/
2021-01-15 QCA Submission Meeting notes

2021-01-15 QCA Submission Meeting notes

Owned by David Dotson, created
Last updated: Apr 20, 2021
3 min read

Date

Jan 15, 2021

Participants

  • @David Dotson

  • @Joshua Horton

  • @Trevor Gokey

  • Ben Pritchard

  • @Pavan Behara

Goals

  • New advancements

  • New submissions

  • Upcoming infrastructure improvements

    • STANDARDSv3 submission machinery in QCSubmit

    • STANDARDSv3 submission machinery in qca-dataset-submission

  • Upcoming science support

    • Enforced c1 symmetry in psi4 is ready

  • Larger advances

    • Automated FF coverage gap identification, torsion prioritization, submission generation

    • Benchmarking (dashboard, etc.)

Discussion topics

Item

Presenter

Notes

Item

Presenter

Notes

QCSubmit - geometric defaults

Josh

  • Optimizations take a lot more steps than they should; reset-hessian = False, should be True.

    • geometric keywords are likely used in the hash for compute (good thing )

  • We’ll re-roll PEPCONF

gau2grid



  • Need to pin gau2grid=2.0.3 on production envs with psi4

PEPCONF issues

Trevor

  • TG: What kinds of SCF changes can we make in QCSubmit?

    • TG: will look for a few settings for DIS, Fock-related things; try out different SCF settings to see what produces more success on problem PEPCONF molecules

  • BP: QCElemental has compute_error, contains error message and error type; also a failed_operation, QCEngine will convert everything that doesn’t succeed into a failed_operation; geomeTRIC stdout is stored in a different place in the QCEngine output for a failed optimization case, but in a different place than it would be for success

Big QCFractal changes

Ben

  • BP: thinking of setting up a test server that features major changes to the server code; is there interest in making each OpenFF submission submit a subsetted dataset submission to this test server as well

    • JH: could have a GHA that pulls qcfractal master as part of this test; could be a GHA on qcsubmit itself as a way to do a canary on qcsubmit for upstream qcfractal changes; could stand up a local server and try to submit to it

New submissions

David

  • TG: QCSubmit question: if I want to add conformations to an already-submitted dataset, is there a good way to do this?

    • JH: slightly awkward with the naming; would have to work out highest conformer number that already exists for each molecule, and start from +1 that; would increment by itself thereafter

    • JH: Load dataset.json as a QCSubmit OptimizationDataset, then add your molecules by looping through the dataset

      • DD: can we get a snippet for how this could work?

      • JH: can do

  • PB: new QCS release?

    • JH: yes, new one with geomeTRIC update

STANDARDSv3 implementation

David

  • TG: Also QCFractal changes possible; are we going to put our custom things into metadata?

    • BP: will be changes coming up, but may not line up into exactly what you’re thinking

  • DD: perhaps we solution around how fractal currently works; only push for fractal changes where we really can’t move forward

    • TG: can we put binary data in string data? arbitrary data structures?

    • BP: valid json works well

Enforced c1 symmetry in psi4 is ready

Josh

  • Just need to try it out and see if it works as expected

    • will do a local test; that would be easiest

    • TG: remember it needs to be a symmetric molecule to take effect

Wavefunctions

Trevor

  • Is the wavefunction reconstruction only in openff-recharge at the moment?

    • JH: yes

    • TG: could wavefunction reconstruction make sense in QC* somewhere (or psi4)?

      • BP: feels like it makes most sense in QCElemental; a single function that consumes a Wavefunction data model and spits out the reconstituted wavefunction

Action items

@David Dotson will re-roll PEPCONF with reset=True, coordsys='dlc'.
@David Dotson will pin gau2grid=2.0.3 on prod envs
@Trevor Gokey will investigate possible SCF parameters to reduce SCF convergence error frequency
@Joshua Horton will cut new QCSubmit release
@Joshua Horton will provide @Trevor Gokey with a code snippet for adding conformations to an already-submitted dataset in a reasonable way
@Joshua Horton will lay out issues, coordinate efforts, for implementing STANDARDSv3 in QCSubmit, qca-dataset-submission (DD will assist)
@Joshua Horton will locally test out enforced c1 symmetry in psi4; determine if it works as expected
@David Dotson will raise issue on wavefunction reconstruction in QCElemental

Decisions