📝 SummarXiv

Abstract

We present jorbit, a python/JAX library designed to enable modern data-driven numerical studies of the solar system. Written entirely in JAX, an auto-differentiable and optionally GPU accelerated language behind many current large-scale machine learning efforts, jorbit includes an independent implementation of REBOUND's IAS15 integrator and the ability to parse precomputed ephemerides such as the JPL DE series. In its default behavior, jorbit maintains ~1 mas agreement with JPL Horizons on ~decade timescales for typical main-belt asteroids, enabling it to fully capitalize on high-precision astrometry and ranging data. We include details of the code's implementation and several worked examples, including illustrations of jorbit's ability to simulate N-body systems, forward model astrometric data, fit orbits, replicate the Minor Planet Center's "MPChecker" service, and contribute to modeling the effect of minor planets on stellar light curves.