Reionization Effective Likelihood

The Reionization Effective Likelihood (RELIKE) is a fast and accurate likelihood code that compresses the final Planck 2018 likelihoods for the purpose of constraining models of the global ionization history.

• Using the python package relike, you can obtain the likelihood of any model of ionization history xe(z) in the range 6 < z < zmax.

• The relike code works by projecting the model onto the principal components (PC) of ionization history in the CMB data for its PC amplitudes mj's, which are used to quickly return the effective likelihood of the model; you may also use it to evaluate the likelihood at multiple points and plot the parameter posteriors (assuming flat prior in the parameter).

• To run an MCMC chain, you can either use the relike python package from inside of a sampler (e.g. Cobaya or Cosmosis), or use our release of CosmoMC-relike which has an implementation of the relike code in fortran (both in KDE and Gaussian modes) used to produce published results.

• Note that there are two modes of effective likelihoods: 1) kernel density estimate (KDE) mode, which is slightly more accurate, and 2) the Gaussian approximation mode, which is good enough for most models. The python relike contains the Gaussian mode only, while the fortran implementation in CosmoMC-relike has both.

Reference: Heinrich & Hu 2021 (arxiv: fill in)

Release Note

v1.0 (coming soon)

• Added relike: a standalone python likelihood package.
• Added CosmoMC-relike: an MCMC implementation using the CosmoMC sampler.
• Supporting arbitrary xe(z) specified by the user between 6 < z < zmax, where zmax = 30; assuming fully reionized hydrogen for z < 6.
• Planck likelihoods used: plik_lite_TTTEEE + lowl + srollv2.

Getting Started

• First, clone all submodules:

  git clone --recurse-submodules https://github.com/chenheinrich/RELIKE.git

• If you don't want CosmoMC-relike for now, just git clone and if you change your mind later, use

  git submodule update --init --recursive

• To stay updated, use git submodule update --remote --merge beside git pull.

Installing relike: Python Likelihood Package

This is a standalone python likelihood package, outputting the Planck likelihood for any global ionization history model xe(z). The functional form of xe(z) between 6 < z < zmax is specified by the user (we only support zmax = 30 for now), and fully reionized hydrogen is assumed for z < 6 with typical helium ionization history.

• Install requirements first:

  (It is recommended to create a virtual environment with python 3.8. See for example https://virtualenvwrapper.readthedocs.io/en/latest/ for getting setup with python virtual envs).

  Activate the virtual environment

  cd RELIKE

  pip install -r requirements.txt (add --user when working on a cluster if you don't have a virual environment activated)

• Install relike:

  pip install .

  For editable install:

  pip install -e .

• To run tests:

  pytest tests

• To play with relike using Jupyter notebooks:

  jupyter notebook examples/example.ipynb .

• To run example script (same as in the demo Jupyter notebook):

  python3 examples/example_likelihood.py

  This will run two examples: 1) print the relative chi-squared of an example tanh model relative to the Planck 2018 best-fit tanh model; 2) calculate and plot the optical depth posterior in the tanh model using relike.

Installing CosmoMC-relike: MCMC Sampler + relike in Fortran

CosmoMC-relike uses the generic sampler of CosmoMC to sample the fortran implementation of the relike likelihood.

You may need to update all submodules recursively like this:

git submodule update --init --recursive

cd CosmoMC-relike

See further installation instructions at CosmoMC-relike