2023-01-10 Meeting notes

Update from MolSSI

• BP is busy

Infrastructure advances

• JW: The OFFTK 0.12 release might’ve accidentally broke qcsubmit and MT fixed it.

• DD: Updating the QCA ML repos with spice metadata. https://qcarchive.molssi.org/apps/ml_datasets/
  

Throughput status

OpenFF PEPCONF OptimizationDataset v1.0

• 3723 → stuck for a few days → 4010 (back to regular pace of around 40 or so calcs a day)

• DD – Also just now submitted a new batch of workers. Are there users?

• PB – YW may be using parts of this dataset, but I haven’t heard from him yet. One external user mentioned that this dataset isn’t complete

User questions/issues

• PB: Just for information, there is a user question about weird energies with psi4, it seems to be a threading issue with in-core algorithm calculations,
  

• PB – I can’t reproduce the submitted issue, but it looks ilke psi4 can be erratic. The QC records show some conformer single point energies computer by UCI cluster being 20-30 hartrees lower than other confs.

• DD – Which psi4 was the compute using?

  • PB – Currently it’s using 1.4 (?)

• DD – How many records show this?

  • PB – User mentioned 40 confs out of 33000.

• JW – Is it possible that the optimization led to a different final geometry in these?

  • PB – These are single points, not opts

• DD – Who is spice set owner?

  • JW – I don’t know. Let me ask lead team.

• PB – I’ll reply that we’re looking into the issue and that it seems to be rooted in psi4.

• DD – We now have SPICE sets listed on ML datasets on QCA webapp, but I have questions about how to fill in metadata.

  • PB – Quality=DFT

Science needs