Optimal Variance in Two-Point Sampling: Determinantal Designs and Strongly Rayleigh Laws
This preprint studies variance minimization for samples of size two with prescribed inclusion probabilities. A four-unit real projection gives an exact counterexample to Loonis's Conjecture 6.1 and improves on the ordered Loonis-Mary design. For arbitrary population sizes, a centered Gram formulation gives an exact semidefinite program and its dual for strongly Rayleigh pair laws. A direct rank reduction gives an optimal real projection for one statistic and an optimal complex projection for two. Sharp thresholds for mixtures with the same marginals and explicit six-unit and ten-unit examples establish the limits of determinantal designs. In the ten-unit example, the strongly Rayleigh minimum of the variance sum is 7/6, while every complex projection design with the same marginals has a variance sum at least 1/243000 greater. The files include the manuscript, LaTeX source, and exact Python computations.
Authors
- Akihiro Koide (ORCID: https://orcid.org/0009-0002-8391-3480)
Publication Details
- Journal
- Figshare
- Published
- 2026-11-15
- DOI
- https://doi.org/10.6084/m9.figshare.35016058
- Primary Topic
- Optimal Experimental Design Methods
- Type
- preprint