We present astronomix, a differentiable astrophysical fluid simulator.
Being written in high-level Python / JAX, astronomix is well-suited
for fast method development and benchmarking. Automatic differentiability
opens the door for gradient-based inverse modeling and sampling as well
as surrogate / solver-in-the-loop training. astronomix features
a high-order finite difference constrained transport scheme as well
as a 2nd order finite volume scheme. Modules for gravity, stellar
winds and cosmic rays are available / actively developed.
