Quantum advantage in postselected metrology

The paper investigates whether postselection can yield a genuinely nonclassical advantage in general quantum parameter estimation. Nonclassicality is taken to mean noncommutation of observables and is quantified via negativity in a quasiprobability representation (an extension of the Kirkwood-Dirac distribution). Quantum metrology aims to estimate unknown parameters with precision beyond classical limits; Fisher information is the key figure of merit, lower-bounding estimator variance via the Cramér-Rao bound. While classical and quantum postselection can increase Fisher information per successful outcome, a long-standing debate surrounds the nonclassical nature and utility of postselected quantum experiments and weak values. The authors ask whether postselection can surpass even the best standard (postselection-free), input-and-measurement-optimized quantum protocols and, if so, what quantum feature enables this. They propose that negativity arising from noncommutation is the resource, and analyze the Fisher information of postselected protocols compared to optimized prepare-measure schemes.

Background includes quantum metrology improvements via entanglement, where error scaling improves from N^-1/2 to N^-1 (Giovannetti, Lloyd, Maccone). Prior debates have focused on weak values, which can lie outside an observable’s spectrum and can offer metrological advantages for weak couplings, yet their classicality and optimality have been contested (e.g., Ferrie & Combes; Combes et al.). Quasiprobability distributions, such as the Wigner function and the Kirkwood-Dirac (KD) distribution, can assume negative or nonreal values, signaling nonclassicality and underpinning phenomena in quantum computation and scrambling. Earlier works linked KD negativity and contextuality to nonclassical resources. The present work generalizes beyond weak-coupling-specific scenarios by employing a doubly extended KD distribution tailored to the experiment’s eigenbases, tying nonclassical values directly to operational settings.

Two metrological procedures are contrasted: (1) Optimized prepare-measure: choose input state and final measurement to maximize Fisher information about a parameter θ governing a unitary evolution U(θ)=e^{-iθÂ}. (2) Postselected prepare-measure: after U(θ), perform a projective postselection {F, 1−F}; upon success, perform an information-optimal final measurement. For a quantum state ρ(θ)=U(θ)ρ0U(θ)†, the quantum Fisher information (QFI) is the Fisher information maximized over all generalized measurements; for pure states it equals I_Q(θ)=4 Var(Â). The maximal QFI over inputs for fixed generator  is (Δα)^2, where Δα is the difference between Â’s maximum and minimum eigenvalues. In the postselected setting, the successful state is ρ_ps=FρF/Tr(FρF). The authors derive an expression for the postselected QFI I_Q(θ|ρ_ps) and recast it in terms of a doubly extended Kirkwood-Dirac quasiprobability distribution defined with respect to the eigenbases of  and F. This representation cleanly connects the attainable postselected QFI to the presence or absence of negativity (or nonreality) in the distribution. They formalize costs: preparation cost C_P, postselection cost C_ps, and final measurement cost C_M, and define information-cost rates R(θ)=I(θ)/(C_P+C_M) and R_ps(θ)=p0 I_ps(θ)/(C_P+C_ps+C_M), where p0 is the postselection success probability. Theorems compare classical commuting theories (where the distribution is nonnegative and real) with quantum noncommuting scenarios.

• In optimized prepare-measure experiments with generator Â, the maximal quantum Fisher information equals (Δα)^2; for pure states I_Q(θ)=4 Var(Â).
• Postselection can increase QFI per successful trial beyond the unpostselected I_Q(θ). Crucially, in quantum settings, I_Q(θ|postselection) can exceed the optimized bound (Δα)^2, which is impossible classically.
• Theorem 1 (classical commuting case): If  and F commute (classically), the doubly extended KD distribution reduces to a real, nonnegative probability distribution, and no postselected prepare-measure experiment can yield more Fisher information than the optimized prepare-measure experiment; consequently, R_ps(θ) ≤ max R(θ) in classical theories.
• Theorem 2 (quantum noncommuting case): An anomalously large postselected Fisher information (exceeding the optimized prepare-measure bound) implies that the QFI cannot be expressed via a nonnegative doubly extended KD distribution; negativity (or nonreality) is necessary for this advantage. Pairwise noncommutation alone is not sufficient; the amount of advantage relates to the degree of negativity.
• Information-cost rate: In quantum mechanics, negativity allows R_ps(θ) > max R(θ) even compared to quantum experiments with optimized inputs, especially when C_M > C_P + C_ps.
• Unbounded conditional QFI: If the generator  has at least three non-identical eigenvalues (M ≥ 3), the postselected Fisher information I_ps(θ) can be made arbitrarily large (unbounded). When C_P and C_ps are negligible relative to C_M, the information-cost rate can, in principle, grow without bound.
• Trade-offs: Generally, I_ps(θ) and I(θ|all trials) remain below (Δα)^2 due to information discarded in failed postselections. However, with doubly degenerate minimum and maximum eigenvalues of Â, the product p·I_ps(θ) can approach (Δα)^2 while the conditional I_ps(θ) diverges, illustrating arbitrarily large per-success information at vanishing success probability.
• Conceptual link: Anomalous postselected QFI is closely connected to anomalous weak values and, counterfactually, to proofs of quantum contextuality via weak-value anomalies.

The results demonstrate a practical quantum advantage for metrology when final measurements are costly: conditioning the costly measurement on successful postselection can yield superior information-cost rates beyond any classical commuting strategy and beyond quantum strategies without postselection. This is especially pertinent for single-particle experiments where unsuccessful postselection may naturally prevent triggering detectors, making postselection essentially costless. Fundamentally, the work attributes the metrological advantage specifically to noncommutation and the associated negativity in a tailored quasiprobability representation, distinguishing it from other quantum resources like entanglement or discord. It reveals that postselection can endow a postselected quantum state with more Fisher information than any standard input state could achieve, and highlights a probabilistic violation of the optimized Cramér-Rao-like bound Var(θ)(Δα) ≥ 1 in the sense of conditional, per-success information and rates. The connection to anomalous weak values and contextuality invites further exploration of the foundational implications of metrological advantages arising from quasiprobability negativity.

The paper establishes that postselection can provide a genuinely nonclassical metrological advantage: in quantum (noncommuting) theories with quasiprobability negativity, postselected experiments can surpass the Fisher information and information-cost rates attainable by optimized postselection-free protocols. It introduces a doubly extended Kirkwood-Dirac representation that directly links achievable postselected QFI to negativity, proves classical impossibility (commuting case), and shows that postselected QFI can be made arbitrarily large under suitable spectral conditions of the generator. Practically, the approach benefits scenarios with expensive final measurements, and conceptually, it ties metrological advantage to noncommutation, negativity, and anomalous weak values. Future work includes experimental demonstrations (e.g., photonic platforms), quantifying the relationship between the magnitude of negativity and achievable metrological gains across broader protocols, and clarifying the foundational ties to contextuality.

• The metrological advantage relies on noncommutation and negativity in the doubly extended KD distribution; commuting (classical) scenarios cannot realize the advantage.
• Pairwise noncommutation without negativity (e.g., purely nonreal but nonnegative real parts) is insufficient to produce anomalous Fisher information.
• Gains pertain to conditional (per-success) information and information-cost rates; overall average information including failed postselections does not increase beyond quantum limits established in prior work.
• Practical benefits depend on relative costs: advantages are most pronounced when the final measurement cost exceeds the sum of preparation and postselection costs (C_M > C_P + C_ps).
• Open questions remain about the precise quantitative relation between negativity and metrological gain across other protocols and operator settings, and about the foundational linkage to contextuality.