Cosmological parameter sampling
This project aims to implement joint power spectrum and cosmological parameter sampling in Commander3 without likelihood approximations, by interfacing Commander directly with the CAMB Boltzmann code. The main goal is cosmological parameter estimation with full end-to-end error propagation, which will be essential for next-generation B-mode experiments such as LiteBIRD.
Illustration of joint cosmological parameter and CMB sky map sampling move
Currently Commander is set up to estimate component maps and the power spectrum of the CMB, but cosmological parameter sampling is performed using CosmoMC independently using resampled CMB estimates. Besides taking extra human time, this induces a longer correlation length, especially for high signal-to-noise multipole regions of the power spectrum.
This problem can be circumvented by doing a joint sampling step, i.e., sampling from the cosmological parameters and the power spectrum simultaneously. With this implemented, we will be able to both reduce the correlation lengths of the Gibbs chains and perform cosmological parameter sampling in a single framework with all other parameters.
Preliminary code: Current GitHub Commander CAMB branch
Open to all interested parties; new collaborators very are very welcome!
- Run CosmoMC on Commander CMB outputs
- Implement internal likelihood-free cosmological sampling in Commander
- Testing, comparing, and predicting the expected uncertainty and correlation lengths
- Interface Commander with CAMB to support direct cosmological parameter estimation
- Joint a_lm + power spectrum sampling, as demonstrated by Racine et al. (2016)
- Estimate tau and r from LFI+WMAP and simulated LiteBIRD data
- Integration of external datasets (BAO, SN, etc.) and priors
- Johannes Røsok Eskilt (Coordinator)
- Duncan Watts
How to get involved:
If you want to get involved, please send Johannes Røsok Eskilt an email