asdex 0.3.3

Automatic Sparse Differentiation in JAX

asdex

Automatic Sparse Differentiation in JAX.

asdex (pronounced Aztecs) exploits sparsity structure to efficiently materialize Jacobians and Hessians. It implements a custom Jaxpr interpreter that uses abstract interpretation to detect sparsity patterns from the computation graph, then uses graph coloring to minimize the number of AD passes needed.

Installation

pip install asdex

Or with uv:

uv add asdex

Example

import asdex
import jax
import jax.numpy as jnp

def f(x):
    return (x[1:] - x[:-1]) ** 2

x_sample = jnp.zeros(50)  # sample input for sparsity pattern detection
jac_fn = jax.jit(asdex.jacobian(f, x_sample))
# ColoredPattern(49×50, nnz=98, sparsity=96.0%, JVP, 2 colors)
#   2 JVPs (instead of 49 VJPs or 50 JVPs)
# ⎡⠙⢦⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⎤   ⎡⣿⎤
# ⎢⠀⠀⠙⢦⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⎥   ⎢⣿⎥
# ⎢⠀⠀⠀⠀⠙⢦⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⎥   ⎢⣿⎥
# ⎢⠀⠀⠀⠀⠀⠀⠙⢦⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⎥   ⎢⣿⎥
# ⎢⠀⠀⠀⠀⠀⠀⠀⠀⠙⢦⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⎥   ⎢⣿⎥
# ⎢⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠙⢦⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⎥   ⎢⣿⎥
# ⎢⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠙⢦⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⎥ → ⎢⣿⎥
# ⎢⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠙⢦⡀⠀⠀⠀⠀⠀⠀⠀⠀⎥   ⎢⣿⎥
# ⎢⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠙⢦⡀⠀⠀⠀⠀⠀⠀⎥   ⎢⣿⎥
# ⎢⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠙⢦⡀⠀⠀⠀⠀⎥   ⎢⣿⎥
# ⎢⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠙⢦⡀⠀⠀⎥   ⎢⣿⎥
# ⎢⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠙⢦⡀⎥   ⎢⣿⎥
# ⎣⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠉⎦   ⎣⠉⎦

for x in inputs:
    J = jac_fn(x)

Instead of 49 VJPs or 50 JVPs, asdex computes the full sparse Jacobian with just 2 JVPs.

Documentation

• Getting Started: step-by-step tutorial
• How-To Guides: task-oriented recipes
• Explanation: how and why it works
• API Reference: full API documentation

Acknowledgements

Adrian Hill gratefully acknowledges funding from the German Federal Ministry of Education and Research under the grant BIFOLD26B.

This package is built with Claude Code based on previous work by Adrian Hill, Guillaume Dalle, and Alexis Montoison in the Julia programming language:

• An Illustrated Guide to Automatic Sparse Differentiation, Hill, Dalle, Montoison (2025)
• Sparser, Better, Faster, Stronger: Efficient Automatic Differentiation for Sparse Jacobians and Hessians, Hill & Dalle (2025)
• Revisiting Sparse Matrix Coloring and Bicoloring, Montoison, Dalle, Gebremedhin (2025)
• SparseConnectivityTracer.jl, Hill, Dalle
• SparseMatrixColorings.jl, Dalle, Montoison
• sparsediffax, Dalle

These works in turn stand on the shoulders of giants, notably Andreas Griewank, Andrea Walther, and Assefaw Gebremedhin.

The asdex logo was designed by @overripemango.

Citation

If you use asdex in your research, please cite:

@software{asdex2026,
  author = {Hill, Adrian},
  title = {asdex: Automatic Sparse Differentiation in JAX},
  url = {https://github.com/adrhill/asdex},
  doi = {10.5281/zenodo.18788242}
}