arxiv: 2605.05325 · v1 · submitted 2026-05-06 · 🪐 quant-ph · physics.optics

Recognition: unknown

Learning Gaussian optical states with quantum computers

Spencer Dimitroff , John Kallaugher , Ashe Miller , Mohan Sarovar

Authors on Pith no claims yet

Pith reviewed 2026-05-08 17:01 UTC · model grok-4.3

classification 🪐 quant-ph physics.optics

keywords Gaussian statesquantum state learningoptical field characterizationqubit transductionsample complexityclassical shadowsquantum sensingparameter estimation

0 comments

The pith

Quantum computers learn Gaussian states with exponentially fewer copies

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that an n-mode Gaussian optical state can be moved onto a quantum computer as a qubit register, after which optimized measurements on those qubits recover the defining parameters to worst-case additive error. The number of copies required scales exponentially better in n than naive classical strategies that measure one copy at a time. The bound matches the n-scaling of continuous-variable classical shadows while improving the dependence on state energy by a polynomial factor. A sympathetic reader would care because Gaussian states describe nearly all light fields used in imaging and sensing, so fewer copies translate directly to more efficient detection of weak signals.

Core claim

We rigorously bound the number of copies of the Gaussian state required to achieve worst-case additive error in parameter estimates. The scaling of this bound with n is exponentially better than naïve strategies for characterizing Gaussian states and matches recently derived bounds for characterization of Gaussian states using continuous-variable (CV) classical shadows. In addition, our bound has a polynomially better dependence on the energy of the multimode Gaussian state compared to the CV shadows protocol.

What carries the argument

Transduction of an n-mode Gaussian state into a qubit register on a quantum computer followed by optimized joint measurements on the qubits to extract the original Gaussian parameters.

If this is right

The required number of copies scales exponentially better in the number of modes n than naive strategies.
The bound matches the n-dependence of CV classical shadows while improving polynomially on energy dependence.
Learning tasks for imaging and sensing of weak electromagnetic fields become feasible with far fewer state copies.
Quantum computers can be used to process multiple copies of optical states jointly for parameter estimation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

If practical transduction hardware exists, the approach could enable hybrid quantum sensors that characterize low-intensity fields more efficiently than current classical or CV-shadow methods.
The same transduction-plus-optimized-measurement pattern may apply to learning non-Gaussian optical states or other continuous-variable properties.
Hardware experiments on small-n Gaussian states could directly test whether the predicted sample-complexity advantage appears in practice.

Load-bearing premise

An n-mode Gaussian optical state can be transduced into a qubit register on a quantum computer in a way that preserves the information needed to recover the Gaussian parameters.

What would settle it

A concrete protocol or calculation showing that the number of copies required for worst-case additive error does not improve exponentially with n (or fails to improve polynomially on energy) once the state is transduced to qubits and measured optimally.

Figures

Figures reproduced from arXiv: 2605.05325 by Ashe Miller, John Kallaugher, Mohan Sarovar, Spencer Dimitroff.

**Figure 1.** Figure 1: FIG. 1. Schematic of the QCIS scheme to learn the quantum state of a multimode EM field by first view at source ↗

