flashquad.integrator

FlashQuad — numerical integrator with pluggable array backends.

Module Contents

Classes

FlashQuad

Numerical integrator bound to a specific array backend, dtype, and device.

API

class flashquad.integrator.FlashQuad(backend: str, dtype: Any = None, device=None)

Numerical integrator bound to a specific array backend, dtype, and device.

Parameters

backend : str Array backend name. One of 'numpy', 'torch', 'cupy', 'jax'. dtype : optional Floating-point dtype that must be compatible with the chosen backend. For 'torch' pass a torch.dtype (e.g. torch.float64); for numpy-compatible backends pass a numpy dtype (e.g. numpy.float64). Defaults to the backend’s float64. device : optional Compute device. For 'torch', pass a string ('cuda', 'cpu', 'cuda:0', …) or a torch.device. When None (the default) the torch backend auto-selects CUDA if available. Ignored for other backends.

Initialization

__repr__()

trapz(func, intervals, num_points, *, params=None, boundary=None)

Integrate using the composite trapezoidal rule.

Args: func: Integrand. Called as func(*mesh_coords) or func(*mesh_coords, params_expanded) when params is given. intervals: Integration bounds per dimension, e.g. [[0, 1], [0, 1]]. num_points: Grid points per dimension, e.g. [101, 101]. params: Parameter array shaped (batch, num_params). Each row is one set of parameters for batched evaluation. boundary: Optional mask function applied to the integrand.

simpson(func, intervals, num_points, *, params=None, boundary=None)

Integrate using composite Simpson’s 1/3 rule.

Args: See :meth:trapz. Each entry in num_points must be odd.

booles(func, intervals, num_points, *, params=None, boundary=None)

Integrate using composite Boole’s rule.

Args: See :meth:trapz. (num_points[i] - 1) must be divisible by 4.

gauss(func, intervals, num_points, *, params=None, boundary=None)

Integrate using Gauss-Legendre quadrature.

Args: See :meth:trapz.

mc(func, intervals, num_points, *, params=None, boundary=None)

Integrate using Monte Carlo sampling.

Args: func: Integrand function. intervals: Integration bounds per dimension. num_points: Total number of random samples (single integer). params: Parameter array shaped (batch, num_params). boundary: Optional mask function.