1 Introduction↩︎

Bell’s theorem [1] showed that no local hidden-variable theory can reproduce all quantum predictions. Subsequent experiments [2], [3] confirmed violations of Bell inequalities, implying that nature is either nonlocal or indeterministic (or both). However, nonlocal deterministic theories (e.g., the de Broglie–Bohm interpretation [4]) remain compatible with Bell experiments, at the price of sacrificing locality.

A different line of attack on hidden variables is provided by the Kochen–Specker theorem [5], which shows that no non-contextual hidden-variable theory (NCHV) can reproduce quantum mechanics. Unlike Bell’s theorem, Kochen–Specker requires no locality assumption, but it does require a set of 117 rays (or 33 vectors in modern versions) and is state-independent. More recently, the KCBS inequality [6] provided a state-dependent test of non-contextuality using a single qutrit and 5 measurements.

In this work, we propose a different test that is also state-dependent and uses only a single qutrit, but requires only two sequential measurements. The key idea is an analogy with the Monty Hall problem [7], [8], reformulated in quantum language. Our protocol yields a simple equality (\(1/3\) vs \(1/6\)) that is violated by quantum mechanics, providing an operational demonstration of the failure of non-contextual determinism that is accessible to a broad audience.

2 Ontological models framework↩︎

We adopt the standard framework of ontological models [9], [10]. Let \(\Lambda\) be the space of ontic states (hidden variables). A preparation procedure \(P\) (here, preparing \(|\psi_0\rangle = \frac{1}{\sqrt{3}}(|A\rangle+|B\rangle+|C\rangle)\)) is represented by a probability distribution \(\rho(\lambda | \psi_0)\) over \(\Lambda\). A measurement \(M\) with outcomes \(k\) is represented by a set of response functions \(\xi(k | \lambda, M)\) satisfying \(\sum_k \xi(k | \lambda, M) = 1\) for all \(\lambda\).

In this work, we consider two types of measurements:

• A projective measurement in the \(\{|A\rangle,|B\rangle,|C\rangle\}\) basis. For a deterministic non-contextual model, there exists a function \(v(\lambda) \in \{A,B,C\}\) giving the outcome.

• A POVM \(\{F_1, F_2\}\) defined below. For outcome \(F_1\), the ontological model includes a stochastic update map \(\Gamma(\lambda' | \lambda, F_1)\) describing how the ontic state changes.

For the initial state \(|\psi_0\rangle\), quantum mechanics predicts: \[\int \rho(\lambda | \psi_0) \, \delta_{v(\lambda), A} \, d\lambda = \frac{1}{3},\] and similarly for \(B,C\). This fixes the initial distribution under the assumption that the model reproduces quantum statistics.

3 Quantum protocol↩︎

Let \(\mathcal{H}_3\) be a Hilbert space with orthonormal basis \(\{\ket{A},\ket{B},\ket{C}\}\). Prepare the initial state: \[\ket{\psi_0} = \frac{1}{\sqrt{3}}\bigl(\ket{A}+\ket{B}+\ket{C}\bigr).\] We choose \(\ket{A}\) as the “bet” (analogous to the initially chosen door).

3.1 Step 1: Coherent discard (generalized measurement)↩︎

We implement a two-outcome POVM with elements \(\{F_1, F_2\}\) defined by: \[F_1 = \frac{1}{2}\bigl(\ket{A}\bra{A} + \ket{C}\bra{C}\bigr),\qquad F_2 = I - F_1.\]

The probability of obtaining outcome \(F_1\) is: \[\bra{\psi_0}F_1\ket{\psi_0} = \frac{1}{3}.\]

To physically realize this POVM, we use a Naimark dilation: introduce an ancilla qubit initially in \(\ket{0}\), apply a unitary \(U\) on \(\mathcal{H}_3 \otimes \mathcal{H}_2\), and then projectively measure the ancilla in the computational basis \(\{\ket{0},\ket{1}\}\). The conditional state of the qutrit after measuring the ancilla in \(\ket{1}\) is proportional to \(M_1\ket{\psi_0}\), where \(M_1\) is a Kraus operator satisfying \(F_1 = M_1^\dagger M_1\).

