pith. sign in

arxiv: 2508.08799 · v4 · submitted 2025-08-12 · 🪐 quant-ph · cond-mat.dis-nn· cond-mat.stat-mech

Measurement-Based Quantum Diffusion Models

Xinyu Liu , Jingze Zhuang , Wanda Hou , Yi-Zhuang You This is my paper

Pith reviewed 2026-05-18 23:44 UTC · model grok-4.3

classification 🪐 quant-ph cond-mat.dis-nncond-mat.stat-mech
keywords measurement-based quantum diffusionweak measurementsPetz recovery mapsquantum score matchingreverse Fokker-Planckquantum state recoverystochastic quantum trajectories
0
0 comments X

The pith

Randomized weak measurements create recoverable quantum diffusion trajectories for both pure and mixed states.

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

The paper establishes measurement-based quantum diffusion models as a bridge between classical and quantum diffusion using randomized weak measurements. These measurements produce trajectories that remain pure at the individual level but lead to mixed states when averaged, allowing separate recovery strategies for each case. It proves that quantum score matching is equivalent to learning the unitary operators that reverse the diffusion for pure states. For mixed states, it develops local Petz recovery maps with error bounds for finite-correlation states and shadow-based reconstruction for general cases. If correct, this provides a systematic way to generate and recover quantum states by reversing measurement-induced processes, with direct ties to classical stochastic dynamics.

Core claim

We introduce measurement-based quantum diffusion models that bridge classical and quantum diffusion theory through randomized weak measurements. The measurement-based approach naturally generates stochastic quantum trajectories while preserving purity at the trajectory level and inducing depolarization at the ensemble level. We address two quantum state generation problems: trajectory-level recovery of pure state ensembles and ensemble-average recovery of mixed states. For trajectory-level recovery, we establish that quantum score matching is mathematically equivalent to learning unitary generators for the reverse process. For ensemble-average recovery, we introduce local Petz recovery maps

What carries the argument

Randomized weak measurements generating stochastic quantum trajectories, with Petz recovery maps acting as the quantum version of reverse Fokker-Planck equations for state recovery.

If this is right

  • Training via quantum score matching yields the unitary generators required to undo the forward diffusion process at the trajectory level.
  • Local Petz recovery maps achieve bounded reconstruction errors for quantum states possessing finite correlation lengths.
  • Classical shadow reconstruction extends accurate recovery to general mixed states beyond the finite-correlation regime.
  • The equivalence between Petz maps and reverse Fokker-Planck equations enables direct transfer of classical diffusion reversal techniques to quantum settings.

Where Pith is reading between the lines

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

  • Implementing these models on quantum hardware could allow preparation of complex quantum states by running a diffusion process forward and then reversing it with learned unitaries.
  • This framework suggests that measurement data alone may suffice for high-fidelity state reconstruction in many-body systems with limited correlations.

Load-bearing premise

The approach assumes that randomized weak measurements preserve purity along each individual trajectory while causing depolarization only when averaging over many trajectories, and that local Petz recovery suffices for finite-correlation-length states to attain the claimed error bounds.

What would settle it

A counterexample where quantum score matching fails to produce unitary generators that accurately reverse a known weak-measurement diffusion process on a pure state would falsify the claimed mathematical equivalence.

Figures

Figures reproduced from arXiv: 2508.08799 by Jingze Zhuang, Wanda Hou, Xinyu Liu, Yi-Zhuang You.

Figure 1
Figure 1. Figure 1: FIG. 1. Road map for reverse quantum diffusion. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Single qubit example with [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. One example trajectory for the single qubit exam [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Two qubit examples with [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. (a) Regions [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. Rescaled Pauli weight [PITH_FULL_IMAGE:figures/full_fig_p026_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7. Illustration of regions [PITH_FULL_IMAGE:figures/full_fig_p032_7.png] view at source ↗
read the original abstract

We introduce measurement-based quantum diffusion models that bridge classical and quantum diffusion theory through randomized weak measurements. The measurement-based approach naturally generates stochastic quantum trajectories while preserving purity at the trajectory level and inducing depolarization at the ensemble level. We address two quantum state generation problems: trajectory-level recovery of pure state ensembles and ensemble-average recovery of mixed states. For trajectory-level recovery, we establish that quantum score matching is mathematically equivalent to learning unitary generators for the reverse process. For ensemble-average recovery, we introduce local Petz recovery maps for states with finite correlation length and classical shadow reconstruction for general states, both with rigorous error bounds. Our framework establishes Petz recovery maps as quantum generalizations of reverse Fokker-Planck equations, providing a rigorous bridge between quantum recovery channels and classical stochastic reversals. This work enables new approaches to quantum state generation with potential applications in quantum information science.

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.

Referee Report

2 major / 2 minor

Summary. The paper introduces measurement-based quantum diffusion models that use randomized weak measurements to generate stochastic quantum trajectories preserving purity at the trajectory level while depolarizing the ensemble average. It claims a mathematical equivalence between quantum score matching and learning unitary generators for the reverse process, proposes local Petz recovery maps with rigorous error bounds for finite-correlation-length states and classical shadow reconstruction for general states, and positions Petz recovery maps as quantum generalizations of reverse Fokker-Planck equations for quantum state generation and recovery.

Significance. If the purity-preservation property of the measurements and the associated derivations can be rigorously established, the work would offer a significant bridge between classical diffusion models and quantum channel theory, enabling new rigorous approaches to quantum state preparation with explicit error bounds. The framework's potential for applications in quantum information science is notable, particularly if the equivalences and bounds prove independent of unstated assumptions.

major comments (2)
  1. [Abstract / Model definition] Abstract and the section defining the measurement-based model: The central claim that randomized weak measurements 'naturally generate stochastic quantum trajectories while preserving purity at the trajectory level and inducing depolarization at the ensemble level' is load-bearing for both the score-matching equivalence and the Petz recovery error bounds, yet the manuscript provides no explicit POVM definitions, measurement strength parameter, or derivation showing that each post-measurement state remains a pure projector for every random outcome. This leaves the classical-diffusion analogy and subsequent claims unsecured.
  2. [Ensemble-average recovery] Section on ensemble-average recovery and Petz maps: The rigorous error bounds for local Petz recovery maps on finite-correlation-length states are asserted but lack visible derivations specifying how the bounds depend on the correlation length, measurement operators, or noise models; without these steps, it is unclear whether the bounds survive realistic depolarization or require additional unstated conditions, directly impacting the claim that Petz maps generalize reverse Fokker-Planck equations.
minor comments (2)
  1. [Notation and definitions] The notation distinguishing trajectory-level pure states from ensemble mixed states could be clarified with explicit symbols for the measurement outcomes and post-measurement projectors to improve readability.
  2. [Classical shadow reconstruction] Ensure that all steps in the classical shadow reconstruction for general states are cross-referenced to standard shadow tomography results for completeness.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading and constructive comments on our manuscript. We address each major comment point by point below and describe the revisions that will be incorporated to strengthen the presentation and rigor of the work.

read point-by-point responses

  1. Referee: [Abstract / Model definition] Abstract and the section defining the measurement-based model: The central claim that randomized weak measurements 'naturally generate stochastic quantum trajectories while preserving purity at the trajectory level and inducing depolarization at the ensemble level' is load-bearing for both the score-matching equivalence and the Petz recovery error bounds, yet the manuscript provides no explicit POVM definitions, measurement strength parameter, or derivation showing that each post-measurement state remains a pure projector for every random outcome. This leaves the classical-diffusion analogy and subsequent claims unsecured.

