chebax

Differentiable Chebyshev approximants for special functions, aimed at JAX and GPUs. A build-time generator (mpmath precision) produces fixed-degree branchless polynomial kernels with exact derivative series, including gradients with respect to function parameters (e.g. the Bessel order). Prebuilt recipes give out-of-the-box special functions; a bake step emits self-contained pure-JAX modules and xsf-style C++ headers.

Early development: the generic core (fit, jax evaluation, exact derivative series, segmentation) and the first recipe are in — besselj(v) for any real order in [0, 10] on x in [0, 8], with dJ/dx via jax.grad and the order gradient via besselj_dnu(v), no mpmath at use time. Tails (x > 8), more recipes, and bake emitters are not yet. Read PROJECT.md for the plan and evidence, CLAUDE.md for how to work here. Grown out of a private research project (the ../bessel/ references in PROJECT.md point there); the two load-bearing measurements are reproduced here in experiments/. BSD-3-Clause.