[Pw_forum] Relax calculation not converging

marsamos at democritos.it marsamos at democritos.it
Mon Oct 12 12:04:44 CEST 2009


Dear Dimpy, the default value for scf steps is 100. The message means 
that the code does not achieve the electronic convergence after 100 scf 
steps. This usually means that your initial electronic configuration is 
too far from the scf one and the algorithm does not manage to converge 
in 100 step. Or, that there exist "two" solutions and the algorithm is 
oscillating. If it is just a question of been to far there are two 
possibilities 1) the algorithm need just more scf steps, so you just 
need to increase electron_maxstep (default=100) or to decrease 
mixing_beta (or both) (attention the decrease of mixing beta might 
produce that the algorithm need also more electron step). 2) you are 
too far and "two" solutions are accessible. This is for instance what 
happens in SiO2 for the charged -1 and -2 oxygen interstitial, or the 
Si vacancy. For treating such a problem you need to first do a 
pre-relaxation with some trick. For the insterstitial it is better to 
start from a neutral relaxation, for the Si vacancy one need to relax 
first with a substituted Al. Sometimes it can also bee usefull to 
rpe-relax with electronic temperature to facilitate convergence (ad a 
broadening).

hope it can help you, bests

Layla

Quoting Dimpy Sharma <dimpy.sharma at tyndall.ie>:

>
> Hi Quantum espresso users,
>
> I have been trying to perform a relax calculation in my system with 
> 99 atoms (semiconductor system) and I am running my calculation in 48 
> processors. The calculation has finished 3 scf steps and finished 100 
> iterations, but after this it stops and it is showing me the 
> following message .
> ' convergence NOT achieved after 100 iterations: stopping'
>
> I have restart my calculation, still I got the same error message 
> after 100 iteration steps.
>
> Can any body give me any suggestion?
>
> Thanks
>
> Dimpy
>



----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.




More information about the users mailing list