[Wien] analysis and optimize.job

[Gerhard Fecher]

Dear Peter,
It was without -orb
convergence criteria were -ec 0.00001 -cc 0.001
I tried both large and small mixing, the new and the old Broyden mixing schemes
I took care about all possible bad QTL values, and changed number of k-values and plane waves,
the result was reproducible.

The reason that I found that was that the solution of a calculation for the optimized lattice parameter
was different from the one obtained in the optimization.

It seems this material (ferrimagnet) has two energy minima close together at Volumes being very close together (within 2%).
The convergence problem is also not that Wien specific, KKR and LMTO have the same problems and it depends very critically on the start parameters.

I nam not that much afraid that the calculation does not guarantee which solution is correct,
as the solutions differ by roughly 1 mu_B in the magnetic moment, I will find the correct state from the experiment, hopefully.

Ciao
Gerhard

PS.: I did not copy the case_initial.struct  in all cases, but only when I know what I like to do the in following.

________________________________________
Von: wien-bounces at zeus.theochem.tuwien.ac.at [wien-bounces at zeus.theochem.tuwien.ac.at] im Auftrag von Peter Blaha [pblaha at theochem.tuwien.ac.at]
Gesendet: Donnerstag, 17. Juli 2008 10:11
An: A Mailing list for WIEN2k users
Betreff: Re: [Wien] analysis and optimize.job

Thank's for reporting the MMTOT problem (it was due to incopatible :MMT and :MMTOT labels)-
(change in SRC_w2web/htdocs/utils/analyse.pl  MMTOT to MMT)

> about the structure after doing a volume optimization:
> I think that one usually does not use the last volume of the series for refinement of an optimization.
> Usually I do some rough steps and then put some finer steps if needed, starting with the initial structure,
> therfore the end of my optimization.job usually looks like:
>
>  ....
>  save_lapw  ${i}
>  cp  case.scfm $i.scfm
>  #    save_lapw  -f -d XXX $i
>  end
> rm case.struct
> cp  case_initial.struct case.struct

I'm not too happy about that !
a) The next "x optimize" will use anyway case_initial.struct (I know that it is broken in the
current w2web version, but should be ok in the next one). So there's no need for that.
b) This would lead to an "incompatible set of files" (struct file does NOT correspond to
clmsum and vector files). Some people may run   x lapw2 -qtl   right after  x optimize (probably
not knowing that this is for the last volume ....)


> I realized also some effects using charge extrapolation.
> It might lead to the fact that one stays in the wrong energy minimum.
> I found different minima and magnetic states ifusing charge extrapolation compared to
> the the scratch version using dstart.
> The result of the charge extrapolation was also different if going from small to large volumes
> or from large to small volumes. (In one case the latter missed completely the energy minimum).
> This might be something very peculiar and in many cases there is no difference,
> but one should take care about the results.

Is this including -orb (LDA+U) ???
Is this because you chose a very small mixing in case.inm ??? I'd expect that the new mixer
does not need that !! and with charge extrapolation you may be so close to the final solution,
that an over damped mixer stops too early (or you need a better conv.criterium).
If this happens with full convergence, who tells (guarantees) that dstart results are ok and the
others are not ?

Best regards

--

                                       P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
Email: blaha at theochem.tuwien.ac.at    WWW: http://info.tuwien.ac.at/theochem/
--------------------------------------------------------------------------
_______________________________________________
Wien mailing list
Wien at zeus.theochem.tuwien.ac.at
http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien