2023-08-09 FF Fitting Meeting

Updates on NEB

TGokey

https://docs.google.com/presentation/d/1FSmTvQNs8sAe0cBAsbH5BINk06M0Cv3GY9GGb6Vkzc4/edit#slide=id.p1

• Slide 9

  • BS: question from Jeff last week: if you integrated this force along the path, do you get the energy difference?

    • TG: I haven’t finished checking the line integrals yet. The improper force is very high, indicating that we want our improper term to push away from planar, which is the opposite of our current situation

  • CC: units? Shouldn’t this have units of energy over length? It’s just eV/degrees here. Force in cartesian space should have energy over length as units, and you’re just projecting it over the dihedrals, so the units on this page should still have energy over length

    • TG: so to clarify here, you’re saying it shouldn’t be eV/degrees?

    • CC: saying it should be eV over A or nm

    • TG: I’ll think about the units on this slide. I used Lee Ping’s code to convert the Cartesians to internal gradients, I’ll ask him about the conversion

    • BS: the generalised force should be energy over degrees or radians

    • TG: No I checked that – Lee Ping used Angstrom and degrees. Regardless, shouldn’t change the directions, just the scales

    • CC: but if it’s a generalised force it’s not actually a projection. I’m not understanding what you’re doing to get the arrows on this plot

    • TG: form the Wilson B-matrix to convert Cartesians to internals; I used Lee Ping’s code, check that for me…

    • CC: if that’s the process, the units are correct for it to be a generalised force

    • TG: why should it be a generalised force?

    • CC: a generalised force is a quantity you can take the dot product of to get the total energy of the system. (…)

• DLM: what’s next?

  • TG: working on line integrals now… if I were to look at something else, I would fit an actual FF, but I don’t expect too much to change since the shapes are generally similar. I’ve been building a small dataset, and I’ll test this approach on those more complicated molecules. So generally scaling up and seeing if NEB gives us something new that TD doesn’t.

• BS: interesting that MM is steeper than QM – any reason you can think of that’s the case?

  • DLM: think I was partially getting at this the other week. We often see that MM impropers end up too stiff since it’s not always well described by a simple harmonic. e.g. the minimum shape can broaden flatly and then go up steeply, or there could be two minima away from zero. That might explain the reason the out of plane bend in MM doesn’t go that far.

  • TG: my question here is if it’s a non-bonded issue or a valence issue. If it’s a valence issue I can fit that.

  • DLM: if you’re interested you could try modifying the functional form of the improper

  • TG: sure – I would need snapshots of the improper bending.

  • DLM: you could do a scan of the improper

  • TG: I wonder if the flexibility of the improper is dependent on temperature

  • PB: aren’t you showing the 2D torsiondrive?

  • DLM: shows example of scan of improper angle at 0 degrees: https://openforcefieldgroup.slack.com/archives/C8P54J2JY/p1691613182717739

  • TG: I do have new molecules with N that I’m hoping to study. But I think the qn is if we really want to fit to 2D scans

  • DLM: Don’t know that we do due to computational expense – but there may be low hanging fruit here that harmonises with your chemical perception work. Possibly, we can use this data to work out how and when to divide improper types. If NEB can help us get this data, that could be compelling reason to try using it. We already know we don’t model pucker very well

  • TG: one thing I want to do is figure out why the pucker is there in the first place. Normally we fit TDs to torsions only. I want to figure out what we want to be fitting to NEBs (e.g. bonds, angles, too)

• TG: I talked with Bill about potential molecules. Are there any potential molecules we’d like to see? I have to manually chose the 2 degrees of freedom we want to scan for each molecule so there is some human expense here

  • e.g. Bill’s given me some peptide amide bonds; ethers and esters; a few different valences of sulfur, phosphorus, and N.

  • DLM: suggests molecules in Slack link above

  • TG: I will need to do NEB on molecules as well, so I’m aiming for 10-20 molecules

Additional SI from DLM, PB, and Jessica Maat

Functional form:

From Jessica's old messages, improper functional form for two minima cases: <Improper smirks="[:1]-#7X3:2-[:4]" id="i5" sqrtk1="2.051309243917e+00" sqrtk2="1.305089015005e-01" k1="4.207869614179e+00" periodicity1="4" phase1="180." k2="-1.703257337088e-02" periodicity2="8.0" phase2="-180." parameterize="sqrtk1, sqrtk2" parameter_eval="k1=PRM['PeriodicTorsionForce/Improper/sqrtk1/[:1]-#7X3:2-[:4]']**2, k2=-1PRM['PeriodicTorsionForce/Improper/sqrtk2/[:1]-#7X3:2-[*:4]']**2"/>