[gpaw-users] different convergence properties depending on LCAO guess

[Ask Hjorth Larsen]

Hi Nick

On Tue, 27 Jul 2010, naromero at alcf.anl.gov wrote:

> Ask,
>
> This calculation only ran for 1 hour but it didn't look like
> it was going to converge. I used basis = 'szp(dzp)'
> http://en.pastebin.ca/1909290
>
> Also, a few comments below:
>
> Nichols A. Romero, Ph.D.
> Argonne Leadership Computing Facility
> Argonne National Laboratory
> Building 240 Room 2-127
> 9700 South Cass Avenue
> Argonne, IL 60490
> (630) 252-3441
>
>
> ----- "Ask Hjorth Larsen" <askhl at fysik.dtu.dk> wrote:
>
>> Hi Nick
>>
>> On Tue, 27 Jul 2010, Nichols A. Romero wrote:
>>
>>> Ask,
>>>
>>> Number of Bands in Calculation:         3456
>>> Bands to Converge:                      Occupied States Only
>>> Number of Valence Electrons:            6208
>>
>> Right, that could be the problem.  In some systems you need as many
>> bands
>> as there are electrons.  This is probably not the case here, but try
>> adding some nonetheless.
> Ooh... that's a lot. I thought JJ's rule of them was to add about 20%
> extra electrons.

Right, that's normally sufficient.  In any case, it's probably easy to 
investigate these things on a smaller system :)

>>> mixer = Mixer(0.15,3),
>>
>> Consider using Mixer(0.1, 5, 100.0) or so.
> I will give it a try.
>
>>
>>>
>>> occupations=FermiDirac(width=0.10, maxiter=1000),
>>
>> It's a bit strange that the FermiDirac is taking that many
>> iterations.
> maxiter = 1000 is the default. When I use the partial wave expansion,
> maxiter = 4000 was needed. But even then it fails to converge.

That's most likely a symptom of something else going wrong.

Regards
Ask