brepax 0.1.0

Differentiable rasterizer for CAD Boolean operations — JAX-native B-Rep kernel with stratum-aware gradients.

BRepAX

Differentiable rasterizer for CAD Boolean operations.

BRepAX is a JAX-native library that enables gradient-based optimization through B-Rep (Boundary Representation) topology changes. It translates contact dynamics formulations from differentiable physics into the CAD domain, providing stratum-aware differentiation for Boolean operations on geometric primitives.

Installation

pip install brepax

With optional dependencies:

pip install "brepax[viz]"          # Visualization
pip install "brepax[persistence]"  # Persistent homology
pip install "brepax[all]"          # Everything

Quick Start

import jax
import jax.numpy as jnp
from brepax.primitives import Disk

# Create two disks
disk1 = Disk(center=jnp.array([0.0, 0.0]), radius=jnp.array(1.0))
disk2 = Disk(center=jnp.array([1.5, 0.0]), radius=jnp.array(1.0))

# Evaluate SDF -- fully differentiable
query = jnp.array([0.75, 0.0])
sdf_value = disk1.sdf(query)

# Gradients flow through everything
grad_fn = jax.grad(lambda q: disk1.sdf(q))
gradient = grad_fn(query)

Documentation

Full documentation: lv416e.github.io/brepax

License

Apache License 2.0. See LICENSE for details.