Refined Arthurs–Kelly Relation

• The refined Arthurs–Kelly relation is a generalized joint-measurement uncertainty bound that incorporates state-dependent corrections and error-spread trade-offs.
• It tightens the original constraints by linking measurement inaccuracies with both intrinsic observable spreads and correlations from entanglement.
• Experimental and theoretical advances validate its universal applicability across different measurement models and multiparameter estimation scenarios.

Searching arXiv for relevant papers on the refined Arthurs–Kelly relation and closely related universal complementarity / joint-measurement uncertainty results. The refined Arthurs–Kelly relation denotes a family of strengthened or generalized joint-measurement uncertainty bounds that extend the original Arthurs–Kelly result beyond its restrictive assumptions. In its 1965 form, the Arthurs–Kelly relation constrains simultaneous approximate measurements of two non-commuting observables through a lower bound on the product of measurement inaccuracies or pointer spreads, but only under conditions such as global unbiasedness or separable probe structure. Subsequent work reformulated the problem in state-dependent, operational, symplectic, and multiparameter-estimation settings, yielding refined inequalities that remain valid for arbitrary joint measurements, incorporate intrinsic spreads or correlation-deficit terms, and in some cases become tight under entanglement-assisted measurement design. The modern notion of a refined Arthurs–Kelly relation is therefore not a single formula but a class of universally valid strengthenings that quantify complementarity more sharply than the original bound (Weston et al., 2012, Lee, 2022, Wang et al., 13 Apr 2025).

1. Original Arthurs–Kelly framework and its limitations

The original Arthurs–Kelly setting concerns joint approximate measurement of two non-commuting observables $A$ and $B$, typically modeled by commuting meter observables that provide simultaneous readouts. In the formulation summarized by Weston, Hall et al., the inaccuracy of an estimator $\hat A_{\rm est}$ of $\hat A$ is quantified by the root-mean-square error

$$\epsilon(A_{\rm est})=\bigl\langle(\hat A_{\rm est}-\hat A)^2\bigr\rangle^{1/2},$$

and similarly for $B$ (Weston et al., 2012). Under the assumption of global unbiasedness, Arthurs and Kelly showed that

$$\epsilon(A_{\rm est})\,\epsilon(B_{\rm est})\ge \frac{c}{2},\qquad c:=|\langle[\hat A,\hat B]\rangle|.$$

This expresses the standard complementarity trade-off: increasing the accuracy of one estimate degrades the accuracy of the other (Weston et al., 2012).

The central limitation is the unbiasedness requirement. In the 2012 experimental complementarity analysis, this assumption is described as extremely restrictive and as failing in EPR-type scenarios where entanglement is exploited to generate highly accurate state-tailored joint estimates (Weston et al., 2012). Related critiques by Fujikawa and Umetsu identify the same structural weakness in modified Arthurs–Kelly-type inequalities: unbiased measurement and unbiased disturbance can fail for bounded observables such as spin, and this failure is directly responsible for experimental violations of conditionally valid error-product relations (Fujikawa et al., 2012, Fujikawa, 2013).

A parallel line of generalization appears in the three-body Arthurs–Kelly measurement model with two probes and one signal. For tripartite separable initial states, the usual twice-Heisenberg form is recovered only as a special case; if the initial probes are entangled, even that generalized separable-probe lower bound can be violated (Hwang et al., 2024). This establishes that the standard Arthurs–Kelly inequality is not universally robust once probe correlations or state-adapted strategies are allowed.

2. Universally valid refinement via error–spread trade-off

A major refinement was derived by Weston, Hall et al. for arbitrary joint measurements, without any unbiasedness hypothesis. Their relation is

$$\epsilon(A_{\rm est})\,\frac{\Delta B_{\rm est}+\Delta B}{2} \;+\; \epsilon(B_{\rm est})\,\frac{\Delta A_{\rm est}+\Delta A}{2} \;\ge\; \frac{c}{2},$$

where $\Delta A$ and $\Delta B$ are the usual quantum spreads of the target observables, $B$0 and $B$1 are the spreads of the estimates, and $B$2 (Weston et al., 2012).

This refinement is stronger than the original Arthurs–Kelly relation in two distinct senses. First, it is universal: it applies to arbitrary joint measurements rather than only globally unbiased ones. Second, it binds inaccuracies to both the spreads of the true observables and the spreads of the estimates themselves, rather than only to the product $B$3 (Weston et al., 2012). The symmetrized factors $B$4 and $B$5 are not ad hoc; they arise directly in the derivation from averaging two commutator bounds.

The derivation begins by representing the joint estimates by commuting operators $B$6 and $B$7, which can always be arranged by a suitable Naimark extension. One then uses the algebraic identity

$B$8

followed by the triangle inequality and Cauchy–Schwarz bounds on each commutator term (Weston et al., 2012). The resulting inequality is therefore structurally tied to commutator decomposition rather than to specific detector models.

Within the same analysis, the refined relation is shown to imply both Hall’s relation

$B$9

and Ozawa’s relation

$\hat A_{\rm est}$0

particularly when optimal estimators satisfy

$\hat A_{\rm est}$1

and the analogous identity for $\hat A_{\rm est}$2 (Weston et al., 2012). In that sense, the refined error–spread relation subsumes earlier universal bounds.

3. Experimental verification and EPR-type joint measurement

The 2012 experiment tested universal complementarity relations using entangled photon pairs in the state

$\hat A_{\rm est}$3

with $\hat A_{\rm est}$4, so that $\hat A_{\rm est}$5 while retaining strong EPR-type correlations (Weston et al., 2012). The target observables on the first qubit were the Pauli $\hat A_{\rm est}$6 and $\hat A_{\rm est}$7 operators.

The joint measurement combined a semiweak measurement of $\hat A_{\rm est}$8 on the first qubit with a direct $\hat A_{\rm est}$9-measurement on that same qubit, while the second qubit was measured in an optimally chosen basis $\hat A$0 to form an estimator $\hat A$1. Two estimators were tested: the simple estimator $\hat A$2 and the optimal state-dependent estimator $\hat A$3 that minimizes $\hat A$4 (Weston et al., 2012). By recording the full joint statistics of the semiweak outcome, the $\hat A$5-outcome, and the $\hat A$6-outcome, and by invoking the contextual-value formalism, the experiment extracted inaccuracies and spreads operationally from data.

The key empirical point is that the product of inaccuracies was sufficiently small to violate the widely used Arthurs–Kelly relation, while the refined universal relation remained valid and was in fact saturated in the entanglement-assisted regime (Weston et al., 2012). This is not merely an experimental anomaly relative to the older bound; it directly exhibits the fact that state-tailored entangled strategies evade global unbiasedness and therefore fall outside the validity domain of the original Arthurs–Kelly theorem.

A common misconception is that such a violation overturns complementarity itself. The opposite conclusion is supported by the experiment: complementarity survives, but its correct universally valid quantitative expression is not the naive Arthurs–Kelly product form. The refined relation preserves a nontrivial lower bound by coupling inaccuracy to the spreads of both the estimates and the underlying observables (Weston et al., 2012).

4. Modified and universally valid variants in operator-based measurement theory

Another strand of refinement emerges from operator-based analyses of measurement error and disturbance. Fujikawa and Umetsu discuss a modified Arthurs–Kelly relation of the form

$\hat A$7

where

$\hat A$8

derived from Robertson’s relation under assumptions of unbiased measurement of $\hat A$9, unbiased disturbance of $$\epsilon(A_{\rm est})=\bigl\langle(\hat A_{\rm est}-\hat A)^2\bigr\rangle^{1/2},$$0, and commutativity $$\epsilon(A_{\rm est})=\bigl\langle(\hat A_{\rm est}-\hat A)^2\bigr\rangle^{1/2},$$1 (Fujikawa et al., 2012). This formulation combines measurement error and intrinsic spread into output standard deviations.

However, neutron spin experiments with $$\epsilon(A_{\rm est})=\bigl\langle(\hat A_{\rm est}-\hat A)^2\bigr\rangle^{1/2},$$2, $$\epsilon(A_{\rm est})=\bigl\langle(\hat A_{\rm est}-\hat A)^2\bigr\rangle^{1/2},$$3, and a detuned measurement direction show that this modified Arthurs–Kelly form can fail both theoretically and experimentally (Fujikawa et al., 2012). The analysis attributes the failure to the inability to maintain unbiased disturbance for bounded observables in strong measurements. In a related treatment, conditionally valid uncertainty relations are described as singular in the precise-measurement limit, because exact readout of one observable is incompatible with strict unbiased disturbance of its conjugate (Fujikawa, 2013).

To avoid such singularities, Fujikawa and Umetsu advocate a universally valid refinement defined through

$$\epsilon(A_{\rm est})=\bigl\langle(\hat A_{\rm est}-\hat A)^2\bigr\rangle^{1/2},$$4

with the inequality

$$\epsilon(A_{\rm est})=\bigl\langle(\hat A_{\rm est}-\hat A)^2\bigr\rangle^{1/2},$$5

This version requires only $$\epsilon(A_{\rm est})=\bigl\langle(\hat A_{\rm est}-\hat A)^2\bigr\rangle^{1/2},$$6 and explicitly retains the intrinsic spreads $$\epsilon(A_{\rm est})=\bigl\langle(\hat A_{\rm est}-\hat A)^2\bigr\rangle^{1/2},$$7 and $$\epsilon(A_{\rm est})=\bigl\langle(\hat A_{\rm est}-\hat A)^2\bigr\rangle^{1/2},$$8, thereby remaining well defined even when one measurement error formally vanishes (Fujikawa, 2013). This suggests a broad conceptual principle: refined Arthurs–Kelly-type relations become universally valid when intrinsic quantum fluctuations are treated as part of the total inaccuracy rather than as extraneous preparation noise.

5. Universal formulations under local representability and joint measurability

A more abstract refinement is developed in the local-representability framework. For a measurement $$\epsilon(A_{\rm est})=\bigl\langle(\hat A_{\rm est}-\hat A)^2\bigr\rangle^{1/2},$$9, a fixed state $B$0, and observables $B$1, Lee defines state-dependent error quantities through pull-back and push-forward maps between quantum observables and classical outcome functions. In this setting the universal uncertainty relation

$B$2

holds, where

$B$3

(Lee, 2022). The original Arthurs–Kelly–Goodman inequality appears as a special case under global unbiasedness.

For joint measurements $B$4 and $B$5 admitting a joint measurement $B$6, a refined Arthurs–Kelly–Goodman theorem is obtained in which the product of the two errors satisfies

$B$7

with

$B$8

(Lee, 2022). Since $B$9, this strictly refines the usual commutator lower bound.

A related formulation in the same program gives, for local representatives $$\epsilon(A_{\rm est})\,\epsilon(B_{\rm est})\ge \frac{c}{2},\qquad c:=|\langle[\hat A,\hat B]\rangle|.$$0 and $$\epsilon(A_{\rm est})\,\epsilon(B_{\rm est})\ge \frac{c}{2},\qquad c:=|\langle[\hat A,\hat B]\rangle|.$$1,

$$\epsilon(A_{\rm est})\,\epsilon(B_{\rm est})\ge \frac{c}{2},\qquad c:=|\langle[\hat A,\hat B]\rangle|.$$2

where $$\epsilon(A_{\rm est})\,\epsilon(B_{\rm est})\ge \frac{c}{2},\qquad c:=|\langle[\hat A,\hat B]\rangle|.$$3 is the quantum covariance and $$\epsilon(A_{\rm est})\,\epsilon(B_{\rm est})\ge \frac{c}{2},\qquad c:=|\langle[\hat A,\hat B]\rangle|.$$4 (Lee, 2022). These results show that the refined lower bound acquires explicitly state-dependent real and imaginary contributions beyond the canonical commutator term. A plausible implication is that “refinement” in this framework means tracking the mismatch between quantum correlations and those reconstructed from measurement outcomes, rather than merely adding correction terms to a product inequality.

