| | @Jessica Maat (Deactivated) | JM: Here are the graphs showing improper angles on the x-axis, wbo values on the y-axis, and I made interactive plotmol plots that shows the structures as well. CB: The structure on the right with wbo=0.96, having two O=S(=O)(CH3) attached will contribute to the pyramidalization of Nitrogen since they’re heavy and do not stay in the plane as they’re being pushed away due to sterics.
Similarly, on the left with wbo=0.9 (lowest) ortho substituents of ethyl will push and makes it slightly pyramidal.
At the highest wbo of 1.03 highly electron withdrawing ortho nitrogen and more planar.
Does this tell us that WBO doesn't have anything to do with this? SB: I kind of agree, it all seems tricky, we get these horrendous pictures of our beautiful hypothesis. I wonder what will happen if we study only simplest molecules and end up similar to the situation with rings, would this create a problem? CB: Yes, lot of our confounding issues should be addressed in other areas of forcefield, why do we need bondorders I can show the motivation for this work,
the outlier in this JACS set is an aryl & thiophene attached, unsubstituted case goes planar when it shouldn’t I believe this is a FF issue
Electronic effects of delocalization matter for entire chemotypes, we see these complications everywhere in our training & testing SB: I agree with that. I worry if we live in this world of simplicity. I wonder if we should look at metrics other than wbo like Danny Cole uses electron density maps. CB: Charge is one metric but charge models may or may not describe conjugation.
Arnie Hagler has been advocating for a crossterm between proper and improper for a long time and I am running far away from that because of the complexity involved.
May be we should do the same for aryl amines as we discussed before with biaryls where we build a simpler pattern that doesn’t have any ortho substituents. SB: Aniline dataset was created with that in mind I guess, JM did you look at that yet? We talked about plots with the spectrum of pyramidalization few weeks back. JM: I didn’t yet, I will do. DLM/SB: Yeah, we are having the same conversation again and again with the sterics, etc., we should avoid them in future and look at the new simpler datasets.
|
|
| @Pavan Behara
| Slides here PB: I have an update on the status of QM theory benchmarking. This is more like literature review and soliciting feedback on the steps ahead. SO, in general QM methods are benchmarked against a combination of datasets with different molecular properties, some of the major ones are Database2015 from Don Truhlar’s, GMTKN55 from Grimme’s, MGCDB84 from Head-Gordon’s groups. From these sets the industry best looks like wB97M-V but we don’t have a faster implementation in psi4 yet since VV10 derivatives are still analytical, there are D3 versions available though. Some other functionals of interest are M06-2X-D3(0), M05-2X-D3(0), M08-HX-D3(0), B3LYP-NL, PW6B95-D3.
For now HJ submitted a different basis sets within B3LYP-D3BJ our current level of theory, we may need some more calcs with other functionals.
Also, need to decide on the baseline to compare, CCSD(T) or its variants. CB: I think you’re doing everything from scratch, IIRC Victoria Lim did something like extrapolating to Complete Basis set limit, or other work five years back.
Also, from the zoo of density functionals we can’t be switching to a new level of theory every 2-3 years throwing away the huge amount of data we already have, I believe we already made an optimal choice when we started this effort. DLM: We can talk to VL and see how far she did. PB: Okay. CB: Can you show me some of the molecules that describe Noncovalent interactions from the benchmark sets you’re talking about? PB: I don’t have them handy, but these are the molecules in the set LPW created to study the correctness of theory we are using now.
I am not proposing any changes right away based on literature review, I am looking at the best functionals they reported but most of the studies are done at large basis sets like Quadruple basis, etc., I am trying to find out how those will translate to our needs with computation time and accuracy in mind. B3LYP has known deficiencies and most papers point that out, in one of the recent papers by Kieron Burke they show how torsional profiles are misfit with B3LYP even with dispersion correction. So, this is to affirm how good we are doing now and still a work in progress. CB: Okay, when I look at LPW’s molecules I can see few molecules that are sterically hindered, I would like to see some more with strong steric interactions since those are the most troublesome for us so far. For example, a cyclic propane with two ethyl substituents, and biaryls, etc.. PB: Sure, will add more such molecules. TG: So, what prompted this study? Do you have a list of molecules from our current training or test sets that show bad behavior? PB: I am not sure about that, LPW & HJ may shed some light on that. TG: Okay, one possibility is that they might have included that in the test set you’re showing.
|