[gpaw-users] different convergence properties depending on LCAO guess

[Nichols A. Romero]

Ask,

Number of Bands in Calculation:         3456
Bands to Converge:                      Occupied States Only
Number of Valence Electrons:            6208

mixer = Mixer(0.15,3),

occupations=FermiDirac(width=0.10, maxiter=1000),

Nichols A. Romero, Ph.D.
Argonne Leadership Computing Facility
Argonne National Laboratory
Building 240 Room 2-127
9700 South Cass Avenue
Argonne, IL 60490
(630) 252-3441


----- "Ask Hjorth Larsen" <askhl at fysik.dtu.dk> wrote:

> Hi Nick
> 
> On Mon, 26 Jul 2010, naromero at alcf.anl.gov wrote:
> 
> > Hi,
> >
> > I have a system where the LCAO guess leads to very different
> behavior in the
> > SCF cycle. This is even with a small mixing parameter of ~ 0.01:
> >
> > 1. basis = 'sz' appears to converge the system, albeit very slowly
> > 2. basis = partial wave expansions causes the system to jump
> suddenly from a negative energy to a positive energy (after iter: 3)
> and eventually fail due to inability to locate the Fermi energy.
> 
> It sounds like a case where the real problem might be a too low number
> of 
> bands, which can really throw the rmm-diis eigensolver off - try 
> increasing nbands.  How many bands and electrons, and what temperature
> are 
> you using?
> 
> >
> > The system is an interface between a semi-conductor and a
> semi-metal. 
> > Are there any particular family of systems that can be very
> sensitive to 
> > the details of the LCAO initialization?
> 
> Nothing in particular that I know of - but let's see if the above
> works 
> before we start worrying too much.
> 
> Regards
> Ask