Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

MonoPartNeRF:Human Reconstruction from Monocular Video via Part-Based Neural Radiance Fields

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Randomized weak measurements make quantum diffusion reversible with provable error bounds.

desk verdict The PDF and metadata are different papers, and the quantum paper's central training loss (Eq. 27) has vanishing gradient with respect to the control Hamiltonian, so the headline equivalence is unsupported as written. read the letter →

arxiv 2508.08798 v1 pith:I43NX2AM submitted 2025-08-12 cs.CV

Yao Lu , Jiawei Li , Ming Jiang This is my paper · ORCID
classification cs.CV MSC 81P6881S25 PACS 03.67.-a03.65.Ta
keywords measurement-basedquantumdiffusionrandomizedweakmeasurementtrajectoryscorematchingPetzrecoverymapclassicalshadowtomographystategenerationLindbladdynamics
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

Measurement-based quantum diffusion claims that the forward noising of a quantum state can be realized physically by randomized weak measurements, and that reverse generation splits into two problems with provable solutions: recovering individual pure-state trajectories and recovering the ensemble-averaged state. For trajectories, quantum score matching is shown to be equivalent to learning a unitary generator, so the missing principled training objective for quantum diffusion becomes a control problem. For ensembles, the paper proves that Petz recovery maps reverse the measurement-induced Lindblad evolution, with local Petz maps giving finite-depth recovery under a finite-Markov-length condition and classical shadows giving an exact inversion of the Pauli-twirled channel. If correct, this gives quantum state generation the same theoretical footing as classical diffusion: derived objectives, certified recovery errors, and a clear path to hardware implementation.

What carries the argument

The machinery has three linked pieces. First, forward diffusion is generated by randomized weak measurements, with per-step Kraus operators that factorize over qubits; because the measurement observable distribution is invariant under local Clifford unitaries, the averaged channel is Pauli-twirled and therefore diagonal in the Pauli basis, so all decoherence information is encoded in closed-form Pauli weights. Second, the reverse pure-state process is a unitary flow, and the classical score–flow identity is lifted to Hilbert space: the score operator $\mathcal{S}_\theta(|\psi\rangle,t)=\exp[(-\gamma\,\delta O^2/2 + 2i H_\theta(|\psi\rangle,t))\,dt]$ implements one reverse step, and optimizing a denoising score-matching loss on trajectories is proved equivalent to learning the generator $H_\theta$. Third, ensemble reversal uses the Petz recovery map, a channel that inverts a quantum evolution using the state itself as a prior, in a twirled integral form; under a finite-Markov-length condition on the conditional mutual information, local Petz maps can be stacked into a finite-depth recovery circuit, while classical shadow tomography provides a device-agnostic alternative via the inverse of the Pauli-twirled measurement-and-prepare channel.

What would settle it

Run the local Petz recovery protocol on a 10-qubit transverse-field Ising chain at the critical point $h_x = 1.0$, where the correlation length diverges and the exponential-CMI assumption fails; a sharp drop in recovery fidelity would confirm the theorem's limitation, while high fidelity would show the assumption is not load-bearing. For trajectory-level recovery, train the control Hamiltonian on two-qubit spin-singlet measurement records and check whether the Wasserstein-1 distance shrinks to zero as the training error goes to zero; if it does not, the score-matching/controller equivalence is falsified.

Watch

Extended reading notes

Core claim

The central claim is that a quantum diffusion model can be built entirely from measurement: randomized weak measurements, drawn from single-qubit Pauli observables, drive any initial ensemble toward the uniform maximally mixed product distribution while each conditional trajectory stays pure; the averaged state obeys the Lindblad master equation $\partial_t \bar\rho = \mathcal{L}[\bar\rho]$ with exponential decay of Pauli weights $w_\mu(t)=\exp(-4\gamma t\, \mathrm{supp}(\mu)/(3n))$. The reverse of a pure-state trajectory must be a deterministic unitary flow, and the paper proves that the denoising score-matching objective over trajectories is equivalent to learning the control Hamiltonian of that flow, with a Wasserstein-1 bound that vanishes as the training error goes to zero. For ensemble-average recovery, the paper introduces local Petz recovery maps, channels that invert the weak measurement step using the reduced state as a prior, and proves that, when every intermediate state has finite Markov length, a sequence of such maps recovers the initial state within trace distance $\epsilon$; for general states, classical shadow reconstruction inverts the measurement-and-prepare channel exactly, with sample complexity governed by a shadow norm. Finally, in the large-spin separable limit the Petz map reduces to the classical backward Fokker–Planck equation, establishing the announced bridge between quantum recovery channels and classical stochastic reversal.

