search

xzackli / Bolt.jl

differentiable boltzmann code
MIT License
43 stars 5 forks source link

Lensing #84

Open jmsull opened 1 year ago

jmsull commented 1 year ago

Marco has graciously provided an initial halofit implementation, we should also make some progress toward CMB lensing

marcobonici commented 1 month ago

Any development on this side? Regarding the lensing of CMB, are you planning to implement something like this ?

jmsull commented 1 month ago

Hi Marco,

Appreciate you checking up on our slow progress =)

I started to do this a few months ago but have not prioritized it since then - if you (or someone you know) wants to open a lensing PR (or update your halo for PR) then feel free, otherwise I’ll move it up the priority list for me!

Thanks!

Jamie

On Mon, Oct 7, 2024 at 11:17 Marco Bonici @.***> wrote:

Any development on this side? Regarding the lensing of CMB, are you planning to implement something like this https://arxiv.org/abs/astro-ph/0502425 ?

— Reply to this email directly, view it on GitHub https://github.com/xzackli/Bolt.jl/issues/84#issuecomment-2397222084, or unsubscribe https://github.com/notifications/unsubscribe-auth/AMCTXDW22D37GCJQM3GNFRDZ2KQYXAVCNFSM6AAAAABPQFT24SVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGOJXGIZDEMBYGQ . You are receiving this because you authored the thread.Message ID: @.***>

marcobonici commented 4 weeks ago

Hi @jmsull , sure! I was planning to do something similar, so might as well give that to you here. The ingredients are:

  1. Halofit (but any other nonlinear recipy would be good), which should be working in the PR I opened a while ago
  2. Compute the lensed Cls using the correlation functions. After this, you should be mostly ready to compare to CAMB/CLASS and maybe run some chains ;)?
marcobonici commented 4 weeks ago

Ok, I should be close to this one. I implemented the lensed Cls as done in CAMB. The results are almost the same; after a bit of cleaning it should be good to go. I would start with "If it runs and the resutls are correct, we are ok", then I would probably implement some plan/caching mechanisms as a decent amount of time is spent computing some cosmology independent quantities. Hopefully you'll hear from me soon.

The actual implementation runs in ~ 0.5 seconds, so it shouldn't be adding a bad overhead to Bolt.jl. But as I said, after we make it work there should be space for performance enhancement!

jmsull commented 4 weeks ago

Amazing!! Thank you for doing this Marco

On Sun, Oct 13, 2024 at 13:19 Marco Bonici @.***> wrote:

Ok, I should be close to this one. I implemented the lensed Cls as done in CAMB. The results are almost the same; after a bit of cleaning it should be good to go. I would start with "If it runs and the resutls are correct, we are ok", then I would probably implement some plan/caching mechanisms as a decent amount of time is spent computing some cosmology independent quantities. Hopefully you'll hear from me soon.

The actual implementation runs in ~ 0.5 seconds, so it shouldn't be adding a bad overhead to Bolt.jl. But as I said, after we make it work there should be space for performance enhancement!

— Reply to this email directly, view it on GitHub https://github.com/xzackli/Bolt.jl/issues/84#issuecomment-2409056997, or unsubscribe https://github.com/notifications/unsubscribe-auth/AMCTXDT4JQOF7MPJYIVKWYTZ3KTR3AVCNFSM6AAAAABPQFT24SVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIMBZGA2TMOJZG4 . You are receiving this because you were mentioned.Message ID: @.***>