---
title: Air-start variable-oxide-width atom-strip Q/V optimizer
---

# Air-start variable-oxide-width atom-strip Q/V optimizer

**Worker:** deliberately not started · **stage:** `prepared` · **status:**
`ready_for_user_authorization`

This fresh campaign is fully prepared while the current density-0.5 optimizer
continues untouched. It uses the same 16 × 2 µm topology region, y-oriented
atom, one-micron immutable air strip, XYZ symmetry, exact temporal discovery,
tracked-pole Q/V objective, component-aware Yee operator, adaptive trust
controller, plateau-driven stage handoffs, gradual beta continuation, and
topology-free smooth-boundary endpoint.

| prepared quantity | value |
|---|---:|
| trainable SiN topology initialization | **density 0 (air start)** |
| fixed feedthrough pin | exact SiN, index 2.0 |
| SiO2 material | **exact index 1.44** |
| initial SiN thickness | 0.100 µm |
| initial SiO2 thickness | 0.300 µm per side |
| initial SiO2 full width | **1.800 µm** |
| SiO2 full-width range | 0.600–2.000 µm |
| material discretization | component-aware Yee |
| spatial filter | none |
| snapshot cadence | every 50 accepted updates |
| scheduled stop | none |

## Material architecture

The SiO2 guide is not represented by a second gray density field. It has only
smooth analytic width and thickness coordinates and therefore remains exact
SiO2 in its bulk and exact air outside its boundary. The SiN core alone has a
2-D freeform density plus a thickness coordinate. The hard atom strip
overrides both materials.

This separation removes the ambiguous question of how one projected scalar
should choose among air, SiO2, and SiN. During binarization, beta acts only on
SiN occupancy. The oxide width/thickness remain analytic subpixel boundaries;
their gray cut cells are physical boundary averages, not unfinished material.

## Prepared stage sequence

1. Air-start raw Adam optimizes the exact absolute atom-field temporal trace;
   all three analytic dimensions are frozen.
2. After temporal convergence and trustworthy Q≥100, adaptive-trust L-BFGS-B
   optimizes direct atom-local Q/V using only raw SiN pixels.
3. At the pixel-only Q/V plateau, SiN thickness, SiO2 thickness, and SiO2
   width become trainable together. Gradual fixed-beta continuation begins,
   with Q/V still the sole objective.
4. At binary readiness and a fixed-beta Q/V plateau, the SiN topology moves
   to a topology-free cubic-spline phase field. The analytic oxide boundaries
   remain analytic and joint Q/V optimization continues indefinitely.

Once launched, this page will be replaced automatically by the live dashboard
with Q, V, Q/V, Purcell, pole wavelength, fit quality, ideal/measured temporal
traces, geometry/difference, fields, trust state, beta, and all three physical
dimensions.

The exact campaign methodology is already stored with the campaign code and
will be attached here automatically at launch.