Finite-Sample Selected Covariance Spectra in Classical Shadows

Masahito Hayashi

Operator-norm bounds show selected covariance estimation in classical shadows can require samples independent of system dimension for local protocols.

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

2026-06-28 18:55 UTC pith:GPM6QEYE

load-bearing objection This paper gives explicit operator-norm bounds on selected centered covariances for classical shadows, with dimension-independent sample size for local protocols and closed-form expressions for biased Pauli cases.

arxiv 2606.00527 v1 pith:GPM6QEYE submitted 2026-05-30 quant-ph

Finite-Sample Selected Covariance Spectra in Classical Shadows

Masahito Hayashi This is my paper

classification quant-ph

keywords classical shadowscovariance matrixfinite sample boundsoperator normlocal Pauli measurementsdimension independencequantum state estimation

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.

The paper proves an operator-norm error bound for the selected sample-centered empirical covariance of classical-shadow outputs that holds for any shadow protocol. When protocol constants in this bound do not depend on system size, the sample complexity becomes independent of dimension. This is verified for local product protocols where bounds depend on support sizes and local coefficients instead of total dimension. For biased local Pauli shadows, closed-form expressions confirm the dimension-independent behavior under uniform bounds on relevant parameters.

Core claim

Our main theorem applies to arbitrary shadow protocols and gives an operator-norm error bound for the selected sample-centered empirical covariance. When the protocol-dependent constants appearing in this bound remain independent of the ambient system size, the required sample size is also independent of the ambient dimension. The proof combines matrix Bernstein concentration, an exact rank-one centering identity, and Weyl and Davis--Kahan perturbation bounds. We verify this bounded-output condition for local measurement settings, leading to dimension-independent selected covariance estimation for general local product shadow protocols with fixed local dimension.

What carries the argument

The operator-norm error bound on the selected sample-centered empirical covariance matrix, obtained via matrix Bernstein concentration together with rank-one centering and perturbation bounds.

Load-bearing premise

The protocol-dependent constants in the operator-norm error bound remain independent of the ambient system size.

What would settle it

A calculation showing that the operator-norm error for selected covariances in a local Pauli shadow protocol grows with the number of qubits, even under fixed local parameters, would disprove the dimension-independent sample complexity.

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

If this is right

Finite-weight product observables in local protocols lead to bounds controlled by support sizes and local reconstruction coefficients.
Uniform bounds on set size, weight, and coefficients imply dimension-independent estimation.
For biased local Pauli shadows, bounds evaluate in closed form from Pauli supports and probabilities.
Exact covariance formula governed by Pauli compatibility shows bias effects on variances and couplings.
Global Clifford shadows do not automatically exhibit this dimension-independent local behavior.

Where Pith is reading between the lines

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

This framework could support efficient covariance analysis in high-dimensional quantum systems using only local measurements.
Similar techniques might apply to other statistical estimators in quantum information processing.
Checking the size-independence of constants for new protocols would determine their suitability for large-scale applications.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

This paper gives explicit operator-norm bounds on selected centered covariances for classical shadows, with dimension-independent sample size for local protocols and closed-form expressions for biased Pauli cases.

read the letter

The core result here is a finite-sample operator-norm bound on the selected sample-centered empirical covariance for arbitrary shadow protocols, plus the observation that local product protocols can keep the required sample size independent of total dimension when certain protocol constants stay bounded.

The work applies matrix Bernstein to the rank-one centered terms, then uses Weyl and Davis-Kahan to control the perturbation from the true covariance. That combination is standard, but the paper works out the exact centering identity for shadow outputs and derives a closed-form covariance expression for biased local Pauli shadows that tracks Pauli compatibility and inverse-probability factors. It also checks the dimension-independence condition explicitly for local protocols: the bounds depend on observable weight, support size, local reconstruction coefficients, and selected-set size, with no n-dependence once local dimension is fixed. The comparison to global Clifford shadows is useful because it shows the local behavior is not automatic.

The soft spots are limited. The main theorem is conditional on the protocol constants remaining dimension-independent, and while the paper verifies this for the local case, the resulting bounds are still expressed in terms of those constants rather than fully simplified numerical values. No numerical experiments appear in the provided material, so practical tightness is not checked. The proof ingredients are off-the-shelf, so the contribution sits in the application and the explicit formulas rather than new concentration machinery.

