Pseudooptimization?

Axel akoh... at gmail.com
Wed Nov 18 16:37:38 CET 2009


ilya,


> What I'd like to do is to calculate/fit MM potential parameters that
> would adequately describe both overall molecule-surface interactions and
> reproduce differences between adsorption sites.

that is a very ambitious goal. depends a lot on what
kind of surface and what kind of molecules.

> So I want to describe short and long-range pairwise Molecule Atoms -
> Surface Atoms interactions by empirical MM potentials, some kind of LJ
> and Coulomb.

not sure if the lennard-jones potential is a good choice
for that. if you want fairly accurate modeling, you may
have to use potentials that include many-body interactions.
there are several approaches to that.

> Using MM fields I could run really long MD trajectories (~2-3*10^3 ps)

a few nanoseconds for LJ+q is not what i would consider very long. ;)

> Hence, I need to estimate atom-atom interactions somehow.
>
> Do you(or any one) know how can I get pairwise atom-atom interaction
> energies and forces in cp2k?

well, you have to keep in mind that _the_ major property (and
major computational challenge) of a quantum mechanical description
of a system of interacting particles is that everything interacts
with everything. so there is no straightforward decomposition
into atom-atom pairwise contributions.

if you had a dilute, weakly interacting system (say liquid argon)
you could use the radial pair distribution functions and reconstruct
the pairwise potentials from them (take the logarithm). in other
cases people use force matching, but for the most part, those are
all non-transferable methods.

one approach to mimic quantum like many-body interactions, that has
seen a lot of interest recently is the REAXX force field, but even
for the few parameterized elements, there seem to be revisions coming
regularly.

> With this I could really break through!

always remember: there ain't no such thing as a free lunch.
if these things were so very easy, they would have been done
long ago. my personal impression is that if you are after an
efficient quantum mechanical-like description of your system,
using (and parameterizing) some semi-empirical method may be
an alternative to think about. with today's computer power and
some smart "linear scaling" algorithms, nanosecond timescales
should be possible.

cheers,
   axel.
> Sincerely,
> Ilya
>



More information about the CP2K-user mailing list