2021-10-08 QC meeting notes

Updates from MolSSI

Ben

• Can run a copy of the DB, test migrations on expanded server storage added this week

• DD: Pavan’s single point dataset somewhat blocked by migration; slow queries that timeout should be addressed by upgrade

  • PB: okay with forward progress at the moment, can compensate with local calculations for now

• PB: am I able to run managers for QCA on my HPC resources?

  • DD: yes, let’s get you set up this afternoon

• QCA main page switched from Netflify to Github Pages

  • may have some questions for OpenFF on their preferences there

  • currently docs for QC* on readthedocs, would like to migrate to using Github Actions and pushing built docs to same repo as the main page

Compute

• JH: Newcastle managers got through XTB tasks

  • have some OpenMM compute up, but not getting tasks even though there should be work to do in the task queue

  • reviewing status, many of the torsiondrives in an incomplete state

  • DD: torsiondrives look like they have exhausted the service limit on public QCA

  • BP: limit at 1500

  • DD: can we double it?

  • BP: can do, set to 3000

New submissions

• dipeptide dataset

  • CC: in order to support non-driven torsions, need support in QCFractal implemented

  • JH: need additional_keywords or similar added to TDKeywords to support injection into the Optimizations

  • BP: not sure where this fits; needs to be unique to the Torsiondrive entry, not the whole collection

  • JH: logic already works for optimizations, so that’s all there

  • DD: I will take this on as a task; this is blocking us on biopolymer work

• RESP dataset (#230)

  • DD: will do final review and merge today; no need for additional keywords

    • SB: was needed for one of the DFT methods, was a pretty loud failure

    • WW: local use of method/basis does not yield errors without these keywords

  • WW: does Simon’s recharge support RESP-style grid?

    • SB: is this the grid ESP is computed on?

      • can’t remember what grid RESP uses, but I think this is a different grid than what recharge uses

      • WW: msk-style is what I need

      • SB: cool, if you can open an issue that could be added pretty easily

• JH: new compute specs on #231

  • hitting validation errors

  • DD: due to pulling in ancient qcportal; need an explicit range-pin in validation conda env

  • JH: also hit issue in production env on Newcastle by pulling in QCElemental 0.23.0, which is incompatible with latest QCFractal

    • DD: pin to 0.22.0; will pin in prod envs

  • SB: how loose is “loose”?

    • JH: using gaussian convergence criteria for geometric

      • loosens everything except energy criteria: https://openff-qcsubmit.readthedocs.io/en/latest/api/generated/openff.qcsubmit.procedures.GeometricProcedure.html#openff.qcsubmit.procedures.GeometricProcedure