2021-06-08 Meeting notes

30 mins

B68 potential parameterization

Daniel Cole

• Go through proposed plan for fitting Buckingham-6-8 type potentials with OpenFF / QUBEKit infrastructure.

DC: Shows slides to introduce the project. The target functional form is the Buckingham damped 6-8 potential for nonbonded interactions with normal bonded terms and electrostatics. We know that the dispersion terms in this form is more physical and introduces more parameters which should allow for more accuracy. This helps direct derivation from QM methods as we normally underestimate C6 parameters but that makes sense as we normally neglect the C8 term.
DC: We already have a proof of concept of how to plugin custom functional forms in the smirnoff format using smirnoff-plugins .
DC: JH has done some testing on this using the Buckingham 68 model on recreating the water properties using the published Rowley model, but we see an error with heat capacity which seems to be consistent between the b68 model and a normal tip3p_fb model.
LP: This may be due to the correction terms which are applied to account for the high frequency vibrations. We need to look into if the correction is being applied or is possibly wrong.
DC: We see that the smirnoff plugin seems to be working as we can recreate the Rowley work to within error bars apart from the heat capacity, so now this works what should we do with the model.
DC: My main idea was to fit a transferable hydrocarbon forcefield using this form as the intermolecular interactions will be dominated by vdW. Can we use evaluator to get a lot of properties to fit to?
DC: We also need some initial values for the b68 form which can come from literature or we can derive them from QM.
LP: I found the thermoML database evaluator draws from are missing a lot of older physical properties, so we may need to do some manual input.
SB: ThermoML might be okay for hydrocarbons but this should be easy to check against.

LP: I think that for hydrocarbons one of the challenges is going to be linear dependencies in the vdW parameters, as there is probably a sub set of space where parameters compensate for each other. We might get around this by being careful with data selection.

DC: restraining to initial QM parameters should also help us stay near the optimum values and avoid this.

DC: Parameter extraction, we can just extract the bond and angle terms from smirnoff or qube maybe with a small refit? Torsions will need a refit but we need to check if we need scaling or not.
LP: have we checked that there is no humps using the damped C6 and exp respulsion term.
DM: We should also check that there is no singalrity at 0 for FEP calculations so we do not have to use a softcore.
DC: The damping function by design should make the dispersion terms go to 0 but we need to check when combined with a lmbda term how fast it goes to zero.
DC: moving on to extracting the dispersion terms we have used the T-S method in the past but we can also use the XDM method to get the C6 and C8 and higher terms.
DC: The repulsive term due to electron density overlap, the b parameter is related to the electron decay rate. If you use an AIM method like DDEC you can get the atomic decay rate.
DC: A combines a lot of quantum effects together but we can derive it from QM using a scaling relation to the free atom volume using a similar method to the T-S method. This does leave some parameters that need to be fit. We can also maybe derive a set of A, b, C6 and C8 which could be transferable. One idea would be to fit a set of bespoke force fields and then look at how similar the parameters are.

LP: Bespoke vs transferable, its hard to define transferability but if we could show some level of transferability that would be fantastic, but this might be hard to do in a short time scale.

LP: I need to look at the Rowley paper but did it perform better? I think that if the water model did not use all of the degrees of freedom it might do better on solution properties as well. We should keep this in mind when optimising the model.

DC: Do we have any data that does not need a free energy.
SB: We have mixture ethalpies available which we can look at.

DC: We need some hydrocarbon physical property data, and look at the liturature for QM starting values and update QUBEKit.