2024-08-12 QCA Dataset submission meeting

Discuss/approve Joshua Horton PR

• LM – Looks fine to me, though I’m not sure what the standards for review are. The dataset creation pipeline was a bit complex.

• LW – I had a quick look at this. Other than the technical side of using DDX, it looks fine to me. Also, the compute tag didn’t like up with what he said it would be (isambard vs mlpepper)

  • JW – I think the tag misalignment isn’t a big deal - Neither of those have been used for our workers or datasets so no risk of error there.

• LW – Other than that, the dataset looks good.

• LM – No objections here. I’ll add tracking and compute-isambard label and merge.

Peptide fragment dataset w/ impl solvent

• CC – I’m planning to use the parameters that JH laid out in this PR as a guide. My big question is whether the QCArchive deduplication will cause a resubmission to reuse the previous torsiondrives, if I just do a 1D scan starting from points in an existing 2D torsiondrive.

  • JW – I’m 70% sure this would be treated as

  • JW--possible that the 1D opt was affected by the other 2D points

  • CC--I think it should have everything it needs to calculate the 1D TD, though

  • Added to BP questions

LW – 2 datasets running on NRP

• One running with a bunch of phosphates, 150ish persistent errors. Most of these have very high magnitude charge. Some opt failures, some segfaults. Is there a way to increase max iterations?

  • JW – I think this is probably on the QCA side.

  • Added to BP questions

Things to ask BP

• If QCA already knows about a 2D torsioindrive, then I submit a 1D slice of that with the same compute spec, will it reuse opts from the 2D torsiondrive?

• What’s the best way to use a higher scf iteration limit for an existing dataset? Should we resubmit with a new compute spec?