Archived axis-biased Fourier seed campaign

Archived axis-biased Fourier seed campaign#

Stopped seed audit

This 200 nm silicon / 1.35 µm run was stopped on 2026-08-24 after 20 continuous-discovery updates. It is preserved but is not the active campaign.

This restart corrected the earlier solid-slab initialization, but its random seed was generated on a rectangular nonnegative mode-index grid. The radial index envelope unintentionally favored modes with either kx=0 or ky=0 and omitted one of the two oblique orientations. The live spectrum made that bias visible before substantial compute was spent.

Frozen outcome#

Quantity	Value
Updates	20
Initial LDOS / vacuum	0.573570
Final and best LDOS / vacuum	0.652525
Projection beta	1.142857
Mean grayness 4ρ(1−ρ)	0.823819
Reduced-design fill	44.540%
Independent Fourier coefficients	13,725

At update 12, 36.7% of the seed power lay exactly on a coordinate axis and 52.3% lay within 15 degrees of one. The optimizer was already correcting the bias—those values fell to 26.3% and 44.6%, while diagonal-sector power rose from 12.8% to 18.9%—but an isotropic seed is a cleaner experiment.

Geometry and history#

Initial and final continuous geometry of the axis-biased seed campaign

LDOS and continuation history of the axis-biased seed campaign

Coefficients and field#

Cross-shaped Fourier spectrum that exposed the seed anisotropy

Final Ez field from the axis-biased seed campaign

The immutable archive is benchmarks/artifacts/invdes_fourier_atom_hole_ldos_spectral_discovery_v2.rejected_axis_biased_seed_20260824.npz with SHA-256 b22dd11ad2f8f94e57ba9e8691fdb10ba7475efca4b3158f6693438e00203b15.