[Wien] structure relaxation requirement for input atom positions

[qiangxu]

Dear Laurence Marks:

    Thanks for your answer, and the clear definition of my problem.

    I found out the crash problem I met is perhaps due to that "DFT energies
are far from perfect" as the second reason you mentioned:  I used
pseudopotential method parkage: VASP instead of WIEN2k (in order to save
computing time). The force between two very close atoms( interatomic
distance far less than normal bond-length) might be not calculated
correctly,  because core electrons, which in this case also contribute to
the interaction of the two atoms, are however not well described in
pseudopontential method.

    For the general remarks about local minimum problem you made , I totally
agree. In fact, this is not a problem for DFT only, but a general one to any
refinement, for instance, structure refinement using electron diffraction or
X-rays diffraction.  How to avoid this problem in solving the structure of a
novel compounds (using diffraction methods or DFT calculation) is quite
interesting issue. I have some results and I am now preparing a paper with
my boss (Henny Zandbergen). After it is ready, I wish I can send it to you
and have your kind commends. Thanks in advance


Best wishes
Qiang


-----Original Message-----
From: wien-bounces at zeus.theochem.tuwien.ac.at
[mailto:wien-bounces at zeus.theochem.tuwien.ac.at]On Behalf Of Laurence
Marks
Sent: 19 January 2007 15:12
To: A Mailing list for WIEN2k users
Subject: Re: [Wien] structure relaxation requirement for input atom
positions


I don't think you have posed the problem well. There are really two
problems:
a) How to find a local minimum from some initial starting point.
b) How to find the global minimum, i.e. the lowest of all possible minima.

A good refinement code will always solve a) and should never crash. It
should always find the "right" positions in the sense that they are
the local minimum in some basin. The minimizer in Wien2k does this
fairly well, with the important caveat that if the DFT problem is
poorly posed/conditioned (e.g. bad RMT choices) it may not be too
accurate.

Doing b) is harder mainly because it takes orders of magnitude more
computer time, in most cases more computer time than is available
except for very simple systems. Simulated annealing or genetic
algorithms coupled with a local search are the two standard
techniques. In some fashion one has to search over starting points to
find the lowest possible energy among all the different local minima.
This is not implimented in WIen2k in any automated fashion.

Cases where DFT does not find the "right" positions are either because
the initial starting point was close to some local minimum, but not
the global minimum or because of the fact that DFT energies are far
from perfect. How big the basin is around each local minimum varies
from structure to structure, so it is almost impossible to generalize.
Sorting this out is not something that (in most cases) a computer can
do for you, you have to use the most important scientific tool above
your eyes.

On 1/17/07, qiang <q.xu at tnw.tudelft.nl> wrote:
> Dear all:
>
> As we know that given a starting structure model, molecular Dynamics can
be
> used to refine atom to their energy (force) equilibrium position.
> Obviously, if the starting structure model is too far away from the
correct
> one,
> the refinement may crash or may not lead to the right positions.
> For a successful structure relaxation, does anyone know the general
> requirement for a starting structure model?  How close to the correct
should
> a starting structure model be? (e.g. the starting atom positions should be
> within 0.5 A away from the correct positions.)  Is there any ref. about
this
> problem?
>
> Best regards
>
> Qiang
>
> _______________________________________________
> 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
http://www.numis.northwestern.edu
_______________________________________________
Wien mailing list
Wien at zeus.theochem.tuwien.ac.at
http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien