Sharp Information Limits for Gaussian Covariance Mixtures under Partial Observation
We determine the small-noise expansion of the maximal relative entropy between a centered Gaussian covariance mixture and the Gaussian law with the same covariance. We prove a formula for the constant term at every positive definite barycenter feasible in a compact semialgebraic covariance set. For a five-cycle of pair observations, we find the optimal rank mixture and characterize saturation under small observation errors. We determine the critical sample scale and optimal limiting power for distinguishing two models with the same full covariance, and analyze uniform smoothing at the singular barycenter.
Authors
- Akihiro Koide (ORCID: https://orcid.org/0009-0002-8391-3480)
Publication Details
- Journal
- Figshare
- Published
- 2026-11-20
- DOI
- https://doi.org/10.6084/m9.figshare.35069191
- Primary Topic
- Bayesian Methods and Mixture Models
- Type
- preprint