Pith. sign in

REVIEW 2 major objections 5 minor 14 references

Procrustes Tomography -- reconstructing noisy quantum channels made easy

T0 review · 2 major / 5 minor · reviewed 2026-07-10 · grok-4.5

Pith's one-line read Procrustes tomography reconstructs noisy quantum channels from reconstructed states, matching or beating standard methods under realistic sampling and preparation noise.

desk verdict Clean pedagogical reformulation of process tomography as a Procrustes problem; competitive numerics, incremental novelty, worth a referee. read the letter →

arxiv 2607.07988 v1 pith:ARUQYT2W submitted 2026-07-08 quant-ph

classification quant-ph
keywords processtomographyProcrustesproblemNISQquantumchannelsChoimatrixlinearinversionstatepreparationerrorsCPTPprojection
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

Cloud NISQ users often need to characterize the actual noisy process their circuits experience, but full process tomography is expensive and the textbook methods are either nonphysical or awkward to implement. This paper introduces Procrustes tomography: prepare a set of input states, run them through the device, reconstruct the output density matrices by ordinary state tomography, pack those matrices into columns of B and the inputs into A, and recover the process matrix as the least-squares map Lambda = B A^+. Optional weighting handles uneven shot budgets, input-state tomography corrects preparation errors, and simple projections restore complete positivity and unitarity when desired. Numerical experiments on two-qubit amplitude-damping-plus-dephasing channels show that the method matches or exceeds Choi-matrix and linear-inversion tomography under even sampling, uneven sampling, and imperfect preparation, while remaining pedagogically transparent and modular.

What carries the argument

The Procrustes fit Lambda = min_L ||L A - B||_F (or its weighted form), solved by the Moore-Penrose pseudoinverse Lambda = B A^+; subsequent spectral truncation of states, Knee et al. CPTP projection, and Kronecker-product SVD for nearest-unitary projection.

What would settle it

Re-run the same two-qubit Lindblad channels, replace the linear-inversion state tomography step with maximum-likelihood or Bayesian state tomography at identical total shot counts, and check whether Procrustes still matches or beats Choi and linear-inversion process tomography on average gate infidelity.

Watch

Extended reading notes

Core claim

Process tomography can be reduced to an ordinary Procrustes least-squares problem on reconstructed density matrices: the process superoperator is recovered by Lambda = B A^+, where the columns of A and B are vectorized input and output states. With optional weighting, input-state tomography, CPTP projection, and unitary projection, this construction matches or outperforms both Choi reconstruction and linear-inversion process tomography for representative noisy two-qubit channels under the sampling and preparation conditions studied.

Load-bearing premise

The method's reported advantage rests on using linear-inversion state tomography (with negative-eigenvalue truncation) as the estimator of the output density matrices; better state estimators are not compared.

Editorial extensions

If this is right

  • Any future improvement in state tomography (readout-error mitigation, machine-learning estimators, etc.) immediately improves process tomography with no change to the Procrustes algorithm.
  • State-preparation errors can be removed by simply performing state tomography on both the prepared inputs and the process outputs, at the cost of roughly twice the experiments.
  • When a gate is known to be nearly unitary, the Kronecker-product SVD projection yields a rank-1 process that saturates at lower infidelity and can be characterized with as few as d+1 states.
  • Uneven sampling of cheap computational-basis preparations versus expensive superpositions can be optimally weighted by the fourth root of the shot counts, reducing over-bias from low-sample states.

Reading between the lines

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

  • Because the method never requires the user to form virtual non-physical operators, it is immediately usable by experimental groups that already have a working state-tomography pipeline.
  • The same A/B construction could be applied to continuous-time process characterization by treating intermediate-time density matrices as additional columns, turning tomography into a trajectory-fitting problem.
  • If cloud providers expose only limited mid-circuit measurement or reset options, the modularity of Procrustes still allows users to characterize the effective channel they actually experience.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The manuscript introduces Procrustes tomography for reconstructing noisy quantum channels: after preparing a set of input states, performing state tomography on the outputs, and assembling the vectorized density matrices into matrices A (inputs) and B (outputs), the process matrix is obtained by the ordinary Procrustes solution Λ = B A⁺ (Eq. 10), optionally with a diagonal weighting matrix W that accounts for uneven shot counts (Eq. 11). The method is compared with the textbook Choi-matrix reconstruction and with linear-inversion process tomography under Lindblad noise (amplitude damping + dephasing) for two-qubit gates. Numerical experiments (Figs. 1–4) examine even and uneven sampling, imperfect state preparation, and optional projection onto the nearest unitary channel; they show that the Procrustes estimator matches or exceeds the fidelity of the two baselines, especially once input-state tomography or unitary projection is included. Finite-sample non-CPTP maps are corrected by the Knee et al. projection for all three methods.