This is for people working on shadow estimation and quantum tomography who need finite-sample covariance guarantees. A reader already using classical shadows for covariance-related tasks would find the local-protocol conditions and the Pauli closed form directly usable.

It deserves a serious referee. The argument structure is consistent and the derivations rest on verifiable external inequalities without circularity.

Referee Report

0 major / 3 minor

Summary. The manuscript develops finite-sample operator-norm error bounds for selected, sample-centered empirical covariance matrices of classical-shadow outputs. The central theorem applies to arbitrary shadow protocols and combines matrix Bernstein concentration with an exact rank-one centering identity and Weyl/Davis-Kahan perturbation bounds. When protocol-dependent constants in the bound are independent of ambient system size, the required sample size is likewise dimension-independent. The authors verify the bounded-output condition for local product protocols (fixed local dimension), derive a closed-form covariance expression for biased local Pauli shadows governed by Pauli compatibility and inverse-probability factors, and contrast the resulting dimension-independent behavior with global Clifford shadows.

Significance. If the main theorem holds, the work supplies a rigorous, conditional criterion for achieving dimension-independent sample complexity in selected covariance estimation via classical shadows. This is significant for scalable quantum state learning, as it isolates the precise protocol features (support size, local reconstruction coefficients, selected-set cardinality) that control the constants. The explicit derivation of the covariance formula directly from measurement probabilities, rather than fitting, and the verification that local protocols can satisfy the dimension-independent condition are concrete strengths. The conditional framing of the claim and reliance on standard concentration tools are appropriately cautious.

minor comments (3)

[Abstract / Introduction] The abstract and introduction would benefit from an early, explicit definition of the selected compression operator and the centering identity before the main theorem is stated.
[Pauli shadows covariance derivation] In the section deriving the exact covariance formula for biased local Pauli shadows, a small worked numerical example (e.g., two-qubit selected set) would help illustrate how bias affects off-diagonal statistical couplings.
[Comparison with global Clifford shadows] The comparison paragraph with global Clifford shadows would be strengthened by a brief remark on whether any global protocol can ever satisfy the bounded-constant condition.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the thorough summary and positive assessment of our work, including the recognition of its significance for scalable quantum state learning and the appropriateness of our conditional framing. The recommendation of minor revision is noted. No major comments were provided in the report.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The derivation relies on external standard results (matrix Bernstein concentration, Weyl/Davis-Kahan bounds) applied to an exact centering identity obtained directly from measurement probabilities. The dimension-independence claim is explicitly conditional on protocol constants being n-independent, which is verified for local protocols via support-size and coefficient bounds without any fitting, self-definition, or self-citation load-bearing. The Pauli covariance formula is derived from compatibility and inverse-probability factors, not renamed or forced by the target result. The argument is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 0 invented entities

The central claim rests on standard mathematical concentration and perturbation results rather than new axioms or fitted parameters. No free parameters or invented entities are introduced.

axioms (3)

standard math Matrix Bernstein concentration inequality
Invoked to obtain the operator-norm deviation bound for the empirical covariance.
standard math Weyl and Davis-Kahan perturbation bounds
Used to control the effect of the centering and selection on eigenvalues and eigenspaces.
standard math Exact rank-one centering identity
Applied to the sample-centered covariance before concentration.

pith-pipeline@v0.9.1-grok · 5750 in / 1378 out tokens · 20779 ms · 2026-06-28T18:55:31.327408+00:00 · methodology

0 comments

read the original abstract

We study finite-sample estimation of selected covariance matrices of classical-shadow outputs. For a general shadow-output vector, we consider its covariance matrix and a fixed selected compression. Our main theorem applies to arbitrary shadow protocols and gives an operator-norm error bound for the selected sample-centered empirical covariance. When the protocol-dependent constants appearing in this bound remain independent of the ambient system size, the required sample size is also independent of the ambient dimension. The proof combines matrix Bernstein concentration, an exact rank-one centering identity, and Weyl and Davis--Kahan perturbation bounds. We verify this bounded-output condition for local measurement settings. For general local product shadow protocols with fixed local dimension, finite-weight product observables lead to bounds controlled by support sizes and local reconstruction coefficients, not by the total number of tensor factors. Hence uniform bounds on selected set size, observable weight, and local reconstruction coefficients imply dimension-independent selected covariance estimation. For biased local Pauli shadows, we evaluate the relevant bound in closed form from the selected Pauli supports and local basis-selection probabilities. We also derive an exact covariance formula governed by Pauli compatibility and inverse-probability overlap factors, showing how measurement bias affects both diagonal variances and off-diagonal statistical couplings. A comparison with global Clifford shadows shows that this dimension-independent local behavior is not automatic for every shadow protocol.

