Campaign: quan_loncar_v1/autonomous_wrapper

Assembled 2026-07-29 15:57 UTC.

Engine: quan_loncar_v1 · path: engines/quan_loncar_v1/campaigns/autonomous_wrapper/

campaign.yaml

# Quan–Lončar autonomous campaign wrapper
name: quan_loncar_autonomous_wrapper
device: quan_loncar_airclad_nanobeam
engine:
  path: engines/quan_loncar/v2.0.0/quan_loncar_optimizer
  version: "2.0.0"
entrypoints:
  campaign: run_campaign.py
docs:
  strategy: README.md

README

Quan--Loncar autonomous campaign

This is a restartable, constrained BFGS campaign around the complete Phase 26 analytic all-hole gradient. It optimizes the 45 direct physical parameters (gaps, rx, ry) of the N=15 symmetric taper and writes an append-only JSONL ledger plus a small live dashboard.

The FDTD checkpoint uses the same 15 optimized taper with ten fixed outer mirror pairs on each side. Those mirrors are intentionally outside the analytic parameterization: their purpose is to suppress finite-domain waveguide leakage, so an FDTD Q is a radiation-Q checkpoint rather than an identical finite-mirror pole calculation. Every paid submission is estimated first and blocked when it would exceed the campaign credit cap.

Run from the repository root:

python quan_loncar_autonomous_optimizer/run_campaign.py --max-iterations 1000
python -m http.server 8080 -d quan_loncar_autonomous_optimizer/runs/<run>/dashboard

The campaign is intentionally single-process. Stop it safely with touch quan_loncar_autonomous_optimizer/runs/<run>/STOP.