Pith. sign in

REVIEW 4 major objections 6 minor 23 references

Capturing Quantum Snapshots from a Single Copy via Mid-Circuit Measurement and Dynamic Circuit

T0 review · 4 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A single unknown quantum state can be reconstructed non-destructively through a guess-and-check loop using SWAP-test fidelity as the only feedback.

desk verdict The headline single-copy snapshot claim is unsupported; what is actually demonstrated is a modest but legitimate fidelity-optimization study. read the letter →

arxiv 2504.21250 v1 pith:ZBVP7G2R submitted 2025-04-30 quant-ph

Debarshi Kundu , Avimita Chatterjee , Archisman Ghosh , Swaroop Ghosh This is my paper · ORCID
classification quant-ph
keywords quantumsnapshotmemorystatetomographymid-circuitmeasurementdynamiccircuitSWAPtestevolutionarystrategyfidelityfeedback
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper proposes a framework for taking a 'quantum snapshot' of an unknown quantum state at an intermediate point in a circuit without consuming the state, using only one copy. A classical model proposes a candidate state, the candidate is compared with the target using a SWAP test, and the measured fidelity is fed back to improve the candidate. Because only the candidate register and the ancilla are reset between trials, the target is never re-prepared, so in principle the whole reconstruction happens against a single copy. The authors report average fidelity above 0.99 in noiseless simulation across 100 random states, and faithful reconstruction of a single-qubit known state on real hardware within three optimization steps; the full mid-circuit implementation is still being developed.

What carries the argument

The mechanism is the SWAP test used as a fidelity oracle inside a hybrid quantum-classical loop. A candidate state generated by a classical model is loaded on one register; an ancilla-controlled swap with the fixed target on another register yields an estimate of $F = 2P(0) - 1 = |\langle \psi | \phi_u \rangle|^2$, which is the only feedback the model sees. Dynamic circuits with mid-circuit reset let the candidate register and ancilla be reinitialized after every evaluation while the target is left untouched, enabling iterative optimization from a single copy. The paper implements the classical side with two optimizers—a gradient-based neural generator with finite-difference gradients and a gradient-free evolutionary strategy called QESwap—and applies them to state-vector, density-matrix, and unitary representations.

What would settle it

Prepare two identical copies of a known pure state. Run the full QSDC loop on one copy while leaving the other untouched, then perform quantum state tomography on both; if the processed copy's density matrix differs from the untouched copy's by more than the noise floor, the non-destructive single-copy premise is refuted.

Watch

Extended reading notes

Core claim

The central claim is that SWAP-test fidelity alone is enough to drive an unknown pure quantum state to near-perfect reconstruction, and that this can be done non-destructively and from a single copy when dynamic circuits allow the candidate register and ancilla to be reset mid-circuit while the target register is preserved. The paper validates the reconstruction step (the learning loop with fidelity feedback) in noiseless and noisy simulation and on real hardware for a single-qubit state; the full mid-circuit, single-copy protocol is presented as the intended architecture, with simulation-based implementation described as ongoing work. A stated limit is that the SWAP test estimates the Hilbert-Schmidt inner product, which makes the method reliable for pure states but not for general mixed states.

Load-bearing premise

The entire non-destructive single-copy scheme rests on the premise that the unknown target state is not disturbed while the SWAP test's ancilla is measured, so the same target can be queried again and again during optimization.

Editorial extensions

If this is right

  • Mid-circuit debugging becomes possible: at any point in a circuit, the current state can be logged without halting or re-running the circuit many times.
  • Reconstructed states stored as classical vectors can be re-prepared later, giving a classical form of quantum memory that could feed QRAM-style retrieval without persistent physical qubit storage.
  • The gradient-free evolutionary variant is the practical choice under noise; the paper reports it converges faster and remains stable where the neural-network approach degrades.
  • Because convergence takes only a few epochs for small states, a single snapshot may be captured within milliseconds, which is compatible with coherence times of some existing or near-future platforms.
  • The approach is confined to pure states by the SWAP test's inability to distinguish mixed states; extending it would require a different fidelity probe.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the target truly survives every SWAP test, the same loop becomes a reusable query oracle for an unknown state; that would open a route to sample-efficient state learning that does not require the exponential number of copies used in standard tomography. The paper does not make this comparison.
  • A natural next experiment is to run the full loop on a known state and then perform tomography on the target register; if the target density matrix is unchanged to within noise, the non-destructive single-copy claim is confirmed, and if not, the protocol will need a query budget.
  • The architecture implies a concrete hardware target: feedback round-trip time plus candidate reset must be much shorter than the target's coherence time. This could be tested on platforms with fast mid-circuit control by measuring the maximum number of successful SWAP queries before fidelity decays.
  • Because the paper's mixed-state analysis shows the feedback signal measures Hilbert-Schmidt overlap rather than true quantum fidelity, a direct extension would be to purify the target and run the same loop; whether that recovers the mixed state is untested.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper proposes Quantum Snapshot with Dynamic Circuit (QSDC), a hybrid quantum-classical framework that aims to reconstruct an unknown quantum state from a single copy at arbitrary points in a circuit, using SWAP-test fidelity as the only feedback signal to train either a neural network or an evolutionary strategy. The reconstructed state is stored classically. The authors validate the optimization component in noiseless and noisy simulations and on IBM hardware, reporting fidelities above 0.99 for up to 3 qubits and successful single-qubit reconstruction on hardware within 3 epochs.

