Jonathon Misiewicz
Jonathon Misiewicz
I can add this to my reviewing queue, but it will be behind the DLPNO PR, since I've already started on that one.
It's worth mentioning that we are hoping to [switch the DFTD3 version](https://github.com/psi4/psi4/issues/2360) we use to one that, incidentally, is already on conda-forge.
Thanks for the investigation. In that case, this is a question of benchmarking the extra zeroing versus not doing things in a loop...
Per conference call, L2 is moved to 1.5.
Is this PR still ready for review? I notice that one of the todo items isn't checked off yet.
> As for the NL and dispersion question, I also thought that was the plan, but I think both @loriab and especially @JonathonMisiewicz would prefer keeping the "old" behaviour. I'm...
> The internal "validated" dictionary is in my view perfectly well documented here: > > https://github.com/psi4/psi4/blob/72dabe0a889b2026a27a70481c79a2e93164f1dc/psi4/driver/driver_cbs.py#L1346-L1365 So, just to be clear, that applies for every stage in the entire dict?...
Yes, you missed [some conversation](https://github.com/psi4/psi4/pull/2132#discussion_r717021335). Lori and I said that would mean psivar definitions would have to change, which I didn't want in this PR. Susi was against the change...
The PR really is getting better with each pass. There's just a _very_ large amount of code debt to understand how the `cbs` code works.
If there are no ROHF-specific formulas, sure. But if a module has specific methods and formulae for ROHF, I think those need to be checked for CUHF compatibility on a...