👀 Overview
Summary | In Approach I it was discovered that the wildly varying pyridine virtual site charge likely came from a couple bugs and from fitting to incomplete data. Re-fitting to the complete set of data, with fixed code, drastically improved the pyridine virtual site, as well as pyridine condensed phase properties. However, the new charge of 0.17 e was now considered potentially too low when presented, and the decision was made to try restraining to initial values in Approach III. |
|---|---|
GitHub repo/branch | |
Status | WON'T PROGRESS |
\uD83D\uDEA9Â Milestones and metrics
Stage | Milestone/Benchmark | Contributors | Deadline | Status |
|---|---|---|---|---|
Train virtual sites and BCCs to existing HF/6-31G* data | Re-fit 2.1.0 BCCs and virtual sites to ESP data, with bugs fixed | Oct 2023 | COMPLETED | |
Re-fit valence and vdW parameters to condensed phase properties | Re-fit FF terms for vsites-candidate, existing vdW sites | Nov 2023 | COMPLETED | |
Re-fit FF terms for vsites-candidate, with N split into two following findings in Approach I | Nov 2023 | COMPLETED | ||
Benchmark | Improved or equivalent performance for molecules with virtual sites added (Cl, Br, pyridines) on training data | Nov 2023 | FAILED |
📊 Progress and findings
Re-fitting to full data
The greatest impact this had was on the pyridine virtual site, which changed from a charge magnitude of 1+e to 0.17e.
Re-fitting to condensed-phase properties
This significantly improved condensed phase properties for pyridine (presentation: https://docs.google.com/presentation/d/1HNK260eEh7dMPGbvQfPHWQ8VPIigvy5tgEXio2bSm-M/edit#slide=id.g25ce1cb38dd_1_26 ), although not more so than simply retraining BCCs. However, it was suggested that the 0.17e virtual site is now lower than expected, so the decision was made to investigate the effect of restraining to the initial guess in Approach III.