Significance. If the central claim were established, a non-destructive single-copy quantum snapshot capability would be highly impactful for circuit debugging, quantum memory, and introspection. The paper makes a useful contribution by demonstrating that evolutionary strategies can optimize quantum state estimates from fidelity feedback, and by providing convergence data across qubit counts and noise settings. However, the physical and experimental basis for the central single-copy, mid-circuit claim is not provided, and the reported experiments validate a different, multi-copy protocol. The significance is therefore conditional on resolving the major gaps below.

major comments (4)
  1. [Abstract and Section IV, first paragraph; Section III.C.1] The abstract and conclusion claim 'single-copy, mid-circuit state reconstruction,' but Section IV explicitly states that the experiments were run 'without mid-circuit measurements, reinitializing the circuit at each step,' and Section III.C.1 labels the dynamic-circuit implementation as 'work in progress.' The reported simulations and hardware results therefore validate a multi-copy optimization loop, not the single-copy protocol. This is a load-bearing discrepancy between the claims and the evidence.
  2. [Section III.A and Section III.B.4] The protocol assumes 'qubit state preservation throughout hybrid optimization,' but the manuscript does not analyze the back-action of the repeated SWAP tests on the target state. In the standard SWAP test described in Section III.B.4, the controlled-SWAP entangles the target with the candidate and the ancilla; measuring the ancilla projects the joint state, and unless the candidate exactly equals the target, the target's reduced state is altered. No derivation, weak-measurement scheme, or modified circuit is provided to justify that the target remains intact for subsequent queries. The single-copy non-destructive claim is therefore physically unsubstantiated.
  3. [Table I and Section IV.B] The caption of Table I states that entries are 'shown only for cases that successfully exceeded the threshold,' but the paper does not report the number or fraction of trials that failed to reach the threshold. This selection bias makes the reported average epochs and fidelities uninterpretable as success rates, and it undermines the convergence claims (e.g., 'within 3 epochs' on hardware). The manuscript should report success counts, including cases that did not converge, for each configuration.
  4. [Section III.B.4 and Section III.B.5] In the description of the gradient-based method, the fidelity is computed directly as |⟨ψ|ϕu⟩|² in the forward pass rather than from measurement statistics of a simulated SWAP test circuit. This means the simulations do not validate that fidelity from the SWAP test can serve as the sole feedback signal; they validate optimization with exact overlap as the loss. The manuscript should either describe how the SWAP test is actually simulated in PennyLane or clarify that the simulation uses direct fidelity, and should temper the claim that the SWAP test is the only feedback.
minor comments (6)
  1. [Abstract] There is a typo in 'evolutionary strategie' — it should be 'evolutionary strategies.'
  2. [Author list] The email for Archisman Ghosh is the same as that for Avimita Chatterjee (amc8313@psu.edu); this appears to be a typo.
  3. [Throughout] The term 'SW AP' appears with a stray space in multiple places (e.g., abstract, Section III.B.4); it should be 'SWAP'.
  4. [Section III.B.2] The name 'Mottonen' should be typeset as 'Möttönen' for consistency with the referenced work.
  5. [Figure 2] The label 'e:#Epoches' appears garbled; it should read 'e: #Epochs'.
  6. [Table I] The 'NA' entries in Table I are unexplained; the manuscript should state why certain methods/configurations are not reported (e.g., density matrix results plateaued at 0.8 fidelity).

Circularity Check

2 steps flagged · score 6.0 of 10

Reported fidelities are the in-sample training objective computed against known targets, and the single-copy non-destructive claim is assumed via 'qubit state preservation' rather than derived or tested.

