WorksPhysics and Astronomy
Experimental constraints on gauge hidden-variable models in Bell tests
Abstract
Quantum mechanics assigns identical statistics to measurement procedures that realize the same final effect, irrespective of the transformations used to reach it. We test this operational equivalence with polarization-entangled photons by implementing one of Bob's analyzer settings either directly or through closed wave-plate trajectories calibrated to produce the same measurement operator. Motivated by ontological descriptions in which a hidden frame is transported between contexts, we formulate a local path-memory model and derive the mixed-context bound . The quantity is a coupling-dependent counterfactual mismatch probability: it bounds a contextual Bell expression, but neither generates nor predicts the quantum Bell correlation. We therefore test path memory through the observable difference between direct and looped implementations, using a no-signaling phenomenological response in which a hidden holonomy modifies one correlation term in . The direct reference violates the Bell inequality, while the looped mixed-context curves remain indistinguishable from it within the experimental precision. Our results support the quantum equivalence of calibrated measurement implementations and constrain path-sensitive hidden-variable extensions that predict an observable memory of the optical history.
Introduction
Bell’s theorem turned the tension between entanglement and local realism into an experimental question [1, 2, 3]. In the Clauser-Horne-Shimony-Holt (CHSH) scenario, Alice chooses between settings and , Bob between and , and the signed combination is formed [4]. Any Bell-local model with setting-independent hidden variables obeys , whereas quantum mechanics allows . Loophole-free experiments have established violations of the Bell-local bound [5, 6, 7]; the question is no longer whether entangled correlations can be explained by a Bell-local model.
Here, we address a different assumption, usually left implicit: what physically constitutes a measurement setting? Quantum theory identifies the setting with its measurement effects. In an optical experiment, however, the same effect can be realized through many sequences of wave plates and polarizers. Quantum mechanics treats these implementations as operationally equivalent. An ontological extension need not do so if an internal degree of freedom retains a memory of the path used to reach the final analyzer.
We formulate this possibility using a gauge hidden-variable (GHV) description of polarization. A pure polarization state defines a point on the Poincaré sphere, and an oriented frame is attached to the corresponding tangent plane. A connection transports the hidden frame along the trajectory traced by the optical transformations. After a closed path, the quantum measurement can return to its initial form while the hidden frame acquires a holonomy set by the enclosed solid angle [8, 9, 10].
This geometry does not evade Bell’s theorem. If the complete response remains Bell factorisable, it obeys

Figure 1. Geometric transport of a hidden tangent frame. a, A pure polarization state is represented by a point on the Poincaré sphere . The orthonormal basis spans the local tangent plane, and the colored circle denotes the fiber of hidden orientations, . b, Bob’s polarization follows a closed path . The hidden frame is transported with successive tangent planes and, after the polarization returns to its initial state, may acquire the holonomy , where is the enclosed solid angle. Operationally equivalent paths can therefore retain different histories in the GHV description.
the Bell bound regardless of the topology of its hidden-variable space. We use the gauge construction for a narrower purpose: to identify a possible path-dependent correction to an already observed entangled correlation. We compare direct and looped implementations of Bob’s analyzer, calibrated to realize the same final effect, and search for a structured deformation of a mixed-context witness. Such a deformation is distinct from a loss of visibility or an ordinary wave-plate calibration error.
Gauge-covariant local response. In both classical and quantum contexts, a polarizer can be characterized by a unit vector on the Poincaré sphere. In quantum mechanics, the action of such a polarizer can be represented by the binary projection operator , , where is the identity matrix, are the Pauli matrices, and describe two orthogonal states of polarization. If we put a sequence, say , of lossless waveplates along the path of light before entering the polarizer, these plates can be represented by a unitary operator , such that . In quantum mechanics, two measurement procedures are considered operationally equivalent if they produce the same measurement operators, and thus, yield identical probabilities for all outcomes. Thus, two waveplates’ sequences and are operationally equivalent when .
Given the Poincaré sphere , every Stokes vector has a tangent space . The unit tangent bundle is the collection of all tangent vectors of length 1 at all points. Here, the GHV ontic space is taken to be the unit tangent bundle . We use the symbol to represent the natural projection map , associated with this bundle, such that . Let the unit vector trace a closed curve on the unit sphere, where is a continuous parameter such that (closure condition). To each value of is associated a tangent space . For such a space we introduce the orthonormal basis . Let be a tangent vector field along , defined by , where . We call the hidden orientation field. The vector is precisely a choice of lift, parameterized by , of the curve into : at each it specifies not just the point but also a direction in the tangent plane, i.e., a point of the fiber . Thus the pair contains all the information of the polarization state characterized by the Stokes vector , except its overall sign.
