Doubt regarding Meam potential

Dear All,

I want to use MEAM potential to define interaction in C-C and Fe-Fe and lj/cut for Fe-C. in a 3 layer composite (Fe-C-Fe: 1-2-3) This is what I currently have:

#Force Field Using MEAM & lj/cut for Fe-C-Fe system

pair_style hybrid meam lj/cut 10.2 lj/cut 10.2
pair_coeff * * meam Fe3C_library.meam Fe C Fe3C.meam Fe C Fe

pair_coeff 1 2 lj/cut 1 0.02495 3.7
pair_coeff 2 3 lj/cut 2 0.02495 3.7

I wanted to ask if my interpretation is correct. Thanks!

Regards,
Jane

1 Like

Dear All,

I want to use MEAM potential to define interaction in C-C and Fe-Fe and
lj/cut for Fe-C. in a 3 layer composite (Fe-C-Fe: 1-2-3) This is what I
currently have:

#Force Field Using MEAM & lj/cut for Fe-C-Fe system

pair_style hybrid meam lj/cut 10.2 lj/cut 10.2
pair_coeff * * meam Fe3C_library.meam Fe C Fe3C.meam Fe C Fe

pair_coeff 1 2 lj/cut 1 0.02495 3.7
pair_coeff 2 3 lj/cut 2 0.02495 3.7

I wanted to ask if my interpretation is correct. Thanks!

the only way to be certain would be to set up careful tests (e.g. one
Fe layer, i.e. type 1, with meam all by itself, then adding a second
layer, i.e. type 2, but with with 2-2 interactions set to "none" and
1-2 to lj/cut and manually check whether the energies and forces of
the Fe-layer are like you expect them to be, and then do the reverse
and so on)

there are two problems with this kind of approach.

1) there is a fundamental issue. when you replace the Fe-C
interactions with lj/cut, you will also remove the contribution of the
C atoms to the embedding energy. which will make the Fe-Fe
interactions in the Fe layers and the C-C interactions behave as if
they were isolated layers and not in the vicinity of other atoms.
using lj/cut for Fe-C cannot compensate for such a manybody
interaction, as it does't affect the Fe-Fe or C-C.

2) in case you don't care about issue 1), the *clean* way to have 3
separate many-body regions and have them interact via lj/cut would
require defining the meam pair style 3 times, but meam doesn't support
this due to being implemented in fortran in a non-object oriented way,
which is incompatible with specifying multiple instances of the pair
style.

Also, are you sure that your meam parameters were tuned for or are
transferable at all towards representing the kind of system, that you
want to model?

axel.

Dear Dr Axel,

Thanks for your prompt inputs on my doubt. I am already running two simulations as you suggested in 1) and am still waiting for results. I’ve read a couple of papers in literature which have used meam (Cu-Cu & C-C) and lj/cut for Cu-C for Cu-C-Cu composite structures. I’m going to compare the results of above with the results I get using lj/cut for Fe-Fe & tersoff for C-C ; however I wanted to first confirm if my above approach makes sense or I am completely off track.

Also, if I define the interactions in the following manner:
pair_style meam
pair_coeff * * meam Fe3C_library.meam Fe C Fe3C.meam Fe C Fe

Does this take the cross interactions between Fe-C into consideration? I assume not as per my understanding from the reference paper [Mullet et al. 2004] I thus decided to define the lj/cuts for 1-2 and 2-3. What would be another alternative if not lj/cut?

What you are doing will include Fe/C cross-interactions

with MEAM, which does not match what you say you want.

I think you should use MEAM in the pair hybrid command

twice, once for just Fe (via Fe NULL Fe) and once for just

C (via NULL * NULL), then add lj/cut for the 1/2 and 2/3

interactions. The pair hybrid doc page has explation/examples

for something like this.

Steve

What you are doing will include Fe/C cross-interactions
with MEAM, which does not match what you say you want.

I think you should use MEAM in the pair hybrid command
twice, once for just Fe (via Fe NULL Fe) and once for just
C (via NULL * NULL), then add lj/cut for the 1/2 and 2/3
interactions. The pair hybrid doc page has explation/examples
for something like this.

but that won't work for MEAM, because it is static Fortran.
it might work for using MEAM for Fe and something else like Tersoff.
however, the general issue of MEAM's embedding function behaving as if
there were no atoms near it remains.

axel.

good point - I forgot about the Fortran issue with MEAM not

allowing it to be used twice with pair hybrid.

If some Fortran guru knows how to restructure the MEAM lib

so that is do-able, that would be great. Or even better,

re-code a version in C(++) …

Steve

good point - I forgot about the Fortran issue with MEAM not
allowing it to be used twice with pair hybrid.

If some Fortran guru knows how to restructure the MEAM lib
so that is do-able, that would be great. Or even better,
re-code a version in C(++) ...

the effort to redo this in fortran to be like C++ would probably be
more than just redoing MEAM in C++ right away.
i've already created an issue for that.
https://github.com/lammps/lammps/issues/174

axel.

I already tried MEAM twice and read in the mailing list about the same issue, so I ruled that out. If I use meam only once and not use lj/cut for cross-interactions, does that (close to correct)ly define the interactions between Fe-Fe , Fe-C & C-C ?

pair_style meam
pair_coeff * * meam Fe3C_library.meam Fe C Fe3C.meam Fe C Fe

(no lj for cross-interactions assuming meam takes care of the cross-interactions)

There are papers with opposing views which confuse me a lot. Some high end journal papers have used MEAM + lj/cut, some have used EAM+lj/cut+tersoff, some even have only lj/cuts.As I am reading more, I am understanding what Dr Axel means that research is not an activated discovery. I really appreciate all the help Dr Steve, Dr Axel and Aidan are providing to us!

Another I have is to use hybrid/overlay with eam, lj/cut and tersoff. I am currently running several tests to find out what’s happening if I use what, to nail down to the best option.

The MEAM potential has built-in rules for cross-interactions, so just using MEAM and nothing else will probably work as well as any MEAM/LJ hybrid. In the end, the electronic structure of a 3-layer Fe-C-Fe composite is so exotic, it is unlikely that any potential fitted to bulk properties will work very well. The best thing to do is try it out and compare your results to high-quality DFT calculations.

Aidan