will,
Axel,
I am modeling the metal (gold) surface with a many-body potential. However,
I also have a situation where a sulfur atom may bond to two adjacent Au
atoms at a "bridge" site on the metallic surface. In this case, I model the
bonded interaction with a Morse potential between each S-Au pair, and then
define a "bond" between the two Au atoms so as to constrain the distance
between these two atoms. This make it a sort of three-body case, I suppose.
yes, very much so. i have some personal experience with this
kind of system: a couple of years ago, i did a very difficult and
time consuming DFT CPMD calculation that was used in combination
with other calculations on smaller systems to help interpreting
experimental results. http://dx.doi.org/10.1126/science.1158532
some people on the project decided to develop a many-body
potential based on the results, but i have not yet heard that they
were successful. it is a very tricky system due to the sulfur atoms'
ability to pull gold atoms out of the flat surface and the dynamic
nature of this system.
Unfortunately, this leads to the situation I am currently having trouble
addressing -- in certain instances I have an Au-S-Au-S-Au configuration,
where each S is bonded to two gold atoms. "Bonds" are also applied between
Au1 and Au3 and Au3 and Au5, so based on the topology of this configuration
LAMMPS does not compute nonbonded interactions for Au1-S4 or Au5-S2, though
I would like for them to be. I am also seeing now that these same two pairs
could be considered not only 1-3 pairs but also 1-4 pairs from their bonding
topology. Eg, Au1-Au3-S4 and Au1-S2-Au3-S4.
I suppose if I could constrain the Au-Au distance within a bridge site in
some way other than having to define a bond, this would also solve my
problem.
hmmm... what you could do is to define a non-bonded potential
to be added to your system via pair_style hybrid/overlay to add
the constraints. based on the fact that, according to the quantum
chemical calculations, your model cannot be overly accurate to
boot, it should be acceptable. you can just have multiple bond
types representing gold, map them all to your gold many-body
potential and then map them selectively with an appropriately
short cutoff to the non-bonded add-on potential. if you want it
to be harmonic, you can just use a tabulated potential.
that may take some effort to set up and will not allow dynamical
changes, but should sidestep the exclusions issue completely.
whether the total model will be useful. is hard to predict.
after my experience with the DFT calculation (i had to run
on a full rack of an IBM blue gene/l for over a month solid)
it will be extremely challenging to get anything with statistical
relevance out of higher level methods.
cheers,
axel.