Michael F. Herbst

Well for me FFT threading suck more.

Thanks for looking into this. I'm also surprised about my bad scaling results for practical examples.

So we just do the scaling manually?

Yes I thought about that. I think a good way forward could be to have some defined input files for the systems (e.g. in QE format or in "our own")...

Hi @dynamic-queries. Thanks for your interest and sorry it took us so long to get back to you. Pretty busy times at the moment ;) Indeed any help on this...

To be tested further (i.e. beyond silicon :smile:).

@antoine-levitt Ready for another look if you get the time.

Could it be that you are running a very large calculation? How many electrons and bands do you have? It could be that our diagtol heuristics is simply very bad...

Actually not only `ScfDiagtol` is a closure, also some of the other callbacks are. It is generally not save to reuse them. Rather you should reuse the arguments to construct...

I think we should add a note to the documentation that makes this statefulness of the callbacks more clear.