Dear group members,
My question is about the possibility of using the pair_style hybrid/overlay command with manybody, like Brenner, potentials. I study the impact of carbon clusters with silicon dia crystals. To do that I would prefer to use Stillinger-Weber (sw) potential for Si, Brenner (airebo) potential for carbon, and Tersoff potential for Si-C interactions. Since the formation of silicon-carbide clusters, the pair-wise potentials for Si-C interactions are not appreciated.
To the best of my knowledge, using (as example) the input lines, as shown below, is NOT allowed in LAMMPS.
pair_style hybrid/overlay sw airebo/morse 4.0 tersoff/zbl
pair_coeff * * sw Si.sw Si Si
pair_coeff * * airebo/morse CH.airebo-m C C
pair_coeff * * tersoff/zbl SiC_Erhart-Albe.tersoffZBL Si C
Is there any solution in LAMMPS for my situation of using manybody potentials that include the same elements (here for cross Si-C interactions) with using overlay or any other LAMMPS commands?
Thank you and kind regards,
Victor
P.S. I did see that my topic is probably similar to the one “Question about pair_style hybrid/overlay” from Jan 2010. I agree with Axel’s response to that one, but I hope that my topic is pretty different and also, probably, there were some upgrades to the issue over the last 15 years.
You cannot do that. All of these potentials are many-body potentials and your decomposition would only work for pairwise additive potentials (like lj/cut or morse).
Also, hybrid/overlay is wrong for such a decomposition since it will double count interactions where the mapping overlap (hybrid would be the correct choice) and your pair_coeff commands are completely bogus since you are mapping either atoms of type 1 and type 2 to any of the 3 atom sub-styles. For example, pair_coeff * * tersoff/zbl SiC_Erhart-Albe.tersoffZBL Si C would cover all interactions between atom types: two body interactions 1 1, 1 2, 2 2, and threebody terms 1 1 1, 1 1 2, 1 2 1, 1 2 2, 2 1 1, 2 1 2, 2 2 1, and 2 2 2.
Since you don’t provide a direct link to that post, it is difficult to confirm.
Yours is a fundamentally flawed conceptual problem and there are no code changes that can “resolve” it.
Dear Axel,
Thank you very much for your reply and the comments.
In fact, I noted in my topic that the pair_coeff * * option prohibit to describe specific atomic group/s of the system with a specific manybody potential. And yes, the option to “link” specific group to specific potential is allowed for pairwise potentials only.
For some scientific problems, like mine, employing pairwise potentials tend to produce physically incorrect calculation results. And therefore, I’m not sure if it’s me who has a fundamentally flawed conceptual problem and not the LAMMPS code…
Of course, one can say, if you don’t like the concept, write you own one code. If there is no option to upgrade the LAMMPS since the conceptual or other problem, I would agree with that claim.
It is very obviously you. I am pointing this out to people in this forum (and previously on the lammps-users mailing list) for more than a decade. If you dig through the archives you’ll find more detailed explanations.
That may very well be true, but what you suggest to do is even worse. You expect that when you have two atom types you can assign a different pair style to the 1 1, 1 2, and 2 2 pairs. That is inconsistent to begin with but moreover ignores that you have to also consider 1 1 1, 1 1 2, 1 2 1, 1 2 2, 2 1 1, 2 1 2, 2 2 1, 2 2 2 triples and with your partitioning the 1 1 assignment would include 1 1 1 but nothing else, the 1 2 includes 1 2 1 and 2 1 2, and the 2 2 assignment 2 2 2, but all other triples are unaccounted for.
Also it is not quite clear which style to assign 1 1 2 to. 1 1 or 1 2?
Bottom line, the by far preferred solution is to use the same pair style for all interactions. Hybrid with manybody can only work if the manybody styles are wholly included and the mixed term must be provided by a pairwise additive potential, e.g. if you have a bulk silicon object and a bulk carbon object and they collide (e.g. in nanomachining simulations)
pair_coeff * * sw Si.sw Si NULL
pair_coeff * * airebo CH.airebo NULL C
pair_coeff 1 2 morse ...
Yes, I fully understand and not just understand but fully agree with the concept where using the same pair style for all interactions is preferable. What to do, real life, and its description, especially at the level of the classical approach, does not always fit. For example, among other potentials, Si dia is best described by sw for us. The airebo (or airebo/morse) is preferable for carbon system. Pair wise (like morse) is good for pairwise interactions and not for silicon carbide, especially big clusters. I agree that employing different potentials for 1 1 1, 1 1 2, …, 222 could not be a best solution. But if the overlay option is not used, I do not see a conceptual problem (especially vs morse) in the approach, where the interactions k,l-i, and j-m,n in a system, where i-j represent, for example, Si-C, and k,l and m,n are Si and C, respectively, will be described by another potential than in the case where i-j are CC and/or i-j are Si-Si. Only comparison of the computational results with appropriate experimental measurements can give the answer, whether the approach is correct or not.
Anyway, thank you very much for your answer to my original question that there are no code changes that can “resolve”.
There is, because the A-A-B interaction is as much an A-B interaction as it is an A-A interaction and you cannot decompose it linearly into a superposition of A-A and A-B. The interaction of either A-A or A-B is not constant but depends on the distance to the third atom.
This inconsistency will lead to inconsistent energies and thus non-conservation of energy if the system deforms or similar. Using a pairwise additive potential for the mixed interaction is the only consistent option. In most cases, it doesn’t have to give a very good representation because what mostly matters is the repulsive branch, so in your example, e.g. using just ZBL for mixed terms might give you a better result.
In an experiment, you cannot decompose the results to individual pairs and triples. In fact, a quick look at quantum chemistry shows that “everything interacts with everything” so you already have some simplification here. The questions now are a) is that simplification suitably accurate and b) how do different aproximations combine with each other. So, in the total result (e.g. total energies or geometries) you can have error cancellation and get the expected result for the wrong reasons.
Or put differently, you may have a conceptually wrong model, but the errors may just not show for some conditions.
Dear Axel,
Thank you very much for your productive comments and discussions.
As always, I really appreciate your feedback.
Kind regards, Victor