2021-10-15 QC meeting notes

Updates from MolSSI

Ben

• BP – Waiting on PR number #693 to finish CI, then can merge and cut release, deploy to production

• DD: should we make the same change on the next branch?

  • BP: will let you know; probably trying a rebase first

Compute

David

• QM and MM workers on Lilac

• ANI and MM workers on Newcastle

• QM, MM, and QM-200g workers on PRP

  • DD – Still needed?

  • PB – ~1000 of my big jobs errored out, I’ll error cycle those manually later today.

• New submissions

• dipeptide dataset – For Chapin Cavender – First PR below allows setting of keywords on collections (will enable constraining chi1 and chi2 on jobs) – Second PR threads the new keywords through.

• imposed electric field in QM; follow-up

  • JHorton implemented new keywords in QCSubmit to allow for perturbed dipole, cut a release as version

  • WW – Which driver is this dataset using? Energy? Optimize? Gradient?

    • DD – This one is an optimization.

    • WW – I was going to see if I could reproduce wavefunction from in-memory and out-of-memory QCEngine calculation. Probably an MP2 problem, since if I change to HF there isn’t a problem. So is this an MP2 issue?

    • DD – Let me make sure I understand – The MP2 calculations with the OpenFF default method completed. Did you see the discrepancy here?

    • WW – I haven’t pulled down the results from QCA. But

    • JW – Clarify the discrepancy?

    • WW – Issue on psi4 – Says that before psi4 1.3.2 has a difference in outcomes for in-memory vs. out-of-memory calculations. But this shouldn’t be affecting me, since all of my calcs are using psi4 1.4.0. But it’s possible that the issue remains for MP2 or multithreaded runs.

    • PB – So this discrepancy is arising from using SBoothroyd’s script to pull down wavefunctions and make ESP grids?

    • WW – It should be the same script to get the in-memory and out of memory calcs. The issue I’m seeing is from MP2.

    • DD – How big is the difference?

    • WW – A normal ESP in atomic units is .0000x, and what I’m seeing is 1.x or 1x.

    • DD – We’re using psi4 1.4 in production

    • WW – I think I’m still seeing this issue when using MP2, will confirm with a clean environment and report what I find in next week’s meeting. Psi4 devs dealt with something similar in the above-linked issues so they will probably be able to fix this quickly if needed.

      • (WW after meeting) – I may have found a relevant post to this here: http://forum.psicode.org/t/hf-esp-slightly-different-as-g09/924/8

  • v1.1 fix:

• (DD merged/submitted above PR)

User questions/issues

Science support needs

Infrastructure needs / advances

• DD – what is the latest QCEngine compatible with prod QCArchive / QCFractal?

  • BP – The latest one. The root cause isn’t QCEngine, it’s QCElemental.

  • DD – We’re having an environment resolution conflict due to

  • BP – Root cause is that QCElemental made a fix that changed a default for serialization. But this requires changes in QCEngine, and then QCFractal. So this needs to bubble up.

  • DD – Ok, so we won’t rush a fix on this.

  • BP – in next, consolidating based on functionality, instead of component; makes adding new functionality easier by not requiring changes in so many distinct places in the codebase

    • recently added the ability to do identifier searches, which would work for e.g. identifiers that are CMILES

  • BP – been talking to Gabe Gomez on ML/quantum, want to use QCArchive

    • Kjell Jorner is a collaborator, works on morfeus

• psi4 on conda-forge