arXiv:2609.26705
When are bosonic Gaussian states classical to learn?
Senrui Chen, Antonio Anna Mele, Francesco Anna Mele, John Preskill
Abstract
A fundamental question in physics is: When does classical behavior emerge from quantum systems? Bosonic Gaussian states provide a natural setting to explore this quantum-classical boundary, as they capture both the classical field behavior and the intrinsic quantum nature of light. Here, we address this problem from a learning-theoretic perspective by asking: When are bosonic Gaussian states classical to learn? That is, under what conditions (if any) can an n-mode bosonic Gaussian state be learned with as few samples, and with operations as simple, as are needed to learn a classical 2n-variate Gaussian distribution? We establish a smooth crossover in learnability governed by the state's thermal fluctuations: - Cold Gaussian states are non-classical to learn: When the covariance matrix satisfies $Σ\le(\frac12+O(\frac1n))I$, i.e. close to the vacuum covariance, tomography under single-copy (i.e., non-entangled) measurements fundamentally requires $Ω(n^3)$ copies, strictly exceeding the sample complexity $Θ(n^2)$ of learning classical Gaussian distributions. We show that this hardness persists even when few-copy entangled measurements are allowed. - Warm Gaussian states are classical to learn: When thermal fluctuations exceed the vacuum noise, parameterized by $Σ\ge(\frac12+ν)I$ for any parameter $ν>0$, we prove that single-copy tomography requires $N=Θ\left(n^2\min(n,1+ν^{-1})\right)$ copies. This bound is tight and is achieved by simple, non-adaptive, unentangled heterodyne measurements. Crucially, for $ν=Ω(1)$, the sample complexity drops to $Θ(n^2)$, matching the classical case. Our results tightly characterize a quantum-to-classical crossover in the learnability of bosonic Gaussian states, reveal a novel connection between fundamental physics and statistical learning theory, and have implications for real-world sensing experiments.
AI Summary
The paper asks when tomography of an $n$-mode bosonic Gaussian state matches learning a classical $2n$-variate Gaussian distribution in sample complexity while using only single-copy measurements. Accuracy is measured in trace distance, with constant success probability; the classical benchmark is $\Theta(n^2/\varepsilon^2)$ samples. For general Gaussian states, any possibly adaptive single-copy protocol requires $\Omega(n^3/\varepsilon^2)$ copies. Hard instances can be passive states satisfying $\Sigma\preceq(1/2+O(1/n))I$, with constant total expected photon number. Measurements on at most $k$ fresh copies, without quantum memory between blocks, require $\Omega(n^3/[\varepsilon^2\min\{\sqrt{k},n\}])$ copies. Under the promise $\Sigma\succeq(1/2+\nu)I$, the single-copy complexity is exactly $\Theta((n^2/\varepsilon^2)\min\{n,1+\nu^{-1}\})$, achieved by nonadaptive heterodyne measurements. Thus the cubic scaling persists for $\nu\lesssim1/n$, crosses through $\Theta(n^2/(\nu\varepsilon^2))$, and becomes quadratic for $\nu=\Omega(1)$. The cold-state bounds use a reduction encoding finite-dimensional mixed states into low-energy passive Gaussian states. Temperature-dependent lower bounds extend Bayesian posterior anti-concentration to bosonic photon-number sectors using symmetric-power representations and Schur polynomials. The upper bound combines empirical heterodyne statistics, covariance inflation to ensure physicality, and a one-sided covariance-to-trace-distance bound. The results concern worst-case full tomography, not structured sensing or estimation of selected properties; the experimental comparisons do not establish corresponding sensing bounds. The $k$-copy lower bound lacks a matching upper bound. The crossover promise bounds ordinary covariance eigenvalues and does not cover squeezed states with quadrature variance below vacuum.
GPT-5.6 Sol and GPT-6 were used to strengthen several of our technical results, and to help check the mathematical arguments and improve the exposition.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- comments welcome