Source-linked AI summary
An experimental test of noncontextuality without unwarranted idealizations
Michael D. Mazurek, Matthew F. Pusey, Ravi Kunjwal, Kevin J. Resch, Robert W. Spekkens
TL;DR
Noncontextuality is a broad operational notion of classicality, but prior tests assumed noiseless measurements and exact operational equivalences. This paper develops tests that avoid those idealizations using generalized probabilistic representations and implements one photonic test, obtaining a strong violation of the noncontextual bound. The work also identifies tomographic completeness as the most significant remaining loophole.
Problem
Noncontextuality has broad scope, but previous experimental tests relied on noiseless measurements and exact operational equivalences that real experiments do not precisely realize.
Method
The paper constructs secondary procedures satisfying operational equivalences and analyzes them using generalized states and effects, assuming three binary-outcome measurements are tomographically complete.
Results
A = 0.99709±0.00007 violates the noncontextual bound of 5/6 ≈0.833 by 2300σ.
Takeaways & Limitations
The techniques convert quantum proofs of noncontextuality failure into experimental tests robust to noise and experimental imprecisions.
Takeaways & Limitations
The most significant loophole is the assumption that three independent binary-outcome measurements are tomographically complete, which no experiment can conclusively test for all possible measurements.
Abstract
from arXiv · showhide
To make precise the sense in which nature fails to respect classical physics, one requires a formal notion of classicality. Ideally, such a notion should be defined operationally, so that it can be subjected to a direct experimental test, and it should be applicable in a wide variety of experimental scenarios, so that it can cover the breadth of phenomena that are thought to defy classical understanding. Bell's notion of local causality fulfills the first criterion but not the second. The notion of noncontextuality fulfills the second criterion, but it is a long-standing question whether it can be made to fulfill the first. Previous attempts to experimentally test noncontextuality have all presumed certain idealizations that do not hold in real experiments, namely, noiseless measurements and exact operational equivalences. We here show how to devise tests that are free of these idealizations. We also perform a photonic implementation of one such test that rules out noncontextual models with high confidence.
I. INTRODUCTION
Noncontextuality offers a broad operational notion of nonclassicality, but prior experimental tests relied on noiseless measurements and exact operational equivalences. The paper introduces theoretical innovations intended to remove these idealizations and reports a photonic test that rules out noncontextual models with high confidence.
- Motivation: Noncontextuality applies beyond Bell’s specialized spacelike-separated scenarios and relates to several phenomena considered distinctly quantum.These include negative quasiprobabilities, quantum advantages in cryptography and computation, and anomalous weak values.
- Experimental gap: Previous tests assumed deterministic measurement responses, which are justified only for noiseless measurements that real experiments never achieve exactly.The paper addresses this measurement-noise problem directly.
- Experimental gap: Real experiments also fail to realize exact operational equivalences required to apply noncontextuality to measurement events.Earlier work did not satisfactorily explain how deviations from strict equivalence should affect interpretation.
- Contribution: The paper develops a test of noncontextuality without the prior idealizations and implements one version using quantum optics.The reported experiment rules out noncontextual models with high confidence.
II. A NONCONTEXUALITY INEQUALITY
The paper formulates noncontextuality as a constraint on ontological models and derives an inequality from operationally equivalent measurements and preparations. Perfect average correlation would require deterministic responses that conflict with the noncontextuality constraints.
- Operational noncontextuality: Noncontextuality requires operationally statistically equivalent procedures to have statistically equivalent representations in the underlying model.For preparations, the ontic state cannot encode which operationally indistinguishable procedure was used.
- Operational noncontextuality: A measurement M* that is operationally equivalent to a fair coin flip must produce uniformly random outcomes for every ontic state.This is the measurement-side application of noncontextuality.
- Operational noncontextuality: Three preparations that are operationally indistinguishable must assign the same operational statistics across all measurements.Each preparation can also be decomposed into an equal mixture of two procedures, enabling the correlation construction.
- Noncontextuality inequality: The quantity A measures average correlation between the measurement outcome and the preparation variable, and noncontextuality imposes a nontrivial upper bound.The proof proceeds by showing that perfect correlation would require deterministic responses incompatible with the measurement mixture.
- Noncontextuality inequality: The resulting bound is tight and is established by a contradiction involving deterministic assignments for the three component measurements.Every deterministic assignment violates the relevant constraint, so perfect average correlation is impossible under noncontextuality.
III. QUANTUM VIOLATION OF THE INEQUALITY
Quantum theory supplies qubit preparations and measurements that satisfy the required operational equivalences and attain the logical maximum of the correlation quantity. The inequality remains applicable with measurement and preparation noise until the average correlation falls below 5/6.
- Quantum construction: Quantum theory predicts qubit preparations and measurements achieving A = 1, the logical maximum.The three measurement axes are separated by 120° in the x-z plane, with preparations given by their eigenstates.
