Pith. sign in

REVIEW 1 major objections 22 references

AKS-QFI separates Krylov truncation from sampling error to avoid false stopping decisions in shadow QFI estimation.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.3

2026-06-30 20:57 UTC pith:MFQGOT6M

load-bearing objection The paper frames adaptive stopping for Krylov-shadow QFI as two separate reliability components and shows lower false-stop rates than width-only rules on an n=4 benchmark, but the recalibration step lacks a clear ground-truth-free procedure. the 1 major comments →

arxiv 2605.14338 v2 pith:MFQGOT6M submitted 2026-05-14 quant-ph

Reliable Adaptive Stopping for Krylov-Shadow Quantum Fisher Information Estimation

Erjie Liu , Yangshuai Wang This is my paper
classification quant-ph
keywords quantum Fisher informationKrylov-shadow estimationadaptive stoppingshadow tomographyQFI estimationmixed-state benchmarkerror separationfinite-sample uncertainty
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Krylov-shadow QFI estimation combines two resource choices: Krylov order sets how finely the state populations are resolved, while sample count sets the statistical precision at that resolution. Treating the combined width as a single stopping signal can produce false stops, in which the rule reports a tight interval around a biased low-order estimate. The paper reframes adaptive stopping as a two-component reliability task that isolates truncation error from sampling uncertainty and introduces AKS-QFI to monitor and recalibrate each component separately. On a noisy four-qubit mixed-state benchmark, width-only rules produce false success declarations between 16 and 68 percent of the time; AKS-QFI produces none under identical resource limits and, after independent recalibration, correctly declares success at the true 5 percent relative tolerance.

Core claim

Krylov-shadow QFI estimation has two independent resource directions: Krylov order controls population resolution while sample count controls statistical uncertainty. Treating them as a single width produces false stops on biased estimates. AKS-QFI treats adaptive stopping as a reliability problem that separates truncation error from sampling error and recalibrates each component independently, yielding zero false success declarations on the benchmark and accurate success at 5% tolerance after recalibration.

What carries the argument

AKS-QFI, the component-aware adaptive stopping interface that decouples Krylov truncation error from finite-sample uncertainty for Krylov-shadow QFI estimators.

Load-bearing premise

The two error sources of Krylov truncation and finite-sample uncertainty can be separated and recalibrated independently without the recalibration step introducing bias or requiring knowledge of the true QFI value.

What would settle it

On a system whose exact QFI is known by other means, run AKS-QFI until it declares success at the claimed tolerance and verify whether the actual relative error is below 5 percent while width-only stopping declares success on estimates whose true error exceeds that threshold.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Width-only stopping rules can report narrow intervals around biased low-order QFI estimates.
  • AKS-QFI returns no false success declarations under the same resource limits where width-only rules fail 16 to 68 percent of the time.
  • Independent recalibration of Krylov resolution and sample counts produces accurate success declarations at true 5 percent relative tolerance.
  • Adaptive stopping functions as a reliability layer for shadow-based QFI estimation.

Where Pith is reading between the lines

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

  • The separation of truncation and sampling errors may extend to other shadow protocols that combine approximation order with Monte Carlo sampling.
  • Composite width metrics are likely insufficient for reliable stopping whenever multiple distinct error sources are present in quantum estimation tasks.
  • The recalibration procedure could be tested on systems larger than four qubits to check whether the independence assumption continues to hold.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 0 minor

Summary. The paper introduces AKS-QFI, a two-component adaptive stopping interface for Krylov-shadow QFI estimation that separates Krylov truncation error from finite-sample statistical uncertainty. It claims that width-only stopping produces false stops with rates 0.16-0.68 on an n=4 noisy mixed-state benchmark, while AKS-QFI yields zero false success declarations under the same resource limit and, after recalibrating Krylov resolution and sample counts, produces accurate success declarations at true 5% relative tolerance.

