[gpaw-users] Convergence of eigenstate and density

[Jun Yan]

Hi, Ask,

    The reason I use rmmdis is that it supports parallelization over 
bands. That's important for my calculation since i have to keep domain = 
1 and if there is not enough kpoints to be parallelized, I have to used 
band parallelization. What you proposed might be good to converge the 
scf steps, but it will not solve my problem :p

All the best,
Jun


Ask Hjorth Larsen wrote:
> Hi
>
> On Sat, 30 Apr 2011, Jun Yan wrote:
>
>> Well, I dont quite understand your question.
>> I use
>>                 convergence={
>>                  'density': 1.0e-6,
>>                  'eigenstates': 1.0e-4,
>>                  'bands': nband}
>>
>> and nband = 260 while 80 occupied bands are included.
>>
>> Jun
>
> How big a difference do you get between the rmm-diis result and the cg 
> (fully converged) result?
> If you perform a modest number of initial iterations with CG, it's 
> possible that the RMM-DIIS can perform the rest without trouble.
>
> You can use maxiter, try/except KohnShamConvergenceError and 
> set(eigensolver=...) to do this.
>
> Regards
> Ask