2020-04-23 QCFractal Meeting notes

Get updates from MolSSI

• Working on lots of management automation

• A bit backlogged on normal tasks, due to high maintenance load

• Feedback on optimizationdataset interface? Considering refactoring

  • Missing functions to query status of collection, etc

  • We can share feedback in #qcfractal channel

  • JH – Have to run status twice on datasets, since first call always fails

• Will connect @David Dotson with Ben, to see about taking over mundane maintenance/task management

Perform manager maintenance if needed

• PRP, Galileo, UCI HPC3 are up

• Lilac is down

Work on QCF integration into relevant code

• Intermediates in geometry opts are taking up lots of space –

  • Geometries aren’t that expensive

  • Text output (STDOUT) is really large –

    • We could keep it in the short term, but delete it after some sunset period

    • Could have option to not delete output if the geometry opt failed

• JH – Can we paralellize hessian calculations? Since they’re done using finite difference, each displacement could be a separate task

  • BP – Not sure that they’re done by finite displacement. But it’s efficient to use output of last wavefunction as initial guess for next one. I’m curious to see how this would work.

  • JH – Can we make use of the option to store the wavefunction?

• JM – When will there be time for more datasets?

  • Queue planning:

    • Finish fitting dataset

      • 750 core hours for remaining 700 hessians = 535000 core hours remain

      • We can continuously use ~2500 cores

      • This is about 8.5 days remaining

    • Produce benchmarking set – Fitting datasets will be deprioritized Apr 30 – Medium priority

      • ~1600 1-d torsions

      • Similar size for opt/hessian dataset

      • This is to-be-decided, will use first benchmark set in the short term

    • Everything else (submit at low priority)

      • HJ – Theory benchmarking set

        • 40 1-D torsions using different levels of theory

      • JM – Grid optimization set

        • Like torsion scan, looking at pyrimidal atom inversion