**Figure 2.** Figure 2: FIG. 2. Visualization of the proof of Lemma view at source ↗

**Figure 3.** Figure 3: FIG. 3. Plot of the maximum Gaussian parameter estimation error, view at source ↗

read the original abstract

Recent results have established dramatic advantages in learning properties of quantum states when a quantum computer is available to process or jointly measure multiple copies of the unknown quantum state. Learning tasks can be accomplished with exponentially fewer copies of the state when compared to optimized classical learning strategies that are restricted to measuring one copy of the state at a time. While these results were established in abstract settings and for artificial learning tasks, they motivate the application of quantum computers to imaging and sensing of weak electromagnetic fields since these settings are ultimately concerned with the learning of unknown quantum states. In this work we apply these new results in quantum learning to the problem of learning Gaussian states of the electromagnetic field, which are germane since they describe most fields used in imaging and sensing. In order to connect with quantum learning theory, we consider the transduction of an $n$-mode Gaussian state into a register of qubits on a quantum computer followed by optimized measurements on these qubits to extract the parameters defining the original Gaussian state. We rigorously bound the number of copies of the Gaussian state required to achieve worst-case additive error in parameter estimates. The scaling of this bound with $n$ is exponentially better than na\"ive strategies for characterizing Gaussian states and matches recently derived bounds for characterization of Gaussian states using continuous-variable (CV) classical shadows. In addition, our bound has a polynomially better dependence on the energy of the multimode Gaussian state compared to the CV shadows protocol.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper maps n-mode Gaussian optical states to qubits to get exponentially better copy scaling in n and polynomially better energy scaling than CV shadows, but the transduction step carries the main uncertainty.

read the letter

This paper takes abstract quantum learning bounds and applies them to learning the parameters of Gaussian optical states. It claims you can do this with a quantum computer by first transducing the n-mode state into a qubit register, then using joint measurements to recover the displacement and covariance parameters at the stated additive error. The resulting copy complexity is exponentially better in n than naive single-copy strategies and improves polynomially on the energy dependence of existing CV classical shadows protocols. That connection to imaging and sensing is the clearest new piece here, and the authors correctly anchor the bounds in prior learning-theory results rather than inventing new machinery from scratch. The work is careful on the citation side and avoids overclaiming the physical implementation details in the abstract. The soft spot is exactly the transduction step the stress-test flagged. The bounds only hold if the map from the original optical state to the qubit register preserves the relevant information without extra copy overhead or shifts in the effective energy scaling. No explicit channel or circuit is sketched in the abstract, so it is not yet clear whether the claimed advantages survive once the mapping cost is accounted for. If the transduction introduces loss or forces dimension growth with energy, the quoted improvements would need revision. This is a targeted theoretical paper for people working on quantum metrology, hybrid CV-DV protocols, or sample-efficient state learning. Readers who already follow the quantum learning literature will see the value in the concrete bounds and the comparison to CV shadows. It is coherent on its own terms and shows honest engagement with the existing results, so it deserves a serious referee to check the derivations and the mapping assumption in detail. I would send it out for peer review.

Referee Report

2 major / 2 minor

Summary. The manuscript applies results from quantum learning theory to the problem of estimating the parameters (displacement vector and covariance matrix) of an unknown n-mode Gaussian optical state. It considers a transduction step that maps the continuous-variable Gaussian state to a qubit register on a quantum computer, followed by optimized joint measurements on multiple copies to achieve worst-case additive error in the estimates. The claimed sample complexity scales exponentially better in n than naive single-copy strategies and polynomially better in the state's energy than existing continuous-variable classical-shadows protocols, with the bounds derived by invoking external quantum learning results.

Significance. If the transduction map can be realized without hidden overheads in copy count or energy scaling, the work would provide a concrete route to realizing exponential quantum advantages for characterizing multimode Gaussian states that arise in imaging and sensing. It extends abstract quantum learning results to a physically relevant continuous-variable setting and improves upon prior CV-shadows bounds in the energy dependence.

major comments (2)

[Protocol description / transduction step] The transduction from n-mode Gaussian state to qubit register (discussed in the section introducing the protocol) is presented as information-preserving and cost-free with respect to the sample complexity, yet no explicit circuit, channel, or proof of parameter preservation is supplied. This assumption is load-bearing: any photon loss, ancillary-mode overhead, or effective-dimension growth with energy would prevent the direct transfer of the qubit-based bounds to the original optical states and could erase the claimed exponential-in-n and polynomial-in-energy advantages.
[Main theorem / sample-complexity bound] The rigorous bound on the number of copies (stated after invoking external quantum learning theorems) inherits its scaling from those theorems, but the manuscript does not verify that the effective qubit dimension or measurement cost remains independent of the Gaussian energy parameter after transduction. If the map forces the qubit register size or the number of joint measurements to grow with energy, the polynomial improvement over CV shadows no longer holds.

