gh-85: add correlation function class within summary statistics#92
Conversation
…-Pos works, She-Pos and She-SHe give weird results
|
Hi @llinke1 , thx for the PR! I'll start reviewing it:) |
|
Thanks @marcobonici! I think I would prefer the FFTLog implementation to be a separate issue. To me it is useful to already have a working (although slow) implementation of correlation functions, so the later FFTLog implementation could be compared against it. I am happy to help with the FFTLog implementation of course, although I am not yet familiar with |
|
Hi @llinke1 , first and foremost sorry for keeping this PR open for a while.
Let me know if you need to discuss anything. |
|
Hi @marcobonici , |
|
Very quickly:
|
|
@llinke1 @marcobonici Laila already have written permissions here, it does not make sense to go with forks :) Just change @llinke1 role in playground |
…implementation to `angular_correlation_function_wigner`
|
Hi @marcobonici ,
Should I create a PR in |
|
Hi @llinke1 sorry for the delay, but I was sick. |
|
@all-contributors please add @llinke1 for code |
|
I've put up a pull request to add @llinke1! 🎉 |
…s_within_summary_statistics gh-85: add correlation function class within summary statistics
…s_within_summary_statistics gh-85: add correlation function class within summary statistics
…s_within_summary_statistics gh-85: add correlation function class within summary statistics
Added the class$\xi_+$ and $\xi_-$ and their counterparts for galaxy clustering and galaxy-galaxy-lensing), using an already defined
CorrelationFunctionto the "summary_statistics" folder as requested in #85This class can compute projected correlation functions (i.e. the cosmic shear functions
AngularTwoPointobject.The Hankel transforms are carried out by explicit summation over the Wigner-$d$-matrices:
$$ \xi^{ab}\pm(\theta) = \sum\ell\frac{2\ell+1}{4\pi},(\pm1)^{s_b}, C^{ab\pm}\ell,d^\ell{s_a,\pm s_b}(\theta) $$
This is the accurate representation without a flat-sky approximation, but a fast Hankel transform with FFTLog might be a lot faster.
I also added the notebook$C(\ell)$ and $\xi$ against $\ell$ modes is required for accurate Hankel transforms (30000 - 60000), but with this high number the $\xi$ are of similar accuracy as the $C(\ell)$ at scales between 5' and 100'. $\xi_-$shows unsurprisingly the worst agreement, as it depends the strongest on the high-$\ell$ modes.
compare_c_ell_ccl.ipynb, which tests theccl.A high number of