⚛️ quantum physics

Privacy in continuous-variable distributed quantum sensing

This paper introduces a continuous-variable distributed quantum sensing protocol that enables a network of three or more parties to estimate a global average phase with Heisenberg-limited precision while keeping individual local phases private, a feat shown to be uniquely achievable for complete privacy only in the two-party Gaussian case due to fundamental no-go theorems for finite-energy probes.

Original authors: A. de Oliveira Junior, Anton L. Andersen, Benjamin Lundgren Larsen, Sean William Moore, Damian Markham, Masahiro Takeoka, Jonatan Bohr Brask, Ulrik L. Andersen

Published 2026-09-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: A. de Oliveira Junior, Anton L. Andersen, Benjamin Lundgren Larsen, Sean William Moore, Damian Markham, Masahiro Takeoka, Jonatan Bohr Brask, Ulrik L. Andersen

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the emerging world of quantum technology, scientists are learning to use the strange rules of the subatomic realm to build sensors of unprecedented sensitivity. These devices do not just measure temperature or magnetic fields; they can detect the faintest shifts in time, distance, or force by tracking how quantum particles change as they interact with their environment. A particularly exciting frontier is the idea of a distributed network, where many separate sensors, perhaps located in different cities or even on different continents, work together as a single, giant instrument. By sharing a special kind of quantum connection known as entanglement, these distant nodes can combine their data to measure a global property, such as the average time across a network of atomic clocks, with a precision that no single sensor could ever achieve alone. However, a critical challenge arises in such a network: how can the group learn the average value without any single participant learning the private, local values of their neighbors? If the sensors are to be useful for tasks like synchronizing clocks or mapping magnetic fields without revealing sensitive local data, the system must be designed so that the global answer is clear, but the individual contributions remain hidden.

A team of researchers has now tackled this problem, asking whether it is possible to build a distributed quantum sensor network that protects the privacy of every local measurement while still delivering a highly precise global result. They focused on a specific type of quantum system that uses light, known as continuous-variable sensing, which is attractive because it can be operated at room temperature and scaled up to large networks. The researchers set up a theoretical framework to test if a network of quantum sensors could estimate an average phase—a measure of how much the light waves have shifted—while ensuring that no single node could figure out the specific phase shift that occurred at any other node. Their investigation led to a surprising and definitive conclusion about the limits of privacy in these systems.

The study reveals a fundamental trade-off that depends entirely on the size of the network. For a network involving only two parties, the researchers found that it is possible to achieve perfect privacy. By using a specific type of entangled light known as a two-mode squeezed state, the two nodes can measure their combined average phase with extreme precision, while the difference between their individual phases remains completely invisible to both of them. In this two-person scenario, the quantum connection is so strong that it effectively hides the individual contributions, allowing the group to know the average without knowing the parts. However, the researchers proved that this perfect privacy is impossible to maintain once the network grows to include three or more parties. They demonstrated mathematically that for any network of three or more nodes using finite-energy light, it is impossible to hide every possible combination of local phases while still measuring the average. In larger networks, while the system can still hide the individual values of each node, it cannot hide all the other ways the local phases might combine.

To address this limitation for larger networks, the team designed a practical protocol that offers the next best thing: weak privacy. In this scheme, a special entangled state is distributed across many nodes using a network of beam splitters, which are optical devices that split and combine light beams. The researchers showed that in this setup, every individual local phase remains private; no single node can deduce the specific phase shift of any other node. However, the system does leak some information about other combinations of the phases, just not the individual ones. The protocol allows the network to estimate the average phase with a precision that improves dramatically as the total amount of light increases, a level of performance known as Heisenberg scaling. This is significantly better than what classical sensors can achieve. The study also examined how real-world imperfections, such as signal loss in optical fibers or unintended shifts in the light's brightness, affect this privacy. They found that while signal loss reduces the overall precision, it does not break the privacy of the individual phases. In contrast, adding extra brightness to the light, known as displacement, improves the precision but destroys the privacy, making it possible to infer the individual values.

The researchers compared their proposed method with other potential quantum resources, such as different types of entangled light patterns, and found that their approach was superior for maintaining privacy. They concluded that while perfect, total privacy is mathematically forbidden for networks of three or more nodes, the proposed protocol offers a robust and experimentally feasible way to protect individual data while still gaining the massive precision benefits of quantum networking. This work provides a clear roadmap for building future quantum sensor networks that are both powerful and secure, defining exactly what is possible and what is fundamentally impossible in the quest for private, distributed quantum sensing.

Technical Summary: Privacy in Continuous-Variable Distributed Quantum Sensing

Problem Statement
The paper addresses the challenge of privacy in distributed quantum sensor networks, specifically within the continuous-variable (CV) regime. The core problem is to estimate a global parameter (the average of local phases imprinted at $M$ nodes) while ensuring that no individual node, colluding subset of nodes, or external observer can reconstruct the specific local phase values ( $\theta_\mu$ ) or non-target linear combinations of these phases. While discrete-variable networks have established protocols for private parameter estimation, the feasibility and limits of achieving "complete privacy" (hiding all phase combinations orthogonal to the target) using finite-energy Gaussian states in CV networks remain unexplored.