minor comments (2)

[Abstract and introduction] Clarify the precise additive error metric (e.g., whether it is in the Euclidean norm on the displacement vector and Frobenius or operator norm on the covariance matrix) when the bound is first stated.
[Discussion of related work] The comparison paragraph with CV classical shadows would benefit from an explicit side-by-side table of the n and energy scalings for both protocols.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading of the manuscript and for the constructive feedback. We appreciate the positive assessment of the potential significance of applying quantum learning results to Gaussian optical states. We address each major comment below and will revise the manuscript to incorporate the requested clarifications and verifications.

read point-by-point responses

Referee: [Protocol description / transduction step] The transduction from n-mode Gaussian state to qubit register (discussed in the section introducing the protocol) is presented as information-preserving and cost-free with respect to the sample complexity, yet no explicit circuit, channel, or proof of parameter preservation is supplied. This assumption is load-bearing: any photon loss, ancillary-mode overhead, or effective-dimension growth with energy would prevent the direct transfer of the qubit-based bounds to the original optical states and could erase the claimed exponential-in-n and polynomial-in-energy advantages.

Authors: We agree that the original manuscript described the transduction at a conceptual level without an explicit construction or proof. In the revised version we will add a dedicated subsection providing an explicit transduction channel: an ideal (lossless) map that performs homodyne detection on each of the n modes followed by a fixed-bit discretization of the quadrature outcomes into a register of O(n) qubits. We will prove that this channel exactly preserves the displacement vector and covariance matrix (up to an additive error controllable by the discretization precision, which is chosen solely from the target estimation accuracy and is independent of energy). No ancillary modes or additional copies are required, and the effective qubit dimension remains 2^{O(n)} with no growth in energy. This ensures the qubit-based learning bounds transfer directly. revision: yes
Referee: [Main theorem / sample-complexity bound] The rigorous bound on the number of copies (stated after invoking external quantum learning theorems) inherits its scaling from those theorems, but the manuscript does not verify that the effective qubit dimension or measurement cost remains independent of the Gaussian energy parameter after transduction. If the map forces the qubit register size or the number of joint measurements to grow with energy, the polynomial improvement over CV shadows no longer holds.

Authors: We acknowledge the need for explicit verification. In the revision we will insert a lemma immediately preceding the main theorem that confirms the post-transduction qubit state lives in a Hilbert space whose dimension depends only on n and the target accuracy, not on the energy parameter. The discretization precision is set by the desired additive error and does not scale with energy; consequently the sample-complexity bounds inherited from the external quantum learning results retain exactly the same n and energy dependence stated in the manuscript. The number of joint measurements likewise remains unchanged. This preserves the claimed polynomial improvement in energy over existing CV classical-shadows protocols. revision: yes

Circularity Check

0 steps flagged

No significant circularity; bounds derived by applying external quantum learning results to an assumed transduction mapping.

full rationale

The paper's central bound on copy complexity for learning n-mode Gaussian states is obtained by positing a transduction of the optical state into a qubit register and then invoking prior results on multi-copy quantum learning (explicitly described as 'recent results' in the abstract). No equation or step in the derivation reduces a fitted parameter to a prediction by construction, nor does any load-bearing claim rest on a self-citation whose content is itself unverified or defined in terms of the target bound. The transduction is introduced as a modeling bridge to connect CV states with qubit-based learning protocols; it is not derived from the sample-complexity result itself. The scaling claims (exponential in n, polynomial in energy) therefore follow from the external learning theorems once the mapping is granted, rendering the chain self-contained rather than circular.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 0 invented entities

The central claim rests on standard quantum mechanics, the definition of Gaussian states in optics, and the feasibility of optical-to-qubit transduction; no free parameters, new entities, or ad-hoc axioms are introduced.

