Causal Fisher-Information Inequalities: Classical Causal Model Falsification and Metrological Advantage
Jeongho Bang, Su-Yong Lee
Abstract Fisher-information inequalities are increasingly used as operational witnesses of nonclassical metrological behavior, yet their interpretation is often tied to special pictures such as segmented dynamics or discrete trajectories. We develop a more general causal account. Given any classical experiment described by a directed acyclic graph, conditional independences, and modular parameter dependence, the associated Fisher informations must satisfy causal Fisher-information inequalities (CFIIs). The key result is a causal-path series law: for an additive parameter transmitted along a classical path A → C → B, inverse Fisher information acts as an information resistance and adds in series. Thus, any CFII violation rules out the entire specified classical causal model class. Moreover, such a violation directly certifies metrological advantage, since it implies a precision unattainable by any model in that class. We trace this advantage to Fisher-information synergy: off-diagonal score correlations excluded by classical modularity. A coherent single-qubit rotation illustrates deterministic CFII violation, estimator-level achievability, robustness to split-optimized classical benchmarks, and chain-amplified gains. Finally, an AI-assisted adversarial finite-data test confirms certifiability under visibility loss and readout error, with optimized modular classical models reaching but never crossing the CFII boundary.
•Jun 19, 2026•0 citations