Load-bearing premise

For the ensemble-recovery half of the paper, the load-bearing premise is that every intermediate state has exponentially decaying conditional mutual information with distance; for general many-body targets that need not hold, and without it the local Petz protocol has no proven error guarantee.

Editorial extensions

If this is right

  • Quantum diffusion models acquire a derived training objective: denoising score matching on pure-state trajectories is equivalent to learning a control Hamiltonian, replacing heuristic loss functions.
  • Ensemble-average recovery can be performed without learning when correlations are short-ranged, using local Petz maps constructed directly from measurement data and knowledge of the forward channel.
  • Classical shadow reconstruction recovers the initial average state from weak-measurement records with rigorous concentration bounds, giving a purely classical post-processing route to quantum state estimation.
  • Petz recovery maps and classical backward Fokker–Planck diffusion coincide in the large-spin separable limit, so quantum and classical denoising share the same reverse dynamics.
  • The error theorems supply conditions under which state-generation fidelity can be certified: sufficiently small training error, sufficiently long diffusion time, and exponentially decaying conditional mutual information along the trajectory.

Reading between the lines

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

  • Editorial inference: the Pauli-twirled channel structure likely extends to other locally scrambled measurement ensembles, such as Clifford or unitary-2-design protocols, with the reconstruction map set by the corresponding frame potential; the paper only sketches this direction.
  • Editorial inference: the finite-Markov-length condition provides a practical diagnostic for whether a target state is locally recoverable; states with long-range entanglement, such as GHZ-type or critical states, should saturate the trace-distance error bound.
  • Editorial inference: the trajectory-level bottleneck is the classical decoder that infers the latent pure state from the measurement record; a quantum controller using the record directly could remove the post-selection overhead, though the paper leaves this implementation open.
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 / 5 minor

Summary. The manuscript arXiv:2508.08799 presents a theoretical framework for measurement-based quantum diffusion models. Randomized weak measurements define a forward process that preserves pure states along individual trajectories while depolarizing the ensemble average. The paper targets two recovery tasks: trajectory-level recovery of pure-state ensembles via a learned control Hamiltonian, which is claimed to be equivalent to quantum score matching, and ensemble-average recovery via classical shadow reconstruction and local Petz recovery maps, with error bounds in both cases. A classical large-spin limit is used to show that Petz recovery reduces to backward Fokker-Planck diffusion. The body of the paper is a quantum-information manuscript with derivations in a supplemental material; the abstract and title, however, describe a computer-vision method for monocular human reconstruction (MonoPartNeRF) with experiments on ZJU-MoCap and MonoCap that do not appear in the body.

Significance. The paper offers several analytically valuable results: closed-form Pauli weights for the measurement and measurement-and-prepare channels (Eqs. (22), (40)), an exact channel inversion for classical shadow reconstruction with a shadow-norm sample complexity (Eqs. (42)-(47)), and a derivation connecting Petz recovery to classical backward diffusion in the large-spin limit (Sec. III C 4). These components are self-contained and largely machine-checkable, and the local Petz protocol is a natural application of finite-Markov-length recovery ideas. However, the central trajectory-level claim—that the training objective Eq. (27) is equivalent to score matching and trains a unitary reverse generator—is not supported by the printed equation, and the error bounds omit the classical decoder's estimation error. The significance of the framework would be high if these gaps are repaired.

