[Wien] Q: charge lost in SO calculation depending on LAPW / APW+lo (2)

[Eeuwe Sieds Zijlstra]

Dear Peter Blaha and Torsten Andersen,

Thank you both for your replies. I would like to add the following.

First of all, yes, I loose charge ONLY for a 6s=LAPW, 6p=APW+lo,  6p-RLO 
calculation! All other combinations are fine!
The 6s=LAPW, 6p=LAPW, 6p-RLO combination is also stable and fine when more 
than one k point is used. 

The :CHA in case.scf is fine:
:CHA  : TOTAL CHARGE INSIDE UNIT CELL =      15.000000
So, the Fermi integration works.

I know that many k-points are required for such a small unit cell and I also 
know that Emax in case.in1 should be larger when spin-orbit is included if 
one wants to obtain converged results (usually I use Emax=3.0Ry and then 
check for convergence). I thought that my question would be clearer / easier 
to answer if I presented the smallest possible calculation that already shows 
this behavior. Sorry for causing confusion this way. 
The important point is that I found the same behavior also when I included 
more k points and a higher Emax.

One more point: If I increase the muffin-tin sphere, the charge loss is 
greater.

Why I play around with this? To be sure that I understand everything. To 
convince myself that the program works correctly. Such checks usually do not 
take much time.

I would be very happy if you could look again at this problem. 

Thanks in advance, 

Eeuwe



Peter Blaha wrote:
> Please check the scf file, if the Fermi integration works ?
> Do you get the proper valence charge
> grep ':CHA  ' case.scf
> or is this value already smaller ?
> 
> Such small unit cells require many k-points (100-10000) !!!! Results with
> Gamma only are nonsense.
> 
> For the second variational procedure you MUST use a larger E-window in
> case.in1. Thus increase the default 1.5 and do NOT decrease it to 0.9
> 
> I do NOT understand the statement, that LAPW for p is "unstable" ? 
> 
> To be sure I understand you correctly: You loose charge ONLY for a
> 6s=LAPW, 6p=APW+lo,  6p-RLO calculation ? All other combinations are fine ?
> 
> After all, why do you change the default in1 file and play with APW/LAPW ?
> If one plays around, than for best convergence use APW+lo for d, but LAPW 
> for s+p