rfx

Differentiable FDTD for RF/Microwave — powered by JAX

rfx is a 3-D finite-difference time-domain (FDTD) electromagnetic simulator built in JAX. It targets RF and microwave engineers who need accurate broadband field solutions, practical RF workflow tooling, and native gradient support for inverse design — without leaving Python.

The docs below cover the stable v1.4.0 package surface and call out selected workflow-hardening and validation updates already merged on main.

Why rfx

Feature	Detail
GPU acceleration	7,309 Mcells/s on RTX 4090, 5,847 on RTX 3090, 5,249 on RTX A6000
Multi-GPU distributed FDTD	`jax.pmap` slab decomposition for larger 3-D jobs on a single multi-GPU host
Autodiff	`jax.grad` through the full time-stepping loop for topology and shape optimization
Topology optimization	Density-based filtering, projection, and beta continuation for inverse design
Time-domain optimization objectives	Use proxy losses such as `minimize_reflected_energy()` and `maximize_transmitted_energy()` inside differentiable loops
Conformal PEC	Dey-Mittra method for 2nd-order accuracy on curved conductors
Material fitting	Debye/Lorentz fitting plus differentiable FDTD-based refinement from S-parameters
Non-uniform mesh + auto-config	Graded z-profiles and `auto_configure()` for thin substrates and broadband setups
Harminv resonance	High-Q mode extraction with sub-bin frequency precision
Multi-mode + Floquet ports	Rectangular TE/TM eigenmodes and phased-array unit-cell workflows with Bloch periodic BC
Cross-validated	Published 5-case RF validation plus earlier cavity/waveguide cross-validation vs. Meep and OpenEMS
500+ tests	Pytest suite with lint + CPU/GPU coverage and advanced-example validation scripts

Quick install

pip install rfx-fdtd

PyPI package name: rfx-fdtd Python import:

import rfx

GPU support (JAX + CUDA):

pip install "jax[cuda12]" rfx-fdtd

Recent highlights

v1.4.0 brings waveguide / Floquet ports into differentiable forward() and optimize() loops, so modal and periodic-port workflows stay inside the gradient-based design path.
v1.4.0 also adds richer export and dataset I/O: geometry JSON, experiment reports, simulation datasets, optimisation trajectories, and broader far-field / optimisation result saves.
v1.4.0 rounds out the advanced solver stack with JAX-differentiable far-field evaluation, experimental ADI 2-D / 3-D paths, SBP-SAT subgridding, and the research-grade example bundle under examples/50_advanced/.
Current main continues to harden the workflow with sim.preflight() validation, a 23-script cross-validation bundle, and newer geometry coverage such as the PolylineWire primitive.

Pages that describe evolving or research-grade workflows should say so explicitly. Use the user guide for the supported surface, and treat advanced or experimental notes as capability previews rather than blanket guarantees.

Ten-line example

from rfx import Simulation, Box, GaussianPulse

# 2.4 GHz patch antenna on FR4
sim = Simulation(freq_max=4e9, domain=(0.08, 0.06, 0.025), boundary="cpml")
sim.add(Box((0.0, 0.0, 0.0), (0.08, 0.06, 0.0016)), material="fr4")
sim.add(Box((0.02, 0.01, 0.0016), (0.049, 0.049, 0.0016)), material="pec")  # patch
sim.add(Box((0.0, 0.0, 0.0), (0.08, 0.06, 0.0)), material="pec")            # ground
sim.add_source((0.029, 0.03, 0.0008), "ez",
               waveform=GaussianPulse(f0=2.4e9, bandwidth=0.8))
sim.add_probe((0.029, 0.03, 0.0008), "ez")

result = sim.run(n_steps=8000)
modes = result.find_resonances(freq_range=(1.5e9, 3.5e9))
print(f"Resonance: {modes[0].freq/1e9:.4f} GHz  Q={modes[0].Q:.0f}")

Key concepts

Simulation ──► add_material / add / add_port / add_probe / add_ntff_box
     │
     └──► run() ──► Result
                       ├── time_series   (n_steps × n_probes)
                       ├── s_params      (n_ports × n_ports × n_freqs)
                       ├── ntff_data     (raw far-field accumulation)
                       └── find_resonances() → [HarminvMode]

Documentation by topic

Getting Started

Installation — Python/JAX install, GPU notes, dev setup
Quick Start — first simulation with the current stable high-level API
Your First Patch Antenna — current-source resonance workflow for a 2.4 GHz FR4 patch

Modeling & Setup

Simulation API — Simulation, Result, auto_configure, ports, NTFF, and design helpers
Materials & Geometry — library materials, Debye/Lorentz, CSG shapes, PCB stackup
Sources & Ports — point sources, lumped/wire ports, multi-mode waveguide ports, Floquet ports
Probes & S-Parameters — DFT probes, S-matrix, Harminv, de-embedding, Smith chart
Non-Uniform Mesh — graded z-profiles for thin-substrate structures
Waveguide Ports — multi-mode waveguide S-matrix with TE/TM eigenmodes
Floquet Ports — phased-array unit-cell analysis with Bloch BC

Analysis & Validation

Cross-Validation & Accuracy — published RF benchmarks plus cross-solver checks
Convergence Study — mesh refinement analysis and accuracy verification
Far-Field & RCS — NTFF radiation patterns and scattering workflows
Antenna Metrics — gain, efficiency, beamwidth, bandwidth, F/B ratio
Visualization & Analysis — plotting, exports, post-processing
Solver Comparison — feature and performance comparison vs. Meep and OpenEMS

Design & Optimization

Inverse Design — autodiff-driven topology, shape, and far-field optimization
Topology Optimization — density-based inverse design with filtering and projection
Parametric Sweeps — sequential sweep and jax.vmap batch evaluation
Material Fitting — CSV import, Debye/Lorentz fitting, differentiable FDTD-based fitting
Patch Antenna Design — full design workflow for a rectangular microstrip patch
Microstrip Filter Design — coupled-line bandpass filter with two-port S-parameters

Advanced & Research Methods

Advanced Features — multi-GPU, mixed precision, proxy objectives, advanced example workflows
Conformal PEC — Dey-Mittra method for curved conductors
SBP-SAT Subgridding — local mesh refinement with JIT performance
Gradient Behavior — where gradients are strong vs noisy
Geometry & Limitations — primitives, supported workflows, and remaining trade-offs

AI Agent Guide

rfx for AI Agents — why the API is agent-friendly
Automatic Simulation Configuration — geometry-to-sim parameter workflow
Prompt Templates for RF Design — reusable agent prompts
Automated Design Workflows — end-to-end design patterns

Project & maintainer

Migration Guide — Meep/OpenEMS workflow mapping
Contributing — maintainer workflow and test expectations
Changelog — release-level capability summary plus current-main notes

Repo examples

The public site focuses on guides. For runnable research-style workflows, see examples/50_advanced/ for patch optimization, waveguide inverse design, broadband matching, array coupling, dielectric-lens shaping, material characterization, and GPU validation scripts.

Guide Index

Open the full guide index