[gpaw-users] convergence criteria as a function of system size

[Ask Hjorth Larsen]

Hi Nick

On Tue, 13 Jul 2010, Nichols A. Romero wrote:

> Hi,
>
> In order for the SCF cycle to be considered converged, they following three
> criteria should be met:
>
> {'energy': 0.001, # eV
> 'density': 1.0e-4,
> 'eigenstates': 1.0e-9,
> 'bands': 'occupied'}
>
> Except for the criteria on the total energy, the other two (density and eigenstates) would seem to be independent of the system size. But are the issues of numerical
> precision that might come into play?
>
> I will ask my question another way: If I am dealing with 3000 - 5000 bands, is the
> default convergence criteria too strict for obtaining converged total energy and forces?

The energy criterion is evaluated per atom.  I.e. larger number of atoms 
need a less converged total energy.  The two others are proportional to 
the valence electron count.  See paw.py around line 460.

Regards
Ask