Thanks for making this clear to me (I misunderstood the "needs" in "If additionally the exchange part of the Fock operator needs to be approximated (RI-SCF calculations), a larger fitting basis is needed.").
Do you (or somebody else) has experience/recommendation when to use dscf or ridft (RI-J-HF (+ marij) or RI-JK-HF (+ marij)) for HF calculations?
Here you wroth that
RI-JK is faster than non-RI calculation only for mid-sized molecules with large basis sets. It has mainly been developed as a pre-step for RI-MP2 and RI-CC2 where large basis sets are the default. If RI-JK is used for small basis sets like SVP, it can be much slower than the non-RI or the RI-J calculation!! [...] Use RI-JK - if the number of basis functions is not too high (depends on the amount of memory you have, roughly 1000-1500 basis sets is usually the limit where the non-RI calculation gets more efficient again).
Is this still up to date? What is about RI-J-HF?
When thinking about RI-HF as first steps for a RI-MP2 or RI-CC2 calculation I'm still concerned about the missing density check (see
here, in my tests this is true for RI-J-HF and RI-JK-HF in parallel mode. Or can somebody show me a way/dispel my doubts) but for "just" HF calculations I don't check the density be default anyway ... So I'm thinking about always using RI-J-HF + marij (like I always use RI-DFT + marij instead of just DFT). Can I do so without worrying about molecule size or is there a region where this is slower than non-RI HF?
I hope it's ok that we discuss about this topic in the Ricc2 subforum. If not you can of course shift this post into a new thread in the dscf/ridft subfolder.