2021-04-29 Force Field Release meeting notes

IC Hessian fitting

HJ: I can present results in the next meeting, almost wrapping it up.

Cutom nonbonded forces/functional form

DC: John mentioned some change to the FF format, any updates on that, JW & SB?

SB/JW: Cutom nonbonded forces is still work in progress, hopefully by next release.

Benchmark infra

JW: SB, you asked about the reproducibility of Lim & Hahn results with the new openff-benhcmark infra, I can go through that.

SB: Yeah, I am looking more for the fractal instance type of doing it.

Sulfonamide

CB: Any progress on the sulfonamide pathologies?

PB: Not yet, I still have to do some refit with the molecules matching those parameters from Gen 2 sets.

CB: The equilibrium angles are converging to 90 in 4-membered ring, 60 in 3-membered ring, etc., so we need to know why it is happening?

DLM: It looks like if there is a difference in equilibrium angles then we may have to split the parameter?

CB: I am concerned that we will be imparting the same pathologies in Trevor’s work, we don’t need the equilibrium angles to be 90 for these parameters.

PB: In FB we’re already tying to reduce the delta between QM and MM geometry

CB: Then, I’m wondering what could drive tetrahedral hypervalent sulfur to have these small valence angles

SB: Maybe we should do a “canary” test using a sp2 carbon.

CB: I think the optimizer will need to kow what a “normal” alkane would look like, so this test case should ahve some inear alkanes as well.

SB: Agree. The above test will tell us whether the ring geometries are driving the problem or now.

TG: Following yesterday’s discussion, that’s one of the experiments that we talked about – see how ring terms affect linear alkane terms. So we’d compare

Openff-benchmark

SB – Will current openff-benchmark reproduce Lim+Hahn study?

JW – It should come close, but likely not exact

JW’s “cheat sheet” for running benchmarking workflow locally/in docker: 2021-01-14 Benchmarking Testing Party Meeting notes

DD’s “general instructions” for external partners: Optimization Benchmarking Protocol - Season 1 | C. Multi node, no server/manager setup

SB: What’re the validation tests for the infrastructure?

JW: Here are some of the tests we did before we put it into the hands of industry people, do you have anything in specific you want to see?

SB: Some sanity checks on the numbers we are getting out by manually looking at a smaller set of molecules or even a single molecule and whether we are getting the right units, etc., like regression tests

JW: Yeah, I think this is a good idea, will add on issue tracker.

SB: Thanks for doing that.

JW: Opened

Enumerating stereochemistry

JW: One thing we learned in discussions with Dalke is not to worry about roundtrips much

QM benchmarking

PB – When I’m looking at possible methods to include in the benchmark, which programs can I use? Can we get new methods added to psi4? What chemical/computational constraints should I consider?

• (General) - Should use psi4 for producitons ets, but it’s OK to use stuff like ORCA to study feasibility of a new method/basis, as long as we can get it added to psi4 in a timely way.

• (General) – We can try to contract psi4 devs to prioritize roadmap items, though we’ve never done this before, so we’d need to know a while ahead of time.

• (General) – We should be able to do molecules with up to 50 heavy atoms, on a 64GB system