When varies between 0 and 1, the basis vectors and undergo a parallel transport along . By definition, , so that is invariant under parallel transport. However, the hidden orientation field rotates by the angle , as , where is the solid angle subtended by the closed curve . If is the Jones vector generating the Stokes vector via the Pauli-matrix map (), here, is the scalar product. Under parallel transport , and . The Stokes vector and the Jones vector belong to different vector spaces and carry different amounts of information, with giving the more complete description of polarization.
In this context, a polarisation analyser– whether Alice’s or Bob’s –is fully characterized, apart from a global sign, by the pair , where and are the Stokes vectors describing the polarization transmitted and reflected by the polarizer, respectively, and the waveplates sequence determines the closed curve .
Suppose the light source emits a pair of photons in the “hidden state” , and is the probability measure assigned to this state. Let be the probability that for a given photon pair emitted in the state , Bob’s detector will fire (), or will not fire (), where specifies the analyzer setting, e.g., the polarizer orientation. A simple local detector model suggests taking , where is a unit vector defined by . For a linear analyzer at angle , where , and since , then
When a waveplates sequence is in place, remains unchanged, and , so that , where we have defined . These results also apply, mutatis mutandis, to , where is Alice’s linear analyzer orientation. Bell locality is imposed by the requirement that the complete model be Bell factorisable, that is
with .
The observed probabilities , , and , , are obtained by averaging , , and , respectively, over the probability distribution for the hidden variables. Choosing a uniform fiber distribution , we find , and . By construction, . The shift preserves the observed probabilities. A hidden frame can therefore retain a local path memory without changing Alice and Bob’s single-photon count rates.
To connect these results with the standard quantum-mechanical prediction, let us define the dichotomic random variables and , which take the values with the joint probability . By definition , and . The correlation is thus given by,
The angular dependence is in agreement with the quantum-mechanical prediction, but the visibility is one half and the largest sinusoidal CHSH value is . More generally, no Bell-factorisable completion can exceed . The tangent-frame model supplies local Malus statistics and a geometric memory; it does not derive the unit-visibility quantum correlation.
Counterfactual mismatch bound. Path memory allows the response associated with the nominal setting to depend on its implementation. For the direct and looped sequences, define the two dichotomic random variables , , as
where , and here and hereafter we use the same notation as in Eq. (1), but now the probabilities , with , can be arbitrary. For a complete hidden state, including any random seed used to make the response deterministic, consider the mixed-context polynomial
where the dichotomic random variables , , and , also take the values and . Adding and subtracting in the equation above gives
The term in parentheses is a standard deterministic CHSH polynomial and has magnitude . Similarly, , so that
Defining
and , yields
(9) is not the standard two-setting CHSH inequality. The two occurrences of are different complete settings, distinguished by their physical implementations. Moreover, only records how often the two counterfactual outcomes differ. It does not determine the sign of the correction or its correlation with the remaining terms. The condition is therefore necessary for the upper bound to accommodate , but it is not sufficient to produce that value. At , one has , and the local construction still obeys the Bell bound; it cannot account for the observed reference violation.
Using the local response model Eqs. (1) and (2), we obtain the probability distributions of and , we can calculate from (8), to obtain
where has been used. This is a property of the chosen local coupling, not a general observable prediction. The Malus marginal alone does not determine , and Eq. (10) should not be identified with the non-factorisable response introduced below.
Operational response tested experimentally. The experiment does not use to generate the baseline Bell violation. Instead, it asks whether the observed entangled correlation acquires an additional dependence on the implementation path. The source produces a state close to
for which the direct correlation is written , where is the correlation visibility. To parameterize a possible path-dependent displacement while preserving unbiased marginals, we use
with
where
and is a phenomenological response coefficient. The value gives path independence and corresponds, up to orientation, to Levi-Civita transport of a tangent vector on the unit sphere. Other values require additional ontological structure and do not follow from the Pancharatnam-Berry phase alone.
Equation (12) is no-signaling but, for Bell-violating correlations, not Bell factorisable. It is a phenomenological joint response used to expose a measurable signature of hidden-frame transport; it is not derived by saturating Eq. (9). In particular, does not determine the visibility or phase of Eq. (13).
The correlation difference is
For , this becomes . The first-order sensitivity is largest near zero crossing of the correlation fringe and vanishes at its extrema. A nonzero would establish operational path dependence, but a null result does not place a model-independent upper bound on the counterfactual quantity .