We choose the Kraus operator: \[M_1 = \frac{1}{\sqrt{2}}\bigl(\ket{A}\bra{A} + \ket{C}\bra{C}\bigr).\]

In the ordered computational basis \(\{\ket{A},\ket{B},\ket{C}\}\), this is represented by the diagonal matrix: \[M_1 = \frac{1}{\sqrt{2}}\begin{pmatrix} 1 & 0 & 0 \\ 0 & 0 & 0 \\ 0 & 0 & 1 \end{pmatrix}.\]

The standard minimal dilation (see [11], Section 2.2.8) gives a unitary matrix of the form: \[U = \begin{pmatrix} \sqrt{I - M_1^\dagger M_1} & -M_1^\dagger \\ M_1 & \sqrt{I - M_1 M_1^\dagger} \end{pmatrix}.\]

Since \(M_1\) is self-adjoint and diagonal, we compute: \[I - M_1^\dagger M_1 = I - F_1 = \begin{pmatrix} 1/2 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 1/2 \end{pmatrix},\] \[\sqrt{I - M_1^\dagger M_1} = \begin{pmatrix} 1/\sqrt{2} & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 1/\sqrt{2} \end{pmatrix}.\]

Using the basis ordering \(\{\ket{A,0}, \ket{B,0}, \ket{C,0}, \ket{A,1}, \ket{B,1}, \ket{C,1}\}\), the explicit \(6 \times 6\) unitary matrix is: \[U = \begin{pmatrix} 1/\sqrt{2} & 0 & 0 & -1/\sqrt{2} & 0 & 0 \\ 0 & 1 & 0 & 0 & 0 & 0 \\ 0 & 0 & 1/\sqrt{2} & 0 & 0 & -1/\sqrt{2} \\ 1/\sqrt{2} & 0 & 0 & 1/\sqrt{2} & 0 & 0 \\ 0 & 0 & 0 & 0 & 1 & 0 \\ 0 & 0 & 1/\sqrt{2} & 0 & 0 & 1/\sqrt{2} \end{pmatrix}.\]

This matrix is real, symmetric, and orthogonal. It consists of \(2\times2\) Hadamard-like blocks for the \(\{A,C\}\) subspace and identity blocks for \(\{B\}\). A direct calculation verifies \(U^T U = I_6\).

Verification of the dilation↩︎

Apply \(U\) to the initial state \(\ket{\psi_0}\otimes\ket{0} = \frac{1}{\sqrt{3}}(\ket{A}+\ket{B}+\ket{C})\otimes\ket{0}\): \[\begin{align} &U\bigl(\ket{\psi_0}\otimes\ket{0}\bigr) = \frac{1}{\sqrt{3}}\begin{pmatrix} 1/\sqrt{2} \\ 1 \\ 1/\sqrt{2} \\ 1/\sqrt{2} \\ 0 \\ 1/\sqrt{2} \end{pmatrix} =\\ & \frac{1}{\sqrt{3}}\left(\ket{B,0} + \frac{1}{\sqrt{2}}\left[\ket{A,0}+\ket{C,0} + \ket{A,1} + \ket{C,1}\right] \right). \end{align}\]

When the ancilla is measured in the computational basis:

• Outcome \(\ket{1}\) occurs with probability \(\left\|\frac{1}{\sqrt{3}}(\frac{1}{\sqrt{2}}\ket{A,1}+\frac{1}{\sqrt{2}}\ket{C,1})\right\|^2 = \frac{1}{3}\). The conditional qutrit state is \(\frac{1}{\sqrt{2}}(\ket{A}+\ket{C}) = \ket{\psi_1}\).

• Outcome \(\ket{0}\) occurs with probability \(\frac{2}{3}\), and the conditional state is orthogonal to \(\ket{A}\).