  1. fitted input called prediction [Section III.B.4-III.B.5 and Section IV (fidelity results)]
    "In the forward pass, the real and imaginary parts of the generated vector form a normalized complex state; fidelity is computed against |ϕu⟩. ... The fidelity score between the generated state |ψ⟩ and the target state |ϕu⟩ defines the loss function L = 1 − F ... After training, the final state is classically compared to the target via |⟨ψ|ϕu⟩|2, with F≈ 1 indicating successful reconstruction."

    The reported reconstruction fidelity is exactly the overlap |⟨ψ|ϕu⟩|2 used as the training loss, evaluated on the same target states |ϕu⟩ that are directly available to the custom autograd function in simulation. Thus the high fidelities in Table I (e.g., 0.9999) report convergence of the optimizer on in-sample targets, not an independent prediction obtained from SWAP-test feedback alone. The 'unknown' state is an input to the gradient computation, so the result is the fitted objective reflected back as a headline metric.

  2. self definitional [Section III.A and Section V Conclusion]
    "The method assumes qubit state preservation throughout hybrid optimization—an aspirational but increasingly feasible condition as coherence times approach the millisecond regime, potentially enabling accurate reconstruction from a single copy on real hardware. ... QSDC enables multiple mid-circuit snapshots within a single copy of quantum circuit and there after classical storage of those quantum states."

    The paper's central claim is non-destructive single-copy state reconstruction, but the mechanism used to establish it is the assumption that the unknown target is preserved throughout repeated SWAP tests and mid-circuit measurements. That is the very property being claimed. The standard SWAP test entangles the target with ancilla and candidate, and measuring the ancilla can swap the candidate into the target register, so preservation is not automatic. The paper provides no back-action analysis. Independently, Section IV states experiments were run 'without mid-circuit measurements, reinitializing the circuit at each step,' so the reported data do not test the assumed preservation.

full rationale

There is no load-bearing self-citation chain: citation [16] by co-authors is used only as a QRAM reference, and no uniqueness theorem or prior-work ansatz is imported to force the QSDC choice. The circularity is internal. First, the simulation protocol uses the target state |ϕu⟩ directly inside the custom autograd function that computes the loss, and the final fidelity is the same expression against the same states; therefore the 'up to 0.999 across 100 random quantum states' claim is an in-sample training objective, not a prediction from an unknown state. Second, the single-copy, non-destructive snapshot capability is not derived from the dynamics of the SWAP test; it is assumed in Section III.A ('The method assumes qubit state preservation throughout hybrid optimization'), and Section IV confirms the experiments reinitialize the circuit at each step, bypassing mid-circuit measurement entirely. Section III.C.1 additionally labels the dynamic-circuit implementation as 'work in progress.' The manuscript is candid about the purity limitation and the mixed-state failure of SWAP-test fidelity, but that honesty does not repair the main assertion. The hardware demonstration on known single-qubit states is a valid proof-of-concept for fidelity-feedback optimization, yet it does not validate single-copy reconstruction. Overall, the central snapshot claim is partially circular: high fidelity reduces to the fitted training objective, and non-destructive single-copy access reduces to an unverified assumption.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The method's single-copy non-destructive claim rests on the unproven assumption that repeated SWAP tests preserve the target state. The optimization hyperparameters are hand-chosen and not shipped, and the dynamic-circuit timing requirements are aspirational.

free parameters (6)
  • ES learning rate alpha = 0.05
    Hand-chosen in Section III.C.2; affects convergence speed and final fidelity.
  • ES noise scale sigma = 0.1
    Hand-chosen in Section III.C.2; controls exploration range.
  • ES population size N = 50
    Hand-chosen in Section III.C.2; larger populations increase cost.
  • NN learning rate = 1e-4
    Hand-chosen in Section III.B.5; no learning rate schedule reported.
  • NN architecture (six linear layers, latent dim 256) = six linear layers, 256-dim latent vector
    Hand-chosen in Section III.B.3; no ablation or justification.
  • gradient scaling_factor = not reported
    Mentioned in Section III.B.5 as necessary for stability but the value is not given.
assumptions (4)
  • ad hoc to paper The unknown target state is preserved throughout repeated SWAP tests and qubit resets.
    Section III.A assumes 'qubit state preservation throughout hybrid optimization'; the standard SWAP test projects the joint state and does not leave the target untouched, so this premise is not established.
  • domain assumption Mid-circuit measurements and resets can be applied to candidate and ancilla qubits without disturbing the target qubits.
    Required by the dynamic circuit protocol in Fig. 2; current SDKs and hardware lack mature support, as the authors state in Section I.C.
  • domain assumption Classical model updates and circuit reinitialization complete within the qubit coherence time.
    The architecture in Fig. 3 requires td_cq + td_qc + tp_c << tau_d; this is an aspirational condition, not demonstrated.
  • standard math The SWAP-test measurement gives F = 2P(0) - 1 = |<psi|phi>|^2.
    Standard result used as the only feedback signal in Section III.B.4; not derived in the paper but widely accepted.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Capturing Quantum Snapshots from a Single Copy via Mid-Circuit Measurement and Dynamic Circuit." pith.science (2026). https://pith.science/paper/ZBVP7G2R

