Implement real space correlation functions
Opened this issue · 3 comments
Currently we only produce angular Cls. we need another integration step to turn these Cls into correlation functions. It's essentially a matter of integration over Bessel functions, it's documented in the CCL paper for example https://arxiv.org/abs/1812.05995 and this method requires an implementation of FFTLog, which can already be found in Python form here: https://github.com/JoeMcEwen/FAST-PT/blob/master/fastpt/HT.py
It's "just" a matter of turning that code to JAX.
It's necessary if we want to fully reproduce smthg like DES Y1, but not a priority as long as we remain at the level demos and forecasts, where things can remain in the angular cls space.
I can copy the wigner transform code from my repo. I think we can go through it first before doing it. FYI, the transform code doesn't need to be differentiable.
Right, yes great, I can add you to this repo, and you can feel free to add it somewhere in a branch, and then we can iterate.
Ahaha, I recognize my buddy @sukhdeep1989 we don't want no approximation! Brute Force all the way!
Lol, ok, we'll need to benchmark this ^^' And thanks for the code!