Inclusion of atoms relevant for electrolyte design.

[CompRhys]

Is your feature request related to a problem? Please describe.
We're looking at using Sage in relation to a project on electrolyte design for li-ion batteries. Currently the force field doesn't have parameters for several atoms of interest namely; Si, B, Cs+ and Mg2+.

Describe the solution you'd like
We'd be interested in finding out more about the process of refitting the forcefield and adding new elements and to what extent we may be able to assist. Generally we're interested in the process based on the potential need to add more atoms or ions over time.

Describe alternatives you've considered
We can populate some of the necessary terms based on heuristics and taking terms from other force fields but the concern would be that without refitting the forcefield this may introduce holes etc that prevent stable simulation or simply be misspecified and give poor results.

[mattwthompson]

Hey @CompRhys - somebody else might flesh out more of the details about how this could happen, but here are a couple of other discussions that add some context of why Sage is missing a few not-so-rare elements

• https://github.com/orgs/openforcefield/discussions/26
• https://github.com/orgs/openforcefield/discussions/19

[lilyminium]

Hi @CompRhys! Just to build off Matt's answer, as demonstrated in those links, we have had some interest in these parameters before (especially boron and silicon).

We'd be interested in finding out more about the process of refitting the forcefield and adding new elements

Briefly, for new elements like B/Si we would need both bonded and non-bonded parameters. For bonded parameters we can develop starting point parameters using a tool such as OpenFF BespokeFit or fitting parameters to computed QM data. For non-bonded parameters it's a bit trickier -- we would need to obtain initial LJ values from somewhere and fit these to physical property data. In addition, our charge model (currently AM1-BCC) does not currently address all these elements, e.g. B. We are however exploring to other models such as to a graph neural network. Here, our initial release is focused on getting the NN to reproduce AM1-BCC, so the chemistries are still limited, but future iterations will look beyond AM1-BCC to fit directly to QM data.

Also, I believe @jthorton has done some work with boron parameters in this space.

to what extent we may be able to assist

As mentioned in Matt's links I'm not sure we have a formal process for this, or for potential collaborations -- @davidlmobley would be best able to advise here. However, any assistance would be very welcome! I do know for example that in the past, people have contributed datasets that would be useful for fitting but are hard to get, e.g. condensed phase data.

[mrshirts]

At CU we are investigating parameterization of consistent ions in the relatively short range timescale (~6-12 months) at OpenFF, though to start just in a LJ 12-6 functional form. We are interested in adding more data to the fitting such as osmotic coefficients which should in theory help get the concentration dependent terms better.

[davidlmobley]

@CompRhys - depending on whether/how much effort you want to put in to this, it's possible we could set up a call to kind of walk you through what would be involved. As noted the largest hurdle for boron and silicon is condensed phase data for boron/silicon parameters. Probably someone wanting to push on this (perhaps you) and a sync with Danny Cole's group at Newcastle might be enough to get this going again, as these would be of interest for us in our mainline FFs and the LJ parameters are the main things blocking us.

Ion parameters will require someone focused on parameter assessment/fitting -- basically doing almost the same thing literature parameters have done, but with our tools/consistent with the rest of our FF. Shouldn't be very hard, but it will take someone working on it a bit. Again, if you want this to be you, we'd be very open to discussing further -- the main thing impeding this is just lack of human time to prioritize it.

[CompRhys]

We've decided to try validate our proposed workflow on a subset of restricted chemistries not including the elements above and then decide based on those results how to move forward. If those results are positive I'll come back here!