[Wien] optimization procedure - questions

[Laurence Marks]

A complex question, without a completely simple answer.

First, both optimization methods are sensitive to the parameters being
used, for instance the number of k-points, TEMPS versus TETRA, how "good"
the RMTs are as well as which version you are using. While you do not need
these to be perfect, they should be reasonably good. Without knowing more
about your parameters it is hard to say. I strongly suggest using TEMPS
with 0.0018 (room T) rather than TETRA. There is some telegraph noise in
TETRA with MSR1a which can be problematic.

It is also not obvious from your email to what level System 2 has not
converged. In many cases MSR1a for non-insulators will converge to 1-2
mRyd/au but can oscillate particularly with the earlier versions. I am
hopeful that the next version will be better, but it is still under
development. In many cases forces at the level of a few mRyd/au do not
matter, they are well below the accuracy of positions due to inaccurate
functionals. I have seen people reporting issues on the mail list when they
try and converge to very high levels -- which is a waste of electrons.

Comparing MSR1a and mini is not simple. With the current released version
mini can be better for systems with soft modes which are not insulators.
However, mini can also be slow. One of the important issues may be how the
Mn d-electrons should be treated. In many cases a better treatment of the
system leads to faster convergence. Perhaps the Mn needs to have a U or an
-eece correction?

N.B., there is not such thing as "the correct method", there is only the
one which works.


On Thu, Feb 5, 2015 at 8:10 AM, Yevgen Melikhov <melikhovyy at yahoo.com>
wrote:

> Dear Prof. Blaha,
> Dear users of WIEN2k,
>
> I have several questions on how best to perform optimization procedure for
> the following problem:
>
> I have a system with 96 atoms (it is relatively big in order to
> accommodate 1% of Mn in GaAs), which I refer to as System 1. The other
> system is the same but with two vacancies, so, overall, I have 94 atoms in
> this System 2.
>
> My first step is to relax both structures (assuming fixed lattice
> constant) before calculating X-ray absorption spectra.
>
> Logically, I do not expect severe changes of atomic positions in System 1.
> However, for the System 2, I expect some severe rearrangements (to be
> confirmed yet).
>
> In the WIEN2k User Guide it is said that there are two methods to solve
> relaxation problems:
>    (i) using min command, and
>    (ii) running run_lapw with MSR1a switch in case.inm file.
>
> Am I correct to assume that for System 1 the method (ii) should work
> fine/faster?
>
> Am I correct to state that usage of the method (ii) for System 2 is wrong
> or at least will take much more time to optimize positions? In fact I tried
> using this method (ii) for System 2 and after 2,000 (!) iterations I gave
> up. I can see that some atoms, which are expected to move, do move. But
> their positions have not converged after so many iterations.
>
> As my system is magnetic, will it be correct to optimize first System 2
> without spin polarization with method (i) and then fine tune the positions
> with spin polarization with method (ii). I think by doing it this way, I
> should speed up my calculations. I saw a discussion on this forum which
> took place some time ago with respect to magnetization and
> relaxation/optimization urging not to do that but, as magnetic atoms (Mn
> only) are quite far from each other this should not lead to a problem for
> my System, should it?
>
> Finally, due to the nature of the LAPW method, could it be better to do
> optimization of atomic position using DFT software with plane waves and
> pseudopotentials and then continue calculations in WIEN2k? In the
> literature I found that some researchers do that but I wonder whether this
> is really that time efficient?
>
> Any comments will be highly appreciated.
>
> Sincerely yours,
>   Yevgen Melikhov
>     Institute of Physics PAN, Warsaw.
>
> P.S. I did try to read <Mixer_Readme> (the section “Parallel Atomic
> Minimization Algorithms (a.k.a. the Energizer Bunny)” specifically) and
> <Optimization Notes> by Prof. Marks.
>
> P.P.S. When running the method (ii) for System 2 for 2,000 (!) iterations,
> I did not forget to decrease RMT by 10%. When I run command nn I did not
> see any two atoms being very close to each other.
>
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> SEARCH the MAILING-LIST at:
> http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html
>



-- 
Professor Laurence Marks
Department of Materials Science and Engineering
Northwestern University
www.numis.northwestern.edu
Corrosion in 4D: MURI4D.numis.northwestern.edu
Co-Editor, Acta Cryst A
"Research is to see what everybody else has seen, and to think what nobody
else has thought"
Albert Szent-Gyorgi
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20150205/c2a87be6/attachment.html>