6. Correlated probes, symplectic formulations, and generalized measurement models

In continuous-variable Arthurs–Kelly models, refinement also arises from explicit dependence on probe correlations, covariance transport, and full Hamiltonian dynamics. For arbitrary correlated probes, the induced system POVM is covariant under phase-space translations, and its marginals are unsharp position and momentum observables with noise distributions $$\epsilon(A_{\rm est})\,\epsilon(B_{\rm est})\ge \frac{c}{2},\qquad c:=|\langle[\hat A,\hat B]\rangle|.$$5 and $$\epsilon(A_{\rm est})\,\epsilon(B_{\rm est})\ge \frac{c}{2},\qquad c:=|\langle[\hat A,\hat B]\rangle|.$$6 (Bullock et al., 2014). The corresponding root-mean-square errors satisfy

$$\epsilon(A_{\rm est})\,\epsilon(B_{\rm est})\ge \frac{c}{2},\qquad c:=|\langle[\hat A,\hat B]\rangle|.$$7

and covariance of the phase-space observable implies

$$\epsilon(A_{\rm est})\,\epsilon(B_{\rm est})\ge \frac{c}{2},\qquad c:=|\langle[\hat A,\hat B]\rangle|.$$8

hence

$$\epsilon(A_{\rm est})\,\epsilon(B_{\rm est})\ge \frac{c}{2},\qquad c:=|\langle[\hat A,\hat B]\rangle|.$$9

(Bullock et al., 2014). Correlated probes can produce “focusing,” meaning sharper marginals than single-probe measurements, but they do not violate the fundamental Heisenberg-type error–error bound.

In a symplectic treatment of the Arthurs–Kelly model with Gaussian product states, the covariance matrix evolves as

$$\epsilon(A_{\rm est})\,\frac{\Delta B_{\rm est}+\Delta B}{2} \;+\; \epsilon(B_{\rm est})\,\frac{\Delta A_{\rm est}+\Delta A}{2} \;\ge\; \frac{c}{2},$$0

and the multimode symplectic-covariant uncertainty principle $$\epsilon(A_{\rm est})\,\frac{\Delta B_{\rm est}+\Delta B}{2} \;+\; \epsilon(B_{\rm est})\,\frac{\Delta A_{\rm est}+\Delta A}{2} \;\ge\; \frac{c}{2},$$1 yields the three-term inequality

$$\epsilon(A_{\rm est})\,\frac{\Delta B_{\rm est}+\Delta B}{2} \;+\; \epsilon(B_{\rm est})\,\frac{\Delta A_{\rm est}+\Delta A}{2} \;\ge\; \frac{c}{2},$$2

(Arvind et al., 2020). This is explicitly stronger than a naive $$\epsilon(A_{\rm est})\,\frac{\Delta B_{\rm est}+\Delta B}{2} \;+\; \epsilon(B_{\rm est})\,\frac{\Delta A_{\rm est}+\Delta A}{2} \;\ge\; \frac{c}{2},$$3, and it makes the apparatus-noise contributions operationally visible through closed-form expressions for $$\epsilon(A_{\rm est})\,\frac{\Delta B_{\rm est}+\Delta B}{2} \;+\; \epsilon(B_{\rm est})\,\frac{\Delta A_{\rm est}+\Delta A}{2} \;\ge\; \frac{c}{2},$$4 and $$\epsilon(A_{\rm est})\,\frac{\Delta B_{\rm est}+\Delta B}{2} \;+\; \epsilon(B_{\rm est})\,\frac{\Delta A_{\rm est}+\Delta A}{2} \;\ge\; \frac{c}{2},$$5.

