Archived solid-start Fourier atom-hole campaign

Archived solid-start Fourier atom-hole campaign#

Stopped initialization study

This optimizer was stopped on 2026-08-24 after 665 updates. Its complete checkpoint is preserved, but it is not the active design campaign.

This run established that a zero-coefficient Fourier initialization cannot be combined safely with immediate exact binarization. Zero latent coefficients decode to density 0.5; the binary tie convention made the nominal design a solid silicon slab, while the initial erosion and dilation thresholds saw nearly opposite material states. That was an initialization artifact, not a physics-motivated seed.

Frozen outcome#

Quantity	Value
Updates	665
Best recorded LDOS / vacuum	0.963590
Final LDOS / vacuum	0.729534
Final silicon fill	73.756%
Independent Fourier coefficients	13,725
Hard atom hole	1 µm diameter
Final forward material	exact binary

The best result remained below vacuum LDOS after hundreds of updates. The experiment therefore provides useful negative evidence: a full Fourier basis does not rescue a poorly posed solid-slab cold start.

Initial and final geometry#

Solid slab initialization and final binary geometry — The immutable atom hole remained clear, but the topology began from an almost entirely solid design.#

Complete optimization history#

LDOS and Fourier optimization history for the stopped solid-start run — All 665 executed updates are retained. Threshold-dependent oscillation is visible rather than removed from the record.#

Final Fourier spectrum and field#

Final Ez field of the stopped solid-start campaign

Reproducibility#

The immutable archive is benchmarks/artifacts/invdes_fourier_atom_hole_ldos_dense_fft_v1.rejected_solid_start_20260824.npz with SHA-256 a2ac63cd230a43de1a42b9ee730758edad53d22a8fe4622656bd3679b61eb0b2. It contains the parameters, optimizer moments, literal densities, Ez field, and complete per-update JSON history.