ExoJAX

Version 1.6 (History)

Note: Paper II will be under peer review. We plan to release version 2.0 at the time of acceptance of the paper II.

ExoJAX provides an auto-differentiable high-resolution spectrum model for exoplanets/brown dwarfs using JAX. ExoJAX enables a fully Bayesian inference of the high-dispersion data to fit the line-by-line spectral computation to the observed spectrum, from end-to-end (i.e. from molecular/atomic databases to real spectra), by combining it with the Hamiltonian Monte Carlo in recent probabilistic programming languages such as NumPyro. So, the notable features of ExoJAX are summarized as

  • HMC-NUTS, gradient-based optimizer available

  • Easy to use the latest molecular/atomic data in ExoMol, HITEMP, HITRAN, and VALD3

  • A transparent open-source project; anyone who wants to participate can join the development!

For a more geek-oriented explanation

ExoJAX is a spectral model based on the Differentiable Programming (DP) paradigm! ExoJAX aims to provide building blocks for retrieval code, much like Minecraft 🧱.

🟢 If you have an error and/or want to know the up-to-date info, visit ExoJAX wiki. Or use the discussions form on github or directly raise issues.

Contents

ExoJAX example (exojaxample)

References

  • Kawahara, Kawashima, Masuda, Crossfield, Pannier, van den Bekerom, ApJS 258, 31 (2022) (Paper I)

  • Kawahara et al., 2024, in prep. (Paper II)

License & Attribution

Copyright 2021-2024, Contributors

  • Hajime Kawahara (@HajimeKawahara, maintainer)

  • Yui Kawashima (@ykawashima, co-maintainer)

  • Kento Masuda (@kemasuda)

  • Ian Crossfield

  • Dirk van den Bekerom (@dcmvdbekerom)

  • Daniel Kitzmann (@daniel-kitzmann)

  • Brett Morris (@bmorris3)

  • Erwan Pannier (@erwanp) and Nicolas Minesi (@minouHub) from RADIS community

  • Stevanus Nugroho (@astrostevanus)

  • Tako Ishikawa (@chonma0ctopus)

  • Yui Kasagi (@YuiKasagi)

  • Shotaro Tada (@sh-tada)

  • Ko Hosokawa (@KoHosokawa)

  • Hibiki Yama

ExoJAX is free software made available under the MIT License. See the LICENSE.