2025-06-25 Mitchell/Wagner Check-in meeting notes

JM

• Had some broad asks from OMSF, generally cool but somewhat open-ended

JW

• Thoughts on reducing anti-recommendations? github.com/orgs/openforcefield/discussions/46

  • JM – There’s a difference between “the infrastructure works” and “use this method for your science problem” - Eg our infra might get the right potential energy, but if we’re recommending a workflow with no minimization/heating/equilihration then folks should do their own homework before doing “science”

  • (General) – Maybe what we want here is a general policy for what we take ownership of and what we don’t - eg dEs from force fields matching QM: yes, free energies from long protocols: no.

    • JM – eg, guarantee a function (coordinates) → potential energy, forces (force field), but not any way to choose coordinates to plug in

    • JW – I think if we’re really strict about that then we’d never really have examples that went past create_openmm_system. But we need to bring utility to users to remain funded, so I wonder what we do about the sorites we take outside our borders and how we describe those

    • JM – But if we’re trying to get rid of anti-recommendations, we can upgrade to saying “nothing” without going all the way to “this is the authoritative way to do this workflow”. And we can either explicitly say that our workflows aren’t best practice, or imply this by having obviously unrealistic settings (like saving trajectory every 10 steps, or running 1 minute walltime sims)

      • JW – These do fly over a few peoples' heads - first year grad students sometimes run production with frames every 100 steps

      • JM – I’d love to make instructional material for basics of MD

    • JW – Could you write up your thoughts on the above discussion (and make sure to include a recommended immediate course of action, even if the conversation continues for longer)?

      • JM – Will do, I’ll think about this more and write something up.

      •  

    •  

• How’s pablo going?

  • JM – Would like some thoughts on API decisions - I have some ideas but they’re somewhat unusual.

  • (JM shows demo of with/protonation variant stuff/CONECT-based loading)

  • JW – Wow, that looks great. The API seems super clear and suited for this use case.

  •