The experimental witness mirrors the mixed-context construction in Eq. (5). The first three correlations use the direct implementation, whereas the final occurrence of Bob’s setting is realized through the loop :
For , , and , Eq. (13) gives
The reference curve, obtained for , is . Taking the signed difference gives the particularly transparent relation
For small , this reduces to . Because only one correlation term is changed, the predicted mixed-context curve is deformed in both position and amplitude; it is not a rigid translation of a standard CHSH fringe. The theoretical curves in Figure 3 use the illustrative choice and . They are predictions of the operational response in Eq. (13), not consequences of assigning a value to .

Figure 3. Direct and looped implementations. a, Closed trajectories tested on the Poincaré sphere; colors and symbols identify the loop parameter . The loops are implemented using two QWPs and a HWP; their orientations are specified in the End Matter. b, Experimentally measured absolute mixed-context parameter as a function of . The dotted lines show the theoretical predictions of quantum mechanics – the detailed derivations can be found in the End Matter. c, Eq. (17) for and . The grey dotted line at 2 and the red dashed line at are shown as references to the standard CHSH and Tsirelson bounds, respectively; the contextual bound for the looped witness is instead given by Eq. (9). Error bars are smaller than the markers.
Experimental implementation. Polarization-entangled photon pairs were generated by degenerate type-I spontaneous parametric down-conversion in two orthogonally oriented BiBO (BiBO) crystals. A quasi-continuous-wave laser at 355 nm delivered an average power of approximately 100 mW. A half-wave plate before the crystals set the pump polarization and coherently excited the two down-conversion pathways. Their relative phase was compensated to produce a state close to with a fidelity of 96.4%. The experimental layout is shown in Figure 2.
Figure 2. Experimental implementation. A 355 nm laser pumps two orthogonally oriented BiBO crystals to generate polarization-entangled photon pairs by spontaneous parametric down-conversion. Alice performs direct linear-polarization measurements. Bob realizes the same nominal analyzer either directly or through a QWP–QWP–HWP sequence that traces a closed trajectory on the Poincaré sphere. The direct and looped sequences are calibrated to produce the same final measurement effects. Quantum mechanics, therefore, predicts identical coincidence statistics; an implementation-dependent difference would signal additional path memory.
Alice and Bob analyzed their photons using a half-wave plate followed by a polarizing beam splitter. The outputs were coupled into single-mode fibers and detected with single-photon avalanche photodiodes. Interference filters centered near 710 nm with 10 nm bandwidth selected the degenerate photons, and coincidences were recorded within a 2 ns window. For each analyzer pair, the four coincidence counts were , , and . The correlation was reconstructed as
Closed wave-plate sequences were then introduced in Bob’s arm. Each sequence returned the analyzer to the same final measurement basis while tracing a path on the Poincaré sphere. Operational equivalence was established at the level of the final measurement effect before the entangled correlations were compared. To evaluate Eq. (16), the correlations , , and were measured with the direct analyzer, while was measured with the looped realization of the same nominal setting .
Results
Figure 3 compares the direct implementation with four closed trajectories in Bob’s analyzer, labeled by sr, sr, sr and sr. Here labels the loop used in the wave-plate sequence; any identification with the hidden rotation is a model assumption. In the illustrative theoretical curves we set , whereas the experiment tests whether the measured correlation retains any dependence on the loop at all.
The theoretical curves in fig:3b show the strong loop-dependent deformation expected if : changing the loop changes both the maximum and the position of . The measurements show a markedly different behavior. All five curves are in line with the predictions of quantum mechanics. The direct curve, , is a standard CHSH test and violates the Bell-local bound. The looped quantities compare distinct complete implementations and should not be read as independent standard CHSH tests; their purpose is to test whether changing the implementation of alters the observed correlation. No loop ordering of the kind predicted by the illustrative GHV response is observed.
The lower panel makes the null comparison directly. The signed residuals fluctuate about zero without the structured dependence on predicted by (18). Excursions in the displayed ratio become less informative near settings for which is small, because division by the reference amplifies both statistical noise and small calibration offsets. A quantitative constraint should therefore be obtained from the unnormalized signed difference, or directly from the four correlations, with visibility, offset and angular calibration treated as nuisance parameters. Such a fit determines each and, once the enclosed areas are specified, the coupling . No loop-dependent phase is resolved in the present data.
