/
2025-02-05 FF fitting meeting

2025-02-05 FF fitting meeting

Participants

  1. @Lily Wang

  2. Bill Swope

  3. @Jennifer A Clark

  4. @David Mobley

  5. @Jeffrey Wagner

  6. @Pavan Behara

  7. @Matt Thompson

  8. @Trevor Gokey

 

Recording: Video Conferencing, Web Conferencing, Webinars, Screen Sharing

Password: 82mQBpu=

Discussion topics

Item

Presenter

Notes

Item

Presenter

Notes

 

TG

  • JR: BS and I had to symmetrize some molecules by hand in a previous project

    • DM: GAFF does this on purpose because otherwise the bridgehead carbons get different parameters

    • TG: just to clarify, did you symmetrize the SMARTS?

    • JR: we went in by hand and ensured h1 got the same parameter as h2 (on “SMARTS mimics GAFF for nearly all types”)

  • (“Results fixed outliers and overall frequencies”)

  • LW: which molecules is mol36?

    • TG: six-membered ring fused to four membered ring

  • JR: how large are the frequencies producing these MAEs?

    • TG: the objective doesn’t prefer higher or lower frequencies and is unitless

  • JR: you’re just fitting to a Hessian, not the next level up in the Taylor series, right?

    • TG: right

  • TG: I’m using a simple class I FF here, we could likely improve results if we extended it

    • JR: agreed

  • BS: the NEB has a stiffness parameter. If that’s looser, would the NEB pathway look more like the others?

    • TG: I don’t think so. The nudged part of NEB is that it projects out the parallel component of the force, so that the force between two images is the spring. The choice of the spring constant shouldn’t matter, as the molecular component is always projected out.

    • BS: So the NEB pathway is meant to be a minimum energy pathway of sorts?

      • TG: yes

      • BS: what about the number of nodes on the band?

      • TG: played with that too, didn’t affect the results

  • JR: …

    • TG: the metadynamics calls it free energy but it’s still gas phase sampling through space, not a gibbs free energy

    • JR: would recommend using a dispersion-modified functional, like B3LYP-D3*.

  • LW: how well do you think NEB would generalize well over a larger dataset?

    • TG: not sure, main barrier would be implementation in QCA, but LPW has a student working on this

  • BS: how do you find the minima to interpolate?

    • TG: RDKit conformer generation, then optimized.

 

 

 

 

 

 

Action items

Decisions