⚛️ quantum physics

Weak Typicality of von Neumann Entanglement Entropy in Gaussian Boson Sampling

This paper proves that the von Neumann entanglement entropy in Gaussian Boson Sampling with Haar-distributed passive interferometers exhibits proportional weak typicality and almost sure convergence to its mean, establishing a volume law and providing explicit variance bounds through a novel proof that regularizes logarithmic singularities and applies concentration inequalities on the unitary group.

Original authors: Hongru Zhao

Published 2026-08-19

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Hongru Zhao

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 quiet, invisible realm of quantum physics, light behaves in ways that defy our everyday intuition. Imagine a beam of light not as a steady stream, but as a collection of individual particles called photons, which can be squeezed and stretched into strange, correlated states. Scientists have long been fascinated by how these particles share information when they are mixed together in complex optical networks. This sharing is known as entanglement, a phenomenon where the state of one particle becomes inextricably linked to another, no matter how far apart they are. To measure how deeply two groups of these particles are connected, physicists use a value called entropy. Think of this not as disorder, but as a precise count of how much information is hidden in the connection between the two groups. When a system is perfectly balanced, this value follows a predictable pattern known as the Page curve, a theoretical map that describes how entanglement grows as you divide a system into parts. However, while this map tells us the average behavior of a vast crowd of quantum systems, it leaves a crucial question unanswered: does every single individual system in that crowd follow the map, or do some wander far off course?

This question lies at the heart of a new study by Hongru Zhao, a researcher at the University of Minnesota. The paper investigates a specific type of quantum experiment called Gaussian boson sampling, where light is prepared in a highly squeezed state and then passed through a random optical mixer. In these experiments, the light modes are divided into two groups, and scientists measure the entanglement between them. Previous work had shown that for certain types of measurements, the results were remarkably consistent across different random setups, a property known as typicality. However, for the most fundamental measure of entanglement—the von Neumann entropy—scientists had only been able to prove this consistency for small, shrinking portions of the system. The big question remained: if you take a large, proportional slice of the system, say half the light modes versus the other half, does the entanglement value stay close to the average, or does it fluctuate wildly?

Zhao's work provides a definitive answer: the entanglement is indeed typical. The study proves that for a system with a fixed amount of squeezing, as the number of light modes grows large, the entanglement value for any proportional division of the system will almost certainly be very close to the predicted average. The researchers did not just show that the average is correct; they demonstrated that the chance of finding a system that deviates significantly from this average is vanishingly small. In fact, the probability of such a deviation drops so fast that it becomes practically impossible to find an outlier in a large enough system. This means that the theoretical Page curve, which was previously just a description of the average, is actually a reliable description of what happens in almost every single instance of the experiment.

To reach this conclusion, the team had to navigate a significant mathematical obstacle. The formula used to calculate entanglement has a sharp, singular point where the calculation becomes unstable, much like a cliff edge that makes it difficult to measure the terrain nearby. The researchers developed a clever technique to smooth out this edge, allowing them to apply powerful statistical tools that describe how random matrices behave. By treating the optical mixer as a random unitary matrix and analyzing the singular values of its components, they were able to bound the fluctuations of the entropy. Their proof shows that the variance, or the spread of possible values, grows very slowly—only as the square of the logarithm of the system size. This is a tiny amount of spread compared to the linear growth of the entanglement itself, confirming that the system is tightly clustered around the mean.

The study also clarifies what does not happen. While the entanglement is typical in a relative sense, meaning it stays close to the average as a percentage, it is not typical in a strict, absolute sense for most divisions of the system. If you look at the difference between the actual value and the average, that difference does not vanish for most ratios of the split. It only vanishes when the system is split exactly in half. This distinction is subtle but important: the system is predictable in its scale, but it still retains a small, persistent wiggle room that prevents it from being perfectly rigid. This finding rules out the idea that the entanglement could fluctuate wildly enough to create entirely different macroscopic behaviors, but it also confirms that the system is not so rigid that it eliminates all natural variation.

The implications of this work extend beyond pure theory. Gaussian boson sampling is a leading candidate for demonstrating quantum advantage, where quantum computers solve problems that are impossible for classical machines. For these machines to be useful, their outputs must be reliable and predictable. By proving that the entanglement entropy is typical, this research adds a layer of rigor to our understanding of these quantum systems. It assures us that the complex web of connections formed in these experiments is not a chaotic mess, but a structured, predictable phenomenon that follows the laws of probability with high precision. The work also includes a formal verification using computer-assisted proof software, ensuring that every step of the mathematical logic holds up under the most rigorous scrutiny.

In the end, this paper closes a gap in our understanding of quantum randomness. It confirms that when you mix light in a random optical network, the resulting entanglement is not a roll of the dice where anything can happen. Instead, it is a highly constrained process where the outcome is overwhelmingly likely to be the one predicted by the average. The study transforms the Page curve from a statistical average into a law of nature for these systems, showing that in the vast landscape of quantum possibilities, the typical path is not just common—it is the only path that matters.

Technical Summary: Weak Typicality of von Neumann Entanglement Entropy in Gaussian Boson Sampling

Problem Statement
This paper addresses the statistical behavior of the von Neumann entanglement entropy ( $S_{1,n}$ ) generated in Gaussian Boson Sampling (GBS) scenarios. Specifically, it investigates the "weak typicality" of this entropy for a system of $n$ equally squeezed input modes ( $s \neq 0$ ) passed through a Haar-distributed passive linear optical interferometer.