Significance. If the reported performance advantage holds under realistic device conditions, the work supplies a lightweight, pedagogically transparent alternative to standard process tomography that is immediately usable by NISQ-cloud users who lack specialized tomography expertise. Strengths include an explicit, parameter-free least-squares estimator, a clean treatment of state-preparation errors (simply double the tomography), a natural weighting scheme for uneven sampling, and a unitary-projection step based on the Kronecker-product SVD. The numerical comparisons under controlled Lindblad dynamics are clear and support the claim within the simulated regime. The paper does not claim asymptotic optimality or experimental validation, so its contribution is best viewed as a practical, extensible reconstruction recipe rather than a fundamental advance in tomography theory.

major comments (2)
  1. Numerical simulations section (paragraph beginning “For the following examples…”) and the discussion after Eq. (10): all Procrustes results are generated with linear-inversion state tomography (plus optional negative-eigenvalue truncation). The paper itself notes that, under pure linear inversion, Procrustes and the linear-inversion baseline become mathematically identical. Consequently the observed advantage is attributable only to the optional PSD truncation, the weighting matrix W, or the input-state tomography step. Without at least one comparison against maximum-likelihood or Bayesian state tomography, it remains unclear whether the claimed superiority survives once a higher-quality state estimator is used for every method. A short additional panel or appendix addressing this point would substantially strengthen the central claim.
  2. Figs. 1–4 and the associated text: all numerical evidence is restricted to two-qubit channels generated by a single Lindblad model (T1/T2 ratios and absolute T2 values listed in the figure captions). While the algebraic construction is dimension-independent, the performance ranking versus Choi and linear inversion could change for larger systems or for noise that includes coherent control errors, leakage, or non-Markovian effects. The manuscript should either (i) state explicitly that the superiority claim is limited to the two-qubit Lindblad setting examined, or (ii) supply at least one higher-dimensional or differently noised example.
minor comments (5)
  1. Abstract and opening paragraph: the rhetorical question “What is the most expensive part…?” and the unqualified claim that Procrustes “outperforms established methods in a number of aspects” should be tempered to match the more careful language used in the concluding remarks.
  2. Eq. (2) and surrounding text: the linear combination that recovers |n⟩⟨m| from physical states is standard, yet the precise coefficients for the two-qubit case used later are never written out; a short explicit formula would aid reproducibility.
  3. Fig. 2 caption and text: the non-monotonic infidelity at low shot counts is attributed to “over-biasing toward Clifford operations,” but no quantitative diagnostic (e.g., diamond-norm distance to the nearest Clifford) is supplied; a one-sentence clarification would help.
  4. Typographical issues: “breifly” (p. 2), “porocess” (p. 4), “Figu. 2” (p. 4), and inconsistent hyphenation of “state-preparation” / “state preparation” appear throughout.
  5. References: several recent works on GST, compressed sensing, and ML-based process tomography are cited, yet the classic maximum-likelihood process tomography literature (e.g., Śšvanda et al., Fiurášek) is under-represented; adding one or two standard citations would improve context.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: Procrustes estimator is ordinary least-squares against independent simulated ground truth; performance metric is external.

full rationale

The paper defines Procrustes tomography as the ordinary least-squares solution Λ = B A^{+} (Eq. 10) of the matrix equation B = Λ A formed from reconstructed input and output density matrices. This is a standard Procrustes / pseudoinverse construction; no free parameters are fitted to the fidelity data that later serve as the figure of merit. The numerical comparisons (Figs. 1–4) evaluate average gate infidelity against independently generated Lindblad channels whose Hamiltonians and jump operators are fixed before any reconstruction is performed. When linear-inversion state tomography is used for both Procrustes and the linear-inversion baseline, the paper itself notes that the two estimators become mathematically identical (text after Eq. 10); residual advantages arise only from optional, explicitly derived extensions (PSD truncation, weighting matrix W, input-state tomography, unitary projection) that the baselines do not receive. Self-citations are limited to prior work on related topics (efficient Hamiltonians, etc.) and do not close any definitional loop. The derivation is therefore self-contained against external benchmarks and exhibits no circular reduction.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on standard linear algebra (Procrustes / Moore-Penrose pseudoinverse), the standard definition of CPTP maps, and a conventional Lindblad noise model used only for numerical illustration. No free parameters are fitted to experimental data; the only modeling choices are the simulation parameters (T1/T2 ratios, shot allocations) that define the test cases, not the estimator itself. No new physical entities are postulated.

