Michael F. Herbst

> I'm not sure I see what you mean : If the δρ0 out of the first apply_χ0_4P is completely screwd up, how can we deal with it ? You...

Maybe the idea of adapting the Sternheimer tolerance depending on the occupation might at least be a way out. Indeed I would expect more issues close to the Fermi level...

I agree, but perhaps for now we should just ensure in https://github.com/JuliaMolSim/DFTK.jl/blob/master/src/response/hessian.jl#L129 to not overdo it as @gkemlin suggested. I actually did some experiments in #737 and found that `tol_sternheimer=tol`...

@antoine-levitt NLSolve I agree, but about the CG let's discuss. @gkemlin Minor remark: That's the reason why it's usually a good idea to keep to one PR per feature :smile:.

For reference: There is https://github.com/JuliaComputing/FourierTransforms.jl, which is pretty much abandoned, however.

closing discussion in favour of #790.

@LaurentVidal95 There is an issue with MPI. Either disable the routines for MPI or fix it (probably as easy as using a `gather_kpts_scfres`). Will do a more careful review later.

Just nits. Otherwise LGTM.

> Just a note to add checks for MPI in the main routines (tbd by Michael or Laurent) Just realised that will actually be taken care off automatically by the...

This is essentially solved by the AtomsBase interface now.