Sharp Label Budgets at a Schur Rank Boundary: Optimal Readout Design and Stratified Information Recovery
We establish an exact connection between Schur jet geometry, nuisance-constrained experimental design, and information loss under label coarsening. The underlying map is the second jet Gram of Sym²(Λ³A) on real 4 × 4 matrices, near A(t) = diag(1, 2, t, 0). Retaining all 136 diagonal Gaussian readouts and all thirteen nuisance coordinates, we allow one additional unit-variance Gaussian readout selected by a recorded, parameter-independent label. Exactly four label values are necessary and sufficient for positive definite information at order t² in three prescribed target coordinates. Over the full ordered entry set, the sharp leading minimum eigenvalue is 8192/41, attained by four explicit entries with probabilities (16, 16, 3, 6)/41. For every deterministic partition of these labels, with k observed groups, the true Gaussian mixture experiment has exactly k − 1 information eigenvalues of order t². The remaining directions reappear at order t⁴, except for one exact score kernel when two specified labels are merged. A positive calibration Gram and the second Gaussian Hermite score determine the fourth-order form. The results distinguish delayed recovery from exact differential blindness and show how observations silent at one layer can become decisive at the next, providing a precise basis for future label-aware readout design.
Authors
- Bin Du
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-21
- DOI
- https://doi.org/10.5281/zenodo.24414553
- Primary Topic
- Optimal Experimental Design Methods
- Type
- preprint