Page Comparison

Updates from MolSSI

Infrastructure advances

• Geometric next version (on the anvil) fixes issue with net intramolecular torque and net translational force, that results in the molecule rotating during optimization without any change in geometry. This was a problem raised by Chapin on his protein torsion datasets where constrained optimizations were failing to converge, and Lee-Ping diagnosed and fixed it.
  

Throughput status

• OpenFF Protein Capped 3-mer Backbones v1.0

  • Opts: 251942 → 258812 (6870)

  • TDs: 4 → 4

• RNA Single Point Dataset v1.0

  • default: 8864 → 8868 (100 error)

  • wb97m-d3bj/def2-tzvppd: 2538 → 3241

User questions/issues

• DD – JC had reached out about adding the openff default compute spec to each of the RNA sets.

  • PB – I’ve got this on my to do list. I’ll do it in a new PR to keep from complicating error cycling.

• DD – Once LPW makes the new GeometTRIC release, I’ll update the production deployment env to have that. Should we do any testing before we roll that out? Or do we trust LPW’s default settings?

  • PB – I think we can just roll it out directly. The new feature is default behavior so it should just start working. The defaults seem reasonable.

  • CC – Right, the rotation is measured as a fraction of the whole gradient, and the new behavior only triggers if the rotation is a dominant component of the total gradient.

  • DD – It may be worth rerunning some of the old failed jobs, since some of the failures may get recovered by this change.

    • CC – Yeah, I have some jobs I’d like to push back into error cycling to see if we can get some things more complete.

• DD – CC, are these jobs moving fast enough for you?

  • CC – Pretty much, yes. I’m interested to resubmit some jobs from older sets since I think they hung due to rotation issues.

  • DD – PB, could you tag me when the new GeomeTRIC release comes out, and I’ll update the QCF workers?

    • PB – Yes, will do.

Science support needs