vdW terms, ML water models | Chris Fennell | LW: The water model you talked about on the last slide: is that still in progress? CF: Yes, I didn’t actually show the best 3-point model, that was a very recent (last week) development that involved using our existing hyper-surface ML model but changing a new factor, the bond length (maybe other things too).
CC: Have you considered changing the combining rules for the FF? LW: Was that 50 or 15 ns? LW: How much do values bounce around during in-serial parameterisation? CF: It depends, you need to choose how to perturb it, is it a 10% change? It’ll diverge. We then make 2% changes and then linearly interpolate the target. Sometimes if you don’t run the simulation long enough the dielectric constant won’t converge. It usually takes 2 to 4 loops.
LW: Did you look at non-alcohols e.g. carboxylic acids? LW: Have you looked at the hydrocarbon parameters? When we refit GAFF the hydrocarbon size decreased but that’s the opposite of what you’re seeing. JAC: It seems like the traditional GAFF FF trends toward experiment but does it become slower than experiment as the hydrocarbon chain continues to increase? JAC: Have you considered testing the viscosity of these last 7 hydroxyl models? There is usually a trade-off with modeling diffusivity and viscosity.
|