While previous works had established proportional weak typicality for integer Rényi entropies ( $\alpha \ge 2$ ) and provided asymptotically exact mean curves for the von Neumann entropy, the question of whether the von Neumann entropy itself exhibits proportional weak typicality (i.e., whether $S_{1,n} / \mathbb{E}[S_{1,n}] \to 1$ in probability for subsystems of size $k_n \sim rn$ ) remained an open problem. The difficulty lies in the logarithmic singularity of the entropy function's derivative as the symplectic eigenvalues approach the pure state limit ( $\nu \to 1$ ), which complicates direct variance calculations and concentration bounds.

Methodology
The author develops a proof strategy that avoids direct variance computation by transforming the entropy into a singular value statistic and applying regularization techniques:

Spectral Reduction: The von Neumann entropy is expressed exactly as a sum of a scalar function $\phi_s$ applied to the singular values $\tau_j$ of a principal $k \times k$ block ( $B_U$ ) of the matrix $UU^T$ , where $U$ is the Haar-distributed unitary interferometer. The symplectic eigenvalues $\nu_j$ are related to these singular values via $\nu_j = \sqrt{1 + \sinh^2(2s)(1-\tau_j^2)}$ .
Endpoint Regularization: The function $\phi_s(t)$ has a logarithmic singularity in its derivative at $t=1$ (corresponding to $\nu=1$ ). To handle this, the author introduces a regularized version $\phi_{s,\eta}$ by truncating the domain at $1-\eta$ . This creates a Lipschitz continuous function $F_{n,\eta}$ that approximates the true entropy.
Concentration Inequalities: The proof leverages the concentration of measure on the unitary group $U(n)$ (specifically, the concentration of Haar measure). By bounding the Lipschitz constant of the regularized statistic $F_{n,\eta}$ with respect to the Hilbert-Schmidt distance, the author derives a tail bound for the deviation of $F_{n,\eta}$ from its mean.
Error Control: The deterministic error introduced by the regularization (the difference between $S_{1,n}$ and $F_{n,\eta}$ ) is shown to vanish as $n \to \infty$ when the cutoff parameter is chosen as $\eta_n = n^{-2}$ .
Mean Lower Bound: A linear lower bound for the expected entropy is derived using Rényi entropy of order 2 and properties of the Circular Orthogonal Ensemble (COE), ensuring the normalization ratio is well-defined.

Key Results
The main theorem establishes a quantitative relative concentration bound for the von Neumann entropy. For a subsystem of $k_n$ modes where $k_n/n \to r \in (0,1)$ , and for any $\varepsilon > 0$ :
$P\left( \left| \frac{S_{1,n}}{\mathbb{E}S_{1,n}} - 1 \right| \ge \varepsilon \right) \le 2 \exp\left[ -c_{s,r} \frac{\varepsilon^2 n^2}{\log^2(en)} \right]$
where $c_{s,r} > 0$ is a constant depending on the squeezing strength $s$ and the ratio $r$ .

From this bound, the paper derives several corollaries:

Weak Typicality: $S_{1,n} / \mathbb{E}S_{1,n} \xrightarrow{P} 1$ .
Almost Sure Convergence: The convergence holds almost surely for any coupling of the sequence of unitary matrices with normalized Haar marginals.
Typical Volume Law: The entropy scales linearly with the subsystem size ( $S_{1,n} = \Theta_P(n)$ ), specifically bounded by the mean Page curve value.
Variance Bound: The variance of the entropy is bounded by $\text{Var}(S_{1,n}) = O_s(\log^2 n)$ , which is significantly smaller than the $O(n^2)$ scale of the mean squared.
Logarithmic Tightness: The fluctuations of the entropy around its mean are of order $O_P(\log n)$ .

Additionally, the paper proves that strong typicality (where the additive deviation $S_{1,n} - \mathbb{E}S_{1,n} \to 0$ ) fails for all proportional ratios $r \neq 1/2$ . This is demonstrated using the central limit theorem for linear statistics of the $\beta$ -Jacobi ensemble, showing that the fluctuations are of order $O(1)$ rather than vanishing.

Significance and Claims
The paper claims to close the specific open problem regarding proportional von Neumann weak typicality in the fixed equal squeezing Haar ensemble, a gap left by previous works (Iosue et al., Youm et al.). The significance of the result lies in:

Rigorous Confirmation: It confirms that the published mean Page curve for GBS is not just an average but a "typical" value observed in almost all realizations of the interferometer.
Methodological Contribution: The proof introduces a reusable strategy of regularizing logarithmic singularities in spectral statistics to apply concentration of measure, avoiding the need for exact variance expansions involving hypergeometric series.
Formal Verification: The entire proof chain, including the reduction to singular values, the regularization, and the concentration arguments, has been formally verified using Lean 4, ensuring mathematical rigor.

The author explicitly notes that the result applies to the ideal, lossless premeasurement state and does not directly address sampling hardness, finite-depth circuits, or unequal squeezing scenarios, though it provides a rigorous benchmark for these more complex settings.

Technical Summary: Weak Typicality of von Neumann Entanglement Entropy in Gaussian Boson Sampling

More like this