Methodology and Framework
The authors develop an information-theoretic framework based on the Quantum Fisher Information Matrix (QFIm), denoted as $Q$ .

Privacy Definition: Privacy is characterized by the kernel (null space) of the QFIm. A parameter direction $u$ is unobservable (and thus private) if $u \in \ker Q$ . A local phase $\theta_j$ is considered private if there exists an unobservable direction with a non-zero component in the $j$ -th entry.
Privacy Measure: The paper utilizes a normalized privacy measure $P(Q, u) = \frac{u^T Q u}{\text{tr}(Q)}$ , where $u$ is the target direction (the average phase). $P=1$ indicates complete privacy (the QFIm is rank-one and aligned solely with the target), while $P < 1$ indicates leakage of information into orthogonal directions.
Gaussian Formalism: The analysis is restricted to pure Gaussian states and unitary local phase encodings ( $\hat{U}(\theta) = \bigotimes e^{i\theta_\mu \hat{n}_\mu}$ ). The authors derive a closed-form expression for the QFIm of pure Gaussian states under local phase shifts (Lemma 2) and a specific formula for the privacy measure (Theorem 3).

Key Contributions and Results

No-Go Theorem for Complete Privacy ( $M \ge 3$ ):
The central theoretical result (Theorem 4) proves that for any network with three or more parties ( $M \ge 3$ ), it is impossible to achieve complete privacy of the average phase using finite-energy Gaussian probes. Specifically, if the QFIm has a range spanned only by the average-phase vector, the matrix must be zero (implying no sensitivity). Consequently, for $M \ge 3$ , there will always be some non-target phase combinations that carry finite Fisher information, preventing the hiding of all orthogonal directions.
The Exceptional Two-Party Case:
The paper identifies the two-party case ( $M=2$ ) as an exception where complete privacy is achievable. A two-mode squeezed state (TMSS) subjected to local phase shifts yields a rank-one QFIm. The relative phase direction lies entirely in the kernel, making both individual phases private while allowing the average phase to be estimated with Heisenberg scaling. The authors analyze the impact of optical loss and displacement on this scenario, showing that while loss degrades precision, it preserves the zero eigenvalue (privacy), whereas displacement breaks the symmetry, introducing information leakage and reducing the privacy measure.
Multipartite Protocol with Weak Privacy:
For $M \ge 3$ , the authors propose an experimentally feasible protocol based on distributing a TMSS through a recursive beam-splitter network.
- Weak Privacy: While complete privacy is impossible, this protocol achieves "weak component privacy." Every individual local phase $\theta_j$ remains private because the kernel of the QFIm contains a vector with non-zero entries for all components.
- Spectrum Analysis: The QFIm spectrum consists of one zero eigenvalue (corresponding to a specific "branch-imbalance" symmetry), an $(M-2)$ -dimensional degenerate subspace with shot-noise scaling, and a single dominant eigenvalue with Heisenberg scaling.
- Asymptotic Behavior: As the squeezing parameter $r$ increases, the ratio of the dominant eigenvalue to the sub-dominant ones grows exponentially ( $\sim e^{2r}$ ). While the sub-dominant eigenvalues do not vanish, the fraction of total Fisher information leaking into non-target directions approaches zero, asymptotically approaching the rank-one limit.
Benchmarking Against Other Resources:
The proposed TMSS-tree protocol is benchmarked against other Gaussian resources, including linear cluster states (generated via CZ operations) and independent single-mode squeezed states.
- Cluster States: These perform poorly, with the privacy measure converging to $1/M$ (the separable baseline) regardless of coupling strength, indicating a fundamental incompatibility between their correlations and phase encoding.
- Independent Squeezers: These distribute information uniformly across all phase combinations, offering no privacy.
- Conclusion: The TMSS-tree protocol is shown to be near-optimal among the tested Gaussian resources for balancing estimation precision and privacy.

Significance and Claims
The paper claims to resolve the open question of whether Gaussian states can achieve complete privacy in distributed sensing. It establishes a fundamental limitation: complete privacy is strictly impossible for $M \ge 3$ under finite energy, identifying the two-party TMSS as the unique Gaussian setting for complete privacy.

The work provides a rigorous operational definition of privacy based on the QFIm kernel, distinguishing between "complete privacy" (hiding all orthogonal combinations) and "component privacy" (hiding individual values). The authors emphasize that their privacy guarantees rely on specific assumptions: a trusted source, no side information about local parameters (nodes cannot query their own channels with auxiliary probes), and adherence to the prescribed protocol.

The paper concludes that while the proposed protocol cannot hide all orthogonal phase combinations for $M \ge 3$ , it successfully ensures that no individual node can reconstruct its own or others' local phases without additional side information. The results highlight a trade-off where increasing squeezing improves the concentration of information on the target, asymptotically approaching the ideal privacy limit, even though a strict zero-eigenvalue for all non-target directions is structurally forbidden for $M \ge 3$ .

Technical Summary: Privacy in Continuous-Variable Distributed Quantum Sensing

More like this