arXiv:2610.16702
Fisher-Information Certification from Bell Statistics: CHSH Bounds, Information Loss, and Finite-Data Guarantees
Juan Sumaya-Martínez, Omar Olmos-López, Jorge Mulia Rodríguez
Abstract
Bell-test data are typically used to certify nonlocal correlations, while quantum metrology is concerned with the sensitivity of measurement statistics to unknown parameters. Here we connect these tasks at the level of experimentally accessible Bell statistics. Using standard moment-based Fisher-information inequalities, we derive lower bounds on the classical Fisher information directly from Bell scores and their parameter dependence, without requiring full state tomography. For CHSH measurements we establish a hierarchy between the information contained in the full outcome record, the set of individual correlators, and the compressed Bell parameter, thereby quantifying the metrological cost of data compression. In polarization-phase sensing, this hierarchy reveals a Bell-metrology tradeoff: the point of maximal CHSH violation need not maximize, and can even suppress, the Fisher information carried by the Bell score itself. We further derive finite-sample lower-confidence bounds from coincidence counts and analyze the effects of biased marginals, accidental coincidences, visibility loss, and multipair emission. The framework provides a practical route for extracting metrological guarantees from Bell-type experiments using only standard correlation measurements.
AI Summary
The paper asks what classical Fisher information (FI) can be certified from parameter-dependent Bell-test statistics, rather than whether Bell violation witnesses quantum Fisher information (QFI) or metrological advantage. It treats the settings and outcomes as a parametric probability distribution and applies the standard covariance inequality between an observable and the likelihood score. For a full-correlation Bell score $B=\sum_{xy}c_{xy}E_{xy}$, sampled with parameter-independent probabilities $p_{xy}$, it derives $F_C\ge B'^2/[\sum_{xy}c_{xy}^2/p_{xy}-B^2]$. Uniformly sampled CHSH gives $F_C\ge S'^2/(16-S^2)$, including biased marginals. With balanced marginals, $F_C=\sum_{xy}p_{xy}E_{xy}'^2/(1-E_{xy}^2)$, yielding a tighter correlator-resolved certificate and a hierarchy of lower bounds. For an ideal polarization phase model with canonical CHSH analyzers, $S=2\sqrt2\cos\phi$ while the complete record attains $F_C=F_Q=1$. Both derivative-based certificates vanish at maximal CHSH violation; at quadrature, the correlator-resolved and score-only bounds are respectively $1$ and $1/2$. An optimized separable probe also reaches QFI $1$ for this local encoding, so no entanglement-enhanced precision is established. The analysis includes visibility loss, accidental backgrounds and biased marginals. Simultaneous Hoeffding intervals and calibrated phase modulation yield finite-sample lower-confidence FI bounds; derivative estimation requires a sinusoidal model or a known smoothness bound. The framework assumes regular, stationary distributions, with independent trials for finite-count guarantees. It is not fully device-independent, and coincidence-conditioned FI is per retained event, not per emitted pair or unit time.
OpenAI ChatGPT was used as an assistive tool during manuscript development to help reorganize text, check the internal consistency of algebraic derivations, and improve language and presentation.
Error found: The paper interprets its hierarchy of lower bounds as quantifying information lost under compression, although its correlator-resolved bound can vanish even when the four correlators preserve all the full-record Fisher information.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.