Updates on NEB | TGokey | Slides will be uploaded Slide 3: dashed line shows driving different atoms for the same bond Slide 4: CB: if you let a ball roll downhill from the saddle point, it wouldn’t follow the TD path BS: isn’t the minimum energy pathway TD? TG: yes CB: I’d be looking at perpendicular to the isocontour JW: are they all minimum pathways? they all go to the same minimum through the same maximum CB: Is the question that we’re asking: why are the two pathways different? TG: yes, in part BS: because there are springs connecting the nodes, my intuition is that there’s tension on the NEB making it more straight CB: but is a bigger question which one gives better inputs for parametrizing FF? TG: yes, we want to know if NEB offers up new information we could use, or generally characterise differences between NEB and TD JW: if I had to guess which one would be more useful for fitting, I’d say TD; NEB doesn’t explore OOP bend at all TG: I haven’t fit any actual FFs on this data, but results of FFs fit to NEB would have the improper be artificially stiff
Slide 8: the arrows reflect the contribution of the angles to the gradient / the arrows are the force projected onto the two torsions BS: are these energy gradients or forces? Why are the arrows pointing uphill? TG: I think what this is saying that it wants the sp2 C to be more sp3-like
DLM: with a FF fit, our impropers might be so coarse that we might see no differences or poor results CB: the data shows the improper and torsion are coupled. The OOP is very soft at 90 degrees, but stiff at 0 degrees. Isn’t this just coupling, which we don’t do BS: at 90 degrees you’re 10 kcal above the minima, it might not matter CB: I agree, it’s more important to have correct behaviour around the minima. But until we couple torsions, not sure how we can see differences from the NEB deformation DLM: might be ok for our impropers to be softer in general. We saw before in general that our impropers were too stiff CB: Yes, but if we have perfect stiffness around the minimum, it still wouldn’t be soft enough around the saddle point DLM: shares heatmaps of valence angle vs improper angle – will share data, this would be important to eventually look at BS: functional form is a single minimum? DLM: yes, but can easily change it to have multiple minima JW: … DLM: we could allow minima away from zero DLM: somewhere I have what we thought the form should be, we don’t need to change FF engines TG – would look at 1-4s with impropers
Slide 9: Slide 5: CB: are we fitting to both energy and gradient? In theory both NEB and TD should give the same FF if we do PB: no, we only fit to energies with TorsionProfile targets. If AbInitio we could do both CB: if we fit the hessian at critical points e.g. 90 degrees, that would be very informative. Also if we fit forces at the same points as energies, in principle it shouldn’t matter if we sample the TD or NEB pathway BS: in fact if we fit to forces you want to be off the minimum to get all the information PB: fitting to ICHs is available in FB TG: probably would want both equilibria and off-equilibria CB: there’s a lot of ways to go off-equilibria, some of which can be suboptimal. The nice thing about your NEB is that the sampling of the pathway is quite good TG: would worry that if we only fit to TD forces, or use TD forces first, we might end up with weird results
BS / DLM: eager to see results of FF fit TG: have been putting this off as not sure how to quantify improvement JW: we’re working on an internal benchmarking package TG: will it include additional benchmarks? JW: not to start with TG: I’d like to be able to fit to a bunch of conformers and get an overall consensus, current benchmarks fit to QM minima BS: one check could be reproducing the contour map surface TG: yes, although the standard set is 1000 molecules. Storage could get expensive DLM: happy to help figure out what success means (in chat)
|