Michael F. Herbst

> These two computations are doing literally the same thing Maybe Julia can be more clever in one case and inline something? But indeed very strange this is not exactly...

Yes the test is too simple. Perhaps just run exactly at the Fermi level keeping the electron count once and once slightly above or below and see whether you get...

@epolack @antoine-levitt I changed the way the electrostatics check is done to unify the assumptions / exceptions we did in both the canonical and the grand-canonical ensembles. I think it...

Sounds like a good idea. Thanks for taking a look at that! I wonder whether this should be in DFTK or outside. I guess one could argue both in this...

> This really only happens for the rHF model You're sure this is not because of this being a metal? **Edit:** This is really because of the way we setup...

I found that a different number of irreducible k-points is used between some of the calculations (8 in the displaced finite-diff, 6 in the displaced and 6 in the forward-diff...

I did some more tests with the fix proposed in #744 (which is definitely the right direction to follow) and I think the issue with the failing tests has to...

Looks terribly also like some uninitialised value or something like that.

... or the perturbation breaks one of the symmetries and we filter that out during the response calculation?

Regarding the large norm in `apply_\chi0` we already do something against this, namely here: https://github.com/JuliaMolSim/DFTK.jl/blob/16c1ef1e214e01ce65a7f67429a0ab31352a84f9/src/response/chi0.jl#L351-L353 It should not be a problem to do something similar in solve_ΩplusK_split as well (or...