Pith. sign in

REVIEW 3 major objections 4 minor 34 references

Sampling Groups of Pauli Operators to Enhance Direct Fidelity Estimation

T0 review · 3 major / 4 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Grouping Pauli operators into commuting families before sampling makes direct fidelity estimation no worse in copy count and, for Haar-random states, lower in variance; 8-qubit numerics show 32% fewer copies and 92–99% lower variance.

desk verdict A useful, reproducible extension of DFE with a solid copy-count bound, but the Appendix B variance proof is flawed and should not be accepted as is. read the letter →

arxiv 2501.19228 v2 pith:FAKOC5ZB submitted 2025-01-31 quant-ph

Júlia Barberà-Rodríguez , Mariana Navarro , Leonardo Zambrano This is my paper · ORCID
classification quant-ph
keywords directfidelityestimationPauligroupingcommutingfamiliesimportancesamplingqubit-wisecommutativityfullvariancereductionmeasurement
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

This paper tries to establish a simple way to speed up direct fidelity estimation: before sampling Pauli operators by importance, sort them into commuting families and sample the families instead. The authors claim that this grouped estimator keeps the same accuracy–confidence guarantee (fidelity within $2\varepsilon$ with probability at least $1-2\delta$) while using no more copies on average than the standard protocol, and that under a concentration condition satisfied by Haar-random states its variance is no larger. Numerical simulations on 8-qubit Haar-random states show a 32% reduction in the mean number of copies and a variance reduction of 92% for qubit-wise commutativity and 99% for full commutativity. If the claim holds, current quantum devices can verify prepared states more cheaply, especially when only local measurements are available.

What carries the argument

The load-bearing object is the regrouped probability distribution: for each commuting family $k$, the group weight $\|\mathbf{b}_k\|^2 = \sum_l b_{kl}^2$ replaces the individual weights $b_{kl}^2$, so the random variable becomes $X_k = (\mathbf{a}_k \cdot \mathbf{b}_k) / \|\mathbf{b}_k\|^2$. Each sampled group is measured in its common eigenbasis; the per-group copy count $m_{k_i} = \lceil 2\|\mathbf{b}_{k_i}\|_1^2 / (\|\mathbf{b}_{k_i}\|^4 d \ell \varepsilon^2) \ln(2/\delta) \rceil$ comes from Hoeffding's inequality, and the Appendix B comparison shows that the conditional second moment of the grouped estimator is no larger than that of the original estimator under the ceiling-number approximation. The quantitative engine is Cauchy–Schwarz, which bounds $|X_k|$ and the cross-terms, plus the concentration of Haar-random coefficients $|b_{kl}| \approx 1/d$, which makes $\|\mathbf{b}_k\|_1^2 / \|\mathbf{b}_k\|^4$ large.

What would settle it

Take a target state with a sharply skewed Pauli spectrum, such as a GHZ state or a product state with one dominant coefficient, and compute the grouped and ungrouped estimator variances using the exact ceiling shot counts from Eqs. (4) and (20); if any instance gives a grouped-estimator variance larger than the original-estimator variance at the same $\varepsilon$ and $\delta$, the claimed variance bound fails. A simpler check is to scan all states up to $n=4$ with exact ceilings and test whether the inequality holds in every case.

Watch

Extended reading notes

Core claim

The central discovery is that DFE's sampling step can operate on groups rather than on individual Pauli strings without losing unbiasedness or the confidence bound. Replacing the per-string distribution $b_k^2$ by the group distribution $\|\mathbf{b}_k\|^2$ turns each sampled unit into a whole commuting family, and because the family shares a common eigenbasis, one measurement setting estimates all its expectation values at once. The paper proves that the expected number of copies is bounded by $1 + 1/(\varepsilon^2\delta) + (2d/\varepsilon^2) \ln(2/\delta)$, the same asymptotic scaling as DFE, and, using Cauchy–Schwarz and Hoeffding, derives a variance comparison under the condition $\|\mathbf{b}_k\|_1^2 / \|\mathbf{b}_k\|^4 \gg 1$. Numerically the variance drops by 92% (QWC) or 99% (FC) for 8-qubit Haar-random states at $\varepsilon = 0.05$, $\delta = 0.05$, and the estimator satisfies $\Pr[|Y_{\mathrm{exp}} - F| \le 2\epsilon] \ge 1-2\delta$.

Load-bearing premise

The variance improvement is proven only under the approximation that rounding the per-group shot count up to the nearest integer is negligible, which the paper justifies by an asymptotic concentration condition for Haar-random states without giving a finite-size threshold.

