2026.18.02 Cole Lab Check-In (Aus) Meeting Notes

Improved conformer energies with presto

FC

• JM: Might be limitations in capturing coupling with MM force field

• DC: Why is Sage so much better on Folmsbee than JACS set?

  • FC: Agree surprising, need to check

• DC: Might also be nice to see molecules which do well

Processing OMol

FC

• LW: Could benchmark on SPICE only to double-check
  
  

NNPOps PME Issue

LW

• LW: was thrown off by the directionality of delta, had assumed it was i → j instead of j → i. Smee may [?] have made a similar assumption, I opened a PR reversing deltas when the indices are reversed. Joao has confirmed it fixes the issue

  • NNPOps implementation internally consistent but confusing.

Smee fits to QCArchive data

LW

• https://docs.google.com/presentation/d/1P97G0eje1ib5rSJYdjB7PI-eVus0Mm0VrvnX1IeztGg/edit?slide=id.g2f3a95763d8_0_883#slide=id.g2f3a95763d8_0_883

• Long story short: trying to fit just to QCArchive optimization + torsiondrive data (no linearised harmonics) does not work well. It results in similar behaviour as we see with Jen’s work, lots of diminishing force constants. Now that linearised harmonics are merged it will be easier to experiment with those

  •