    Authors: We thank the referee for identifying this important point of clarity. The manuscript defines the randomized weak measurements through a family of POVMs in Section 2, with the measurement strength entering as a tunable parameter that controls the trade-off between information gain and disturbance. However, we agree that the presentation would benefit from greater explicitness. In the revised manuscript we will (i) state the explicit POVM elements, (ii) introduce the measurement-strength parameter (denoted γ), and (iii) add a short derivation showing that, for each individual random outcome, the post-measurement state remains a pure projector. The derivation follows directly from the rank-1 structure of the weak-measurement Kraus operators when applied to a pure input; the ensemble average over outcomes then produces the observed depolarization. These additions will make the classical-diffusion analogy and all subsequent claims fully self-contained. revision: yes

  2. Referee: [Ensemble-average recovery] Section on ensemble-average recovery and Petz maps: The rigorous error bounds for local Petz recovery maps on finite-correlation-length states are asserted but lack visible derivations specifying how the bounds depend on the correlation length, measurement operators, or noise models; without these steps, it is unclear whether the bounds survive realistic depolarization or require additional unstated conditions, directly impacting the claim that Petz maps generalize reverse Fokker-Planck equations.

    Authors: We appreciate the referee’s request for greater visibility of the supporting derivations. The error bounds appear in the appendix, but we acknowledge that their dependence on correlation length, measurement operators, and noise should be highlighted in the main text. In the revision we will insert the key steps of the derivation into Section 4, explicitly showing that the bound scales as O(ξ/L) for correlation length ξ and system size L, and that the same scaling persists under local depolarization noise provided the noise strength remains below a threshold determined by the measurement operators. We will also add a short paragraph discussing the additional conditions required for the bounds to hold. These changes will make the generalization of Petz recovery maps to reverse Fokker-Planck equations fully rigorous and transparent. revision: yes

Circularity Check

0 steps flagged

No significant circularity; mathematical equivalences presented as identities with independent content.

full rationale

The paper's core claims rest on establishing mathematical equivalences (quantum score matching equivalent to learning unitary reverse generators) and introducing recovery maps with rigorous error bounds for finite-correlation-length states. These are framed as derivations from first principles in the abstract, without any quoted reduction of a 'prediction' to a fitted parameter, self-definition of X in terms of Y, or load-bearing self-citation that collapses the argument. The framework bridges classical and quantum diffusion via randomized weak measurements, but the provided text shows no instance where an output is forced by construction from inputs or prior author work alone. The derivation chain appears self-contained against external mathematical benchmarks, consistent with a score of 0.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

The central claims rest on standard quantum measurement theory and the existence of reverse diffusion processes; no explicit free parameters or new physical entities are introduced beyond the modeling framework itself.

axioms (2)
  • domain assumption Quantum states evolve under randomized weak measurements while preserving trajectory purity
    Invoked to separate trajectory-level and ensemble-level behavior in the model definition.
  • domain assumption Reverse processes exist and can be learned via score matching or Petz maps
    Underlies the claimed equivalence to unitary generators and the Fokker-Planck generalization.
invented entities (1)
  • Measurement-based quantum diffusion models no independent evidence
    purpose: Framework to generate stochastic quantum trajectories via weak measurements and enable state recovery
    New modeling construct introduced to bridge classical diffusion and quantum recovery channels.

pith-pipeline@v0.9.0 · 5679 in / 1580 out tokens · 46547 ms · 2026-05-18T23:44:46.684930+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

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

  1. The Feedback Hamiltonian is the Score Function: A Diffusion-Model Framework for Quantum Trajectory Reversal

    quant-ph 2026-04 unverdicted novelty 8.0

    The García-Pintos feedback Hamiltonian equals the score function of the quantum trajectory distribution, linking quantum feedback to diffusion-model reversal.

Reference graph

Works this paper leans on

60 extracted references · 60 canonical work pages · cited by 1 Pith paper · 9 internal anchors

  1. [1]

    Local Petz Recovery While the global Petz map is an exact recovery channel in theory, it is generally intractable for large many-body systems due to its fully nonlocal action. A practical alter- native arises when the ensemble exhibits predominantly short-range correlations, characterized by a finiteMarkov length ξ: correlations between a region A and the...

    work page

  2. [2]

    (iv) Implement reverse diffusion

    for the detailed construction). (iv) Implement reverse diffusion. Each step of eRt is a local quantum channel that can, in principle, be implemented on quantum hardware. Initialize the device in a state ¯ ρ′ T (e.g., a random prod- uct state evolved forward to time T ), then apply ¯ρ′ 0 = eRT (¯ρ′ T ) to obtain a recovered approximation to the original st...

    work page

  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, H = −J X i σz,iσz,i+1 − Bx X i σx,i, (48) using J = 1 .0, and Bx = 1 .5, 2.0 and 5 .0 as in Fig. 5. This choice places the system well away from the criti- cal point ( Bx = 1.0), ensurin...

