Faking entanglement with imperceptible measurement deviations

Technical Summary: Faking Entanglement with Imperceptible Measurement Deviations

Problem Statement
Quantum entanglement is a fundamental resource for emerging technologies, including quantum communication, computation, and networks. Accurate verification of entanglement, particularly in high-dimensional systems, is critical. However, standard verification methods, such as entanglement witnesses, rely on the assumption that prescribed quantum observables (measurements) can be implemented with perfect accuracy. In reality, as the dimensionality of a quantum system increases, the complexity of implementing ideal measurements grows, leading to inevitable deviations between theoretical models and experimental implementations. While device-independent tests exist, they are experimentally demanding. Current device-dependent methods often assume that measurement imperfections are small and stochastic, but they lack rigorous analysis regarding whether systematic, adversarial deviations—even if imperceptibly small—could compromise the validity of entanglement certification in high-dimensional regimes.

Methodology
The authors investigate the vulnerability of high-dimensional entanglement verification to adversarial measurement deviations. They focus on a widely used entanglement witness based on correlations measured in two complementary bases: the computational basis and its Fourier transform (mutually unbiased bases).

1. Theoretical Attack Construction: The authors construct explicit measurement bases that are $\epsilon$-close to the target bases (where $\epsilon$ represents the measurement infidelity) but are specifically tailored to maximize the witness value for separable (product) states. They analyze three distinct error models to quantify measurement fidelity:

   • Average Fidelity Model: The average overlap between target and implemented measurement outcomes is bounded by $1-\epsilon$.
   • Worst-Case Fidelity Model: Every individual measurement outcome satisfies the fidelity bound $1-\epsilon$.
   • Worst-Case with Individual Projections: A model relevant to photonics where multi-outcome measurements are implemented as a sequence of single projections, allowing independent deviations for each projector.

   For each model, they derive optimal attack strategies using classical product states. They demonstrate that by introducing specific, small perturbations to the measurement basis vectors, a separable state can yield a witness value ($W_d$) that exceeds the theoretical bound for separable states, falsely indicating the presence of high-dimensional entanglement.

2. Experimental Demonstration: The authors experimentally validate these theoretical predictions using classical photonic states encoded in the transverse spatial degree of freedom (macro-pixel basis).

   • Setup: An attenuated pulsed laser serves as a source of uncorrelated photon pairs (simulating a product state). Spatial Light Modulators (SLMs) are used to generate specific product states and to implement the perturbed measurement bases.
   • Implementation: The experiment spans dimensions up to $d=61$. The authors introduce controlled $\epsilon$-perturbations into the measurement apparatus. They account for intrinsic experimental imperfections, specifically crosstalk between modes, by characterizing the crosstalk matrix and incorporating it into the theoretical bounds.
   • Verification: They measure the entanglement witness $W_d$ for the product state under these perturbed conditions and compare the results against theoretical predictions that include experimental crosstalk.

Key Results

• Theoretical Findings: The study proves that arbitrarily small measurement errors, when adversarially encoded, can lead to the false certification of arbitrarily high levels of entanglement in separable states. The magnitude of the "faked" entanglement (quantified by the Schmidt number $K$) increases sharply with the system dimension $d$ for a fixed error $\epsilon$. For instance, in a 61-dimensional system, an error of just $\epsilon = 0.7\%$ is sufficient to fake entanglement in $K=47$ dimensions.
• Experimental Validation: The experiments successfully demonstrated the faking of substantial entanglement in product states.
  • In a 31-dimensional system with $\epsilon = 0.7\%$, the measured witness value was $W_d = 1.36 \pm 0.03$, corresponding to a falsely certified Schmidt number of $K=11$.
  • In a 61-dimensional system with the same error rate ($\epsilon = 0.7\%$), the experiment successfully faked a Schmidt number of $K=26$.
  • The experimental data, when corrected for intrinsic crosstalk (measured at a ratio of approximately 1:700 to 1:760), aligns well with the theoretical predictions of the attack models.
• Robustness of the Vulnerability: The vulnerability persists across all three error models (average fidelity, worst-case fidelity, and individual projections). The results indicate that the flaw is intrinsic to the conventional paradigm of device-dependent entanglement witnessing, rather than a specific artifact of a particular experimental implementation or fidelity quantifier.

Significance and Claims
The paper claims to uncover a fundamental vulnerability in current methods for high-dimensional entanglement detection. The authors argue that as quantum systems become more complex (higher dimensionality), the experimental challenge of realizing ideal measurements increases, which paradoxically simplifies the execution of adversarial attacks. Small, carefully tailored deviations in measurement devices can simulate the statistical signatures of maximal entanglement using entirely separable, classical states.

The significance of these findings is twofold:

1. Security Risk: It highlights the susceptibility of complex quantum devices to small adversarial perturbations, suggesting that current verification protocols may be insufficient for securing quantum technologies against sophisticated attacks.
2. Methodological Necessity: The results underscore the urgent need for developing secure verification methods for quantum information that are robust to bounded discrepancies between theory and experiment. The authors suggest that future verification schemes must operate under general bounded error assumptions, potentially moving towards device-independent or semi-device-independent frameworks that do not rely on the assumption of perfect measurement implementation.

The paper concludes that while the specific level of faked entanglement depends on how errors are quantified, the qualitative behavior—that small errors can generate large false positives in high dimensions—is a universal feature of the current entanglement witnessing paradigm.