free parameters (3)
  • T1/T2 ratios and absolute T2 values used in simulations = T1=5 T2; T2 in {10,100,1000}
    Chosen by hand to define 'high-quality' and 'low-quality' gates (e.g. T1=5 T2, T2=1000 or 10); they control the difficulty of the reconstruction task but are not fitted to real data.
  • shot-allocation ratio for uneven sampling = Ni = 10 N_pm
    Computational-basis states sampled at 10x the rate of superposition states; an arbitrary experimental design choice used only in Fig. 2.
  • weighting exponent for Procrustes weights = 1/4
    wi = N_i^{1/4} is motivated by standard-error scaling but is an ad-hoc choice of the paper; other exponents would change the weighted estimator.
assumptions (4)
  • domain assumption Completely-positive trace-preserving maps are linear and can be represented by a process matrix Lambda acting on vectorized density operators.
    Standard quantum-channel theory; used throughout the Essentials section.
  • standard math The Frobenius-norm least-squares solution Lambda = B A^+ is the unique minimizer of the Procrustes objective when A has full row rank (or the minimal-norm solution otherwise).
    Classical matrix analysis; cited via Gower & Dijksterhuis and standard pseudoinverse theory.
  • ad hoc to paper Negative eigenvalues of reconstructed density matrices may be zeroed and the matrix re-normalized without destroying the subsequent process estimate.
    Common practical fix, but not rigorously justified for the Procrustes pipeline; stated in the Procrustes tomography paragraph.
  • domain assumption The Knee et al. projection maps any estimated superoperator onto the nearest CPTP map in a way that preserves the ranking of reconstruction methods.
    Imported algorithm applied uniformly to all three methods; correctness of the ranking therefore inherits any bias of that projection.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Procrustes Tomography -- reconstructing noisy quantum channels made easy." pith.science (2026). https://pith.science/paper/ARUQYT2W

@misc{pith2026260707988,
  author       = {Pith},
  title        = {Pith review of: Procrustes Tomography -- reconstructing noisy quantum channels made easy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ARUQYT2W}},
  note         = {Machine review of arXiv:2607.07988}
}
read the original abstract

What is the most expensive part of quantum device characterization? Clearly, the answer is quantum process tomography. However, especially for noisy intermediate-scale quantum (NISQ) computers, a comprehensive understanding of the noisy quantum dynamics is essential in interpreting the computational output. In this work, we introduce an efficient method -- Procrustes tomography -- that outperforms established methods in a number of aspects. After a pedagogical and constructive introduction, we demonstrate the utility of the method for representative examples of noisy quantum channels.

Figures

Figures reproduced from arXiv: 2607.07988 by the authors.

Figure 1
Figure 1. Infidelity between the true and reconstructed two-qubit [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 3
Figure 3. Infidelity between the true and reconstructed two [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Infidelity between the true and the Procrustes recon [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

14 extracted references · 14 canonical work pages

  1. [1]

    [1]LaRose R.,Quantum,3(2019)

  2. [2]

    Aquila: QuEra's 256-qubit neutral-atom quantum computer

    [2]Ser vices A. W.,Amazon braket supported regions and deviceshttps://docs.aws.amazon.com/braket/latest/ developerguide/braket-devices.htmlaccessed: 2026- 07-04 (2026). [3]IonQ,Introduction – ionq quantum cloud documentation https://docs.ionq.com/user-manual/introduction accessed: 2026-07-04 (2026). [4]Quantinuum,Quantinuum technical documentation https:/...

