The goal of this repo is to provide an open-source and easy-to-use testbed for comparing full-field weak lensing constraints under an LPT gravity model, lognormal convergence model, and gaussian convergence model.
A guiding principle of this repo is that everything should be runnable within a few hours on colab.
We are adopting the following setting:
- patch size: 5 x 5 deg^2
- pixel size: 5 arcmin^2 (which gives us an lmax of around 3000)
- galaxy number density: 27 gals/arcmin^2
- galaxy ellipticity dispersion: 0.3
We check that our LPT model has sufficient resolution to reproduce the linear theory angular power spectra (i.e. without halofit).
We obtain this match to theory using the following lightcone volume and resolution:
- lightcone volume: 400 x 400 x 4000 Mpc/h
- lightcone resolution: 300 x 300 x 256 voxels
Given that we have a matching theory for our numerical simulator, we can easily compute the 2pt posterior of a simulated map, as well as the full-field posterior by HMC.
Preliminary (unconverged) results:
Here, we tune our simulator to match the setting of the series of papers led by Natalia Porqueres:
- Field-level inference of cosmic shear with intrinsic alignments and baryons https://arxiv.org/abs/2304.04785
- Lifting weak lensing degeneracies with a field-based likelihood https://arxiv.org/abs/2108.04825
The main simulation setting in this series of papers uses the following setup:
- patch size: 16 x 16 deg^2
- pixel size: 15 arcmin^2 (which gives us an lmax of around 1300)
- galaxy number density: 30 gals/arcmin^2
- galaxy ellipticity dispersion: 0.3
- lightcone volume: 1 x 1 x 4.5 Gpc/h
- lightcone resolution: 64 x 64 x 128 voxels
We find that this resolution is not quite sufficient for the LPT model to converge to expected linear theory, but note that is a detail, mostly irrelevant to the story of this line of papers.
Within this model, we compare the full field constraints against a 2pt-function equivalent by doing SBI from the power spectra