Skip to content

Latest commit

 

History

History
75 lines (46 loc) · 2.43 KB

README.md

File metadata and controls

75 lines (46 loc) · 2.43 KB
Raw

DISCO-EB - The DISCO-DJ Einstein-Boltzmann module

Implementation of a differentiable linear Einstein-Boltzmann solver for cosmology in JAX -- a module of the DISCO-DJ framework (DIfferentiable Simulations for COsmology, Done with Jax).

Note that this program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY.

Currently supported features (the list is growing, so check back):

  • Autodifferentiable via JAX
  • Standard LCDM model with Quintessence DE fluid (w0,wa) and massive neutrinos (one species)
  • Thermal history solver based on a simplified Recfast implementation in the module
  • Numerous example jupyter notebooks, e.g. for Euclid-like Fisher forecasts

New modules/plugins can be easily added (see how to contribute in CONTRIBUTING.md file). We are enthusiastic if extensions/improvements that are of broader interest are re-integrated into the master branch and DISCO-EB grows as a community effort.

Currently work in progress:

  • performance improvements

Installation

Simple Install

pip install git+https://github.com/ohahn/DISCO-EB.git

For Development

In a fresh virtual environment:

git clone https://github.com/ohahn/DISCO-EB.git
cd DISCO-EB
pip install -e .

Running the tests

First install the additional dependencies with the [dev] flag:

cd DISCO-EB
pip install -e '.[dev]'

Then run pytest to start the tests. This should be run on GPU platforms.

Getting started

Start with the sample notebook nb_minimal_example.ipynb in the notebooks subdirectory. It explains how to compute a matter power spectrum and take a derivative w.r.t. a cosmological parameter.

Benchmarks

See BENCHMARKS.md for benchmark results on different hardware.

Contributing and Licensing

See file CONTRIBUTING.md on how to contribute to the development.

The software is licensed under GPL v3 (see file LICENSE).

Citing in scientific publications or presentations

If you use DISCO-EB in your scientific work, you are required to acknowledge this by linking to this repository and citing the relevant papers:

Required Python Packages

  • JAX
  • diffrax==0.4.1
  • equinox
  • jaxtyping
  • jax_cosmo