Hamiltonian Learning at Heisenberg Limit for Hybrid Quantum Systems
Lixing Zhang 1 , Ze-Xun Lin 2 , Prineha Narang 2,3 , Di Luo 3∗ · 2025
Read Hamiltonian Learning at Heisenberg Limit for Hybrid Quantum Systems by Lixing Zhang 1 , Ze-Xun Lin 2 , Prineha Narang 2,3 , Di Luo 3∗ (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-2502-20373.txt.
Introduction
Zhang and co-authors prove a learn plan for mixed quantum sets at the best pace cap. Mixed sets pair unlike grain kinds in one quantum frame. Such sets ground work in stuff and facts fields in labs. Their plan reads an unknown spin-boson drive form in full. It hits the Heisenberg cap for all tie marks to a set flaw. Whole time runs as one past the flaw in their proof.
Read counts run as a soft log of that flaw count in step. Small prep and read flaws fail to break the plan in tests. A twin plan spreads reads to cut time per read a lot. Tests span mixed drive learn and range learn in the work. Use spans light, stuff, and high-force fields in the close.
Catalog record
| CREATOR | Lixing Zhang 1 , Ze-Xun Lin 2 , Prineha Narang 2,3 , Di Luo 3∗ |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 66 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-20373
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Hamiltonian Learning at Heisenberg Limit for Hybrid Quantum Systems Lixing Zhang 1 , Ze-Xun Lin 2 , Prineha Narang 2,3 , Di Luo 3∗ Abstract Hybrid quantum systems with different particle species are fundamental in quantum materials and quantum information science. In this work, we establish a rigorous theoretical framework proving that, given access to an unknown spin-boson type Hamiltonian, our algorithm achieves Heisenberg-limited estimation for all coupling parameters up to error ϵ \epsilon with a total evolution time 𝒪 ( ϵ − 1 ) \mathcal{O}(\epsilon^{-1}) using only 𝒪 ( polylog ( ϵ − 1 ) ) \mathcal{O}({\rm polylog}(\epsilon^{-1})) measurements. It is also robust against small state preparation and measurement errors. In addition, we provide an alternative algorithm based on distributed quantum sensing, which significantly reduces the evolution time per measurement. To validate our method, we demonstrate its efficiency in hybrid Hamiltonian learning and spectrum learning, with broad applications in AMO, condensed matter and high energy physics. Our results provide a scalable and robust framework for precision Hamiltonian characterization in hybrid quantum platforms. Introduction—. The study of hybrid quantum systems of different particle species is essential for both understanding fundamental interactions in nature and advancing quantum engineering. In condensed matter physics, electron-phonon interaction historically played a pivotal role in the breakthrough understanding of BCS superconductivity Bardeen et al. 1957 . More broadly, fermion interactions with gauge bosons have been a unifying theme in quantum field theory Feynman 2018 ; Schwinger 1951 ; Yang and Mills 1954 . In particular, (2+1)-dimensional quantum electrodynamics (QED 3 ) provides valuable insights into the cavity-enhanced fractional quantum Hall effect Enkner et al. 2024 , topological quantum matter Song et al. 2023 , and quantum spin liquids Savary and Balents 2016 . In quantum information science, hybrid quantum systems facilitate information transfer between superconducting qubits and infrared photons in quantum transduction Xu et al. 2024 and enable long-distance communication via optical fiber-based quantum networks Stolk et al. 2024 . Recently, hybrid qauntum computing platforms based on Rydberg atoms, ion traps, and superconducting qubit-oscillator Konno et al. 2024 ; Lee et al. 2024 ; Araz et al. 2024 have emerged, providing novel avenues for fault-tolerant quantum simulation and algorithmic advancements. † † *Email: [email protected] To bridge theory and experiment, the characterization of Hamiltonian is crucial, which helps to derive accurate theoretical models Santagati et al. 2017 , calibrate quantum devices Guo et al. 2025 , and design error-mitigation algorithms Dutt et al. 2021 . This necessity has led to the field of Hamiltonian learning, which reconstructs a quantum system’s Hamiltonian from experimental measurements. According to the central limit theorem, achieving a root-mean-square error (RMSE) of \epsilon in any measured quantities requires a scaling of \mathcal{O}(\epsilon^{-2}) in total evolution time and number of measurements, defining the standard quantum limit (SQL). On the other hand, as predicted by the uncertainty principle, the optimal scaling in total evolution time is \mathcal{O}(\epsilon^{-1}) , known as the Heisenberg limit. Recently, Heisenberg limit hamiltonian learning is achieved for qubit systems Huang et al. 2023 ; Bakshi et al. ; Hu et al. 2025 and further generalized to bosonic and fermionic hamiltonians Li et al. 2024 ; Mirani and Hayden . Despite the above advancement, efficient