Formally, the operator \(\bra{1}U\ket{0}\) (the submatrix coupling ancilla \(0\) to ancilla \(1\)) equals \(M_1\): \[\bra{1}U\ket{0} = \begin{pmatrix} 1/\sqrt{2} & 0 & 0 \\ 0 & 0 & 0 \\ 0 & 0 & 1/\sqrt{2} \end{pmatrix} = M_1,\] and \(M_1^\dagger M_1 = F_1\). This confirms that the dilation implements the desired POVM.

3.2 Step 2: Projective measurement↩︎

Immediately after the discard, we perform a projective measurement of the projector \(P_A = \ket{A}\bra{A}\).

The joint probability of obtaining \(F_1\) in step 1 and then \(\ket{A}\) in step 2 is: \[P_{\text{QM}}(F_1, A) = \bra{\psi_0}F_1\ket{\psi_0} \times \bigl|\langle A | \psi_1 \rangle\bigr|^2= \frac{1}{6}.\]

If outcome \(F_2\) occurs, the probability of obtaining \(\ket{A}\) in step 2 is zero because the post-measurement state is orthogonal to \(\ket{A}\).

Therefore, the total probability of obtaining \(\ket{A}\) in the second measurement (averaged over both branches) is: \[Q_{\text{QM}} \equiv P_{\text{QM}}(A) = \frac{1}{6}.\]

4 Deterministic non-contextual prediction↩︎

Consider a deterministic non-contextual ontological model. The system has an ontic state \(\lambda\) with initial distribution \(\rho(\lambda | \psi_0)\) satisfying: \[\int \rho(\lambda | \psi_0) \, \delta_{v(\lambda), A} \, d\lambda = \frac{1}{3},\] and similarly for \(B,C\). For simplicity, we take the minimal model where \(\Lambda = \{A,B,C\}\) and \(\rho(\lambda | \psi_0) = 1/3\) for each, but the argument generalizes to any \(\Lambda\) with a function \(v:\Lambda \to \{A,B,C\}\).

The discard procedure implements the POVM \(\{F_1, F_2\}\). In the ontological model, this corresponds to a stochastic update map \(\Gamma(\lambda' | \lambda, F_1)\) for the ontic state when outcome \(F_1\) occurs. The Monty Hall condition demands that this update never changes the value of \(v(\lambda)\) when outcome \(F_1\) is obtained: \[\Gamma(\lambda' | \lambda, F_1) > 0 \quad \Rightarrow \quad v(\lambda') = v(\lambda).\] That is, the discard never eliminates the true state.

Given this condition, the probability that after the discard the system has \(v = A\) is simply: \[P_{\text{det}}(v=A \text{ after discard}) = \int \rho(\lambda | \psi_0) \, \delta_{v(\lambda), A} \, d\lambda = \frac{1}{3}.\]

The subsequent projective measurement of \(P_A\) merely reveals this pre-existing value. Hence: \[Q_{\text{det}} \equiv P_{\text{det}}(A \text{ in second measurement}) = \frac{1}{3}.\]

5 Equality and quantum violation↩︎

Define \(Q\) as the probability of obtaining \(\ket{A}\) in the second measurement after the coherent discard.

Therefore, if an experiment measures \(Q = 1/6 \pm \epsilon\) with \(\epsilon < 1/12\), the deterministic non-contextual prediction is ruled out. More generally, any measurement of \(Q\) significantly below \(1/3\) (e.g., \(Q < 0.3\) with high confidence) contradicts the NCHV bound.

6 Comparison with other no-go theorems↩︎

• Bell theorem [1]: Assumes locality and two separated systems. Our test uses a single system and makes no locality assumption.

• Kochen–Specker theorem [5]: Proves that no non-contextual hidden-variable theory can reproduce quantum mechanics, but requires a set of 117 rays (or 33 vectors in modern versions) and is state-independent. Our test is state-dependent (only works for \(|\psi_0\rangle\)) and requires only two sequential measurements on a single qutrit, providing a simpler operational demonstration of non-contextuality violation.

• KCBS inequality [6]: Also tests non-contextuality with a single qutrit, but requires 5 measurements with a cyclic condition. Our test uses only 2 measurements and is structurally different, being inspired by the Monty Hall problem.