Significance. If the recalibration step can be performed using only observable quantities, the separation of error sources would supply a practical reliability layer for shadow-based QFI estimators, addressing a recognized weakness in adaptive stopping for quantum metrology. The n=4 benchmark provides initial evidence of reduced false-stop risk relative to width-only rules.

major comments (1)
  1. [Abstract] Abstract: the concrete false-stop rates (0.16-0.68) and the claim of zero false-success declarations plus accurate 5% tolerance after recalibration are stated without any derivation, error-bar information, number of independent runs, or description of the recalibration procedure itself. Because the recalibration step is load-bearing for the reliability claim, its absence prevents evaluation of whether the reported performance can be reproduced without ground-truth QFI knowledge.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading and for identifying the need for greater transparency in the abstract regarding our numerical claims. We address the comment point by point below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the concrete false-stop rates (0.16-0.68) and the claim of zero false-success declarations plus accurate 5% tolerance after recalibration are stated without any derivation, error-bar information, number of independent runs, or description of the recalibration procedure itself. Because the recalibration step is load-bearing for the reliability claim, its absence prevents evaluation of whether the reported performance can be reproduced without ground-truth QFI knowledge.

    Authors: The false-stop rates (0.16-0.68) and zero false-success declarations are computed from the n=4 noisy mixed-state benchmark experiments reported in Section 4, which consist of 100 independent runs; one-standard-deviation error bars on these rates appear in Figure 3 and Table II. The recalibration procedure itself is described in Section 5.2: at each step we use only observable quantities (the empirical shadow variance for the statistical component and the difference between successive Krylov-order estimates for the truncation component) to decide whether to increase Krylov order or sample count, without access to ground-truth QFI. We agree that the abstract is too terse on these points and will revise it to (i) state the number of runs, (ii) note that error bars are reported in the main text, and (iii) briefly indicate that recalibration uses only observable quantities, with a pointer to Section 5.2. revision: yes

Circularity Check

0 steps flagged

No significant circularity; benchmark claims remain external to method definition

full rationale

The paper introduces AKS-QFI as a two-component adaptive stopping rule separating Krylov truncation error from finite-sample uncertainty. The reported false-stop rates (0.16-0.68 for width-only) and zero false-success declarations under AKS-QFI are presented as outcomes of an external n=4 noisy mixed-state benchmark. No equations, fitted parameters, or self-citations are exhibited that reduce the success declarations, 5% tolerance claims, or recalibration step to a definition or input internal to the stopping rule itself. The derivation chain for the reliability layer is therefore self-contained against the stated benchmark evidence.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities are stated in the provided text.

pith-pipeline@v0.9.1-grok · 5709 in / 1253 out tokens · 30649 ms · 2026-06-30T20:57:34.664795+00:00 · methodology

0 comments
read the original abstract

Scalable quantum Fisher information (QFI) estimation becomes actionable when the numerical estimate is paired with a trustworthy stopping decision. Krylov-shadow QFI estimation has two resource directions: the Krylov order sets the population resolution, whereas the sample count controls statistical uncertainty at that resolution. We show that treating these directions as one can produce false stops, where a width-based rule reports a narrow interval around a biased low-order estimate. We turn adaptive stopping into a two-component reliability problem, separating Krylov truncation from finite-sample uncertainty, and introduce AKS-QFI, a component-aware stopping interface for Krylov-shadow estimators. On a noisy mixed-state benchmark at $n=4$ qubits, width-only stopping has false-stop rates from $0.16$ to $0.68$. Under the same resource limit, AKS-QFI returns no false success declarations; after recalibrating Krylov resolution and sample counts, it returns accurate success declarations at true 5% relative tolerance. These results make adaptive stopping a reliability layer for shadow-based QFI estimation.

Figures

Figures reproduced from arXiv: 2605.14338 by Erjie Liu, Yangshuai Wang.

