Advanced Features

This page collects features beyond the minimal builder workflow. The current advanced surface spans distributed execution, radiation-aware optimisation, material fitting, nonlinear materials, and a research-grade example bundle.

Multi-GPU Distributed FDTD

rfx can distribute the Yee update across multiple GPUs on a single host via jax.pmap.

import jax
from rfx import Simulation

sim = Simulation(freq_max=10e9, domain=(0.2, 0.04, 0.02), boundary="cpml")
result = sim.run(n_steps=2000, devices=jax.devices()[:2])

Current distributed support is strongest for slab-decomposed 3-D problems with PEC/CPML boundaries, soft sources, probes, dispersive media, and lumped-port workflows. Unsupported configurations are designed to fail clearly or fall back conservatively.

NTFF-Aware Far-Field Optimisation

The optimiser now supports objectives that consume raw NTFF accumulation data. If your objective accepts an ntff_box keyword argument, optimize() will build the far-field box and pass it into the loss.

import jax.numpy as jnp
from rfx import Simulation, compute_far_field_jax

sim = Simulation(freq_max=10e9, domain=(0.12, 0.06, 0.04), boundary="cpml")
sim.add_ntff_box(
    corner_lo=(0.02, 0.01, 0.01),
    corner_hi=(0.10, 0.05, 0.03),
    freqs=[10e9],
)
grid = sim._build_grid()
theta = jnp.linspace(0.0, jnp.pi, 181)
phi = jnp.array([0.0])

def objective(result, ntff_box=None):
    ff = compute_far_field_jax(result.ntff_data, ntff_box, grid, theta, phi)
    broadside_power = jnp.abs(ff.E_theta[0, 90, 0]) ** 2 + jnp.abs(ff.E_phi[0, 90, 0]) ** 2
    return -broadside_power

Use this pattern for beam steering, broadside-gain maximisation, or null placement objectives. In practice you usually capture the design grid in the objective closure before calling optimize().

Time-Domain Proxy Objectives for Differentiable Loops

Inside optimize() and topology_optimize(), the most robust built-in losses are the time-domain proxy objectives:

from rfx import minimize_reflected_energy, maximize_transmitted_energy

obj_reflect = minimize_reflected_energy(port_probe_idx=0)
obj_transmit = maximize_transmitted_energy(output_probe_idx=-1)

These avoid assuming that a full post-processed S-parameter matrix is available inside the JIT-traced forward pass.

Dispersive Materials and Material Fitting

Debye / Lorentz models

from rfx import DebyePole, LorentzPole

sim.add_material("water", eps_r=5.0, debye_poles=[
    DebyePole(eps_s=80.0, tau=9.4e-12)
])

sim.add_material("resonant", lorentz_poles=[
    LorentzPole(eps_s=2.0, omega_0=2e10, delta=1e8, omega_p=3e10)
])

Differentiable fitting from S-parameters

from rfx import differentiable_material_fit

fit = differentiable_material_fit(...)

This is one of the most distinctive rfx workflows: use FDTD itself as the forward model while fitting Debye/Lorentz poles to measured or synthetic fixtures.

Nonlinear Materials and RIS Workflows

rfx includes a Kerr nonlinear material path via ADE and a dedicated RIS unit-cell workflow for programmable-surface experiments. These are advanced research features rather than beginner-path tutorials, but they are part of the current public capability surface.

Non-Uniform Mesh, Lumped RLC, and Geometry Helpers

Non-uniform z mesh

For thin substrates and layered RF structures, rfx supports graded z spacing:

sim = Simulation(
    freq_max=4e9,
    domain=(0.08, 0.06, 0.0),
    dx=5e-4,
    dz_profile=dz_profile,
)

See Non-Uniform Mesh for the recommended workflow.

Lumped RLC elements

sim.add_lumped_rlc(
    position=(0.02, 0.02, 0.01),
    component="ez",
    R=50.0,
    L=1e-9,
    C=1e-12,
    topology="series",
)

Via and CurvedPatch helpers

from rfx import Via, CurvedPatch

These expose convenient PCB / conformal-antenna primitives at the top level.

Mixed Precision and Field Animation

from rfx import save_field_animation

save_field_animation(result, "field_evolution.gif")

For memory-constrained runs, mixed-precision field storage is also available in current stable releases.

Advanced Example Bundle

The repo ships a research-style example bundle under examples/50_advanced/:

• patch bandwidth optimisation
• waveguide filter inverse design
• broadband matching
• array mutual coupling
• dielectric-lens beam shaping
• S-parameter-based material characterisation
• visualization showcase

The matching examples/50_advanced/gpu_validation/ scripts provide a compact regression layer for these workflows.