large” errors.  (The errors were about Os.)
>
> So I tried instead to run the calculation with the original RMTs and the
> ‘.lcore’ file created by init.  This way, the SCF converged without any
> errors or warnings; ‘lcore’ and ‘dstart -lcore’ were executed at every
> step.  The band structure looks reasonable as well.
>
> What does this lcore stuff imply for my calculation?  Should I consider
> the results suspect?
>
>
> Thanks,
>
>      Elias
>

-- 

                                       P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-165300             FAX: +43-1-58801-165982
Email: blaha at theochem.tuwien.ac.at    WWW: 
http://info.tuwien.ac.at/theochem/
--------------------------------------------------------------------------