Editorial extensions

If this is right

  • For any pure target state, the grouped protocol's expected total copy count has the upper bound $1 + 1/(\varepsilon^2\delta) + (2d/\varepsilon^2) \ln(2/\delta)$, matching standard DFE's scaling, so it never asymptotically needs more copies.
  • The guarantee $\Pr[|Y_{\mathrm{exp}} - F| \le 2\epsilon] \ge 1-2\delta$ carries over to the grouped estimator, so accuracy and confidence are preserved while resources drop.
  • With QWC grouping, only single-qubit basis rotations are needed, so the reduction is implementable on near-term devices; FC grouping trades two-qubit gates for a larger variance reduction.
  • The numerical advantage grows with system size: copy reduction rises from roughly 6% for $n=2$ to about 32–33% for $n=8$ Haar-random states, and variance reduction grows from 53%/63% to 92%/99% for QWC/FC.

Reading between the lines

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

  • The analytical variance bound relies on the unproven finite-size regime where rounding $m_{k_i}$ up to the nearest integer changes nothing, $\lceil m_{k_i}\rceil \approx m_{k_i}$; a finite-$n$ search over sparse target states such as GHZ or product states could reveal whether the variance advantage holds exactly or only asymptotically.
  • The paper leaves open whether overlapping grouping schemes such as ICS/IMA would give further copy reduction if their classical preprocessing cost were accepted; the same group-resampling logic would apply directly to DFE.
  • A natural testable extension is to run the protocol with exact ceilings and realistic readout error on hardware to see whether the 32% copy saving and the 92–99% variance saving survive finite shot noise and gate errors.
  • Because the grouping depends only on the target state $\rho$, the polynomial preprocessing cost is paid once and amortized over repeated verifications of the same state, the typical pattern in device benchmarking.
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

3 major / 4 minor

Summary. The manuscript proposes an enhanced direct fidelity estimation (DFE) protocol in which the Pauli strings of the target pure state are first grouped into commuting families, the groups are sampled with probability proportional to ||b_k||^2, and each sampled group is measured in its common eigenbasis. The authors derive a Hoeffding-based bound on the per-group copy number (Eq. (20)), an expected-total-copies bound (Eq. (24)), and claim in Appendix B that the grouped estimator has variance no larger than standard DFE. They benchmark the protocol numerically on 8-qubit Haar-random, W, and GHZ states, reporting a ~32% copy reduction and variance reductions of 92% (QWC) and 99% (FC) for Haar-random states.

Significance. The basic idea of grouping Pauli operators before importance sampling is natural and practically motivated, and the numerical simulations cover several state families and include a repository link, which is a concrete strength. If the copy-count bound is valid for the non-overlapping groupings actually used, the protocol is a sensible practical variant of DFE with the same asymptotic copy scaling and improved constants in numerics. However, the analytical variance-dominance claim in the abstract and conclusions is not established: Appendix B contains a normalization inconsistency, omits non-sign-definite same-round covariance terms, and relies on an unquantified asymptotic approximation. The contribution would be solid as a copy-count bound plus a numerical study, but the variance theorem needs repair or a weakened statement.

major comments (3)
  1. [Sec. 3, Eq. (24)] The step 'sum_{k=1}^g g_k is at most d^2 for an arbitrary grouping technique' is false for overlapping groupings. In the overlapping framework described in Sec. 2.2 (ICS/IMA), a Pauli operator can appear in multiple groups, so sum_k g_k is not bounded by d^2. Consequently Eq. (24) proves the copy-count bound only for non-overlapping groupings such as the SI partition used in the numerics. The general claim in Sec. 5 that the total copy number is upper-bounded by that of standard DFE should be restricted to non-overlapping groupings, or a separate argument must be supplied for overlapping cases.
  2. [Appendix B, Eq. (30)] The derivation of E[tilde X^2] is internally inconsistent and incomplete. The text states E[c_{k_i l}^{(r_j)}] = a_{k_i l}, whereas Eq. (14) gives E[c] = sqrt(d) a_{k_i l}; this normalization error propagates through the second-moment computation. More importantly, the expansion of (sum_{j=1}^m C_j)^2 keeps the diagonal terms sum_j sum_l E[(c_{j,l} b_l)^2] and the cross-round terms sum_{j != j'} sum_{l,t} E[c_{j,l}] E[c_{j',t}] b_l b_t, but drops the same-round off-diagonal terms sum_j sum_{l != t} E[c_{j,l} c_{j,t}] b_l b_t. For a common-eigenbasis measurement these terms are not equal to sum_{l != t} E[c_{j,l}] E[c_{j,t}] b_l b_t and are not sign-definite, so the inequality 'less than or equal to' in Eq. (30) does not follow. As written, Appendix B does not prove Var(tilde Y) <= Var(hat Y).
  3. [Appendix B, final paragraph] The variance comparison also relies on the approximation ceil(m_k) approx m_k, controlled only by the condition ||b_k||_1^2 / ||b_k||^4 >> 1, with the assertion that Haar-random states satisfy |b_{k l}| approx 1/d. No finite-size threshold in n, epsilon, or delta is derived, and the appendix explicitly works in this asymptotic regime. Since the abstract and Sec. 5 claim an analytical upper bound on the variance, this limitation should be quantified or the claim should be downgraded to an asymptotic or empirically supported statement.
minor comments (4)
  1. [Sec. 3, Eq. (21)] The probability statement should involve the absolute value: Pr[|tilde Y - F| <= 2 epsilon] >= 1 - 2 delta.
  2. [Sec. 2.1, text after Eq. (1)] The dimension is written as 'd = 2 n'; this should be d = 2^n.
  3. [Fig. 3 caption] The caption says DFE is blue, QWC is red, and FC is blue; the color coding should be made unambiguous, especially since the same color name is used twice.
  4. [Code availability] The text contains the typo 'Pyhton'; it should be 'Python'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the derivation is self-contained and the numerical gains are simulated outcomes rather than fitted inputs.