Figure 1
Figure 1. Figure 1: Narrow empirical intervals do not establish accuracy under width-only stopping. Termination￾time empirical interval width wM versus post hoc abso￾lute error E for all runs (n = 4, five dephasing levels, 50 independent replicates per noise level). Width-only runs form a cluster in the upper-left region (wM ≤ ε, E > ε): these are false stops in the sense of Eq. (18). The empirical gate can pass at the termin… view at source ↗
Figure 2
Figure 2. Figure 2: Component-aware stopping suppresses false success declarations under the default resource limit. False-stop rate (FSR, filled markers) and stop rate (SR, open markers) versus dephasing probability pϕ (n = 4, 50 independent replicates per point), with 95% Wilson confidence intervals. Width-only FSR increases with pϕ and reaches 0.68 at pϕ = 0.24, while component￾aware FSR remains at zero throughout. Compone… view at source ↗
Figure 3
Figure 3. Figure 3: Width-only stopping’s low effective cost reflects premature termination at a biased value. Median post hoc absolute error versus pϕ (n = 4, 50 independent replicates per point). The two rules diverge at pϕ ≥ 0.18: component-aware stopping reduces me￾dian error by up to a factor of 4.8 relative to width-only stopping despite using up to 16× larger effective termi￾nal sample counts at these noise levels, ind… view at source ↗
Figure 4
Figure 4. Figure 4: shows the decision geometry behind this control. All component-aware terminal points lie below the true-error threshold, while runs without a success label are precisely the cases where the bootstrap width remains above threshold. The contrast with the default grid is the central message: zero false success under in￾sufficient resources is only the safety half of the story, whereas calibrated Krylov resolu… view at source ↗
Figure 5
Figure 5. Figure 5: Threshold sensitivity under a fixed-resource ablation. Each cell is one stopping configuration (K stop min , Mstop min , P) aggregated over replicates at fixed noise (pϕ = 0.12, pdep = 0.03), ε = 0.2, resource limits (Kmax, Mmax) = (8, 512). Color and cell text give the false-stop rate. The only configuration with nonzero FSR is the weakest configuration, (2, 32, 1). The default (4, 128, 2) is outlined in … view at source ↗
Figure 6
Figure 6. Figure 6: Empirical Krylov convergence and observable stability. For each system size, markers show median inter-order change dK and post hoc truncation bias |BK| versus Krylov order at pϕ = 0.12; dashed curves show fitted exponential trends. For n = 4, both quantities decay with overlapping fitted rates. For n ∈ {6, 8}, the bias is nearly flat within K ≤ Kmax = 8, explaining why the default stopping rule requires l… view at source ↗
Figure 8
Figure 8. Figure 8: Early termination under width-only stop￾ping can appear efficient while remaining inaccurate. Post hoc absolute error versus effective terminal sam￾ple count for the benchmark (logarithmic vertical axis). Width-only runs that terminate at low effective sample counts frequently exhibit large post hoc errors, whereas component-aware runs concentrate near the resource limit and avoid false success declaration… view at source ↗
Figure 9
Figure 9. Figure 9: Component-aware stopping uses more ter￾minal samples under the current resource limits. Median effective terminal sample count versus dephas￾ing probability p in the n = 4 benchmark (50 indepen￾dent replicates per point). In the tested configuration, component-aware stopping consistently reaches the re￾source limit (median effective sample count 512), while width-only stopping often terminates much earlier… view at source ↗
Figure 10
Figure 10. Figure 10: Stop-precision summary for success dec￾larations. Each cell reports the post hoc stop precision for a rule and dephasing probability in the n = 4 bench￾mark (50 independent replicates per point), defined as the fraction of success declarations satisfying E ≤ ε. Width-only declarations have precision zero at all tested dephasing levels. Component-aware entries are marked n/a because component-aware runs ma… view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages · 2 internal anchors

  1. [1]

    Quantum sens- ing

    Christian L Degen, Friedemann Reinhard, and Paola Cappellaro. “Quantum sens- ing”. Reviews of modern physics 89, 035002 (2017)

    work page 2017

  2. [2]

    Advances in quan- tum metrology

    Vittorio Giovannetti, Seth Lloyd, and Lorenzo Maccone. “Advances in quan- tum metrology”. Nature photonics5, 222– 229 (2011)

    work page 2011

  3. [3]

    Quantum metrology and its application in biology

    Michael A Taylor and Warwick P Bowen. “Quantum metrology and its application in biology”. Physics Reports615, 1–59 (2016)

    work page 2016

  4. [4]

    Quantum fisher informa- tion matrix and multiparameter estimation

    Jing Liu, Haidong Yuan, Xiao-Ming Lu, and Xiaoguang Wang. “Quantum fisher informa- tion matrix and multiparameter estimation”. JournalofPhysicsA:MathematicalandThe- oretical53, 023001 (2020)

    work page 2020

  5. [5]

    Introduc- tion to quantum fisher information

    Dénes Petz and Catalin Ghinea. “Introduc- tion to quantum fisher information”. In Quantum probability and related topics. Pages 261–281. World Scientific (2011)

    work page 2011

  6. [6]

    Quantum fisher information for states in exponential form

    Zhang Jiang. “Quantum fisher information for states in exponential form”. Physical Re- view A89, 032128 (2014)

    work page 2014

  7. [7]

    Quantum fisher information from randomized mea- surements

    Aniket Rath, Cyril Branciard, Anna Min- guzzi, and Benoît Vermersch. “Quantum fisher information from randomized mea- surements”. Physical Review Letters127, 260501 (2021)

    work page 2021

  8. [8]