@misc{pith2026250421250,
  author       = {Pith},
  title        = {Pith review of: Capturing Quantum Snapshots from a Single Copy via Mid-Circuit Measurement and Dynamic Circuit},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZBVP7G2R}},
  note         = {Machine review of arXiv:2504.21250}
}
read the original abstract

We propose Quantum Snapshot with Dynamic Circuit (QSDC), a hardware-agnostic, learning-driven framework for capturing quantum snapshots: non-destructive estimates of quantum states at arbitrary points within a quantum circuit, which can then be classically stored and later reconstructed. This functionality is vital for introspection, debugging, and memory in quantum systems, yet remains fundamentally constrained by the no-cloning theorem and the destructive nature of measurement. QSDC introduces a guess-and-check methodology in which a classical model, powered by either gradient-based neural networks or gradient-free evolutionary strategie, is trained to reconstruct an unknown quantum state using fidelity from the SWAP test as the sole feedback signal. Our approach supports single-copy, mid-circuit state reconstruction, assuming hardware with dynamic circuit support and sufficient coherence time. We validate core components of QSDC both in simulation and on IBM quantum hardware. In noiseless settings, our models achieve average fidelity up to 0.999 across 100 random quantum states; on real devices, we accurately reconstruct known single-qubit states (e.g., Hadamard) within three optimization steps.

Figures

Figures reproduced from arXiv: 2504.21250 by the authors.

Figure 1
Figure 1. Overview of Quantum Snapshot: Quantum states are observed (snapshots) at various points within a single circuit at multiple intersections, without collapsing the original system. This enables non-destructive access to quantum information during circuit execution. arXiv:2504.21250v1 [quant-ph] 30 Apr 2025 [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Quantum-Centric Supercomputing Architecture for Single-Copy State Reconstruction. This architecture enables non-destructive quantum state estimation by tightly integrating quantum and classical processors. An unknown quantum state is compared against a generated guess using a SWAP test. Fidelity feedback is processed classically to refine the guess, all within the qubit coherence time τd. Key time components include… view at source ↗
Figure 4
Figure 4. Dynamic Quantum Circuit with mid-circuit measurement circuits interleave quantum gates, mid-circuit measurements, and classical processing within the hardware’s coherence time, enabling conditional branching and greater expressiveness. B. Mid-Circuit Measurement Mid-circuit measurement [19] involves measuring selected qubits during the execution of a quantum circuit, rather than only at the end. These measurements p… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 15 canonical work pages

  1. [1]

    Quantum memories: emerging applications and recent advances,

    K. Heshami, D. G. England, P. C. Humphreys, P. J. Bustard, V . M. Acosta, J. Nunn, and B. J. Sussman, “Quantum memories: emerging applications and recent advances,” Journal of modern optics , vol. 63, no. 20, pp. 2005–2028, 2016

    work page 2005

  2. [2]

    Feedback control of a solid-state qubit using high-fidelity projective measurement,

    D. Rist `e, C. Bultink, K. W. Lehnert, and L. DiCarlo, “Feedback control of a solid-state qubit using high-fidelity projective measurement,” Physical review letters, vol. 109, no. 24, p. 240502, 2012

    work page 2012

  3. [3]

    Character- izing large-scale quantum computers via cycle benchmarking,

    A. Erhard, J. J. Wallman, L. Postler, M. Meth, R. Stricker, E. A. Martinez, P. Schindler, T. Monz, J. Emerson, and R. Blatt, “Character- izing large-scale quantum computers via cycle benchmarking,” Nature communications, vol. 10, no. 1, p. 5347, 2019

    work page 2019

  4. [4]