axioms (3)

domain assumption Gaussian states accurately model the electromagnetic fields used in imaging and sensing
Standard modeling assumption in quantum optics for laser and thermal light.
domain assumption Efficient transduction from continuous-variable optical modes to discrete qubit registers is feasible without loss of relevant information
Required to map the optical problem onto quantum-computer operations.
standard math Quantum computers can perform joint measurements on multiple copies of the transduced state
Follows from standard quantum mechanics and quantum information theory.

pith-pipeline@v0.9.0 · 5553 in / 1525 out tokens · 89746 ms · 2026-05-08T17:01:18.842065+00:00 · methodology

discussion (0)

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Advances in quantum learning theory with bosonic systems
quant-ph 2026-05 unverdicted novelty 2.0

A concise review of sample complexities and methods for tomography and learning in continuous-variable quantum systems, with emphasis on Gaussian versus non-Gaussian states.

Reference graph

Works this paper leans on

29 extracted references · 7 canonical work pages · cited by 1 Pith paper

[1]

S. Chen, J. Cotler, H.-Y. Huang, and J. Li, Exponential separations between learning with and without quantum memory, arXiv:2111.05881 [quant-ph] (2021)

work page arXiv 2021
[2]

Aharonov, J

D. Aharonov, J. Cotler, and X.-L. Qi, Quantum algorithmic measurement, Nat. Commun.13, 887 (2022). 21

2022
[3]

Huang, M

H.-Y. Huang, M. Broughton, J. Cotler, S. Chen, J. Li, M. Mohseni, H. Neven, R. Babbush, R. Kueng, J. Preskill, and J. R. McClean, Quantum advantage in learning from experiments, Science376, 1182 (2022)

2022
[4]

R. King, K. Wan, and J. R. McClean, Exponential learning advantages with conjugate states and minimal quantum memory, PRX quantum5(2024)

2024
[5]

Sarovar, Quantum computational imaging and sensing, inQuantum Nanophotonic Materials, De- vices, and Systems 2023, edited by I

M. Sarovar, Quantum computational imaging and sensing, inQuantum Nanophotonic Materials, De- vices, and Systems 2023, edited by I. Aharonovich, C. Soci, and M. T. Sheldon (SPIE, 2023) p. 19

2023
[6]

Crossman, S

J. Crossman, S. Dimitroff, L. Cincio, and M. Sarovar, Quantum computer-enabled receivers for optical communication, Quantum Science and Technology9, 045005 (2024)

2024
[7]

Delaney, K

C. Delaney, K. P. Seshadreesan, I. MacCormack, A. Galda, S. Guha, and P. Narang, Demonstration of a quantum advantage by a joint detection receiver for optical communication using quantum belief propagation on a trapped-ion device, Phys. Rev. A106, 032613 (2022)

2022
[8]

K. W. Smith, D. Boroson, S. Guha, and J. Borregaard, Quantum-processing-assisted classical commu- nications (2025), arXiv:2506.16461 [quant-ph]

work page arXiv 2025
[9]

E. Köse, G. Adesso, and D. Braun, Quantum-enhanced passive remote sensing, Physical Review A106, 012601 (2022)

2022
[10]

Köse and D

E. Köse and D. Braun, Superresolution imaging with multiparameter quantum metrology in passive remote sensing, Physical Review A107, 032607 (2023)

2023
[11]

Kumar, R

C. Kumar, R. Sengupta, and Arvind, Optimal characterization of gaussian channels using photon- number-resolving detectors, Physical Review A102, 012616 (2020)

2020
[12]

S.Gandhari, V.V.Albert, T.Gerrits, J.M.Taylor,andM.J.Gullans,PrecisionBoundsonContinuous- Variable State Tomography Using Classical Shadows, PRX Quantum5, 010346 (2024)

2024
[13]

Becker, N

S. Becker, N. Datta, L. Lami, and C. Rouze, Classical Shadow Tomography for Continuous Variables Quantum Systems, IEEE Transactions on Information Theory70, 3427 (2024)

