Skip to content

Source

Engine design23_v1 · package package/design23_recentered_optimizer · assembled 2026-07-29 15:57 UTC.


Design 23 validation record

Geometry reconstruction

The seed is reconstructed directly from the uploaded design_23 function and notebook cell. It contains 42 reflected hole pairs (84 holes), 20 mirror pairs, 22 taper pairs, a 234.6254295066 nm shifted mirror pitch, a 556.2636519035 nm SiN width, and a 300 nm SiN thickness. The starting holes are circular. The optimizer stores separate longitudinal and transverse radii, so ellipses can evolve after the first step.

Resonance identification

The supplied FDTD result reports 795.692787 nm and Q = 11812.391474. The analytical probe first found the passive parity branch in P0 at 796.4111681202 nm and Q = 6189.018830. Its eigenvalue residual magnitude was 4.108e-10 and the smallest singular value was 2.615e-12.

The P0 eigenvector was then lifted into the P1 basis to prevent an unrelated upper-half-plane root from being selected. The production seed is

omega / omega_ref = 1.0024398920474082 - 0.00010659850238332136j
wavelength         = 793.7560988092 nm
Q                  = 4701.941724

The pole is passive and lies on the intended branch. The P1 model is used as a local direction generator, not as an absolute-Q substitute for FDTD.

Full analytical-step check

A complete 128-coordinate step was run through the same public optimizer path:

quantity initial accepted candidate
analytical Q 4701.941724 4949.390553
wavelength (nm) 793.756099 793.754806
normalized mode volume 218.347389 220.804971
TE0 decay fraction 0.130231 0.137061
width (nm) 556.263652 556.263628
thickness (nm) 300.000000 300.000017

Maximum coordinate motion was 0.0010963 nm and branch overlap was 0.999998886. This exercised the hole adjoint, the independent width/thickness operator derivatives, nonlinear pole correction, mode-volume residue, and guided-mode coupling calculation.

Eight offline package tests pass. A live cloud submission is intentionally not part of package testing; run the free local preflight, then the baseline-only checkpoint described in README.md before starting optimization.