JAX port of GalSim, for parallelized, GPU accelerated, and differentiable galaxy image simulations.

Disclaimer: This project is still in an early development phase, please use the reference GalSim implementation for any scientific applications.

You can find the most up-to-date documentation for the project here.

The goal of this library is to reimplement GalSim functionalities in pure JAX to allow for automatic differentiation, GPU acceleration, and batched computations.

• Strive to be a drop-in replacement for GalSim, i.e. provide a close match to the GalSim API.
• Each function/feature will be tested against the reference GalSim implementation.
• This package will aim to be a subset of GalSim (i.e. only contains functions with a reference GalSim implementation).
• Implementations should be easy to read and understand.
• Code should be pip-installable on any machine, no compilation required.
• Any notable differences between the JAX and reference implementations will be clearly documented.

• JAX arrays are immutable. Thus, in-place operations on images are not possible and a new image is returned instead. Also, the RNG classes cannot fill arrays and instead return new arrays.
• JAX arrays do not support all the kinds of views that numpy arrays support. This means that some operations that work in GalSim may not work in JAX-GalSim (e.g., real views of complex images).
• JAX-GalSim uses a different random number generator than GalSim. This leads to different results in terms of both the generated random numbers and in terms of which RNGs have stable discarding.

Everyone can contribute to this project, please refer to the CONTRIBUTING.md document for details.

In short, to interact with the project you can:

• Ask or Answer questions on the Discussions Q&A page
• Report a bug by opening a GitHub issue
• Open a GitHub issue or Discussions to ask for feedback on a planned contribution.
• Submit a Pull Request to contribute to the code.

Issues marked with contributions welcome or good first issue are particularly good places to start. These are great ways to learn more about the inner workings of GalSim and how to code in JAX.

See the corresponding documentation page for a list of what is currently implemented.