[Wien] Problem with mixer

[Laurence Marks]

OK, the ratio of CLMs to PWs is such that I think the current version
can be a bit slow.

The number of iterations to converge reflects in part how well the
Hamiltonian is constructed. For instance, if you run non-sp a problem
which is really sp the convergence may be anomalously slow.

On 10/22/07, Björn Hülsen <huelsen at fhi-berlin.mpg.de> wrote:
> Dear L. Marks,
>
> I investigate the interface between an insulator and a magnetic material, looking for the stable structures
> for different terminations and registries. So there are certainly some "unphysical" configurations in the
> candidate pool. Maybe I can compare the stability of the structures with the number of iteration needed to
> converge them when all calculations finished.
>
> At the moment I try different (smaller) values for the mixer which seem to help a bit.
>
> There are 7500 PW's (from case.outputd*) and around 33000 CLM's (number of lines in case.clmsum*). According
> to your second statement this causes troubles.
> Since finally everything converges I would rather wait until the final release of the next version.
>
> Best regards
> Bjoern
>
> Laurence Marks wrote:
> > I can only suspect a problem at 40 iterations.
> >
> > 50-70 is a very large number, and normally convergence takes < 30. The
> > default 0.40 in the mixing is a bit greedy, and 0.3 or 0,2 might be
> > better. You may also (perhaps more importantly) be incorrectly posing
> > the physics or using -it when :DIS is small (which can have some
> > problems and lead to poor convergence),
> >
> > Do you have a relatively small number of PW's which are > 1D-12
> > compared to the number of CLM's? If so I think I know why your
> > convergence is slow but you will have to wait until the next release
> > unless you are having pathalogical problems and want to test a beta
> > version.
> >
> > On 10/22/07, Björn Hülsen <huelsen at fhi-berlin.mpg.de> wrote:
> >> The error occurred in the 40th iteration.
> >> But this problem appears somewhat randomly. I had several (order of 50) other calculations running for 50 - 70
> >>   iterations without it.
> >>
> >> Thank you for the hint.
> >>
> >> Best regards
> >> Bjoern
> >>
> >> Laurence Marks wrote:
> >>> On 10/22/07, Björn Hülsen <huelsen at fhi-berlin.mpg.de> wrote:
> >>>> Dear L .Marks,
> >>>>
> >>>> thanks for the code.
> >>>> But, as I wrote in my first email, I used the default input for the mixer with a history setting of 8.
> >>>>
> >>>> MSEC1  0.0   YES  (BROYD/PRATT, extra charge (+1 for additional e), norm)
> >>>> 0.40            mixing FACTOR for BROYD/PRATT scheme
> >>>> 1.00  1.00      PW and CLM-scaling factors
> >>>> 9999  8         idum, HISTORY
> >>>>
> >>>> Best regards
> >>>> Bjoern
> >>>>
> >>> I missed that. Did it occur at the 20th iteration, or perhaps the
> >>> 40th? If it did I think I know what it is. Unfortunately I cannot
> >>> currently reach my cluster to check. As a patch you can remove the
> >>> history files (rm *bro*) and run with the option "-r 99999". There is
> >>> an obsolete option in the runXX_lapw to reset the mixing after some
> >>> number of iterations which should not be there (with the new mixer)
> >>> and might be causing problems.
> >>>
> >> --
> >> Björn Hülsen
> >> Fritz-Haber-Institute
> >> Theory Department
> >> Faradayweg 4-6
> >> 14195 Berlin
> >>
> >> phone:  +49 30 8413-4863
> >> email:  huelsen at fhi-berlin.mpg.de
> >> www:    www.fhi-berlin.mpg.de
> >> _______________________________________________
> >> Wien mailing list
> >> Wien at zeus.theochem.tuwien.ac.at
> >> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> >>
> >
> >
>
> --
> Björn Hülsen
> Fritz-Haber-Institute
> Theory Department
> Faradayweg 4-6
> 14195 Berlin
>
> phone:  +49 30 8413-4863
> email:  huelsen at fhi-berlin.mpg.de
> www:    www.fhi-berlin.mpg.de
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>


-- 
Laurence Marks
Department of Materials Science and Engineering
MSE Rm 2036 Cook Hall
2220 N Campus Drive
Northwestern University
Evanston, IL 60208, USA
Tel: (847) 491-3996 Fax: (847) 491-7820
email: L-marks at northwestern dot edu
Web: www.numis.northwestern.edu
Commission on Electron Diffraction of IUCR
www.numis.northwestern.edu/IUCR_CED