# Directional coupler — broadband through/cross parity

<span class="validation-badge">validated · 2026-08-28</span>

<div class="case-meta"><span>forward</span><span>resolution 1</span><span>device</span><span>real_world</span><span>pic</span><span>coupling</span><span>tidy3d_golden</span><span>official_notebook</span></div>

This case compares a local FDTDX result with pinned reference data generated by Tidy3D. The originating Tidy3D example is [`DirectionalCoupler.ipynb`](https://github.com/flexcompute/tidy3d-notebooks/blob/c37c785d52e9258c9d048a781524b8e8d7c758ca/DirectionalCoupler.ipynb), pinned at commit `c37c785d52e9`.

::::{grid} 1

:::{grid-item-card} Open the complete executed tutorial →
:link: ../notebooks/directional_coupler
:link-type: doc
:class-card: device-ladder-card

Parameters → exact geometry → voxelized material → local FDTDX run → simulated field → quantitative analysis. Every code cell and retained output is visible, and the `.ipynb` source is downloadable.
:::

::::

This page is the compact benchmark record. The linked notebook is the primary scientific documentation and contains the actual simulation evidence.


## What is exercised

Compact reconstruction of the official DirectionalCoupler notebook. Two 450 nm x 220 nm silicon guides approach through shared raised-cosine bends, are separated by a 200 nm gap over a 4 um coupling section, then separate to distinct ports. One automatic pulsed run records the fundamental through and cross modes at 1.50, 1.55, and 1.60 um. The fixed 25 nm uniform grid resolves the silicon material wavelength with about 18 cells, closely matching the notebook's 20-step automatic grid target.

This case runs at its declared full benchmark resolution (resolution factor 1.0).

## Reconstruction choices

The official device dimensions, material stack, excitations, and measured observable are retained where they determine the physics. The local scene uses **496 × 216 × 72 cells** and the common port/source normalization machinery. The shared energy-decay cutoff stopped the run after **8,392 of 41,960** allowed steps. Geometry is rasterized as connected material before placement when touching polygons would otherwise introduce grid-snapping seams.

The field panel above is explicitly a schematic of the device and propagation path. The curves and scalar values in the result panel are executed benchmark outputs: solid circles are FDTDX; dashed crosses are the frozen Tidy3D reference.


## Recorded result

| Metric | Observed / error | Reference / limit | Error | Effective budget | Result |
|---|---:|---:|---:|---:|---:|
| `through_spectrum` | 0.038508 | 0.2 | 0.038508 | 0.2 | <span class="metric-pass">PASS</span> |
| `cross_spectrum` | 0.0111273 | 0.04 | 0.0111273 | 0.04 | <span class="metric-pass">PASS</span> |
| `total_power` | 0.0293169 | 0.1 | 0.0293169 | 0.1 | <span class="metric-pass">PASS</span> |

The table is rendered from `progress.json`; it is not a hand-written success claim. For metrics that report an error directly, the “observed” column repeats that error and the reference column is the acceptance threshold.

## Reproduce

```bash
uv run fdtdx-bench run --case device_directional_coupler
```

Implementation and provenance: `benchmarks/cases/device_directional_coupler/case.yaml`, `benchmarks/cases/device_directional_coupler/run.py`, `benchmarks/goldens/device_directional_coupler.npz`.

[Return to the example atlas](../index.md)