full rationale

The paper's central estimators are defined by an identity: Eq. (6) rewrites the fidelity as E[X] under the distribution ||b_k||^2, and Eq. (8) defines X_k so that unbiasedness is automatic. This is a construction, not a circular prediction. The copy-count bound in Eqs. (22)-(24) follows from the triangle inequality ||b_k||_1 ≤ sqrt(g_k)||b_k|| and the counting bound Σg_k ≤ d^2; no fitted constant enters. The variance comparison in Appendix B uses the same protocol parameters m_ki and the stated condition ||b_k||_1^2/||b_k||^4 >> 1; while the proof has correctness gaps (the passage from the penultimate to the final line of Eq. (30) is not a valid upper bound, and the same-round covariance terms are not controlled), those are proof errors, not instances of fitting or self-reference. The reference list contains no work by the present authors, so no load-bearing self-citation exists. The reported 92%/99% variance reductions and 32% copy reduction are measured simulation outputs for the stated protocol at fixed ε=0.05, δ=0.05, not parameters fitted to make the prediction come true. Therefore no circularity is established; the Appendix B concern should be treated as a correctness risk, not a circularity.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The protocol introduces no new physical entities and fits no free parameters. Its analysis relies on standard quantum measurement facts and on a concentration assumption about Haar-random state coefficients. The main caveat is that the copy-count bound assumes non-overlapping groupings while the text states it for arbitrary ones.

assumptions (5)
  • domain assumption The target state rho is pure, so the squared Pauli coefficients sum to 1.
    Standard DFE requirement, stated in Sec. 2.1 and used to define the importance sampling distribution in Eqs. (2) and (6).
  • standard math Pauli operators in a commuting group can be measured simultaneously in a common eigenbasis.
    Basis of the grouping advantage, used in Sec. 3 around Eqs. (12)-(14).
  • domain assumption For Haar-random pure states, Pauli coefficients concentrate near |b_kl| approx 1/d, giving ||b||_1^2 / ||b||^4 approx d^2 >> 1.
    Justifies the ceil(m) approx m approximation in App. B; stated in Sec. 3 and App. B without finite-size quantification.
  • domain assumption The grouping is non-overlapping, so the sum of group sizes equals d^2.
    Used in Eq. (24) to convert sum g_k to d^2; the text overclaims this holds for 'arbitrary' groupings, but overlapping groupings violate it.
  • standard math Chebyshev's and Hoeffding's inequalities apply to the sums of independent measurement rounds.
    Used for the error bounds in Eqs. (11) and (17).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Sampling Groups of Pauli Operators to Enhance Direct Fidelity Estimation." pith.science (2026). https://pith.science/paper/FAKOC5ZB

@misc{pith2026250119228,
  author       = {Pith},
  title        = {Pith review of: Sampling Groups of Pauli Operators to Enhance Direct Fidelity Estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FAKOC5ZB}},
  note         = {Machine review of arXiv:2501.19228}
}
read the original abstract

Direct fidelity estimation is a protocol that estimates the fidelity between an experimental quantum state and a target pure state. By measuring the expectation values of Pauli operators selected through importance sampling, the method is exponentially faster than full quantum state tomography. We propose an enhanced direct fidelity estimation protocol that uses fewer copies of the experimental state by grouping Pauli operators before the sampling process. We derive analytical bounds on the measurement cost and estimator variance, showing improvements over the standard method. Numerical simulations validate our approach, demonstrating that for 8-qubit Haar-random states, our method achieves a one-third reduction in the required number of copies and reduces variance by an order of magnitude using only local measurements. These results underscore the potential of our protocol to enhance the efficiency of fidelity estimation in current quantum devices.

