Michael F. Herbst

Yes, see [tests](https://github.com/JuliaMolSim/DFTK.jl/blob/master/test/PspHgh.jl#L129)

> compare against another DFT and hope for the best! I think ABINIT has TM pseudos, which has the advantage that we have parsers for their densities (i.e. you could...

Key point: Make hamiltonian fully immutable, i.e. change both potential and density arrays *at each update* with a new density. In brief: Cut `update_hamiltonian!` completely for `update_hamiltonian`.

> Should we try and get rid of mutation as far as we can? Yes we should, but to be fair quite a few of the patterns you list above...

Super awesome! I have not had a look at the code, but the interface is pretty much exactly how I'd like to have it.

> If julia and BLAS threading are not active at the same (which I think is already the case right now) No you have to actively make sure about that....

Ah I see. Yes that's true.

For sure that part is not parallel.

I'd be careful with the perm due to the closure issue in the `Threads.@threads`

My desktop (openblas, fftw): ``` 1 Julia threads 1 FFT threads 1 OP FFT (ms, nothread): 5.699495 1 OP FFT (ms, thread): 5.786136 1 IP FFT (ms, nothread): 5.621367 1...