It is very clear that with pair_style hybrid, we can create
Si-Si, C-C, Si-C, Si-O interactions. But to create C-O interactions I have
no, no and NO!
it is not very clear and while you may be technically able to do what you
say, it would be a big mistake to do so. your simulation is practically
*guaranteed* to be wrong.
you are looking at this from the wrong perspective. you cannot just mix and
match any kind of potentials for any pairs of elements as you like.
interactions have to be *balanced*. particularly when using manybody
you are not doing ab initio quantum chemistry here (and even there you need
some degree of consistency), so you have to respect for which kind of
systems parameters were derived and how you can use them to build you
now you have a bit of a problem with your setup, since you want to model a
hybrid compound where you have well established potentials for each part of
your hybrid, but no single and consistent parameter set that describes them
this is where the hybrid pair style in LAMMPS comes to your rescue, it
allows you to run each compound with the preferred potential type and
thus the remaining problem is how to describe the interactions *between*
those two compounds. if you want a clean model, this *has* to be done with
a pairwise additive potential type (i.e. no many-body, no long-range
electrostatics), hence the use of CHARMM parameters in the paper you
quoted. this gives you a clean separation and no inconsistencies when
computing the potential for each compound, the cross terms between them and
adding everything up.
there is only one drawback: models like CHARMM are less sophisticated than
manybody potentials, and thus you may not get the level of detail and
accuracy that you want. but since you seem to be new to the business, i
*strongly* urge you to set that up first and study how bad (or how good) it
is for your needs. having a baseline with a consistent interaction model is
always a good thing and helps to tell, if more complex approaches actually
provide and improvement or not.
if you want to have a many-body description for *all* components, you will
have to look for a potential type, that has parameters for *all* atom types
in your system. it is a bad idea to use a many-body potential for
interactions between two types of atoms since it will neglect many-body
terms that would be included, if you had a potential suitable for all atom
types. this inconsistency will likely render your simulation worse than
using the CHARMM parameters. however, using two types of tersoff, and REBO
in this weird way is a certain way to doom.
please spend some more time on the concepts of what you are doing and less
on what is technically possible. in classical MD, many things are
technically possible that are complete and utter garbage. many of the older
papers give good introductions into the motivation why certain combinations
of potentials were used. study those and learn from them (but do not
blindly copy them, since things have changed since).