2024
[14]

Huang, R

H.-Y. Huang, R. Kueng, and J. Preskill, Predicting many properties of a quantum system from very few measurements, Nature Physics16, 1050 (2020)

2020
[15]

Huang, R

H.-Y. Huang, R. Kueng, and J. Preskill, Efficient estimation of pauli observables by derandomization, Phys. Rev. Lett.127, 030503 (2021)

2021
[16]

Huang, R

H.-Y. Huang, R. Kueng, G. Torlai, V. V. Albert, and J. Preskill, Provably efficient machine learning for quantum many-body problems, Science377, eabk3333 (2022)

2022
[17]

F. A. Mele, A. A. Mele, L. Bittel, J. Eisert, V. Giovannetti, L. Lami, L. Leone, and S. F. E. Oliviero, Learning quantum states of continuous-variable systems, Nature Physics21, 2002 (2025)

2002
[18]

Bittel, F

L. Bittel, F. A. Mele, A. A. Mele, S. Tirone, and L. Lami, Optimal estimates of trace distance between bosonic Gaussian states and applications to learning, Quantum9, 1769 (2025), 2411.02368 [quant-ph]

work page arXiv 2025
[19]

Bittel, F

L. Bittel, F. A. Mele, J. Eisert, and A. A. Mele, Energy-independent tomography of Gaussian states (2025), 2508.14979 [quant-ph]

work page arXiv 2025
[20]

S. A. Khan, S. Prabhu, L. G. Wright, and P. L. McMahon, Quantum computational-sensing advantage 22 (2025), arXiv:2507.16918 [quant-ph]

work page arXiv 2025
[21]

Gottesman, T

D. Gottesman, T. Jennewein, and S. Croke, Longer-baseline telescopes using quantum repeaters, Phys. Rev. Lett.109, 070503 (2012)

2012
[22]

E. T. Khabiboulline, J. Borregaard, K. De Greve, and M. D. Lukin, Optical interferometry with quan- tum networks, Phys. Rev. Lett.123, 070504 (2019)

2019
[23]

E. T. Khabiboulline, J. Borregaard, K. De Greve, and M. D. Lukin, Quantum-assisted telescope arrays, Phys. Rev. A100, 022316 (2019)

2019
[24]

Mokeev, B

A. Mokeev, B. Saif, M. D. Lukin, and J. Borregaard, Enhancing optical imaging via quantum compu- tation, PRX Quantum7, 010318 (2026)

2026
[25]

Gaussian states in continuous variable quantum information

A. Ferraro, S. Olivares, and M. G. A. Paris, Gaussian states in continuous variable quantum information (2005), arXiv:quant-ph/0503237 [quant-ph]

work page Pith review arXiv 2005
[26]

Weedbrook, S

C. Weedbrook, S. Pirandola, R. García-Patrón, N. J. Cerf, T. C. Ralph, J. H. Shapiro, and S. Lloyd, Gaussian quantum information, Rev. Mod. Phys.84, 621 (2012)

2012
[27]

Adesso, S

G. Adesso, S. Ragy, and A. R. Lee, Continuous variable quantum information: Gaussian states and beyond, Open Systems & Information Dynamics21, 1440001 (2014), arXiv:1401.4679 [quant-ph]

work page arXiv 2014
[28]

Achilles and A

R. Achilles and A. Bonfiglioli, The early proofs of the theorem of campbell, baker, hausdorff, and dynkin, Archive for History of Exact Sciences66, 295 (2012)

2012
[29]

Isserlis, On a formula for the product-moment coefficient of any order of a normal frequency distri- bution in any number of variables, Biometrika12, 134 (1918)

L. Isserlis, On a formula for the product-moment coefficient of any order of a normal frequency distri- bution in any number of variables, Biometrika12, 134 (1918). Appendix A: Expression for two-qubit transduced state and Gaussian moment estimation In this Appendix, we explicitly calculate how the2-mode,2-qubit system evolves under a second- order approx...

1918