Figures

Figures reproduced from arXiv: 2501.19228 by the authors.

Figure 1
Figure 1. Diagram illustrating the different frameworks [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Improved protocol for estimating the fidelity between an unknown experimental state [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. The plots are obtained using 1000 samples of 8-qubit states with parameters set to [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 32 canonical work pages

  1. [1]

    D. F. V. James, P. G. Kwiat, W. J. Munro and A. G. White, Measurement of qubits, Phys. Rev. A64, 052312 (2001)

    work page 2001

  2. [2]

    R. T. Thew, K. Nemoto, A. G. White and W. J. Munro, Qudit quantum-state tomog- raphy, Phys. Rev. A66, 012303 (2002)

    work page 2002

  3. [3]

    S. T. Flammia and Y.-K. Liu, Direct Fidelity Estimation from Few Pauli Measurements, Phys. Rev. Lett.106 (2011)

    work page 2011

  4. [4]

    M. P. da Silva, O. Landon-Cardinal and D. Poulin, Practical Characterization of Quantum Devices without Tomography, Phys. Rev. Lett.107, 210404 (2011)

    work page 2011

  5. [5]

    Fawzi, A

    O. Fawzi, A. Oufkir and R. Salzmann, Op- timal Fidelity Estimation from Binary Mea- surements for Discrete and Continuous Vari- able Systems (2024), arXiv:2409.04189

    arXiv 2024

  6. [6]

    Huang, R

    H.-Y. Huang, R. Kueng and J. Preskill, Pre- dicting many properties of a quantum sys- tem from very few measurements, Nat. Phys. 16, 1050 (2020)

    work page 2020

  7. [7]

    Huang, R

    H.-Y. Huang, R. Kueng and J. Preskill, Effi- cient Estimation of Pauli Observables by De- randomization, Phys. Rev. Lett.127, 030503 (2021)

    work page 2021

  8. [8]

    Bharti et al., Noisy intermediate-scale quantum algorithms, Rev

    K. Bharti et al., Noisy intermediate-scale quantum algorithms, Rev. Mod. Phys. 94, 015004 (2022)

    work page 2022

Show all 34 references
  1. [9]

    Kandala et al., Hardware-efficient vari- ational quantum eigensolver for small molecules and quantum magnets, Nature 549, 242 (2017)

    A. Kandala et al., Hardware-efficient vari- ational quantum eigensolver for small molecules and quantum magnets, Nature 549, 242 (2017)

    2017

  2. [10]

    A. F. Izmaylov, T.-C. Yen and I. G. Ryabinkin, Revising the measurement pro- cess in the variational quantum eigensolver: is it possible to reduce the number of sepa- rately measured operators?, Chem. Sci. 10, 3746 (2019)

    2019

  3. [11]

    Verteletskyi, T.-C

    V. Verteletskyi, T.-C. Yen and A. F. Iz- maylov, Measurement optimization in the variational quantum eigensolver using a min- imum clique cover, J. Chem. Phys. 152 (2020)

    2020

  4. [12]

    A. Jena, S. N. Genin and M. Mosca, Opti- mization of variational-quantum-eigensolver measurement by partitioning Pauli opera- tors using multiqubit Clifford gates on noisy intermediate-scale quantum hardware, Phys. Rev. A 106, 042443 (2022)

    2022

  5. [13]

    T.-C. Yen, V. Verteletskyi and A. F. Iz- maylov, Measuring all compatible operators in one series of single-qubit measurements using unitary transformations, J. Chem. Theory Comput. 16, 2400 (2020)

    2020

  6. [14]

    A. F. Izmaylov, T.-C. Yen, R. A. Lang and V. Verteletskyi, Unitary partitioning ap- proach to the measurement problem in the variational quantum eigensolver method, J. Chem. Theory Comput.16, 190 (2019)

    2019

  7. [15]

    Zhao et al., Measurement reduction in Accepted inQuantum 2025-06-25, click title to verify

    A. Zhao et al., Measurement reduction in Accepted inQuantum 2025-06-25, click title to verify. Published under CC-BY 4.0. 9 variational quantum algorithms, Phys. Rev. A 101, 062322 (2020)

    2020

  8. [16]

