2020-06-11 QCFractal Meeting notes

Get updates from MolSSI

• Still working on frontend refactor, API may change. Backend may chang but that will all be hidden.

Perform manager maintenance if needed

• Consolidate managers on PRP while we are working on debugging

• JW will ask John to double check lilac is totally off

Work on QCF integration into relevant code

• MM in QCEngine

  • Blocked by the issue where molecules that have previously been submitted for QM and don’t have CMILES are immutable, so we can not run MM operations

  • No clear solution rght now, but at the very least we can resubmit the same molecuels for MM calcs by perturbing the coordinates a little bit. Not sure how to ahndle this moving forward.

• DD – Some issies on qca-ds-sub issue tracker

  • • DD – Could resubmit. JM what do you think?

    • JM – Yes, resubmit.

    • JW – I wonder if this is a consistent problem with input, where we’re picking tricky molecules for MM force fields, so our initial conformations are going to be far from ideal

    • HJ – Maybe we could increase the number of iterations during optimization, so we don’t run into iteration limits.

    • DD – We might consider making a 2.0 of this dataset with more carefully-generated geometries. So we could, in order:

      • Resubmit

      • Increase number of iterations

      • Make 2.0 dataset

    • HJ – I’ve seen these convergence errors before. I used level-shifting to resolve that, and I wonder if we should consider other settings that affect convergence.

    • DD – Is it possible to change these settings on an existing calculation?

    • BP – Changing level shifting is a modification to the calculation, so that would need to be unique. It’s hard to separate theoretical from technical configurations in calculations, since they all affect results.

  • DD – Also looking at

    • DD – Will resubmit and see if errors resolve

    • HJ – That sounds good. I recall seeing some convergence errors here.

    • DD – DGS mentioned that it was many of the larger molecules that were failing

    • HJ – This may indicate that we don’t want to consider torsions in large (>30 heavy atoms) molecules, since those would be fragmented anyway. So we should give other errors priority over these

    • DD – I’ll report back the results of resubmitting these. Later we could also make a 2.0 dataset that drops these tough ones.

  • DD will draft a “standard playbook” for how to handle submissions and communication, and will post it to the qca-ds-sub README

  • DD - Is QCSubmit mature? Should everyone use this for OpenFF submissions?

    • JW – I don’t think we should push adoption of QCSubmit yet. It’s going to change a lot and I don’t want Josh’s changes to slow everyone else down or break their scripts. Maybe our next dataset submissions should use QCSubmit, but people should schedule calls with Josh to ensure they’re using it correctly.

    • HJ – I’d be willing to spend a few hours with JH to do my next submission using QCSubmit.

Zombie lilac manager

• JW will also ping @John Chodera to try and track down remaining lilac manager