Sharp calibration of bounded E-value products under mutual information constraints
We determine the exact universal denominator for the product of two bounded e-values under a mutual information budget. The optimization allows arbitrary marginal distributions subject to upper bounds and mean inequalities. Separate Bernoulli randomizations saturate the means without increasing mutual information or reducing the product moment, yielding an exact one-dimensional reduction. For arbitrary, possibly unequal upper bounds and every information budget, we exhibit an attaining law and identify the entropy threshold at which the correction for arbitrary dependence becomes necessary. We also derive the first two correction terms near independence, characterize convergence to saturation, and give a Gaussian example with an analytically guaranteed information budget.
Authors
- Masanari Kimura (ORCID: https://orcid.org/0000-0002-9953-3469)
Institutions
- The University of Melbourne (AU)
Publication Details
- Journal
- Statistics & Probability Letters
- Published
- 2026-11-17
- DOI
- https://doi.org/10.1016/j.spl.2026.111011
- Primary Topic
- Probability and Statistical Research
- Type
- article
- Field-Weighted Citation Impact
- 0.00