    Real-time dynamics of open quantum spin systems driven by dissipative processes,

    F. Hebenstreit, D. Banerjee, M. Hornung, F.-J. Jiang, F. Schranz, and U.-J. Wiese, “Real-time dynamics of open quantum spin systems driven by dissipative processes,” Physical Review B, vol. 92, no. 3, p. 035116, 2015

    work page 2015

  5. [5]

    Efficient Quantum State Sample Tomography with Basis-Dependent Neural Networks,

    A. W. R. Smith, J. Gray, and M. S. Kim, “Efficient Quantum State Sample Tomography with Basis-Dependent Neural Networks,” PRX Quantum, vol. 2, p. 020348, June 2021. Publisher: American Physical Society

    work page 2021

  6. [6]

    Quantum decoherence,

    M. Schlosshauer, “Quantum decoherence,” Physics Reports , vol. 831, pp. 1–57, 2019

    2019

  7. [7]

    What are the ultimate limits of photonic quantum memories?,

    M. G ¨undo˘gan and D. K. Oi, “What are the ultimate limits of photonic quantum memories?,” Research Directions: Quantum Technologies , vol. 1, p. e8, 2023

    work page 2023

  8. [8]

    The quantum internet,

    H. J. Kimble, “The quantum internet,” Nature, vol. 453, no. 7198, pp. 1023–1030, 2008

    2008

Show all 23 references
  1. [9]

    Quantum repeaters based on atomic ensembles and linear optics,

    N. Sangouard, C. Simon, H. De Riedmatten, and N. Gisin, “Quantum repeaters based on atomic ensembles and linear optics,” Reviews of Modern Physics, vol. 83, no. 1, pp. 33–80, 2011

    2011

  2. [10]

    Quantum internet: A vision for the road ahead,

    S. Wehner, D. Elkouss, and R. Hanson, “Quantum internet: A vision for the road ahead,” Science, vol. 362, no. 6412, p. eaam9288, 2018

    2018

  3. [11]

    What is quantum-centric supercomput- ing?,

    J. Schneider and I. Smalley, “What is quantum-centric supercomput- ing?,” 2024. Accessed: Apr. 12, 2025

    2024

  4. [12]

    Classical feedforward and control flow (a.k.a. dynamic circuits),

    IBM Quantum, “Classical feedforward and control flow (a.k.a. dynamic circuits),” 2025. Accessed: Apr. 12, 2025

    2025

  5. [13]

    Repeat until success,

    IBM Quantum, “Repeat until success,” 2025. Accessed: Apr. 12, 2025

    2025

  6. [14]

    Introduction to mid-circuit measurements,

    PennyLane Team, “Introduction to mid-circuit measurements,” 2025. Accessed: Apr. 12, 2025

    2025

  7. [15]

    Quantum random access memory,

    V . Giovannetti, S. Lloyd, and L. Maccone, “Quantum random access memory,” Physical Review Letters , vol. 100, p. 160501, Apr. 2008. arXiv:0708.1879 [quant-ph]

    2008 arXiv

  8. [16]

    Quantum Random Access Memory For Dummies,

    K. Phalak, A. Chatterjee, and S. Ghosh, “Quantum Random Access Memory For Dummies,” May 2023. arXiv:2305.01178 [quant-ph]

    2023 arXiv

  9. [17]

    Introducing Catalyst: quantum just-in-time compilation | PennyLane Blog

    “Introducing Catalyst: quantum just-in-time compilation | PennyLane Blog.”

  10. [18]

    Classical feedforward and control flow (a.k.a. dynamic circuits)

    “Classical feedforward and control flow (a.k.a. dynamic circuits).”

  11. [19]

    How to create dynamic circuits with mid-circuit measure- ments,

    D. Wierichs, “How to create dynamic circuits with mid-circuit measure- ments,” PennyLane Demos, May 2024. Publisher: Xanadu

    2024

  12. [20]

    M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information: 10th Anniversary Edition . Cambridge University Press, 2010

    2010

  13. [21]

    Transformation of quantum states using uniformly controlled rotations,

    M. Mottonen, J. J. Vartiainen, V . Bergholm, and M. M. Salomaa, “Transformation of quantum states using uniformly controlled rotations,” July 2004. arXiv:quant-ph/0407010

    2004 arXiv

  14. [22]

    Evolution Strategies as a Scalable Alternative to Reinforcement Learning,

    T. Salimans, J. Ho, X. Chen, S. Sidor, and I. Sutskever, “Evolution Strategies as a Scalable Alternative to Reinforcement Learning,” Sept

  15. [2017]

    arXiv:1703.03864 [stat]

    arXiv

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.