When free evolution is included in the full Hamiltonian, further positive corrections appear. For minimum-uncertainty Gaussian states, the pointer variances acquire spreading terms $$\epsilon(A_{\rm est})\,\frac{\Delta B_{\rm est}+\Delta B}{2} \;+\; \epsilon(B_{\rm est})\,\frac{\Delta A_{\rm est}+\Delta A}{2} \;\ge\; \frac{c}{2},$$6 due solely to the free Hamiltonian, leading to refined bounds such as

$$\epsilon(A_{\rm est})\,\frac{\Delta B_{\rm est}+\Delta B}{2} \;+\; \epsilon(B_{\rm est})\,\frac{\Delta A_{\rm est}+\Delta A}{2} \;\ge\; \frac{c}{2},$$7

as well as analogous corrections for retrodictive and predictive error products (Mendoza-Fierro et al., 2021). These refinements quantify the extent to which the idealized instantaneous-measurement limit understates realistic measurement noise.

A distinct generalization, focused on initial-state propagation, considers a tripartite system with two probes and one signal and the impulsive Hamiltonian

$$\epsilon(A_{\rm est})\,\frac{\Delta B_{\rm est}+\Delta B}{2} \;+\; \epsilon(B_{\rm est})\,\frac{\Delta A_{\rm est}+\Delta A}{2} \;\ge\; \frac{c}{2},$$8

For tripartite separable initial states one obtains

$$\epsilon(A_{\rm est})\,\frac{\Delta B_{\rm est}+\Delta B}{2} \;+\; \epsilon(B_{\rm est})\,\frac{\Delta A_{\rm est}+\Delta A}{2} \;\ge\; \frac{c}{2},$$9

with

$\Delta A$0

which reduces to $\Delta A$1 when each local state satisfies $\Delta A$2 (Hwang et al., 2024). Entangled probes can violate this separable-probe bound via nonzero cross-correlations $\Delta A$3, again illustrating that refinement depends critically on which structural assumptions are retained.

7. Multiparameter quantum estimation and the entanglement-dependent refined bound

A recent and conceptually distinct use of the term arises in multiparameter quantum estimation. For simultaneous estimation of time delay $\Delta A$4 and angular frequency $\Delta A$5, Wang et al. derive a tight analytical trade-off in terms of the classical Fisher information $\Delta A$6 and quantum Fisher information $\Delta A$7. After reparametrizing so that $\Delta A$8, the incompatibility of the two parameters is encoded in $\Delta A$9, and for two parameters one obtains

$\Delta B$0

(Wang et al., 13 Apr 2025).

In the quantum-radar application, entangled signal-idler bi-photons are characterized by a correlation parameter $\Delta B$1. For separable photons, one has $\Delta B$2, recovering $\Delta B$3. For entangled bi-photons, $\Delta B$4, which yields the refined Arthurs–Kelly relation

$\Delta B$5

At $\Delta B$6, the original bound is recovered; as $\Delta B$7, the right-hand side tends to $\Delta B$8, corresponding to the infinite-entanglement limit in which the two parameters become compatible in the quantum–Cramér–Rao sense (Wang et al., 13 Apr 2025). The same result is restated in a 2026 summary of the method, which emphasizes that the bound is tight and saturable for every $\Delta B$9 (Wang et al., 22 May 2026).

The significance of this development is twofold. First, it recasts Arthurs–Kelly-type complementarity as a Fisher-information trade-off induced by SLD incompatibility. Second, it provides an explicit optimal-measurement construction: optimal commuting approximations to the SLDs are formed, the vectors $B$00 are orthogonalized by Gram–Schmidt, and a real orthogonal rotation determines a projective measurement basis that saturates the bound (Wang et al., 13 Apr 2025). In the radar examples, the previously proposed “time–time” versus “frequency–frequency” joint measurement is shown analytically to attain the refined limit for arbitrary $B$01 (Wang et al., 13 Apr 2025).

This suggests a broader unification: refined Arthurs–Kelly relations can be understood as tight trade-off laws for forcing incompatible observables, estimators, or SLDs into a single commuting measurement architecture. Across the different formalisms, the same structural message recurs. The original product bound is valid only under restrictive assumptions; once those assumptions are relaxed, the correct quantitative statement of complementarity must incorporate additional state, spread, covariance, or entanglement data, and the sharpest such statements are often saturable in explicitly constructed measurement schemes (Weston et al., 2012, Lee, 2022, Wang et al., 13 Apr 2025).