Discussion and conclusion
The local tangent-frame construction and the experimentally tested joint response play different roles. The former shows how a hidden variable can retain geometric path memory while leaving the local marginal unchanged. It also defines the coupling-dependent mismatch probability . Because the construction remains Bell factorisable, however, it cannot reproduce the reference Bell violation. (9) is consequently a resource bound on a mixed-context counterfactual expression, not a derivation of the quantum value.
The operational model in Eqs. (12) and (13) begins instead from the observed non-factorisable correlation and asks whether a hidden holonomy changes the looped term in the mixed-context witness. The data address this second question. They constrain and, once the geometric areas are specified, the coupling . They do not determine without an additional coupling that consistently embeds the non-factorisable response. In particular, a null does not imply that all counterfactual hidden outcomes are unchanged.
A complete GHV account of the quantum Bell correlation would therefore require nonlocal or otherwise non-Bell-factorisable structure. One possible direction is to supplement the local fiber variables with a relational variable connecting Alice’s and Bob’s frames. Such a theory would explicitly give up Bell locality while retaining operational no-signaling after averaging. The present experiment does not test an arbitrary relational construction: only Bob’s implementation path is varied. A two-sided test, in which Alice’s and Bob’s trajectories are changed independently, would be required to isolate contributions depending on their joint geometry.
The experiment addresses a question distinct from Bell locality: whether a measurement is fully specified by its final quantum effect or whether its implementation history leaves an additional trace in entangled correlations. Within the present precision, calibrated paths leading to the same analyzer give the same mixed-context witness. The result agrees with quantum mechanics and places direct constraints on path-sensitive hidden-variable models that predict an observable memory of the optical trajectory.
Acknowledgments: The authors acknowledge fruitful discussions with Justin Dressel, Eliahu Cohen, Lukas Burns and Marko Cosić. A large language model is used for language editing and proofreading. The authors take full responsibility for the scientific content and conclusions of the manuscript.
END MATTER
We consider polarization-entangled photon pairs initially prepared in the Bell state . In Bob’s arm, two quarter-wave plates (QWPs) followed by a half-wave plate (HWP) are introduced, with optical axes oriented at , , and , respectively. The resulting Jones matrix is
This transformation returns horizontal and vertical polarizations to themselves while introducing a relative phase . Consequently, the initial Bell state becomes, up to a global phase,
For linear polarization measurements at angles and , the correlation function is
where .
Using the experimental settings , , , and , the mixed-context Bell parameter, with the waveplates applied only to Bob’s fourth measurement, becomes
For the direct measurement, . The difference is therefore
References
- [1]A. Einstein, B. Podolsky, and N. Rosen, Can quantum-mechanical description of physical reality be considered complete?, Physical Review 47, 777 (1935).DOI
- [2]E. Schrödinger, Discussion of probability relations between separated systems, Mathematical Proceedings of the Cambridge Philosophical Society 31, 555 (1935).DOI
- [3]J. S. Bell, On the Einstein Podolsky Rosen paradox, Physics 1, 195 (1964).
- [4]J. F. Clauser, M. A. Horne, A. Shimony, and R. A. Holt, Proposed experiment to test local hidden-variable theories, Physical Review Letters 23, 880 (1969).
- [5]B. Hensen, H. Bernien, A. E. Dräüta, A. Reiserer, N. Kalb, M. S. Blok, J. Ruitenberg, R. F. L. Vermeulen, R. N. Schouten, C. Abellán, et al., Loophole-free bell inequality violation using electron spins separated by 1.3 kilometres, Nature 526, 682 (2015).
- [6]M. Giustina, M. A. M. Versteegh, S. Wengrowsky, J. Handsteiner, A. Hochrainer, K. Phelan, F. Steinlechner, J. Kofler, J.-Å. Larsson, C. Abellán, et al., Significant-loophole-free test of bell’s theorem with entangled photons, Physical Review Letters 115, 250401 (2015).
- [7]L. K. Shalm, E. Meyer-Scott, B. G. Christensen, P. Bierhorst, M. A. Wayne, M. J. Stevens, T. Gerrits, S. Glancy, D. R. Hamel, M. S. Allman, et al., Strong loophole-free test of local realism, Physical Review Letters 115, 250402 (2015).DOI
- [8]C. N. Yang and R. L. Mills, Conservation of isotopic spin and isotopic gauge invariance, Physical Review 96, 191 (1954).DOI
- [9]M. V. Berry, Quantum phase factors accompanying adiabatic changes, Proceedings of the Royal Society A 392, 45 (1984).
- [10]M. Nakahara, Geometry, Topology and Physics, 2nd ed. (Taylor & Francis, 2003).
