[XtalOpt-devel] shell model in GULP

[Wei Lai]

Hi Dave,

Thanks for your comments. I guess I was looking at a system that could
benefit from a partially evolutionary algorithm.

Wei

On Wed, May 16, 2012 at 10:23 AM, David Lonie <loniedavid at gmail.com> wrote:
> Hi Wei,
>
> On Tue, May 15, 2012 at 10:01 PM, Wei Lai <laiwei at msu.edu> wrote:
>> 1) It seems "gulpFracShell" forces both Mg and O to adopt core/shell
>> models. How can we use core for Mg and core/shell for O?
>
> This is not supported at this time -- the keyword replacement will
> either print core/shell for all atoms or none. I've not used
> core/shell calculations much, but there may be a way to "trick" GULP
> into neglecting the Mg shells.
>
>> 2) If we already know the positions of Mg and we just want to relax O
>> positions, how can we apply fixed positions (fixed fractional
>> coordinate) to Mg only?
>
> I've typically used XtalOpt on completely unknown systems, and as a
> result never implemented a way to fix the coordinates. I'm not really
> sure how well this work with an evolutionary/genetic algorithm either,
> as I've only tested the operators used to generate new structures in
> cases where all atoms are free to move. If you have a mostly solved
> structure with just a few unknown atoms, I think the best way to use
> XtalOpt would be to generate several guesses at the structure using
> the information that you know, and then use these to seed the first
> generation of the search. Otherwise, it may be best to come up with
> some way to randomly generate and optimize trial structures that
> explore only the unknown structural parameters.
>
> Probably not the answers you were looking for, but I hope this helps!
>
> Dave