Hi-
I run a study-group of data-scientists plus some math-and-physics people who’ve decided to learn molecular modeling techniques. We’re beginners, but slowly making progress.
I’m stuck on our current self-assigned exercise: successfully model molecular bond lengths for a range of diatomics, and am hoping I could use ASE + OpenKim to get to a result. I have LAMMPS running as well, but thought I’d focus on ASE for the time-being.
I’ve successfully run a basic model for O2 using the model
“LJ_ElliottAkerson_2015_Universal__MO_959249795837_003”
…but the bond length is way off so I’m thinking I need a better force field.
In other molecular modeling frameworks (e.g Avogadro) i’ve gotten accurate results for O2 with MMFF94 , so was hoping I could easily find the OpenKim equivalent of this Merck model.
After searching via openkim.org and kim_query.get_available_models() i’ve still failed to spot this model.
Does anyone know if MMFF94 is supported by OpenKim, and if so what its official name is?
best wishes,
Kevin Rowney
Hi Kevin,
I wish you the best with your education!
MMFF94 is not supported in OpenKIM, as it is a bonded force-field. We are in the early stages of the process of extending OpenKIM to support these types of models. For now the only one in our system that I know of is the IFF force field, and I do not think using it is straightforward – I do not have experience using it, and a naive attempt to plug it into one of my ASE calculation causes an error.
On that note – you can use any potential that is labeled as “LAMMPS” in ASE! They will work the same as any portable model. It is unsurprising that LJ_ElliottAkerson_2015_Universal__MO_959249795837_003 gave you incorrect results, as it is a very simple model that really only works with noble gases. The fact that it works with every type of atom is more of a utility/testing type of thing.
So I would recommend trying any of the other models that include oxygen. Do not be surprised if many of them give you incorrect answers – a model that is parametrized to simulate a metal oxide, for example, was likely never meant to be used for an oxygen dimer. My first guess would be that some of the ReaxFF models may be good, but I could be wrong. Best of luck!
Cheers, ilia Nikiforov
1 Like
Ilia-
Thanks very much. You’re helping me see something, from a beginners perspective, is a bit overwhelming. I suspect for more advanced students its intuitively clear, but for me the rather large catalog of different categories of force fields is a bit overwhelming.
I’m only beginning to see the outlines ow, but if you have any other clues you can offer me it would help me out a lot. I gather there are different types of parameterizations and different types of underlying models that are best-suited to different specific cases of chemistry.
I see, for instance on wikipedia’s Force Field entry taxonomy of at least 8 different categories (e.g. Classical, Polarizable, Reactive, Coarse Grained, Machine-Learning, Water, etc…)
Am curious to hear how you and the OpenKim project think of the different categories of these models. Sounds like you’ve made choices on prioritizing coverage on some with the intent to cover others later. It’s kind of an open-ended question, but am still curious to know how you think about the taxonomy of these force-fields and am interested to know more about the OpenKim roadmap in these regards.
Thanks again, for you first response to my question.
Kevin
Hi Kevin,
I think you are catching on quite well. While the classification is fuzzy and there is overlap, the main category of models that are in OpenKIM are not on that Wikipedia page. They are described on the Wikipedia page for “interatomic potential”, under “many-body potentials”. I don’t want to state things too broadly, as the categories are not well-defined in the first place and there is a lot of crossover, but a very oversimiplified categorization would be that generally “interatomic potentials” are used by materials scientists to model inorganic solids, while “bonded force-fields” are used for modeling molecules, e.g. drug design.
Generally, the inputs to the two types of models are quite different. While interatomic potentials (IPs) need only the coordinates and types of atoms, bonded force-fields require one to explicitly state which atoms are bonded to each other (take all this with a grain of salt, as I’ve only worked with IPs, not force fields). Because of this, it is nontrivial to extend a framework originally designed with IPs in mind to bonded force-fields. Nevertheless, it is VERY high on our priority list and should be coming soon.
Cheers, ilia
1 Like