API reference

flashquad.integrator.FlashQuad

The main entry point. Create an instance bound to a backend, then call any integration method.

from flashquad import FlashQuad

fq = FlashQuad(backend="numpy")
fq = FlashQuad(backend="torch", dtype=torch.float32, device="cuda:0")

Parameters:

• backend (str) — One of "numpy", "torch", "cupy", "jax".

• dtype (optional) — Floating-point dtype compatible with the backend. Defaults to float64.

• device (optional) — Compute device (relevant for "torch" and "jax"). Auto-selects GPU when available.

Integration methods

All methods share a common signature:

fq.method(func, intervals, num_points, *, params=None, boundary=None)

Method	Description	Constraints
trapz	Composite trapezoidal rule	—
simpson	Composite Simpson’s 1/3 rule	num_points must be odd
booles	Composite Boole’s rule	(num_points - 1) divisible by 4
gauss	Gauss-Legendre quadrature	—
mc	Monte Carlo sampling	num_points is a single int

Common arguments:

• func — Integrand callable. Called as func(*coords) or func(*coords, params).

• intervals — Integration bounds per dimension, e.g. [[0, 1], [0, 1]].

• num_points — Grid points per dimension (list), or total samples (int, for MC methods).

• params (optional) — Array shaped (batch, num_params) for batched evaluation.

• boundary (optional) — Mask function to restrict the integration domain.

For full auto-generated module documentation, see the package index.