search

xzackli / Bolt.jl

differentiable boltzmann code
MIT License
43 stars 5 forks source link

Scalar C\ell accuracy test #74

Open jmsull opened 2 years ago

jmsull commented 2 years ago

We did the power spectrum - we should do C\ell_TT,_TE,_EE (since there are now functions for those). I will take a first pass at this without digging too deeply into precision settings.

jmsull commented 2 years ago

This could be a separate issue but should also clean up naked factors of c hanging around the source functions...

jmsull commented 2 years ago

Started this - there is a test (with an inadequately coarse tolerance of 11%) here. ~10% seems pretty high given our agreement on P(k) and (at least nonnu) perturbations.

A first glance at this seems to me like we have a k-dependent issue that gets worse at higher k. Perhaps our k grid is not big enough? I would need to look at CAMB/CLASS LOS routines to see if we need to shift/expand our range.

bolt_ttteee_camb_ratio

Also updated the CAMB vs CLASS notebook to produce C_\ells using our existing setup (though I think I have a more recent CLASS version than Zack). Have not been super duper careful with CAMB but seeing agreement at 0.1% in (unlensed) TT,EE ratios after forcing reion taus to match. (TE is obviously crossing zero so will need to take another perspective there). camb_class_TTTEEE_raio

jmsull commented 2 months ago

As a point of reference, currently we are finding a small discrepancy in the $E_{2}(k)$ scalar polarization transfer function versus CAMB at k~ 5e-4 Mpc^-1. @xzackli You have the plot of this if you want to post it here

This is a minor issue we should return to in time, after looking more seriously at CAMB precision settings for TTTEEE