CMBLensing.py¶
CMBLensing.py is the JAX port of the Julia package CMBLensing.jl. It
- simulates flat-sky CMB temperature and polarization fields, lensing potentials and noisy lensed data maps (
load_sim), - lenses, inverse-lenses and adjoint-lenses fields with the LenseFlow ODE (
lense_flow), - reconstructs the unlensed field and the lensing potential by joint MAP optimization (
map_joint), - samples the LCDM parameters jointly with the field and the lensing potential by Gibbs sampling (
sample_joint), extending the algorithm developed by Millea et al. 2020, - forecasts the sampler's posterior widths with a Gaussian Fisher forecast (
fisher_forecast).
All array operations run in JAX (float64), so the code runs on CPU and GPU and is differentiable where it needs to be.
Where to start¶
| I want to... | Go to |
|---|---|
| install the package | Installation |
simulate a map and run map_joint |
MAP Estimation Tutorial |
| learn more about sampling the LCDM parameters | Sampling Tutorial, LCDM Sampling |
| download the chains and other data behind the results | Data |
| run a full experiment on a cluster | HPC Setup |
| compare sampler data to the Fisher forecast | Fisher Forecast, Chain Analysis |
| change the code | Testing |
Repository layout¶
cmb_lensing/ the Python package
camb_splines/ the 5D CAMB grid spline (gitignored, downloaded on first use)
camb_grid_generation/ scripts that build the 5D grid
get_lcdm_samples/ LCDM chains on the HPC (analysed by cmb_lensing/chain_analysis.py)
measure_phi_noise/ phi_noise_analysis.py (the N_phi spectra plots) +
get_map_joint_phi_noise/ map_joint's per-mode N_phi
get_sampler_phi_noise/ the sampler's per-mode N_phi
get_delensed_field_covariance/ the empirical delensed T covariance stencil
runtime_comparison/ map_joint timing, Python vs Julia
tests/ A/B tests against CMBLensing.jl + the CAMB grid test
docs/ this site (Markdown pages + tutorial notebooks)
paper/ the paper source