search

adematti / cosmopipe

Other
4 stars 3 forks source link

What is going on here in nbodykit_survey.py? #3

Open pmcdonal opened 3 years ago

pmcdonal commented 3 years ago

I edited it to put in zeff, so... not comfortable not understanding what I did... See comment "what is going on here with the multiplication when there is more than one catalog"

https://github.com/adematti/cosmopipe/blob/6564c973459d8783c6083eaf5bbab39cdcecc686/cosmopipe/estimators/power_spectrum/nbodykit_survey.py#L72-L86

adematti commented 3 years ago

Multiple catalogs are only used for cross-correlations, so the actual question is: how to define an effective redshift for a cross-correlation? In eBOSS we have followed https://arxiv.org/pdf/2007.09011.pdf, e.q. 8, that extends straightforwardly to cross-correlations.

pmcdonal commented 3 years ago

I didn't mean with zeff - I put that in - I mean with what was there, i.e., with the product computing wdata2 within the loop. The problem is, when reading this from my point of view, just trying to get going with a simple power spectrum, I have no reason to know when/why there might be more than one catalog. E.g., aren't there actually 2 when the randoms are fed in to compute window? (i.e., like 2 copies of the randoms - I haven't figured out the reason for that) About the zeff though: yes, I'm not surprised you have a formula like that in the paper - I just put in the simplest thing I could do to get going.

adematti commented 3 years ago

Yes, there are 2 when computing the window, as the last one receives the 1/distance^n of the wide-order development (in https://github.com/adematti/cosmopipe/blob/e0ae22b434c415f7fb8028c26276b52974a539ee/cosmopipe/estimators/window_function/fft_window_function.py#L116 - the reason I am changing weight_fkp is because nbodykit tolerates two input catalogs only if they have the same positions and weight_comp - should be changed when revisiting nbodykit's algorithm). Leaving open til I write some comments.