major comments (4)
  1. [Title/Abstract vs. Body] The abstract and title describe 'MonoPartNeRF: Human Reconstruction from Monocular Video via Part-Based Neural Radiance Fields' and promise experiments on ZJU-MoCap and MonoCap, but the body is a quantum physics paper with no such dataset experiments. As submitted, the manuscript does not support its stated abstract claims, and the quantum results are not discoverable from the title and abstract. This must be corrected before the paper can be considered.
  2. [§III A 2, Eq. (27) and SM Eq. (A36)] The printed mean-infidelity loss L_θ = ∫dt E[1 - |⟨ψ_{t+dt}| S_θ(ψ̂_{t+dt}) |ψ_{t+dt}⟩|^2] with S_θ = exp[(-γ/2 δO^2 + 2iH_θ)dt] is stationary in H_θ to leading order in dt: the anti-Hermitian part contributes only an imaginary phase to the overlap, so its squared modulus is independent of H_θ at O(dt). The optimum of this objective is thus H_θ = 0, and it cannot implement the claimed equivalence between score matching and learning reverse unitary generators. Moreover, S_θ is non-unitary, so the claim that the reverse process is generated by applying S_θ^† does not follow. The hypothesis of Theorem 1 (SM Eq. A39) is stated for a learned unitary U_θ, but the object trained by Eq. (27) is not unitary and is not shown to satisfy that assumption. If the bra in Eq. (27) is intended to be ⟨ψ_t| rather than ⟨ψ_{t+dt}|, the argument may be repairable; as printed, the central trajectory-level guarantee is unsupported.
  3. [§III A 3 and SM Theorem 1] The Wasserstein error bound does not include the classical decoder's estimation error. The loss in Eq. (27) and the control model are functions of the decoded state ψ̂_{t+dt}, while the theorem's per-step fidelity ε and Lipschitz condition are stated for the true forward-diffused state ψ_{t+dt}. No term bounds the distance between ψ̂_{t+dt} and ψ_{t+dt} in terms of the measurement record or the shadow estimator. The end-to-end guarantee is therefore incomplete; the authors note this in the limitations, but the theorem as stated is stronger than what is proven.
  4. [§III C 2 and SM Eq. (D14)] The local Petz recovery guarantee assumes an exponential CMI decay bound I(A:C|B) ≤ poly(...) e^{-dist(A,C)/ξ} for every intermediate state ρ_t (0≤t≤T) and every subsystem S_j. This is not established for general target states nor for the measurement dynamics; for example, a measurement on a qubit can temporarily increase certain bipartite correlations, and the paper does not prove that the finite-Markov-length property propagates. The protocol's step (ii) also constructs the recovery maps from a shadow estimate ρ̃_0 rather than the true ρ_0; the resulting estimation error is not included in Theorem 2. Thus the claim that the final reconstructed state satisfies ||ρ̃_0 - ρ_0||_1 ≤ ε is conditional on an unverified locality assumption plus an additional estimation error.
minor comments (5)
  1. [§II E and §III B 3] The notation w_P^F(t) and w_P^M(t) are both called 'Pauli weight' in different sections; please rename one (e.g., measurement-channel weight vs. measurement-and-prepare-channel weight) to avoid ambiguity.
  2. [§III A 4, Fig. 4] The caption for Fig. 4(b) mentions a 'thermal state ensemble of a two-qubit Heisenberg model' but does not specify the inverse temperature or the sampling procedure; please add details so the numerical experiment is reproducible.
  3. [§III A 2 and SM Eq. (A36)] The object S_θ in Eq. (28) is called a 'score operator,' but it is not a score function in the classical sense; please clarify its relation to the classical score or rename it to avoid conceptual confusion.
  4. [Introduction, Note Added] The complementary work by Hu et al. [41] is mentioned only in the Note Added; it should be cited and discussed in the introduction where the local recovery mechanisms are introduced.
  5. [References] Reference [12] in the bibliography appears to cite a classical arXiv paper (2505.18621) that is not discussed in the text; please verify that all references are cited in the main text.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the derivation chain is self-contained, with Pauli weights, shadow norms, and Petz-recovery guarantees derived analytically or from externally supported results.

full rationale

The paper's central quantities are derived, not fitted: the Pauli weights and shadow norms follow from an explicit Pauli-transfer-matrix calculation (SM Appendix B), the measurement channel is diagonalized analytically, and the classical-shadow reconstruction map is the closed-form inverse of that channel. The trajectory-level 'quantum score matching to unitary generator' claim is a reparameterization of the probability-flow ODE in which the flow velocity is written as a control Hamiltonian; while the printed mean-infidelity loss in Eq. (27) raises a genuine correctness concern about the leading-order gradient with respect to H_theta, that is a mathematical-support issue rather than a circularity, since the objective is not a fitted parameter renamed as a prediction. The Petz-recovery results are obtained from the standard Petz formula, with the finite-Markov-length assumption stated explicitly and the locality proof following the external Sang--Hsieh result (Ref. [40]); no uniqueness theorem or load-bearing premise is imported solely from the authors' own prior work. Self-citations to locally scrambled dynamics and classical shadow tomography ([28], [29], [43]) are used as supporting tools whose relevant properties are re-derived in the Supplemental Material, so they are not load-bearing. Overall, the derivations are self-contained and no circular reduction of a central claim to its own inputs is present.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The central derivation is analytic and uses no fitted parameters. The key extra-mathematical assumptions are the finite Markov length condition for local Petz recovery (Theorem 2) and the standard large-spin classical limit used to establish the Petz-reverse-Fokker-Planck correspondence.