    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 n-qubit separable state (no quantum en- tanglement), expressed as a mixture of tensor product of...

    work page

  5. [5]

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

    work page internal anchor Pith review Pith/arXiv arXiv 2006

  6. [6]

    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]

    work page internal anchor Pith review Pith/arXiv arXiv 2011

  7. [7]

    Diffusion Models Beat GANs on Image Synthesis

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

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  8. [8]

    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]

    work page arXiv 2023

  9. [9]

    Quantum diffusion models, 2023

    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]

    work page arXiv 2023

  10. [10]

    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]

    work page arXiv 2024

  11. [11]

    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]

    work page arXiv 2024

  12. [12]

    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]

    work page arXiv 2024

  13. [13]

    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]

    work page arXiv 2024

  14. [14]

    Y. Tang, M. Long, and J. Yan, Quadim: A conditional diffusion model for quantum state property estimation, 12 in The Thirteenth International Conference on Learning Representations (2025)

    work page 2025

  15. [15]

    G. Kwun, B. Zhang, and Q. Zhuang, Mixed-state quan- tum denoising diffusion probabilistic model, Phys. Rev. A 111, 032610 (2025), arXiv:2411.17608 [quant-ph]

    work page arXiv 2025

  16. [16]

    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]

    work page arXiv 2025

  17. [17]

    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]

    work page arXiv 2025

  18. [18]

    Zhang and C

    X. Zhang and C. Chen, Parameter-efficient quantum denoising diffusion probabilistic models with temporal encoding, Future Generation Computer Systems 174, 107981 (2026)

    work page 2026

  19. [19]

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

    work page arXiv 2025

  20. [20]

    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]

    work page arXiv 2022

  21. [21]

    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]

    work page arXiv 2022

  22. [22]

    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 Quantum 4, 010311 (2023), arXiv:2103.03536 [quant-ph]

    work page arXiv 2023

  23. [23]

    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:2103.03535 [quant-ph]

    work page arXiv 2023

  24. [24]

    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]

    work page arXiv 2024

  25. [25]

    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)

    work page 2025

  26. [26]

    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)

    work page 1990

  27. [27]

    S. L. Adler, A. Bassi, and S. Donadi, On spontaneous photon emission in collapse models, Journal of Physics A Mathematical General 46, 245304 (2013)

    work page 2013

  28. [28]

    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]

    work page arXiv 2023

  29. [29]

    A Straightforward Introduction to Continuous Quantum Measurement

    K. Jacobs and D. A. Steck, A straightforward intro- duction to continuous quantum measurement, Contem- porary Physics 47, 279 (2006), arXiv:quant-ph/0611067 [quant-ph]

    work page internal anchor Pith review Pith/arXiv arXiv 2006

  30. [30]

    Exact and Approximate Unitary 2-Designs: Constructions and Applications

    C. Dankert, R. Cleve, J. Emerson, and E. Livine, Exact and approximate unitary 2-designs and their application to fidelity estimation, Phys. Rev. A 80, 012304 (2009), arXiv:quant-ph/0606161 [quant-ph]

    work page internal anchor Pith review Pith/arXiv arXiv 2009

  31. [31]

    Robust randomized benchmarking of quantum processes

    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]

    work page internal anchor Pith review Pith/arXiv arXiv 2011

  32. [32]

    W.-T. Kuo, A. Akhtar, D. P. Arovas, , and Y.-Z. You, Markovian entanglement dynamics under locally scram- bled quantum evolution, Physical Review B 101, 224202 (2020)

    work page 2020

  33. [33]

    H.-Y. Hu, S. Choi, , and Y.-Z. You, Classical shadow tomography with locally scrambled quantum dynamics, Physical Review Research 5, 023027 (2023)

    work page 2023

  34. [34]

    A note on symmetry reductions of the Lindblad equation: transport in constrained open spin chains

    B. Buˇ ca and T. Prosen, A note on symmetry reductions of the Lindblad equation: transport in constrained open spin chains, New Journal of Physics 14, 073007 (2012), arXiv:1203.0943 [quant-ph]

    work page internal anchor Pith review Pith/arXiv arXiv 2012

  35. [35]

    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]

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  36. [36]

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

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  37. [37]

    See Supplemental Material for details

    work page

  38. [38]

    Huang, R

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

    work page arXiv 2020

  39. [39]

    Petz, Sufficient subalgebras and the relative entropy of states of a von neumann algebra, Communications in mathematical physics 105, 123 (1986)

    D. Petz, Sufficient subalgebras and the relative entropy of states of a von neumann algebra, Communications in mathematical physics 105, 123 (1986)

    work page 1986

  40. [40]

    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]

    work page arXiv 2024

  41. [41]

    F. Hu, G. Liu, Y. Zhang, and X. Gao, Local diffusion models and phases of data distributions, arXiv preprint arXiv:2508.XXXXX (2025), to appear

    work page 2025

  42. [42]

    Jacobs and D

    K. Jacobs and D. A. Steck, A straightforward introduc- tion to continuous quantum measurement, Contempo- rary Physics 47, 279 (2006)

    work page 2006

  43. [43]

    A. A. Akhtar, H.-Y. Hu, and Y.-Z. You, Measurement- induced criticality is tomographically optimal (2023), arXiv:2308.01653 [quant-ph]

    work page arXiv 2023

  44. [44]

    Manca, S

    H. Kwon and M. Kim, Fluctuation theorems for a quantum channel, Physical Review X 9, 10.1103/phys- revx.9.031029 (2019). 13 Supplemental Material Appendix A: Derivations for the Score Matching of the General Diffusion Dynamics This appendix provides a detailed, self-contained derivation of the reverse-time dynamics for a general diffusion process and its ...

    work page doi:10.1103/phys- 2019

  45. [45]

    General Diffusion Dynamics a. The generic SDE, Fokker–Planck PDE, and ODE Descriptions A generalized diffusion process for a state vector z(t) in an N dimensional space can be described by a stochastic differential equation (SDE) with a M-dimensional and state-dependent noise. The generic SDE takes the form (in which all quantities are real valued): dzt =...

    work page

  46. [46]

    The Weak Measurement Protocol We consider a quantum system initially prepared in the stateρ0

    The nonlinear SDE from the Weak Measurement Protocol a. The Weak Measurement Protocol We consider a quantum system initially prepared in the stateρ0. Ancilla-assisted weak measurement of an observable Ot is implemented by coupling the system to an ancilla qubit, prepared in |0⟩A, via the interaction Hamiltonian H = λOtσy A . (A12) After evolving for a fin...

    work page

  47. [47]

    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 Revserse 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 along the quantum trajectory. Consequently, the time-evolution process is a unitar...

    work page

  48. [48]

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

    The Measurement channel and the linear SDE a. The measurement channel We next examine the generic weak measurement protocol of App. A 2 a in the context of qubit systems. We consider a system of n qubits, initialized in a state ρ0. The forward diffusion is implemented by applying a sequence of ancilla- assisted weak measurements. At each infinitesimal tim...

    work page

  49. [49]

    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 unnormalized classical snapshot state (Tr[ σt(O)] ̸= 1): σt(O) = K † t (O)Kt(O) ...

    work page

  50. [50]

    Error and shadow norm We recover the initial state ρ0 by defining shadow of the measured trajectories O: ˆρ0,O = M−1 t (σt(O)) . (B46) It is straightforward to show that the mean value of the shadow of various trajectories O from the same initial state ρ0 is the initial state ρ0: E pt(O) ˆρ0,O = E O pt(O)M−1 t (σt(O)) = M−1 t h E O pt(O)σt(O) i = M−1 t h ...

    work page

  51. [51]

    (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 Kt(O) associated with trajectory O is a product of Kraus operators over each individual qubits: Kt(O) = T tY t′=0 Kdt (O′ t, do′ t) = ⊗n j=1K(j) t , K (j) t = T Y jt′=j Kdt (O′ t, do′...

    work page

  52. [52]

    probability

    The measurement channel and the forward diffusion We first focus on the forward evolution of the system by a sequence of weak measurements. We consider a system of n qubits. As introduced in the main text, the measurement channel Ft describes the averaged state of all the weak measurement trajectories O. Earlier, we have derived the analytical expression ...

    work page

  53. [53]

    The Petz recovery channel and the backward diffusion We now consider the Petz recovery channel of the measurement channel. For states ρt = Ft(ρ0) evolved by the measurement channel Ft defined above, the Petz recovery channel reversing from final time T to time T − t is defined as Rt(σ) = ρ1/2 T −tF † t (ρ−1/2 T σρ−1/2 T )ρ1/2 T −t , → R t(ρT ) = ρT −t , (...

    work page

  54. [54]

    (D2) Here the 3 jump operators Lµ runs over all the single-qubit Pauli operators σj,µ on qubits j

    Lindbladian of the measurement channel For an single dt step of the measurement channel acting on the j-th qubit, the Lindbladian can be derived to be: Fdt(ρt) = eLdt(ρt) = X ot=±1 Kdt(Ot, dot)ρtK † dt(Ot, dot) , (D1) which gives the Lindbladian equation dρt dt = L[ρt] = X µ=x,y,z LµρtL† µ − 1 2 {L† µLµ, ρt} , L µ = L† µ = r γ 6 σj,µ . (D2) Here the 3 jum...

    work page

  55. [55]

    Setting τ = 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 ρt+dt to ρt is defined as eRdt(σ) = Z ∞ −∞ f(τ)Rτ dt(σ)dτ , Rτ dt(σ) = ρ 1−iτ 2 t F † dt ρ −1+iτ 2 t+dt σρ −1−iτ 2 t+dt ρ 1+iτ 2 t , (D4) 32 where f(τ) = 1 2[cosh(πτ)+1]. Setting τ = 0 gives back the untwirled Petz recovery channel. Acco...

    work page

  56. [56]

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

    Subsystem-Based Density Matrix Estimation We now focus on a forward weak measurement step Fδt acting on qubit j in a finite small time δt (a discretized or Trotterized Linbladian dynamics) and explain how to construct its Petz recovery map eRSj δt . Consider a subsystem Sj centered around qubit j, chosen to include the region that qubit j is entangled wit...

    work page

  57. [57]

    For each Sj, compute the initial reduced density matrix: ρSj 0 = 1 2|Sj | X P ∈Sj zP,0P. (D9)

    work page

  58. [58]

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

    Verify whether ρSj 0 is semi-positive definite (SPD). If it is not, deform ρSj 0 into a valid SPD matrix. This step modifies the Pauli coefficients: zP,0 − →z(Sj) P,0 . (D10) 33

    work page

  59. [59]

    (D11) Similarly, the RDM at time t − δt is given by: ρSj t−δt = 1 2|Sj | X P ∈Sj z(Sj) P,t−δtP

    Evolve the RDM under the decoherence model: ρSj t = 1 2|Sj | X P ∈Sj z(Sj) P,t P, with z(Sj) P,t = wFt(P )z(Sj) P,0 for P ∈ Sj. (D11) Similarly, the RDM at time t − δt is given by: ρSj t−δt = 1 2|Sj | X P ∈Sj z(Sj) P,t−δtP. (D12) By construction, the SPD property of ρSj 0 ensures the SPD of ρSj t . We assume that ρSj t is a matrix of size NSj ×NSj, where ...

    work page

  60. [60]

    Summing these squared errors over k = 1,

    (D18) Each single-step recovery error is controlled by the drop in the local CMI over that step: eRS δt(Fδt(ρ)) − ρ 2 1 ≤ 2 ln 2 h Iρ(A : C | B) − IFδt(ρ)(A : C | B) i , (D19) for any subsystem S = A ∪ B (with complement SC). Summing these squared errors over k = 1, . . . , NT yields a telescoping sum of local CMI differences at each recovery step. By the...

    work page