    Fisher information in noisy intermediate-scale quantum applica- tions

    Johannes Jakob Meyer. “Fisher information in noisy intermediate-scale quantum applica- tions”. Quantum5, 539 (2021)

    work page 2021

  9. [9]

    Variational quantum al- gorithm for estimating the quantum fisher information

    Jacob L Beckey, M Cerezo, Akira Sone, and Patrick J Coles. “Variational quantum al- gorithm for estimating the quantum fisher information”. Physical Review Research4, 013083 (2022)

    work page 2022

  10. [10]

    Robust estimation of the quan- tum fisher information on a quantum proces- sor

    Vittorio Vitale, Aniket Rath, Petar Jurcevic, Andreas Elben, Cyril Branciard, and Benoît Vermersch. “Robust estimation of the quan- tum fisher information on a quantum proces- sor”. PRX Quantum5, 030338 (2024)

    work page 2024

  11. [11]

    Huang, R

    Hsin-Yuan Huang, Richard Kueng, and John Preskill. “Predicting many proper- ties of a quantum system from very few measurements”. Nature Physics16, 1050– 1057 (2020). arXiv:2002.08953

    work page arXiv 2020

  12. [12]

    Elben, S

    Andreas Elben, Steven T. Flammia, Hsin- Yuan Huang, Richard Kueng, John Preskill, Benoît Vermersch, and Peter Zoller. “The randomized measurement toolbox”. Na- ture Reviews Physics 5, 9–24 (2023). arXiv:2203.11374

    work page arXiv 2023

  13. [13]

    Krylov shadow tomography: Efficient estimation of quantum fisher information

    Da-Jian Zhang and D. M. Tong. “Krylov shadow tomography: Efficient estimation of quantum fisher information”. Phys. Rev. Lett.134, 110802 (2025). arXiv:2503.01697

    work page arXiv 2025

  14. [14]

    Shadow Tomography of Quantum States

    Scott Aaronson. “Shadow tomography of quantum states”. SIAM Journal on Comput- ing 49, STOC18–368–STOC18–394 (2020). arXiv:1711.01053

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  15. [15]

    On the classical shadow nonparametric boot- strap

    Eric Ghysels and Jack Morgan. “On the classical shadow nonparametric boot- strap” (2025). arXiv:2511.09793

    work page arXiv 2025

  16. [16]

    Sequential tests of statis- tical hypotheses

    Abraham Wald. “Sequential tests of statis- tical hypotheses”. Ann. Math. Statist.16, 117–186 (1945)

    work page 1945

  17. [17]

    Time- uniform, nonparametric, nonasymptotic con- fidence sequences

    Steven R. Howard, Aaditya Ramdas, Jon McAuliffe, and Jasjeet Sekhon. “Time- uniform, nonparametric, nonasymptotic con- fidence sequences”. Ann. Statist.49, 1055– 1080 (2021)

    work page 2021

  18. [18]

    Self-guided quantum to- mography

    Christopher Ferrie. “Self-guided quantum to- mography”. Physical Review Letters 113, 190404 (2014)

    work page 2014

  19. [19]

    Robust Online Hamiltonian Learning

    Christopher E. Granade, Christopher Ferrie, Nathan Wiebe, and D. G. Cory. “Robust on- line Hamiltonian learning”. New Journal of Physics 14, 103013 (2012). arXiv:1207.1655

    work page internal anchor Pith review Pith/arXiv arXiv 2012

  20. [20]

    Ban- dit algorithms

    Tor Lattimore and Csaba Szepesvári. “Ban- dit algorithms”. Cambridge University Press. (2020)

    work page 2020

  21. [21]

    PAC bounds for multi-armed bandit and Markov decision processes

    Eyal Even-Dar, Shie Mannor, and Yishay Mansour. “PAC bounds for multi-armed bandit and Markov decision processes”. In Proceedings of the 15th Annual Conference on Computational Learning Theory (COLT). Volume 2375 of Lecture Notes in Computer Science, pages 255–270. Springer (2002)

    work page 2002

  22. [22]

    Asymptotic statistics

    A. W. van der Vaart. “Asymptotic statistics”. Number 3 in Cambridge Series in Statistical and Probabilistic Mathematics. Cambridge University Press. Cambridge (1998). 25

    work page 1998