[Octopus-users] strange convergence problems

[Xavier Andrade]

On Tue, 28 Apr 2009, Pavel Redkin wrote:
>
> Dear Octopus users! I noticed a strange convergence behavior of many 
> small test systems, opposite to the one pointed out in the manuals. 
> Namely, these systems (mostly of carbon atoms) could give a converged 
> ground state only for an interval of spacings (e.g. from 0.8 to 1.2 
> angstrom for a diatomic system presented here and from 0.4 to 0.7 
> angstrom for C60 system and various boxes). In addition, the gs runs 
> with smaller spacings needed more iterations to converge. Could someone 
> explain the underlying physics? Or this is an algorithmic problem (not a 
> bug, as nothing crashes)? Also, how can one thus determine the quality 
> of system's representation in case further spacing reduction makes the 
> system non-convergeable?
>

Hi Pavel,

What do you call a converged ground state? Convergence with respect to 
which parameter? spacing? scf iterations? In any case the spacings you 
mention are too large, the default value for carbon is 0.23 angstrom and 
still a smaller spacing is required in some cases. If you have a problem 
with the SCF convergence it is probably not related to the spacing.

Cheers,

Xavier