- Quantum construction: The average density operator for each preparation pair is identical, so the pairs produce the same statistics for all measurements.This establishes the preparation operational equivalence used by the construction.
- Noise tolerance: The inequality accommodates noise in measurements and preparations while the average p(X = b|Mt, Pt,b) remains at least 5/6.Thus the test does not presume noiseless measurements.
IV. CONTENDING WITH THE LACK OF EXACT OPERATIONAL EQUIVALENCE
To handle imperfect operational equivalences, the paper constructs secondary procedures from experimentally inferred generalized states and effects, enforcing the required equivalences while keeping them close to the realized procedures. The approach uses tomographic completeness and supplementary y-axis procedures to address out-of-plane deviations.
- Secondary procedures: The experiment’s primary preparations and measurements deviate from ideal procedures, so their mixtures fail to satisfy the required equalities exactly.The method therefore searches among probabilistic mixtures of primary procedures for secondary procedures satisfying the equivalences.
- Secondary procedures: Secondary preparations are selected so that the mixtures of the three preparation pairs coincide, restoring the operational equivalence without requiring the completely mixed state.The same construction is extended to secondary measurements.
- Experimental refinement: Out-of-plane deviations are handled by supplementing the ideal preparation and measurement sets with y-axis eigenstates and the observable σ · ŷ.Without this refinement, suitable secondary procedures close to the ideal versions may not exist.
- Model-independent analysis: The analysis uses generalized states and effects rather than assuming that the data admit a quantum model.Three binary-outcome measurements are assumed only to be tomographically complete, a weaker assumption than identifying the system as a qubit.
- Experimental implementation: The experimental setup uses heralded photons, polarization preparation, spatial filtering, compensation waveplates, and two-outcome polarization measurements.The spatial filter decouples beam deflections from detector coupling efficiency.
V. EXPERIMENT
The experiment implements a photonic test of noncontextuality using primary and engineered secondary preparations and measurements. The measured correlations violate the noncontextual bound by 2300σ.
- Experimental setup: A heralded single-photon source and polarization optics implement eight preparations and four measurements for the noncontextuality test.The preparations use a polarizer, quarter-wave plate, and half-wave plate; measurements use polarization optics and a polarizing beamsplitter.
- Data analysis: Raw experimental data are fit to GPT states and effects using three tomographically complete binary-outcome measurements.The fit produces estimates of eight primary preparations and four primary measurements, with average χ2 = 3.9±0.3 over 100 runs.
- Operational construction: The primary measurement and preparation estimates closely match the raw data, but they do not themselves satisfy the required operational equivalences.This motivates constructing secondary procedures from probabilistic mixtures of the primary procedures.
- Operational construction: The secondary preparations achieve CP = 0.9969 ± 0.0001, close to the maximum of 1, and the secondary procedures are close to ideal.The construction enforces the operational equivalences required for the test.
- Violation of noncontextuality: The average correlation is A = 0.99709±0.00007, exceeding the noncontextual bound 5/6 ≈0.833 by 2300σ.The six individual degrees of correlation are all above 0.995 across the 100 experimental runs.
VI. DISCUSSION
The paper develops noise-robust experimental tests of noncontextuality while identifying experimental and conceptual limits on operational equivalence and tomographic completeness. It concludes that tomographic completeness remains the most significant loophole for such tests.
- The techniques convert proofs of quantum failure of noncontextuality into experimental tests robust to noise and experimental imprecisions.
- The experiment’s source has a multi-photon component, but its estimated effect changes A by at most 10^-6 and does not affect the conclusions.The measured g^(2)(0) is 0.0105 ± 0.0001, with a 1:4000 ratio of multi-photon to single-photon heralded detection events.
- Polarization–spatial coupling can violate tomographic assumptions, as omitting the spatial filter produced large χ2 values attributed to waveplate-induced beam deflections.
- The most significant loophole is that no experiment can conclusively vindicate the assumption that three independent binary-outcome measurements are tomographically complete for photon polarization.The experiment found a good χ2 fit using that assumption, but all possible measurements cannot be tested.
- Operational noncontextuality requires operationally equivalent procedures to have equivalent representations in the underlying ontological model.
- Exact operational equivalence cannot be realized experimentally because procedures never produce precisely identical statistics.
1. Derivation of bound
The derivation bounds the quantity A by translating preparation and measurement noncontextuality into geometric constraints on response-function triples. The allowed triples form a polygon whose extreme points suffice to establish the bound.
- Preparation noncontextuality makes the ontic distribution μ(λ|P_t) independent of t, enabling a common distribution ν(λ) in the bound.
- Measurement noncontextuality supplies the constraint that restricts the response-function triple for each ontic state λ.
- The response-function triple lies in the unit cube and, under the linear constraint, is confined to a two-dimensional plane.
- Because the relevant function is convex in the response-function triple, it suffices to evaluate it at the polygon’s extreme points.