assumptions (3)
  • standard math Randomized weak measurements in the continuous-time limit follow a nonlinear SDE with Wiener noise.
    Standard quantum trajectory theory (Refs. [22-25]); used to derive forward diffusion in Sec. II.
  • domain assumption All intermediate states rho_t along the diffusion have finite Markov length: CMI decays exponentially with distance.
    Assumed in Theorem 2 (Sec. III C, Eq. D14 of SM) to justify local Petz recovery; not proven for general states.
  • standard math In the large-spin limit, coherent states on different Bloch sphere directions become orthogonal, and Glauber-Sudarshan P multiplication reduces to ordinary multiplication.
    Used in SM Appendix C to take the classical limit of the Petz map; a standard asymptotic approximation in quantum optics.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MonoPartNeRF:Human Reconstruction from Monocular Video via Part-Based Neural Radiance Fields." pith.science (2026). https://pith.science/paper/I43NX2AM

@misc{pith2026250808798,
  author       = {Pith},
  title        = {Pith review of: MonoPartNeRF:Human Reconstruction from Monocular Video via Part-Based Neural Radiance Fields},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I43NX2AM}},
  note         = {Machine review of arXiv:2508.08798}
}
read the original abstract

In recent years, Neural Radiance Fields (NeRF) have achieved remarkable progress in dynamic human reconstruction and rendering. Part-based rendering paradigms, guided by human segmentation, allow for flexible parameter allocation based on structural complexity, thereby enhancing representational efficiency. However, existing methods still struggle with complex pose variations, often producing unnatural transitions at part boundaries and failing to reconstruct occluded regions accurately in monocular settings. We propose MonoPartNeRF, a novel framework for monocular dynamic human rendering that ensures smooth transitions and robust occlusion recovery. First, we build a bidirectional deformation model that combines rigid and non-rigid transformations to establish a continuous, reversible mapping between observation and canonical spaces. Sampling points are projected into a parameterized surface-time space (u, v, t) to better capture non-rigid motion. A consistency loss further suppresses deformation-induced artifacts and discontinuities. We introduce a part-based pose embedding mechanism that decomposes global pose vectors into local joint embeddings based on body regions. This is combined with keyframe pose retrieval and interpolation, along three orthogonal directions, to guide pose-aware feature sampling. A learnable appearance code is integrated via attention to model dynamic texture changes effectively. Experiments on the ZJU-MoCap and MonoCap datasets demonstrate that our method significantly outperforms prior approaches under complex pose and occlusion conditions, achieving superior joint alignment, texture fidelity, and structural continuity.

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Executive Accountability Systems and the Environmental Violations of State-Owned Enterprises in China

    econ.GN 2025-08 unverdicted novelty 5.0 of 10

    China's executive accountability policy (ASIOI) reduced recorded environmental violations by state-owned enterprises, with effects attributed to internal controls, green investment, and green innovation.

Reference graph

Works this paper leans on

61 extracted references · 24 canonical work pages · cited by 1 Pith paper

  1. [1]

    J. Ho, A. Jain, and P. Abbeel, Denoising Diffusion Proba- bilistic Models, arXiv e-prints , arXiv:2006.11239 (2020), arXiv:2006.11239 [cs.LG]

    arXiv 2020

  2. [2]

    Y. Song, J. Sohl-Dickstein, D. P. Kingma, A. Kumar, S. Ermon, and B. Poole, Score-Based Generative Mod- eling through Stochastic Differential Equations, arXiv e-prints , arXiv:2011.13456 (2020), arXiv:2011.13456 [cs.LG]

    arXiv 2020

  3. [3]

    Numerical Demonstration To illustrate the proposed protocol, we perform a nu- merical experiment on a 10-qubit transverse-field Ising chain with open boundary conditions, �=�� � � ��������+1 �� � � � �����(49) using�= 1�0, and� � = 1�5�2�0 and 5�0 as in Fig. 5. This choice places the system well away from the criti- cal point (� � = 1�0), ensuring a finit...

