Installation

Requirements

• Python ≥ 3.12, < 3.14
• JAX (CPU or GPU/TPU backend)

---

Install from GitHub

The simplest way to install SpectralDiffX:

pip install git+https://github.com/jejjohnson/spectraldiffx

---

Development Setup

We use uv for dependency management.

Clone and install

git clone https://github.com/jejjohnson/spectraldiffx.git
cd spectraldiffx
uv sync --group dev

Install with all optional dependencies

To install the full suite including tests, docs, and example scripts:

uv sync --all-extras

---

Optional Dependencies

Extra	Contents
test	pytest, pytest-cov
docs	mkdocs-material, mkdocstrings, mkdocs-jupyter
examples	diffrax, cyclopts, xarray, matplotlib, tqdm

Install a specific extra:

pip install "spectraldiffx[examples] @ git+https://github.com/jejjohnson/spectraldiffx"

---

Verify Installation

import jax
import jax.numpy as jnp
from spectraldiffx import FourierGrid1D, SpectralDerivative1D

jax.config.update("jax_enable_x64", True)

grid = FourierGrid1D.from_N_L(N=64, L=2 * jnp.pi)
deriv = SpectralDerivative1D(grid=grid)

u = jnp.sin(grid.x)
du_dx = deriv(u, order=1)

error = jnp.max(jnp.abs(du_dx - jnp.cos(grid.x)))
print(f"Max error: {error:.2e}")  # should be < 1e-12

---

Precision

JAX computes in float32 unless 64-bit mode is enabled. The test suite runs in float64 (tests/conftest.py), and tests/test_float32.py checks the float32 default. Measured relative difference between a float32 result and the same computation in float64 (CPU):

Operation	N	float32 vs float64
DCT/DST types I–IV, forward + inverse	64²	1.7e-7 – 2.3e-7
solve_poisson_dst (random or smooth RHS)	64² – 1024²	1.5e-7 – 3.2e-7
build_capacitance_solver, every base_bc, λ ∈	24 × 20	1.4e-7 – 4.1e-7

These are a few times float32 machine epsilon (1.2e-7) and do not grow with N over this range. Enable float64 when you need more digits, or when a downstream computation (e.g. long time integration) amplifies round-off:

import jax

jax.config.update("jax_enable_x64", True)