[Octopus-devel] [Octopus-notify] svn commit: r3195 - in trunk: src/main by marques
marques at teor.fis.uc.pt
Mon Sep 10 15:13:51 WEST 2007
On Mon, 2007-09-10 at 13:54 +0200, Xavier Andrade wrote:
> Shall we leave this method as default, then?
I don't have a strong opinion about it... Actually, I think that there
are other very important things to do in this geometry optimization
business before like:
*) What is the reasonable accuracy to ask for? ABINIT says around 2.5e-3
eV/A in the forces, but do we really have that accuracy?
*) When doing geometry optimization, the most reasonable stopping
criterium for the SCF is convergence in the forces (10 times smaller
than the tolerance for the geometry optimization, for example). This
would probably save us many SCF iterations.
*) Benchmark all the GO methods, and figure out it they are really
working (has to be a molecule with at least 10 degrees of freedom ;)
Anybody wants to volunteer?
More information about the Octopus-devel