[Wien] mixing vs. lapw2 CPU time

[Steven Homolya]

On Thu, 21 Aug 2003, Stefaan Cottenier wrote:

> 
> > > Using the "-ec 0.0001 -cc 0.0001 -fc 0.5" options is a bit of paranoia
> on my
> > > part. I don't quite trust the results unless I see that everything has
> > > converged.
> 
> > I don't think this works. Only the LAST criterium will be checked, since
> > we have in the run_lapw script:
> 
> I'm 'suffering' from this kind of paranoia as well (and I hope many of us do
> ;-) ).
> 

With very tight tolerances, I have seen the energy converge to within
0.00005 Ry without the charge having converged at all.

A few more iterations and the charge starts to converge and the energy
jumps around some more, until both settle. I've also seen it go the other
way aroud, i.e., charge appears converged but the energy is still deciding
where to go.

Force seems even more touchy (which makes sense, since it's determined by
the detailed structure of the charge density).

So, if I want to know the force I want to make sure both energy and charge
are converged.

Thank you to Peter Blaha for pointing out where and how the convergence is
tested. If I can get my head around C-shell scripting, I'm going to modify
the script to combine (AND) all convergence criteria, the way I mistakenly
believed it was doing.

-- 
Steven Homolya
School of Physics and Materials Engineering
Monash University, VIC 3800
Australia
Tel: +61 3 9905 3694
Fax: +61 3 9905 3637