[Q-e-developers] Mixing and optimization collaborators ?

[Laurence Marks]

There is no generic version yet, it is not be too hard for me to
modify the existing version. The complicated part where collaboration
matters is designing how it couples to DFT codes plus of course the
testing.

My preliminary thinking is several components:
a) The core algorithm which is mainly lapack calls for relatively
small matrices in most cases.
b) DFT code dependent routines to pack/unpack the basis set, atomic
positions, Hubbard U (if used) and anything else.
c) DFT code dependent routines for trust region control, I.e. ones
that tell the algorithm if a step is not viable, for instance moves
the atoms too much or (APW specific) would lead to overlapping
spheres.
d) Some method for the DFT code to tune some parameters plus user
control. There should be very few user control parameters.
e) Some sort of user output.

Planning before coding is smart although we dont always do it.

On Thu, Oct 24, 2013 at 1:00 AM, Paolo Giannozzi
<paolo.giannozzi at uniud.it> wrote:
> Interesting. Is there a generic implementation of the mixer/optimizer
> that can be adapted to different codes, or should it be written from
> scratch?
>
> Paolo
>
> On Wed, 2013-10-23 at 10:01 -0500, Laurence Marks wrote:
>> For some years I have been working on better optimizers and mixers,
>> mainly for Wien2k but I also made a small contribution to QE in the
>> curvature trapping for your BFGS optimizer for the atom positions. I
>> have a new algorithm (actually one proposed by Bendt and Zunger in
>> 1983) which is mature enough in Wien2k that it is now the recommended
>> optimizer -- it simultaneously "mixes" the density and atomic
>> positions. I am attaching the reprint (which might get filtered out);
>> it is a bit complicated although the fortran code is not as grim as
>> one might think. Alas JCTC mangled horribly the nomenclature section
>> at the end.
>>
>> I am interested in exploring how this algorithm would work in other
>> codes. Unfortunately I do not have the time (or money) to invest in
>> learning all the internal details of other codes. Perhaps someone
>> would be interested in collaborating to try and implement it in QE?
>>
>> _______________________________________________
>> Q-e-developers mailing list
>> Q-e-developers at qe-forge.org
>> http://qe-forge.org/mailman/listinfo/q-e-developers
>
> --
> Paolo Giannozzi, Dept. Chemistry&Physics&Environment,
> Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
> Phone +39-0432-558216, fax +39-0432-558222
>



-- 
Professor Laurence Marks
Department of Materials Science and Engineering
Northwestern University
www.numis.northwestern.edu 1-847-491-3996
"Research is to see what everybody else has seen, and to think what
nobody else has thought"
Albert Szent-Gyorgi