learning for hybrid Hamiltonians remains an open challenge. The coupling between discrete and continuous degrees of freedom (DOFs) in hybrid systems leads to complex dynamic phenomena across different parameter regimes. However, the fundamental asymmetry between discrete states and continuous modes presents significant theoretical and computational hurdles, as traditional methods struggle to capture both quantization and infinite-dimensional dynamics. Even for the simplest hybrid system, a general solution was elusive until Braak 2011 , highlighting the greater complexity of hybrid Hamiltonian learning. Figure 1: Hybrid Quantum systems Hamiltonian learning. Left panel: Schematic of the learning protocol presented in this work. Right panel: Properties and applications of the main algorithm. In this paper, we present the first algorithm that learns hybrid Hamiltonians of spin-boson at the Heisenberg limit. In learning all Hamiltonian coefficients to an RMSE of \epsilon , the proposed algorithm only uses \mathcal{O}(\epsilon^{-1}) total evolution time and \mathcal{O}({\rm polylog}(\epsilon^{-1})) measurements, while maintaining robustness against small state preparation and measurement (SPAM) errors. Besides the main algorithm, we also provide an alternative scheme based on distributed quantum sensing (DQS), which significantly reduces the maximum evolution time per measurement. We numerically verify the scaling of our approaches and provide concrete examples that demonstrate its applicability in near-term quantum devices and spectrum learning problems. Hybrid Hamiltonian learning —We consider Hamiltonian that has the following form: \displaystyle{\hat{H}}={\hat{H}}_{\rm spin}+{\hat{H}}_{\rm boson}+{\hat{H}}_{\rm int} (1) where {\hat{H}}_{\rm spin}=\sum_{a}\xi_{a}\hat{E}_{a} , {\hat{H}}_{\rm boson}=\sum_{n}\omega_{n}{\hat{b}}_{n}^{\dagger}{\hat{b}}_{n} and {\hat{H}}_{\rm int}=\sum_{n,a}\lambda_{a}^{n}\hat{E}_{a}{\hat{X}}_{n} . For the n^{th} bosonic mode, {\hat{b}}_{n}^{\dagger}({\hat{b}}_{n}) is the creation (annihilation) operator, whereas {\hat{X}}_{n}\equiv({\hat{b}}_{n}^{\dagger}+{\hat{b}}_{n})/\sqrt{2} is the position operator. \hat{E}_{a} is a multi-qubit Pauli string, and the spin part of {\hat{H}}_{\rm int} is assumed to be k -local, i. e. |{\rm supp}(E_{a})|\leq k . For example, a 5-qubit spin Hamiltonian with k=3 , \hat{E}_{a} could be XIZYI . The form of {\hat{H}} covers a wide range of models, including the Holstein model and the Su–Schrieffer–Heeger (SSH) model. Explicit expressions for these models are provided in sup . The algorithm proposed in this work achieves the following properties: Theorem 1 . Given a unitary dynamics access to arbitrary Hamiltonian in the form of Eq. ( 1 ), there is an algorithm \mathcal{A} that can estimate \xi_{a} , \lambda_{a}^{n} , \eta_{a}^{n} and \omega_{n} up to a RMSE \epsilon such that: 1. \mathcal{A} takes a total evolution time of T\sim\mathcal{O}(\epsilon^{-1}) to measure all of the coefficients. 2. \mathcal{A} uses \mathcal{O}({\rm polylog(\epsilon^{-1}))} measurements to learn all of the coefficients. 3. \mathcal{A} is robust under small SPAM error. To achieve Theorem 1 , we first cancel {\hat{H}}_{\rm int} via random unitary transformation (RUT). This allow us to learn {\hat{H}}_{\rm spin} and {\hat{H}}_{\rm boson} independently. Compared to Huang et al. 2023 , our algorithm for {\hat{H}}_{\rm spin} works without the low-intersection assumption, making it scalable for general k -local spin Hamiltonians. To learn {\hat{H}}_{\rm int} , we use the learnt coefficients of {\hat{H}}_{\rm boson} as input (the error induced by doing this can be suppressed sup ) to reshape the total Hamiltonian as a displacement channel. Homodyne measurement on the momentum quadrature of the resulting coherent state creates a quantum signal that grows linearly over time with a fixed variance. This achieves the Heisenberg limit. Implementing the robust frequency estimation (RFE) Li et al. 2024 on the quantum signal reduces the number of measurements to \mathcal{O}({\rm polylog(\epsilon^{-1}))} , while making the algorithm robust under small SPAM error. We further develop an alternative scheme based on distributed quantum sensing (DQS) to {\hat{H}}_{\rm int} that also achieves the Heisenberg limit, but with lower maximum evolution time. The general scheme of our approach is provided in Fig. 1 with a detailed pseudo-code in the Supplementary Materials sup . Learning pure spin and boson coefficients — We begin by introducing the learning protocols for terms that only include pure spin or boson operators, which are characterized by \xi_{a} and \omega_{n} . A key algorithm we used is RUT, which reshape {\hat{H}} by inserting random unitary sequence \mathbb{U}(\bm{\theta})=\prod_{j}U_{j}(\theta_{j}) between R segments of e^{-i{\hat{H}}\tau} ( \tau\equiv T/R , and \bm{\theta}\equiv(\theta_{1},\theta_{2},...,\theta_{j}) .). By sampling