  3. [3]

    W., Regnault N

    [13]Choo K., Von Keyserlingk C. W., Regnault N. andNeupert T.,Phys. Rev. Lett.,121(2018) 086808. p-6 Procrustes Tomography – reconstructing noisy quantum channels made easy [14]Iqbal M., Tantiv asadakarn N., Verresen R., Campbell S. L., Dreiling J. M., Figgatt C., Gae- bler J. P., Johansen J., Mills M., Moses S. A.et al.,Nature,626(2024)

  4. [4]

    J., Lidar D

    [15]Bando Y., Susa Y., Oshiyama H., Shibata N., Ohzeki M., Gómez-Ruiz F. J., Lidar D. A., Suzuki S., Del Campo A.andNishimori H.,Phys. Rev. Res., 2(2020) 033369. [16]Robertson R., Doucet E., Spicer E.andDeffner S.,Entropy,27(2025)

  5. [5]

    J.andDeffner S.,arXiv preprint arXiv:2603.03801, (2026)

    [17]Robertson R., Consiglio M., Stevens J., Doucet E., Apollaro T. J.andDeffner S.,arXiv preprint arXiv:2603.03801, (2026) . [18]Irf an A. K.andYu C.,arXiv preprint arXiv:2509.24213, (2025) . [19]Bautra E., Dimitrijevs M.andYakaryilmaz A., arXiv preprint arXiv:2603.14607, (2026) . [20]Di Bartolomeo G., Vischi M., Cesa F., Wixinger R., Grossi M., Donadi S....

  6. [6]

    J., Acharya A., Carleo G., Carrasquilla J.andAolita L.,Nat

    [25]Torlai G., Wood C. J., Acharya A., Carleo G., Carrasquilla J.andAolita L.,Nat. Commun.,14 (2023)

  7. [7]

    [26]Gaikw ad A., Bihani O., Ar vindandDorai K.,Phys. Rev. A,109(2024) 012402. [27]Ahmed S., Quijandría F.andKockum A. F.,Phys. Rev. Lett.,130(2023) 150402. [28]Altepeter J. B., Branning D., Jeffrey E., Wei T., Kwiat P. G., Thew R. T., O’Brien J. L., Nielsen M. A.andWhite A. G.,Phys. Rev. Lett.,90(2003) 193601. [29]Xue S., W ang Y., Zhan J., W ang Y., Zeng...

  8. [8]

    B., Goss N., Marinelli B., Naik R

    [40]Hashim A., Nguyen L. B., Goss N., Marinelli B., Naik R. K., Chistolini T., Hines J., Marceaux J. P., Kim Y., Gokhale P.et al.,PRX Quantum, 6(2025) 030202. [41]Innan N., Siddiqui O. I., Arora S., Ghosh T., Koçak Y. P., Paragas D., Galib A. A. O., Khan M. A.-Z.andBennai M.,Quantum Mach. Intell.,6(2024)

Show all 14 references
  1. [9]

    A., Phys

    [42]Wei V., Coish W., Ronagh P.andMuschik C. A., Phys. Rev. Res.,6(2024) 023250. [43]Lohani S., Kirby B. T., Brodsky M., Danaci O.and Glasser R. T.,Mach. Learn.: Sci. Technol.,1(2020) 035007. [44]Schmale T., Reh M.andGärttner M.,npj Quantum Inf.,8(2022)

  2. [10]

    S., Di Giov anni A., Rotzinger H., Usti- nov A

    [45]Aasen A. S., Di Giov anni A., Rotzinger H., Usti- nov A. V.andGärttner M.,Commun. Phys.,7(2024)

  3. [11]

    T., Gambetta J

    [46]Merkel S. T., Gambetta J. M., Smolin J. A., Po- letto S., Córcoles A. D., Johnson B. R., Ryan C. A.andSteffen M.,Phys. Rev. A,87(2013) 062119. [47]Greenbaum D.,arXiv preprint arXiv:1509.02921, (2015) . [48]Huang T., Gaikw ad A., Moskalenko I., Aggar- w al A., Abad T., Kuzm...

  4. [12]

    [52]Hradil Z.,Phys. Rev. A,55(1997) R1561(R). [53]Aditi K.andBecker S.,Phys. Rev. A,112(2025) 052436. [54]Loan C. F.,J. Comput. Appl. Math.,123(2000) 85 nu- merical Analysis

  5. [13]

    Phys.,24(2022) 055002

    [56]Aifer M.andDeffner S.,New J. Phys.,24(2022) 055002. [57]Stevens J.andDeffner S.,Quantum Sci. Technol.,10 (2025) 04LT03. [58]Higham N. J.,Functions of matrices : theory and com- putation(Cambridge University Press, Cambridge, TAS, Australia)

  6. [14]

    M.andKoch C

    [59]Romer A., Reich D. M.andKoch C. P.,arXiv preprint arXiv:2507.17648, (2025) . p-7

Pith tools

Reviewed July 10, 2026 · model on record in the stance chip above.