Dear community,
This is my first post in this forum, so I will try to be as brief and clear as possible. I recently started with MD simulations using LAMMPS and my goal is to learn how to simulate glassy and ceramic systems. I still have no experience in looking for interatomic potentials beyond those provided by LAMMPS, so I hope you will be patient if my questions are from a beginner.
The system I want to model is a glass composed of K2O-Na2O-CaO-SiO2. Searching the literature I found papers where potentials are reported. In particular, I found this reference: https://www.sciencedirect.com/science/article/pii/S0022309318302643#s0085.
This paper reports the pair interaction potential for more elements than I need since they work with a borosilicate glass. Since this reference has the interaction pairs, I wonder if it makes sense to work with those values for the small number of atoms that my problem has.
I would appreciate the help. Best regards.
The number of atoms is irrelevant though your system should contain more than 10s of atoms to avoid strong finite system effects.
Some factors that are relevant for forcefield selection: atom types (chemical elements), thermodynamic conditions (for instance temperature and pressure), mixture composition and crystal/amorphous phases.
If your system setup does not coincide with the parameterization range of the material then you should better look for another force field.
HTH
Evangelos
Dear Evangelos,
Thanks for the reply. Indeed, I have experimental data to corroborate the results of the simulations. The point is that in my system, I have only 5 types of atoms present. I do not have boron, titanium, iron or magnesium, which are the additional elements reported in the paper. In this case, I would like to use this potential with less element. Anyway, thanks for the reply again.
Regards,
Alfredo.
It’s not a problem when you use a subset of elements that were parametrized in the same force field. The problem is when you mix different parameters from different force fields.
1 Like
Thank @hothello for your comment. I will take care by modelling glasses with others elements.
Best.
I would be careful making such a blanket statement. The paper describes potentials for borosilicate glasses, so it is to be assumed that there are significant amounts of silicate and borate in the system. It is not safe to assume, that the potential works well for silicate alone (or borate) unless the publication explicitly has some tests for that. This story is different for atom types that are only present in smaller amounts: here the omission of those atom types is usually safe; a concern would be when simulating systems, where these atom types are present in large quantities.
1 Like
Sure, the scope of the force field depends on the parametrization strategy. As the authors state that the potentials are transferable, I assume it is safe to use a subset of atoms to study similar compounds, i.e. ceramic oxides.
As stated many times over, this is advice from a random guy on an internet forum. It doesn’t ensure that I read the article, the references therein, and I am familiar with the topic of the new research. This is clearly a job for @alfredo_sanchez 
1 Like
Thanks to both @hothello and @akohlmey. Indeed, the paper performs simulations where there is no presence of B2O3 or SiO2, and where there is (Table 1). I understand @akohlmey’s point, but I wouldn’t know how to prove whether the quantity of a particular atom within the elements that make up this potential is sufficient or not. I imagine that, at the end of the day, the experiment will decide the validity of the model.
Just because a force field (FF) is available with the atoms for your intented application, it DOES NOT mean it is transferable if the training set did not include configurations similar to your intented application. For example, there are many reaxff potentials with C/H/O/N atoms but not all have the pi-bond parameters trained so a benzene molecule might behave in a completely unphysical manner, ie hydrolyze within 1 ps when benzene in groundwater has a half life measured in days or weeks (thats why its a such an environmental problem)
The Potential Energy Surface (PES) is an insanely immense mathematical monster. PES of a system with N atoms doesn’t have N points, it has 3N dimensions! The odds are infinitesimal that someone else before you visited the same tiny slice of subspace you’re interested in and made a Machine-Learning Inter-Atomic-Potential (ML-IAP) or a Reax/… Force Field (FF) for you already. Stop looking for potentials from somewhere else, except to practice and learn to maybe get close to what you’re doing.
BOTTOM LINE: For original research there’s no way around having to generate your own DFT and/or experimental data to train a new MLIAP or FF.
plenty of authors make plenty of claims. the unavoidable tradeoff is that the more general a potential is, the less accurate it is, eg. the Universal Force Field (UFF)…
ive always wanted to quote this to somebody …
“You want it to be one way… But it’s the other way…”
you need to validate a model first before trusting any data coming out of an experiment