I am aware of and have read many previous discussions here on the problematic use of many-body potentials in a hybrid fashion, especially concerning embedded atom potentials. However, I have a question and an idea that I would like to know your opinion on.
My research involves mechanical modeling with a focus on deformations and stress analyses. I have a simulation box containing PURE aluminum polycrystalline (Al type 1 atom), which is reinforced with two types of particles: silicon (Si type 2 atom) and alumina particles (Al2O3: O type 3 and Al type 4). I intend to use EAM potential for 1-1, SW for 2-2, COMB3 for 1-3, 1-4, 3-3, 3-4, 4-4, and ADP for 1-2. I am considering that there is no interaction between particles 2-3 and 2-4.
My question is whether it is possible to set up the hybrid potentials in an acceptable manner. Please let me know.
Based on what I read in the LAMMPS documentation, I have the following idea (which I am not particularly proud of). The idea is based on using overrides to remove undesirable interactions while applying many-body potentials to the entire atoms. Please let me know if this is a worthwhile idea to investigate.
pair_coeff * * adp Si_Au_Al.adp Al Si NULL NULL
pair_coeff 1 1 none
pair_coeff 2 2 none
pair_coeff * * sw MODSi.sw NULL Si NULL NULL
pair_coeff * * comb3 ffield.comb3.NiAlO Al NULL O Al
pair_coeff 1 1 none
pair_coeff * * eam/fs Al1.eam.fs Al NULL NULL NULL
pair_coeff 2 3 none
pair_coeff 2 4 none
Thank you for your reply.
Any additional details on why it is not a viable option would be appreciated.
Do you think the simulation I explained is possible, and do you know any way to implement it using the hybrid setup? Consider that there is no unique potential out there for modeling this problem.
First and foremost, you are treating many-body potentials as pair-wise additive. Internally, there are not only interactions determined by pairs, but also by triples and sometimes even larger tuples. When you cancel pair-wise interactions between atom types, you take out too many.
Second, combining pair styles that have very different methodology is always a big problem. E.g. you have pair styles with charge equilibration, with embedding and so on. Those are not set up to have part of their neighbors missing.
Next, if you combine multiple manybody potentials they must wholly encompass the elements they describe, you cannot have the same element elsewhere. Also potentials used to describe interactions between different sub-groups must be pair-wise additive like lj/cut or morse.
Finally, using a hybrid pair style with many-body potentials is acceptable when the manybody potential parts should describe a whole object, not mixtures and alloys. Then the limitations of using hybrid with manybody are small or sometimes even negligible.
Please note that these points have all been explained at great length in previous discussions and are also to a large degree mentioned in the LAMMPS manual.
So, frankly speaking, what you propose is complete chaos and utter nonsense and thus is best forgotten and never thought of or mentioned again.
That doesn’t make your “hack” any better. You have three options:
pick a different research topic
use quantum chemistry to compute your interactions
learn how to create reliable parameterizations and create your potential file (for “everything”)
Option 3. usually requires a significant amount of 2. anyway, and with everybody being crazy about machine learning these days, you could go down that route. There is a whole “zoo” of machine learning potentials around that promise different levels or accuracy and speed as well as different strategies of training.
Have a nice day, and thank you for the clarification and for being patient with my questions. Cheers!