Andy Jiang

> Yup, creeping numerical noise. What was the behavior before this change? Before the change, the SCF took 40 iterations on a larger system. After the change, it only takes...

> Yeah, for larger / more diffuse basis functions, it's often hard to converge w/ incremental fock. I've already mentioned this in the [docs](https://psicode.org/psi4manual/master/scf.html#cosx-exchange). > > In general, incremental fock...

> I'm slightly confused. So COSX and Direct have different "versions" of incremental Fock build, and this PR has them use the same incremental Fock code? Yes this is correct

> I can start a review once @andyj10224 fixes merge conflicts. Done

> Good work Andy. I like the intent of this PR, which is to consolidate incremental Fock code between different derived `JK` classes. I think that if the incremental Fock...

> > I agree with your points. However, making that change would require a refactor of `DirectJK` (which zero out the J/K matrices at the beginning of every iteration) as...

> Yeah, it's perfectly fine to split things up into multiple PRs. One sensible order I can come up with is: > > 1. Edit the incremental Fock in the...

> > Yeah, it's perfectly fine to split things up into multiple PRs. One sensible order I can come up with is: > > > > > > 1. Edit...

> Immediate questions: @davpoolechem - Is JK stable enough for this PR, or do you need a hold on this while refactoring? @jturney - Does this sound more like a...

> Per the last Psi4 call, I'd like to challenge you to think about how much of libcfmm could be factored out of Psi4. Many other codes might be interested...