    work page

  4. [4]

    Petz Recovery as Time-Reversed Diffusion We establish that the Petz recovery map provides a quantum generalization of classical time-reversed diffu- sion under weak measurements, and that in the classical large-spin limit the two become equivalent. Consider an�-qubit separable state (no quantum en- tanglement), expressed as a mixture of tensor product of ...

    work page

  5. [5]

    Cacioppo, L

    A. Cacioppo, L. Colantonio, S. Bordoni, and S. Gi- agu, Quantum Diffusion Models, arXiv e-prints , arXiv:2311.15444 (2023), arXiv:2311.15444 [quant-ph]

    arXiv 2023

  6. [6]

    Dhariwal and A

    P. Dhariwal and A. Nichol, Diffusion Models Beat GANs on Image Synthesis, arXiv e-prints , arXiv:2105.05233 12 (2021), arXiv:2105.05233 [cs.LG]

    arXiv 2021

  7. [7]

    Parigi, S

    M. Parigi, S. Martina, and F. Caruso, Quantum-Noise- Driven Generative Diffusion Models, arXiv e-prints , arXiv:2308.12013 (2023), arXiv:2308.12013 [quant-ph]

    arXiv 2023

  8. [8]

    Zhang, P

    B. Zhang, P. Xu, X. Chen, and Q. Zhuang, Genera- tive Quantum Machine Learning via Denoising Diffu- sion Probabilistic Models, Phys. Rev. Lett.132, 100602 (2024), arXiv:2310.05866 [quant-ph]

    arXiv 2024

Show all 61 references
  1. [9]

    C. Chen, Q. Zhao, M. Zhou, Z. He, Z. Sun, and H. Situ, Quantum Generative Diffusion Model: A Fully Quantum-Mechanical Model for Generating Quan- tum State Ensemble, arXiv e-prints , arXiv:2401.07039 (2024), arXiv:2401.07039 [quant-ph]

    2024 arXiv

  2. [10]

    K¨ olle, G

    M. K¨ olle, G. Stenzel, J. Stein, S. Zielinski, B. Om- mer, and C. Linnhoff-Popien, Quantum Denoising Dif- fusion Models, arXiv e-prints , arXiv:2401.07049 (2024), arXiv:2401.07049 [quant-ph]

    2024 arXiv

  3. [11]

    G. Kwun, B. Zhang, and Q. Zhuang, Mixed-state quan- tum denoising diffusion probabilistic model, Phys. Rev. A111, 032610 (2025), arXiv:2411.17608 [quant-ph]

    2025 arXiv

  4. [12]

    Y. Zhu, T. Chen, E. A. Theodorou, X. Chen, and M. Tao, Quantum State Generation with Structure-Preserving Diffusion Model, arXiv e-prints , arXiv:2404.06336 (2024), arXiv:2404.06336 [quant-ph]

    2024 arXiv

  5. [13]

    Y. Tang, M. Long, and J. Yan, Quadim: A conditional diffusion model for quantum state property estimation, inThe Thirteenth International Conference on Learning Representations(2025)

    2025

  6. [14]

    Zhang and C

    X. Zhang and C. Chen, Parameter-efficient quantum denoising diffusion probabilistic models with temporal encoding, Future Generation Computer Systems174, 107981 (2026)

    2026

  7. [15]

    J. Liu, R. Li, H. Wang, Z. Yu, C. Liu, J. Ding, and Y. Li, Beyond Equilibrium: Non-Equilibrium Founda- tions Should Underpin Generative Processes in Complex Dynamical Systems, arXiv e-prints , arXiv:2505.18621 (2025), arXiv:2505.18621 [cs.CE]

    2025 arXiv

  8. [16]

    Huang, C

    H. Huang, C. Chen, and Q. Zhao, Continuous-variable Quantum Diffusion Model for State Generation and Restoration, arXiv e-prints , arXiv:2506.19270 (2025), arXiv:2506.19270 [quant-ph]

    2025 arXiv

  9. [17]

    W. W. Ho and S. Choi, Exact Emergent Quantum State Designs from Quantum Chaotic Dynamics, Phys. Rev. Lett.128, 060601 (2022), arXiv:2109.07491 [quant-ph]

    2022 arXiv

  10. [18]