- The plane–cube intersection is a polygon with six vertices given by permutations of (1, 1/2, 0).
- Substituting the resulting bound into Eq. (B10) yields the stated result for A.
2. Tightness of bound: two ontological models
The authors construct ontological models showing that the noncontextuality inequality is tight and that preparation noncontextuality alone does not enforce its bound. The models reproduce the required operational equivalences while differing in measurement contextuality.
- Tightness of the noncontextual bound: The first ontological model saturates the inequality at A = 5/6, proving that the noncontextual bound cannot be reduced.Its operational statistics realize the bound exactly.
- Preparation versus measurement noncontextuality: The second model is preparation noncontextual but measurement contextual, and exceeds the inequality bound with A = 9/10.Thus, measurement noncontextuality is required for the stated bound.
- Operational equivalences: Both models reproduce the operational equivalences: P1, P2 and P3 have identical statistics, while M∗ is indistinguishable from a fair coin flip.These equivalences are verified from the models’ operational probabilities.
- Preparation versus measurement noncontextuality: Only the first model is measurement noncontextual, requiring ξ(0|M∗, λ) = 1/2 for every ontic state λ.The second model can reproduce fair-coin statistics operationally without uniformly random responses at each ontic state.
- Secondary procedures: The secondary procedures are optimized to remain close to the primary ones while satisfying the needed equivalences, yielding CP = 0.9969 ± 0.0001 and CM = 0.9976 ± 0.0001.The construction supplements the primary preparations with σ·ŷ eigenstates and allows motion toward the required equivalences.
1. Fitting the raw data to a generalised probabilistic theory
The raw experimental data are fit to a generalised probabilistic theory rather than assumed to have a qubit description. A weighted hyperplane fit tests whether three binary-outcome measurements are tomographically complete.
- Raw-data representation: The data comprise four measurements performed on each of eight input states, arranged as a 4 × 8 raw-data matrix Dr.Rows correspond to M1 through M4, and columns correspond to the eight preparations Pt,b.
- GPT fitting: The analysis fits Dr to a GPT in which three two-outcome measurements are tomographically complete, replacing the stronger assumption that the system is a qubit.The fitted primary matrix Dp is constrained to lie on a three-dimensional hyperplane in four-dimensional data space.
- GPT fitting: The best-fitting GPT is obtained by weighted total least squares, mapping each raw-data column to its closest point on the fitted hyperplane.Measurement uncertainties are estimated from Poissonian counting statistics.
- Optimization: The hyperplane parameters are found by reducing the optimization from 36 variables to a four-variable numerical problem over {a, b, c, d}.The reduction first minimizes each column’s weighted distance to the hyperplane, then solves for the hyperplane parameters.
- Fit assessment: The fit returns a mean χ2 of 3.9 ± 0.3 across 100 runs and χ2 = 4.33 with p-value 36% after aggregating all runs.These values are consistent with the assumed GPT and Poissonian counting statistics.
- Fit assessment: Tomographic completeness is necessary because statistical equivalence under the performed measurements otherwise need not imply equivalence under all measurements.The experiment therefore adopts the stronger assumption that three measurements are tomographically complete and fits the data accordingly.
3. Analysis of statistical errors
The analysis estimates statistical uncertainty directly from repeated experimental runs rather than imposing a noise model, and applies this treatment to the optimized quantity A. The measured value lies 2300σ above the noncontextual bound, ruling out noncontextual models with high confidence.
- GPT states and effects fitted to finite raw-data samples estimate the true primary preparations and measurements.
- A linear program estimates the mixing weights that generate secondary procedures from the primary procedures.
- Statistical uncertainty in A is estimated from 100 distinct experimental runs, each independently recomputing A.
- The adopted resampling approach avoids relying on assumptions about the experiment’s noise distribution.
- 2300σ above the noncontextual bound, A indicates a very low likelihood that a noncontextual model better fits the true probabilities.
- A is analyzed analogously to a Bell quantity: it is ultimately a function of true probabilities, with linear optimization incorporated into its definition.
Appendix E: Experimental methods
The experiment generates and manipulates polarization-encoded photon pairs, using waveplates and polarization optics to prepare states and perform measurements. Spatial-mode filtering suppresses unwanted preparation information and ensures measurement responses depend on polarization as intended.
- A 20-mW, 404.7-nm diode laser produces photon pairs through spontaneous parametric down-conversion in a type-II PPKTP crystal.
- A quarter- and half-waveplate set the signal-photon polarization to one of eight states.
- A single-mode fibre filters spatial modes so preparation-waveplate angles are not encoded spatially and measurement responses depend only on polarization.
- Measurements use four polarization bases set by waveplates, with a second Glan–Taylor PBS separating the outputs.
- Each measurement records both PBS output ports in two stages to account for unequal coupling and detection efficiencies.