How to enable double precision¤

JAX creates 32-bit arrays by default. spexial's accuracy figures all assume 64-bit, so turn double precision on before you create any array.

In a script or notebook¤

Set it before anything else runs:

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

>>> import jax.numpy as jnp
>>> jnp.zeros(1).dtype
dtype('float64')

The setting is process-global and order-sensitive. Arrays created before the call stay float32, so put it above your other imports if any of them build arrays at import time.

From the environment¤

If you would rather not touch the code — or you need it to apply from interpreter start, before any import can run:

JAX_ENABLE_X64=1 python your_script.py

To apply it to a whole session, export it:

export JAX_ENABLE_X64=1

Under pytest¤

If your project already uses pytest-env, set it in pyproject.toml so every test runs in 64-bit without a per-test fixture:

[tool.pytest_env]
JAX_ENABLE_X64 = "True"

Check it worked¤

Any array will tell you:

>>> jnp.zeros(1).dtype
dtype('float64')

If you get float32, the setting was applied too late.

Watch for NumPy inputs¤

Check the dtype of what comes out of your first spexial call, not what goes in — a NumPy float64 array does not guarantee a float64 result. About precision explains why, and why it matters more here than in most JAX code.