    Z. Cui, P. Zhang, and Y. Tang, Quantum Flow Matching, arXiv e-prints , arXiv:2508.12413 (2025), arXiv:2508.12413 [quant-ph]

    2025 arXiv

  11. [19]

    L. Yang, Z. Zhang, Y. Song, S. Hong, R. Xu, Y. Zhao, W. Zhang, B. Cui, and M.-H. Yang, Diffusion Mod- els: A Comprehensive Survey of Methods and Ap- plications, arXiv e-prints , arXiv:2209.00796 (2022), arXiv:2209.00796 [cs.LG]

    2022

  12. [20]

    Zhang, P

    B. Zhang, P. Xu, X. Chen, and Q. Zhuang, Holographic deep thermalization for secure and efficient quantum ran- dom state generation, arXiv e-prints , arXiv:2411.03587 (2024), arXiv:2411.03587 [quant-ph]

    2024 arXiv

  13. [21]

    J. S. Cotler, D. K. Mark, H.-Y. Huang, F. Hern´ andez, J. Choi, A. L. Shaw, M. Endres, and S. Choi, Emer- gent Quantum State Designs from Individual Many- Body Wave Functions, PRX Quantum4, 010311 (2023), arXiv:2103.03536 [quant-ph]

    2023 arXiv

  14. [22]

    J. Choi, A. L. Shaw, I. S. Madjarov, X. Xie, R. Finkel- stein, J. P. Covey, J. S. Cotler, D. K. Mark, H.-Y. Huang, A. Kale, H. Pichler, F. G. S. L. Brand˜ ao, S. Choi, and M. Endres, Preparing random states and benchmarking with many-body quantum chaos, Nature (London)613, 468...

    2023 arXiv

  15. [23]

    S. L. Adler, A. Bassi, and S. Donadi, On spontaneous photon emission in collapse models, Journal of Physics A Mathematical General46, 245304 (2013)

    2013

  16. [24]

    G. Lima, E. Filatovas, M. Marcozzi, and R. Paulaviˇ cius, A review of quantum-based diffusion models in genera- tive ai, Vilnius University Open Series , 109 (2025)

    2025

  17. [25]

    G. C. Ghirardi, P. Pearle, and A. Rimini, Markov pro- cesses in Hilbert space and continuous spontaneous lo- calization of systems of identical particles, Phys. Rev. A 42, 78 (1990)

    1990

  18. [26]

    Dankert, R

    C. Dankert, R. Cleve, J. Emerson, and E. Livine, Exact and approximate unitary 2-designs and their application to fidelity estimation, Phys. Rev. A80, 012304 (2009), arXiv:quant-ph/0606161 [quant-ph]

    2009 arXiv

  19. [27]

    Giachetti and A

    G. Giachetti and A. De Luca, Elusive phase transi- tion in the replica limit of monitored systems, arXiv e-prints , arXiv:2306.12166 (2023), arXiv:2306.12166 [cond-mat.stat-mech]

    2023 arXiv

  20. [28]

    Jacobs and D

    K. Jacobs and D. A. Steck, A straightforward intro- duction to continuous quantum measurement, Contem- porary Physics47, 279 (2006), arXiv:quant-ph/0611067 [quant-ph]

    2006 arXiv

  21. [29]

    H.-Y. Hu, S. Choi, , and Y.-Z. You, Classical shadow tomography with locally scrambled quantum dynamics, Physical Review Research5, 023027 (2023)

    2023

  22. [30]

    Magesan, J

    E. Magesan, J. M. Gambetta, and J. Emerson, Scal- able and Robust Randomized Benchmarking of Quan- tum Processes, Phys. Rev. Lett.106, 180504 (2011), arXiv:1009.3639 [quant-ph]

    2011 arXiv

  23. [31]

    W.-T. Kuo, A. Akhtar, D. P. Arovas, , and Y.-Z. You, Markovian entanglement dynamics under locally scram- bled quantum evolution, Physical Review B101, 224202 (2020)

    2020

  24. [32]

    V. V. Albert, Lindbladians with multiple steady states: theory and applications, arXiv e-prints , arXiv:1802.00010 (2018), arXiv:1802.00010 [quant-ph]

    2018 arXiv

  25. [33]

    Buˇ ca and T

    B. Buˇ ca and T. Prosen, A note on symmetry reductions of the Lindblad equation: transport in constrained open spin chains, New Journal of Physics14, 073007 (2012), arXiv:1203.0943 [quant-ph]

