The ABINIT Users Mailing List ( CLOSED )

[mmikami@rc.m-kagaku.co.jp]

Hello, I could have some suggestions based on my (limited) experiences, 
although I am not sure ...

1) Have you tried the non-spin-polarized calculation ? Is it converged ?
In non-spin-polarized cases, you could rely on "iprcel" :
http://www.abinit.org/ABINIT/Infos_v4.2/vargs.htm#iprcel
When I tried a metal slab model, the option of iprcel (e.g. 45) worked 
for convergence.
Then, you could switch to spin-polarized calculation with the initial 
DEN/WFK calculated with the non-spin-polarized case...

2) How about the geometry ? If the geometry might be impractical for 
the convergence of SCF, the SCF would fail ...  It might be more 
contracted, with the bonds shorter than those in the bulk ... (When I 
calculated the metal slab, the surface atoms moved inner, kind of 
"contraction")
Once the geometry could be optimized under the condition 1), you could 
switch to spin-polarized case ...
(NB: Similar difficulties may appear in other cases, e.g. defect 
calculations:
http://www.abinit.org/wws/arc/forum/2003-07/msg00040.html
In this case, the geometry needed the different charge state for SCF 
convergence;
the different geometry would allow different charge states, etc.)

Initial wavefunction/density might be crucial to get (fast) SCF 
convergence...
I heard from Prof. Scheffler that in their metal slab models some trick 
about initial wavefunction worked well. The reference might be :
J. Neugebauer and C. G. Van de Walle, Proc. Mater. Res. Soc. Symp. 408, 
43 (1996)
(If the other reference would be cited, could someone correct me ? I 
just checked
http://www.fhi-berlin.mpg.de/th/fhi98md/references.html )

In passing, I have heard that some techniques (to get good an initial 
density for SCF) are under development in biochemical fields, e.g. 
"quasi-canonical localized orbital" (QCLO):
H. Kashiwagi et al. "Convergence process with quasi-canonical localized 
orbital in alll-electron SCF calculation on proteins", Mol. Phys. 101, 
p.81-p.86 (2003)
From the talk, SCF convergence has never been achieved for proteins, 
when the initial density was just the sum of the atomic densities. 
Figure 7 may be also interesting.

Hope this helps,
Masayoshi

On 2003.10.16, at 03:00 Asia/Tokyo, Kristopher Andersen wrote:

> I'm beating my head against a wall trying to converge a cluster of
> niobium atoms. It seems like this would be a relatively easy system,
> since there are only 4 atoms, but the CG minimization fails to
> converge.
>
>  scfcge : ERROR -
>   Potential-based CG line minimization not converged after 13 restarts.
>
> I've tried raising the number of bands (nband), switching to Anderson
> mixing (iscf=3), and raising the number of line minimizations (nline).
> All to no avail. I've also tried all sorts of different occupation
> options and broadening temperatures (from tsmear = 0.1 mHa to 0.1 Ha),
> since there does seem to be a near-degeneracy near the HOMO.
>
> I've attached the input files, including the pseudopotential. If you
> can offer any advice, I'd really appreciate it. There will be a round
> of beer with your name on it at the March Meeting.
>
> --
> Kristopher Andersen               http://yclept.ucdavis.edu/~andersen
> Department of Physics                    andersen@physics.ucdavis.edu
> University of California at Davis                   +1 (530) 752-0446