Calibration and Identification of Determinant–Cofactor Moments under Unknown Gaussian Noise
This paper studies determinant-cofactor Hermite moments of a random matrix observed in independent additive Gaussian noise with unknown entry covariance. A shared calibration sample gives an exact unbiased polynomial inverse and a sharp lower bound on the calibration covariance. Partial transpose characterizes when the inverse is linear in the covariance. For three-by-three matrices, the paper classifies the largest mixed covariance spaces on which the quadratic term vanishes. It also determines how singularity, spectral separation and signal alignment affect the number of calibration observations required. Without a calibration sample, a quartic invariant characterizes identification under second-moment constraints and gives an exact minimax ambiguity radius. The accompanying Python code reproduces the algebraic identities and numerical examples.
Authors
- Akihiro Koide (ORCID: https://orcid.org/0009-0002-8391-3480)
Publication Details
- Journal
- Figshare
- Published
- 2026-11-21
- DOI
- https://doi.org/10.6084/m9.figshare.35083495
- Primary Topic
- Matrix Theory and Algorithms
- Type
- preprint