    2012 arXiv

  26. [34]

    V. V. Albert and L. Jiang, Symmetries and conserved quantities in Lindblad master equations, Phys. Rev. A 89, 022118 (2014), arXiv:1310.1523 [quant-ph]

    2014 arXiv

  27. [35]

    See Supplemental Material for details

  28. [36]

    A. Y. Tong, N. Malkin, K. Fatras, L. Atanack- ovic, Y. Zhang, G. Huguet, G. Wolf, and Y. Bengio, Simulation-Free Schr¨ odinger Bridges via Score and Flow Matching, inProceedings of The 27th International Con- ference on Artificial Intelligence and Statistics, Vol. 238 (2024) p...

    2024 arXiv

  29. [37]

    C. Lu, K. Zheng, F. Bao, J. Chen, C. Li, and J. Zhu, Maximum Likelihood Training for Score-based Diffu- sion ODEs by High Order Denoising Score Matching, inProceedings of the 39th International Conference on Machine Learning, Vol. 162 (2022) pp. 14429–14460, arXiv:2206.08265 [cs.LG]

    2022 arXiv

  30. [38]

    Kwon and M

    H. Kwon and M. Kim, Fluctuation theorems for a quantum channel, Physical Review X9, 10.1103/phys- revx.9.031029 (2019)

    2019 doi

  31. [39]

    Huang, R

    H.-Y. Huang, R. Kueng, and J. Preskill, Predicting many properties of a quantum system from very few measure- 13 ments, Nature Physics16, 1050 (2020), arXiv:2002.08953 [quant-ph]

    2020 arXiv

  32. [40]

    (iv)Implement reverse diffusion.Each step of �� � is a local quantum channel that can, in principle, be implemented on quantum hardware

    for the detailed construction). (iv)Implement reverse diffusion.Each step of �� � is a local quantum channel that can, in principle, be implemented on quantum hardware. Initialize the device in a state ¯� � � (e.g., a random prod- uct state evolved forward to time�), then appl...

  33. [41]

    Petz, Sufficient subalgebras and the relative entropy of states of a von neumann algebra, Communications in mathematical physics105, 123 (1986)

    D. Petz, Sufficient subalgebras and the relative entropy of states of a von neumann algebra, Communications in mathematical physics105, 123 (1986)

    1986

  34. [42]

    H. Kwon, R. Mukherjee, and M. S. Kim, Reversing Lind- blad Dynamics via Continuous Petz Recovery Map, Phys. Rev. Lett.128, 020403 (2022), arXiv:2104.03360 [quant- ph]

    2022 arXiv

  35. [43]

    Sang and T

    S. Sang and T. H. Hsieh, Stability of mixed-state quantum phases via finite markov length (2024), arXiv:2404.07251 [quant-ph]

    2024

  36. [44]

    F. Hu, G. Liu, Y. Zhang, and X. Gao, Local Diffusion Models and Phases of Data Distributions, arXiv e-prints , arXiv:2508.06614 (2025), arXiv:2508.06614 [stat.ML]

    2025 arXiv

  37. [45]

    Jacobs and D

    K. Jacobs and D. A. Steck, A straightforward introduc- tion to continuous quantum measurement, Contempo- rary Physics47, 279 (2006)

    2006

  38. [46]

    A. A. Akhtar, H.-Y. Hu, and Y.-Z. You, Measurement- induced criticality is tomographically optimal (2023), arXiv:2308.01653 [quant-ph]. 14 Supplemental Material Appendix A: Derivations for the Score Matching of the General Diffusion Dynamics This appendix provides a detailed, ...

    2023 arXiv

  39. [47]

    General Diffusion Dynamics a. The generic SDE, Fokker–Planck PDE, and ODE Descriptions A generalized diffusion process for a state vector�(�) in an�dimensional space can be described by a stochastic differential equation (SDE) with a�-dimensional and state-dependent noise. The...

  40. [48]

    The Weak Measurement Protocol We consider a quantum system initially prepared in the state�0

    The nonlinear SDE from the Weak Measurement Protocol a. The Weak Measurement Protocol We consider a quantum system initially prepared in the state�0. Ancilla-assisted weak measurement of an observable � � is implemented by coupling the system to an ancilla qubit, prepared in�0...

