Python API

Each public module of treams_rs has one page with the signature and docstring of every name it exports. Classes list their methods and attributes, including those inherited from other treams-rs classes.

A Pullback line names the method of the returned context that turns the gradient of a real loss with respect to the output into gradients with respect to the inputs (see the glossary).

Module	Summary
treams_rs	T-matrix scattering with Rust numerics and analytic first-order derivatives.
treams_rs.special	Special functions of multipole waves, with NumPy broadcasting.
treams_rs.sw	Coefficients of spherical waves: translation, rotation and periodic arrays.
treams_rs.cw	Coefficients of cylindrical waves: translation, rotation and periodic arrays.
treams_rs.pw	Coefficients of plane waves: translation, expansion and axis permutation.
treams_rs.lattice	Lattice sums of spherical and cylindrical waves, and lattice geometry.
treams_rs.coeffs	Mie and Fresnel coefficients in the helicity basis.
treams_rs.misc	Refractive indices, normal wavenumbers, mode selection and Brillouin zones.
treams_rs.ebcm	Surface integrals of the extended boundary condition method (EBCM).
treams_rs.io	T-matrix files in the tmat.h5 layout, and Gmsh meshes of sphere clusters.
treams_rs.operators	Basis and field operators, as functions and as operator objects.
treams_rs.iterative	Matrix-free scattering by finite clusters of homogeneous spheres in vacuum.
treams_rs.parallel	CPU threads of the native numerical work.
treams_rs.diff	Records: values with analytic first-order pushforward and pullback contexts.
treams_rs.testing	Finite-difference checks of first-order gradients, pullbacks and pushforwards.
treams_rs.advect	Advect adapter: physics objects and records differentiated by Advect, on the CPU and in first-order forward and reverse mode.
treams_rs.jax	JAX adapter: physics objects and records on the CPU with first-order forward and reverse mode, jit and sequential vmap.
treams_rs.torch	PyTorch adapter: physics objects and records as autograd operations on the CPU, in first-order forward and reverse mode.
treams_rs.autograd	HIPS Autograd adapter: physics objects and native records on the CPU.

Returned types lists the objects that records and solvers return.

How native calls run

• numeric_precision: native float64 / complex128; framework outputs follow adapter precision; metadata uses integer and boolean arrays
• native_pairing: dL = Re(sum(conj(g) * dx))
• derivative_order: 1
• static_parameters: mode counts, integer labels, topology; see each operation's docstring
• context_lifetime: Native recorded contexts are reusable: pushforwards and pullbacks share immutable residuals.
• python_execution: CPU

Backend cpu: precision float64, complex128.

support_catalog() and python -m treams_rs --format json report the installed optional frameworks. An installed framework does not mean that treams-rs supports its devices: every adapter runs on the CPU.