k-space TDA

[obackhouse]

Implements k-space TDA.

This PR also changes exactly what is outputed by qsGW. The gf attribute of a qsGW calculation is now the self-consistent Green's function in a larger space than the physical space, the size of which is dictated by nmom_max as in other methods. The quasiparticle energies which previously were stored in this Green's function can be recovered with the qsgw.qp_energy property. This property applies for all momentGW methods, and for other solvers it simply selects the best overlapping poles of the Green's function with each MO.

To-do:

• Compression
  • Not the same as supercell!
• Fock loop
• Self-consistency
  • evGW
  • qsGW
    • Need to check this more thoroughly though
  • scGW
• Parallelism
• Interpolation
• Finite-size q=0, G=0 correction: Should be extra simple for TDA

[obackhouse]

The SRG-qsGW test is still failing on zombie, but on my local machine it agrees with the CI... so i've set it to that value for now

[obackhouse]

Interpolation is now added, but the theory should be discussed - it's definitely not true interpolation and there are a few asterisks, and it's yet to be tested for non-toy systems in more than 1D.

[obackhouse]

The weird qsGW implementation matches the supercell just fine, I'm considering compression to be done even though it doesn't match the supercell as I can see how it would break the k-point symmetry.

[obackhouse]

I've made sure compression is off by default for solids, it should really be benchmarked to work out whether or not it's properly implemented. Same goes for interpolation. Since these are not core features I'm going to merge this.

[ghb24]

Sounds good to me. I'm keen to check the quality of the interpolation for more complicated (3D) systems and compression though. Obviously, FSC will need to be looked at more carefully, as well as the small issue of checking D<3 systems...?

[obackhouse]

Yeah, got all of that in mind