  41. [49]

    From learning unitary to learning score function For an ensemble of pure states, the forward diffusion process implemented by weak measurement must be purity- preserving

    The Unitary Reverse Procedure for Pure States a. From learning unitary to learning score function For an ensemble of pure states, the forward diffusion process implemented by weak measurement must be purity- preserving. As is clear from Eq. (A19), a pure state remains pure alo...

  42. [50]

    The measurement channel We next examine the generic weak measurement protocol of Sec

    The Measurement channel and the linear SDE a. The measurement channel We next examine the generic weak measurement protocol of Sec. A 2 a in the context of qubit systems. We consider a system of�qubits, initialized in a state� 0. The forward diffusion is implemented by applyin...

  43. [51]

    The Measurement-and-prepare channel a. Definition of the channel We now turn to the measurement-and-prepare channel, which is essential for our method to extract weak measure- ment shadow tomography from the outcomes gathered through weak measurements. We define the following ...

  44. [52]

    Error and shadow norm We recover the initial state� 0 by defining shadow of the measured trajectories�: ˆ�0�� =� �1 � (��(�))�(B46) 28 � � � � � � ��� ��� ��� ��� ��� ��� ������� ����� � ���� ������ ���� ��� ����� � ��� ��� ��� � ��� ��� ��� ��� ��� ��� ��� ��� ��� ��� ��� ���...

  45. [53]

    (B46) can be calculated efficiently due to the nice properties we showed earlier

    Efficient calculation of the shadow The shadow in Eq. (B46) can be calculated efficiently due to the nice properties we showed earlier. First, recall that the Kraus operator� �(�) associated with trajectory�is a product of Kraus operators over each individual qubits: � �(�) =�...

  46. [54]

    probability

    The measurement channel and the forward diffusion We first focus on the forward evolution of the system by a sequence of weak measurements. 30 We consider a system of�qubits. As introduced in the main text, the measurement channel� � describes the averaged state of all the wea...

  47. [55]

    The Petz recovery channel and the backward diffusion We now consider the Petz recovery channel of the measurement channel. For states� � =� �(�0) evolved by the measurement channel� � defined above, the Petz recovery channel reversing from final time�to time���is defined as � ...

  48. [56]

    Note that there is no Hamiltonian in the Lindbladian of the measurement channel

    Lindbladian of the measurement channel For an single��step of the measurement channel acting on the�-th qubit, the Lindbladian can be derived to be: � ��(��) =� ��� (��) = � ��=�1 � ��(� �� ���)��� � ��(� �� ���)�(D1) which gives the Lindbladian equation ��� �� =�[� �] = � �=�...

  49. [57]

    Setting�= 0 gives back the untwirled Petz recovery channel

    Lindbladian of the twirled Petz recovery channel The generic infinitesimal twirled Petz recovery channel from a generic� �+�� to� � is defined as �� ��(�) = � � �� �(�)� � ��(�)�� �� � ��(�) =� ���� � � � � �� � � ����� � �+�� �� ����� � �+�� � � ���� � � �(D4) where�(�) = 1 2...

  50. [58]

    Consider a subsystem � � centered around qubit�, chosen to include the region that qubit�is entangled with (which is denoted as region �)

    Subsystem-Based Density Matrix Estimation We now focus on a forward weak measurement step� �� acting on qubit�in a finite small time��(a discretized or Trotterized Linbladian dynamics) and explain how to construct its Petz recovery map �� � � �� . Consider a subsystem � � cent...

  51. [59]

    For each� �, compute the initial reduced density matrix: �� � 0 = 1 2��� � � ��� � ���0��(D9)

  52. [60]

    If it is not, deform� � � 0 into a valid SPD matrix

    Verify whether� � � 0 is semi-positive definite (SPD). If it is not, deform� � � 0 into a valid SPD matrix. This step modifies the Pauli coefficients: ���0 � �� (� � ) ��0 �(D10)

  53. [61]

    We assume that� � � � is a matrix of size� � � �� � � , where� � � = 2��� �

    Evolve the RDM under the decoherence model: �� � � = 1 2��� � � ��� � �(� � ) ��� ��with� (� � ) ��� =� � �(�)� (� � ) ��0 for��� ��(D11) Similarly, the RDM at time����is given by: �� � ���� = 1 2��� � � ��� � �(� � ) ������ ��(D12) 35 By construction, the SPD property of� � �...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.