Metal-organic Frameworks

Dear LAMMPS users,

I am beginning some simulations of metal-organic frameworks, and I am trying to figure out the best way to do this in LAMMPS. My system consists of the elements C, H, N, O, Cu, and Cl. My initial hope was to use ReaxFF because of its accuracy; however, perusing the author’s website (as suggested in the LAMMPS documentation) seems to indicate that a parameterization of these particular elements has not been created.

So I believe the universal force field (UFF) is my next best option, but it will take me a while to input these parameters by hand from the 1992 paper into a LAMMPS input file. I was hoping someone on here who is more familiar with these sorts of simulations could suggest a better option if there is one. I believe I have covered the LAMMPS documentation fairly thoroughly and have found a few tools for creating input files, but any knowledge from someone else would be greatly appreciated!

Thank you,
Nick

Dear Nick,

   The real question is what you hope to get out of the simulation.

   The UFF parameters are available with OpenBabel (www.openbabel.org) as a file UFF.prm, so you won't have to type them in. However, you should check that Lammps supports the UFF functional forms - my recollection is that UFF uses fourier series for angle bending and I don't find that in the the Lammps angle terms.

   You might write to Adri van Duin and see if there is any recent work with Cu so you could use ReaxFF.

   Or, depending upon the type of Cu you want to model you could add copper parameters to one of the standard force fields.

Kevin

Dear LAMMPS users,

dear nick,

I am beginning some simulations of metal-organic frameworks, and I am trying
to figure out the best way to do this in LAMMPS. My system consists of the
elements C, H, N, O, Cu, and Cl. My initial hope was to use ReaxFF because
of its accuracy; however, perusing the author's website (as suggested in the
LAMMPS documentation) seems to indicate that a parameterization of these
particular elements has not been created.

please note that reaxff is not parameterized to elements in general,
but specific sets of elements under specific conditions. those parameters
are not necessarily transferable to other systems with the same set
of elements. if you want that kind of flexibility, you have to use ab initio
methods (in fact, that is the whole point of ab initio methods, that you
don't make any assumptions).

So I believe the universal force field (UFF) is my next best option, but it
will take me a while to input these parameters by hand from the 1992 paper
into a LAMMPS input file. I was hoping someone on here who is more familiar
with these sorts of simulations could suggest a better option if there is
one. I believe I have covered the LAMMPS documentation fairly thoroughly

i think the problem is more that you should spend some more time
on understanding how classical MD works and what its limitations are.
for example parameters are usually not given on a per element basis,
but rather on a per "atom type" basis, i.e. a given element under specific
boundary conditions. so for carbon atoms there are 10s of different types
in a typical force field. you *can* have a more generic description, but that
comes at the expense of transferability. in other words, the less specific
the atom type, the less accurate the results.

and have found a few tools for creating input files, but any knowledge from
someone else would be greatly appreciated!

building inputs for custom systems requires writing/adapting customs tools.
only for very formalized processes, e.g. proteins, one can can automate the
process somewhat using a fragment/residue database. but then again, in
many of those cases, still some customizations are required.

axel.

Hey Nick,
Here are a few more tips in addition to the ones you got from Axel and Kevin.
Reaxff is inherently not more accurate than other potentials. As a
reactive potential it will allow you to model bond breaking/formation
but just because of that feature, doesn't mean that it can yield other
properties such as elastic constants better than other potentials.
Furthermore, are you expecting your MOFs to change the bond
connectivity during your simulations? If you are intending to simulate
gas adsorption in MOFs, in general, the gas molecules do not tend to
form bonds with the MOFs but they weakly adsorb due to the long-range
and LJ interactions. Those are the interactions that matter the most
for the MOF-gas system. If indeed gas adsorption is your goal, you
should try starting by simulating the process in a rigid MOF
(framework atoms do not move). The flexibly of the framework in many
occasions only adds a second order correction to the uptake in the
adsorption isotherms. Keep in mind that MOFs tend to have unit cells
on the order of hundreds of atoms so, you really don't want to use
Reaxff unless it is the only way to attack the problem. Its
computational cost for such systems can be almost two orders of
magnitude greater than that of other force fields.
Carlos

I'll just add that there are several Fourier versions of bonded
interactions in the USER-MISC package.

Steve