Nonparametric Learning Non-Gaussian Quantum States of Continuous Variable Systems
Liubov Markovich, Xiaoyu Liu, Jordi Tura · 2025
Read Nonparametric Learning Non-Gaussian Quantum States of Continuous Variable Systems by Liubov Markovich, Xiaoyu Liu, Jordi Tura (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-06431.txt.
Introduction
Liubov Markovich, Xiaoyu Liu, and Jordi Tura wrote a 2025 paper on learning the state of a quantum system from noisy measurements. Continuous-variable systems, where quantities vary smoothly, underpin quantum computing, communication, and sensing. Writing their full state in standard form is often impractical.
The authors work in the tomographic picture, which recasts a quantum state as a classical probability distribution. They build a kernel-based estimator that needs no prior guess about the state. From noisy data it recovers the density matrix in several bases, plus quantities like purity and overlap. Its error shrinks at a near-optimal rate as measurements grow. The method is built for messy, non-Gaussian states. Such states are exactly the ones quantum technology needs to certify.
Catalog record
| CREATOR | Liubov Markovich, Xiaoyu Liu, Jordi Tura |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 104 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-06431
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Nonparametric Learning Non-Gaussian Quantum States of Continuous Variable Systems Preprint: APS/123-QED Liubov Markovich, Xiaoyu Liu, Jordi Tura Abstract Continuous-variable quantum systems are foundational to quantum computation, communication, and sensing. While traditional representations using wave functions or density matrices are often impractical, the tomographic picture of quantum mechanics provides an accessible alternative by associating quantum states with classical probability distribution functions called tomograms. Despite its advantages, including compatibility with classical statistical methods, tomographic method remain underutilized due to a lack of robust estimation techniques. This work addresses this gap by introducing a non-parametric kernel quantum state estimation (KQSE) framework for reconstructing quantum states and their trace characteristics from noisy data, without prior knowledge of the state. In contrast to existing methods, KQSE yields estimates of the density matrix in various bases, as well as trace quantities such as purity, higher moments, overlap, and trace distance, with a near-optimal convergence rate of O ~ ( T − 1 ) \tilde{O}\bigl(T^{-1}\bigr) , where T T is the total number of measurements. KQSE is robust for multimodal, non-Gaussian states, making it particularly well suited for characterizing states essential for quantum science. Introduction. Continuous-variable (CV) quantum states describe systems with observables taking values in continuous spectra Serafini 2017 . In quantum optics, their realizations, known as qumodes, serve as CV counterparts to qubits, as light fields naturally exhibit continuous quadrature degrees of freedom. CV platforms support a wide range of quantum information tasks, including computation, communication, and sensing Lau and Plenio 2016 ; Arrazola et al. 2019 ; Killoran et al. 2019 ; Bromley et al. 2020 ; Jia et al. 2025 ; Nielsen et al. 2025 . As a result, significant effort has been devoted to learning CV quantum states Wu et al. 2024 ; Mele et al. 2024 ; Holevo 2024 , with particular progress for Gaussian and mildly non-Gaussian states Walschaers 2021 , establishing tight bounds on the trace distance in terms of first moments. Alternatively, shadow tomography for truncated CV systems or energy constraint states is used Becker et al. 2024 ; Aaronson 2018 ; Volkoff et al. 2021 . Due to the continuous nature of qumodes, the density matrix formalism is often impractical. However, CV states can be described in phase space by joint quasiprobability distribution functions (QPDFs), such as the Wigner, Q and P‐functions Wigner 1932 ; Husimi 1940 ; Cahill and Glauber 1969 . These QPDFs are related via one-to-one transforms Man’ko and Markovich 2020 providing a complete description of the state. However, since position \bm{q} and momentum \bm{p} operators cannot be measured simultaneously, we lack a pair of observable random variables to reconstruct their joint QPDFs. Instead, quantum state tomography Smithey et al. 1993 ; Lvovsky and Raymer 2009 ; Babichev et al. 2004 ; Wallentowitz and Vogel 1995 measures a finite number of rotated quadratures \bm{X}_{\mu,\nu}=\mu\bm{q}+\nu\bm{p} at various phase-space directions \mu,\nu\in\mathbb{R} . Using these statistics, the desired QPDF is reconstructed. The Wigner function is reconstructed by inverting the Radon transform of quadrature distributions estimated via histograms and maximum-likelihood estimation (MLE) Leonhardt and Paul 1995 ; Vogel and Risken 1989 ; Banaszek 1998 . This process is sensitive to sampling noise and the inversion of the Radon transform can be ill-posed, necessitating the use of