All pair coeffs are not set with pair_style hybrid command

Dear Axel,

My understanding of principle of EAM potential is that this potential functions with the combination of embedding function and density function inputting parameters of each elements according to distance between two atoms.

Thus, EAM potential is empirically optimized for certain alloy system and it doesn’t make sense using two different EAM potentials or one EAM potential and Lj potentials together, right?

At first, I took a simple approach from adding two types of potentials, EAM and LJ potentials, ignoring the operating principle of two different potentials.

Best regards,
Woo Cheol Jeon

Dear Axel,

My understanding of principle of EAM potential is that this
potential functions with the combination of embedding function and density
function inputting parameters of each elements according to distance
between two atoms.

Thus, EAM potential is empirically optimized for certain alloy system and
it doesn't make sense using two different EAM potentials or one EAM
potential and Lj potentials together, right?

​no. the issue is, that the embedding function depends on the density
contributions of *all* neighboring atoms​. when using pair_style hybrid,
you compute each subsystem separately, so your embedding term will be way
off, if you combine many of those or replace EAM with LJ. the error caused
by this kind of inconsistency is small, if you have two distinct
subsystems, e.g. a metal droplet on a silicon or carbon surface, where you
would approximate the inter-subsystem interactions with LJ or Morse. but
inside an alloy, this is bound to lead to problems. this is a very well
understood issue in the context of QM/MM coupling for force fields.

​the bottom line: there are many mixed interactions models that can be
built using pair style hybrid or hybrid/overlay. a large part of those will
result in calculations that do not crash or are grossly incorrect, but that
doesn't make that a good or even an acceptable solution.

​in my personal opinion, the whole justification of the paper you quoted is
flawed. the fact, that a suitable parameterization isn't readily available
is no justification for using a known to be flawed approach, even more so,
if the validation tests have flaws, too. also, there is a straightforward
approach to rectify it (i.e. build a combined potential file). otherwise,
the saying "if you cannot stand the heat, get out of the kitchen" applies.

axel.​

Dear Axel,

I understood your saying on why I can’t use more than two EAM potentials or EAM and LJ parameters together with pairs_style hybrid command.

The points I missed are that this command works on the system composed of two separated subsystems and the embedding function is effected by the contributions on density of every neighboring atoms in whole system.

Forgive my ignorance on the principle of embedding function.

I’m so sorry, but I have one more question.

Is there any tool to aid combining more than two eam potential into one eam potential file related to LAMMPS?

Thank you

Best regards,
Woo Cheol Jeon

Dear Axel,

I understood your saying on why I can't use more than two EAM potentials
or EAM and LJ parameters together with pairs_style hybrid command.

The points I missed are that this command works on the system composed of
two separated subsystems and the embedding function is effected by the
contributions on density of every neighboring atoms in whole system.

Forgive my ignorance on the principle of embedding function.

I'm so sorry, but I have one more question.

Is there any tool to aid combining more than two eam potential into one
eam potential file related to LAMMPS?

​it doesn't make too much sense to combine existing potentials, as they are
usually parameterized with different strategies and for different
conditions. but rather one would write a code to generate a custom
potential file from the original parameters of _matching_ parameterizations
in the respective publication(s). one example for that is in the
tools/eam_database folder.​ ​you may also want to look at the
tools/eam_generate folder.

axel.

Dear Axel,

OK. I will try to look both eam_database and eam_generate…

Your comments really helped me a lot!

Thanks

Best regards,
Woo Cheol Jeon