More flexible mixing interface #225
Conversation
|
I don't think we should have that logic here. Instead this can be implemented as a custom mixing provided it has sufficient info (like the iteration number). This we can do eg with an info named tuple like we do for the stopping criterion |
|
Sounds reasonable. Thanks for the suggestion! |
src/scf/mixing.jl
Outdated
| Gsq = [sum(abs2, basis.model.recip_lattice * G) | ||
| for G in G_vectors(basis)] | ||
| ρin = ρin.fourier | ||
| ρout = ρout.fourier | ||
| ρnext = @. ρin + m.α * (ρout - ρin) * Gsq / (m.G0^2 + Gsq) | ||
| ρnext = @. ρin + T(params.α) * (ρout - ρin) * Gsq / (T(params.kF)^2 + Gsq) |
There was a problem hiding this comment.
something that's beginning to irritate me is that we manipulate densities in at least three different formats (fourier, real, and RFA). Maybe we should begin thinking of a convention on how to call them to make it clear? Maybe not...
There was a problem hiding this comment.
I agree with that, but for the moment I don't see a simple solution but to annotate the types. Like if it is an RFA, then we annotate, otherwise not. I'll leave it for now as is.
Brought forward from mixing. Needs to be polished.