• Leggett–Garg inequalities [12]: Assume macrorealism (macroscopic distinctness and non-invasive measurability). Our test uses microscopic qutrits and allows invasive measurements.

• De Broglie–Bohm theory [4]: This theory is deterministic but contextual (the outcome of a measurement depends on the full experimental arrangement, including the quantum potential). Our theorem only rules out non-contextual deterministic hidden-variable theories. Bohmian mechanics is contextual and therefore not ruled out by our test. This is a feature, not a bug: our test clarifies that the Monty Hall protocol detects contextuality, not determinism per se.

Thus, our main contribution is a new state-dependent non-contextuality inequality with an operational interpretation rooted in classical probability puzzles, making it accessible to a broader audience.

7 Experimental proposal and loopholes↩︎

We propose an implementation with photonic qutrits using path and polarization degrees of freedom [13], [14]:

1. Prepare a symmetric superposition of three spatial modes using a tritter (tricotomizer) [13].

2. Implement the coherent discard \(F_1\) via a partial projection: a beamsplitter that couples mode \(A\) and \(C\) with reflectivity \(1/2\), followed by a postselection on the reflected path. This realizes the POVM element \(F_1\) and produces the conditional state \(\frac{1}{\sqrt2}(\ket{A}+\ket{C})\).

3. Perform the second measurement: a projective filter selecting \(\ket{A}\) (e.g., a polarizer plus single-mode fiber).

4. Count coincidences to estimate \(Q\).

7.1 Loopholes and experimental imperfections↩︎

The main loophole for this test is the detection loophole: if the efficiency of detectors is low, one might postselect only on events where both measurements yielded a result, biasing the statistics. However, since our protocol uses only a single qutrit and no entanglement, high-efficiency detectors (\(>90\%\)) are available for photons, and trapped ions have near-unit detection efficiency. Thus, the detection loophole can be closed with current technology.

The coincidence loophole (or time loophole) is irrelevant for a single system.

7.2 Tolerance to noise↩︎

Let the initial state be a mixed state \(\rho = (1-\epsilon)|\psi_0\rangle\langle\psi_0| + \epsilon I/3\) (white noise). Then the probability \(Q\) becomes: \[Q_{\epsilon} = (1-\epsilon)\frac{1}{6} + \epsilon \cdot \frac{1}{3} = \frac{1}{6} + \epsilon \cdot \frac{1}{6} = \frac{1}{6}(1+\epsilon).\] The deterministic bound \(1/3\) is violated as long as \(\epsilon < 1\). Even with 50% noise (\(\epsilon=0.5\)), \(Q = 0.25\), still below \(0.333\). The test is remarkably robust.

8 Conclusion↩︎

We have derived a Monty-Hall-type equality that distinguishes non-contextual deterministic hidden-variable theories from quantum mechanics using a single qutrit. The quantum prediction \(Q=1/6\) violates the NCHV bound \(Q=1/3\), providing a new state-dependent non-contextuality inequality. Unlike Kochen–Specker or KCBS, our test requires only two sequential measurements and has a clear operational interpretation rooted in the classical Monty Hall puzzle. An explicit \(6 \times 6\) unitary matrix for the Naimark dilation of the required POVM is provided, demonstrating that the protocol is physically realizable. An experimental test is feasible with present-day photonic or trapped-ion qutrits.

Data Availability↩︎

Data sharing is not applicable to this article as no new data were created or analyzed in this study.

Declarations↩︎

The author declares no competing interests.

Acknowledgments↩︎

Jorge Meza-Domínguez thanks SECIHTI-México for the doctoral scholarship No. 1235731. This work was also partially supported by SECIHTI México under grants SECIHTI CBF-2025-G-1720 and CBF-2025-G-176. The author is grateful for the computing time granted by LANCAD and CONACYT in the Supercomputer Hybrid Cluster “Xiuhcoatl” at GENERAL COORDINATION OF INFORMATION AND COMMUNICATIONS TECHNOLOGIES (CGSTIC) of CINVESTAV. URL: http://clusterhibrido.cinvestav.mx/ and to Hector Oliver Hernandez for his help with the code installations.

References↩︎