regularization Vogel and Risken 1989 ; Lvovsky and Raymer 2009 . Detector inefficiencies blur nonclassical features such as Wigner negativity D’Ariano et al. 2007 . These propagate into the reconstructed density matrix, leading to loss of nonclassicality, reduced purity, or even nonphysical states Lvovsky and Raymer 2009 ; Hradil 1997 . To avoid such instabilities, modern experiments reconstruct the state directly from quadrature data using parametric methods that guarantee physicality Christ et al. 2009 ; Wu et al. 2024 ; Flühmann and Home 2020 ; Jia et al. 2025 . The quadrature X_{\mu,\nu} is a random variable dependent on the measurement settings \mu,\nu , and its PDF \mathcal{W}(x|\mu,\nu) , called the symplectic tomogram, is informationally complete Amosov et al. 2008 , uniquely specifying the density operator \bm{\rho} . Hence, every CV state can be fully described by the classical PDF, i.e., the tomogram, of its quadratures Mancini et al. 1996 ; Ibort et al. 2009 ; Man’ko and Man’ko 1997 ; Dodonov et al. 1974 ; Chernega and Man’ko 2023 ; Dudinets et al. 2024 ; Man’ko and Markovich 2020 . In this Letter, we propose a state reconstruction and trace characteristic estimation method based on nonparametric kernel density estimation (KDE) Silverman 2018 and a kernel characteristic function estimation (KCFE), called kernel quantum state estimation (KQSE). KQSE estimates the state characteristics, using their integral representations via the tomogram and characteristic functions, achieving \tilde{O}(T^{-1}) in the uniform and L_{2} norms, where T is the total amount of measurements with no prior assumptions on the state (Theorem 1 ). This approach is robust to measurement noise and performs equally well for Gaussian and non-Gaussian states. We first prove that the trace distance D(\bm{\rho}_{1},\bm{\rho}_{2})=\tfrac{1}{2}\|\bm{\rho}_{1}-\bm{\rho}_{2}\|_{1} between any quantum states \bm{\rho}_{1} , \bm{\rho}_{2} is lower bounded by the total variation between the corresponding tomograms (Theorem 6 ). In Theorem 3 we show that achieving precision \varepsilon in trace distance using KQSE requires O(\varepsilon^{-5/2}) measurements. While slightly looser than the O(\varepsilon^{-2}) rate known for Gaussian states Holevo 2024 ; Mele et al. 2024 ; Devroye et al. 2023 , our bound holds universally without assumptions on the state. Next, we use our previous work Markovich et al. 2024 , where the characteristic function (CF) \phi(t;\mu,\nu) is defined as the Fourier transform of the tomogram. Theorem 4 provides necessary and sufficient conditions for a CF to define a valid quantum state. Moreover, the CF enables not only the reconstruction of the density matrix \rho(y,y^{\prime})=\langle y|\bm{\rho}|y^{\prime}\rangle in various bases, but also allows for direct analytic evaluation of quantities such as \text{tr}{(\bm{\rho}^{d})} (d\in\mathbb{Z}^{+}) , and state distinguishability measures like the trace distance D(\bm{\rho}_{1},\bm{\rho}_{2}) and overlap \text{tr}{(\bm{\rho}_{1}\bm{\rho}_{2})} . We show that this CF coincides with that of the Wigner function Wu et al. 2024 ; Flühmann and Home 2020 . Fig. 1 illustrates the theoretical and empirical correspondence among the tomogram, CF, density matrix, and Wigner function Man’ko and Markovich 2020 . Finally, we introduce the new nonparametric KCFE that achieves a mean squared error (MSE) convergence rate of {O}(1/n) , where n is the amount of measurements for the fixed \mu,\nu , matching optimal parametric rates of MLE (Theorem 5 ). In signal processing Stefanski and Carroll 1990 ; Sinitsyn and Yanovsky 2005 ; Fan 1991 ; Delaigle and Gijbels 2004 , PDFs of the signals are often estimated via the inverse Fourier transform of a CF, as this approach empirically yields more stable results. The KQSE method uses a collection of KCFEs \{\hat{\phi}_{X}(1;\mu_{i},\nu_{j}\} , i\in[1,N_{\mu}] , j\in[1,N_{\nu}] to estimate the state and its trace characteristics from the noisy data: Theorem 1 . (Convergence of KQSE) Let \widehat{\rho(y,y^{\prime})} and \widehat{\mathrm{tr}(\bm{\rho}_{1}\bm{\rho}_{2})} denote the KQSE estimators of the density matrix \rho(y,y^{\prime}) and the overlap \mathrm{tr}(\bm{\rho}_{1}\bm{\rho}_{2}) , respectively, constructed from T_{\mu}=nN_{\mu} and T_{\mu,\nu}=nN_{\mu}N_{\nu} quantum state preparations, where n samples are collected per setting \{\mu_{i},\nu_{j}\} , N_{\mu},N_{\nu}=\log{n} . Then the estimator achieves: \displaystyle L_{\infty}(\widehat{\rho}(y,y^{\prime}))\equiv\sup\limits_{y,y^{\prime}}\mathbb{E}|\rho(y,y^{\prime})-\widehat{\rho(y,y^{\prime})}|^{2}=\tilde{O}(1/T_{\mu}), (1) \displaystyle\mathrm{MSE}\big(\widehat{\mathrm{tr}(\bm{\rho}_{1}\bm{\rho}_{2})}\big)\!\equiv\!\mathbb{E}\big[|\mathrm{tr}(\bm{\rho}_{1}\bm{\rho}_{2})-\widehat{\mathrm{tr}(\bm{\rho}_{1}\bm{\rho}_{2})}|^{2}\big]\!=\!\tilde{O}(1/T_{\mu,\nu}). Using the KQSE to estimate the pure state \bm{\rho} with \bm{\rho}_{n} we get the optimal rate \mathrm{MSE}(\widehat{D^{2}(\bm{\rho},\bm{\rho}_{n})})=\tilde{O}(T_{\mu,\nu}^{-1}) ,