    Hamamura and T

    I. Hamamura and T. Imamichi, Efficient evaluation of quantum observables using en- tangled measurements, npj Quantum Inf.6, 56 (2020)

    2020

  9. [17]

    Crawford et al., Efficient quantum mea- surement of Pauli operators in the presence of finite sampling error, Quantum 5, 385 (2021)

    O. Crawford et al., Efficient quantum mea- surement of Pauli operators in the presence of finite sampling error, Quantum 5, 385 (2021)

    2021

  10. [18]

    T.-C. Yen, A. Ganeshram and A. F. Izmaylov, Deterministic improvements of quantum measurements with grouping of compatible operators, non-local transforma- tions, and covariance estimates, npj Quan- tum Inf. 9, 14 (2023)

    2023

  11. [19]

    Hadfield, S

    C. Hadfield, S. Bravyi, R. Raymond and A. Mezzacapo, Measurements of quantum Hamiltonians with locally-biased classical shadows, Commun. Math. Phys. 391, 951 (2022)

    2022

  12. [20]

    Gresch and M

    A. Gresch and M. Kliesch, Guaranteed effi- cient energy estimation of quantum many- body Hamiltonians using ShadowGrouping, Nat. Commun. 16 (2025)

    2025

  13. [21]

    Wu, J.Sun, Q

    B. Wu, J.Sun, Q. Huangand X. Yuan, Over- lapped grouping measurement: A unified framework for measuring quantum states, Quantum 7, 896 (2023)

    2023

  14. [22]

    Kliesch and I

    M. Kliesch and I. Roth, Theory of Quan- tum System Certification, PRX Quantum2, 010201 (2021)

    2021

  15. [23]

    Yen and A

    T.-C. Yen and A. F. Izmaylov, Cartan sub- algebra approach to efficient measurements of quantum observables, PRX Quantum 2, 040320 (2021)

    2021

  16. [24]

    Shlosberg et al., Adaptive estimation of quantum observables, Quantum 7, 906 (2023)

    A. Shlosberg et al., Adaptive estimation of quantum observables, Quantum 7, 906 (2023)

    2023

  17. [25]

    Preskill, Quantum Computing in the NISQ era and beyond, Quantum 2, 79 (2018)

    J. Preskill, Quantum Computing in the NISQ era and beyond, Quantum 2, 79 (2018)

    2018

  18. [26]

    Leone, S

    L. Leone, S. F. E. Oliviero and A. Hamma, Nonstabilizerness determining the hardness of direct fidelity estimation, Phys. Rev. A 107 (2023)

    2023

  19. [27]

    Barends et al., Superconducting quan- tum circuits at the surface code threshold for fault tolerance, Nature508, 500 (2014)

    R. Barends et al., Superconducting quan- tum circuits at the surface code threshold for fault tolerance, Nature508, 500 (2014)

    2014

  20. [28]

    M. C. Smith, A. D. Leu, K. Miyanishi, M. F. Gely and D. M. Lucas, Single-Qubit Gates with Errors at the 10−7 Level, Phys. Rev. Lett. 134, 230601 (2025)

    2025

  21. [29]

    Y. Chen, M. Farahzad, S. Yoo and T.-C. Wei, Detector tomography on IBM quan- tum computers and mitigation of an imper- fect measurement, Phys. Rev. A100, 052315 (2019)

    2019

  22. [30]

    F. B. Maciejewski, Z. Zimborás and M. Osz- maniec, Mitigation of readout noise in near- term quantum devices by classical post- processing based on detector tomography, Quantum 4, 257 (2020)

    2020

  23. [31]

    Bravyi, S

    S. Bravyi, S. Sheldon, A. Kandala, D. C. Mckay and J. M. Gambetta, Mitigating mea- surement errors in multiqubit experiments, Phys. Rev. A103, 042605 (2021)

    2021

  24. [32]

    Seshadri, M

    A. Seshadri, M. Ringbauer, J. Spainhour, T. Monz and S. Becker, Theory of versa- tilefidelityestimationwithconfidence,Phys. Rev. A 110, 012431 (2024)

    2024

  25. [33]

    S. Chen, W. Yu, P. Zeng and S. T. Flammia, Robust Shadow Estimation, PRX Quantum 2, 030348 (2021)

    2021

  26. [34]

    Hu et al., Demonstration of robust andefficientquantumpropertylearningwith shallow shadows, Nat

    H.-Y. Hu et al., Demonstration of robust andefficientquantumpropertylearningwith shallow shadows, Nat. Commun.16 (2025). Accepted inQuantum 2025-06-25, click title to verify. Published under CC-BY 4.0. 10 A Sorted insertion In this section, we present the pseudocode for the so...

    2025

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.