discussion (0)

Reference graph

Works this paper leans on

24 extracted references · 6 canonical work pages · 1 internal anchor

[1]

This remaining issue is done in the next subsection. F. Empirical mean control and sample-centering transfer The centering identity in Lemma 2 reduces the pas- sage from true-centered to sample-centered covariance to a bound on∥¯ˆzN∥2
[2]

We now bound the selected empirical mean

We control this quantity using the selected centered-radius bound(72), together with coordinatewise Hoeffding inequalities and a union bound over thelselected coordinates. We now bound the selected empirical mean. The rank- one correction is controlled by its only nonzero eigenvalue, ∥¯ˆzN ¯ˆz⊤ N∥op =∥¯ˆzN∥2 2.The coordinates of ˆzi need not be independen...
[3]

T. W. Anderson,An Introduction to Multivariate Statis- tical Analysis, 3rd ed. (Wiley, 2003)

2003
[4]

K. V. Mardia, J. T. Kent, and J. M. Bibby,Multivariate Analysis(Academic Press, 1979)

1979
[5]

I. T. Jolliffe,Principal Component Analysis, 2nd ed. (Springer, 2002)

2002
[6]

J. A. Tropp, User-friendly tail bounds for sums of random matrices, Foundations of Computational Mathematics12, 389 (2012)

2012
[7]

R. A. Horn and C. R. Johnson,Matrix Analysis, 2nd ed. (Cambridge University Press, 2012)

2012
[8]

Davis and W

C. Davis and W. M. Kahan, The rotation of eigenvec- tors by a perturbation. III, SIAM Journal on Numerical Analysis7, 1 (1970)

1970
[9]

G. W. Stewart and J.-G. Sun,Matrix Perturbation Theory (Academic Press, 1990)

1990
[10]

Y. Yu, T. Wang, and R. J. Samworth, A useful variant of the Davis–Kahan theorem for statisticians, Biometrika 102, 315 (2015), arXiv:1405.0680 [math.ST]

work page internal anchor Pith review Pith/arXiv arXiv 2015
[11]

Vershynin,High-Dimensional Probability: An Intro- duction with Applications in Data Science(Cambridge University Press, 2018)

R. Vershynin,High-Dimensional Probability: An Intro- duction with Applications in Data Science(Cambridge University Press, 2018)

2018
[12]

Hadfield, S

C. Hadfield, S. Bravyi, R. Raymond, and A. Mezzacapo, Measurements of quantum hamiltonians with locally- biased classical shadows, Communications in Mathemati- cal Physics391, 951 (2022). 30

2022
[13]

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)

2020
[14]

Elben, S

A. Elben, S. T. Flammia, H.-Y. Huang, R. Kueng, J. Preskill, B. Vermersch, and P. Zoller, The random- ized measurement toolbox, Nature Reviews Physics5, 9 (2023), arXiv:2203.11374 [quant-ph]

work page arXiv 2023
[15]

K. Bu, D. E. Koh, R. J. Garcia, and A. Jaffe, Classi- cal shadows with pauli-invariant unitary ensembles, npj Quantum Information10, 6 (2024)

2024
[16]

Innocenti, S

L. Innocenti, S. Lorenzo, I. Palmisano, F. Albarelli, A. Fer- raro, M. Paternostro, and G. M. Palma, Shadow tomog- raphy on general measurement frames, PRX Quantum4, 040328 (2023), arXiv:2301.13229 [quant-ph]

work page arXiv 2023
[17]

Ippoliti, Y

M. Ippoliti, Y. Li, T. Rakovszky, and V. Khemani, Opera- tor relaxation and the optimal depth of classical shadows, Physical Review Letters130, 230403 (2023)

2023
[18]

Zhou and P

T.-G. Zhou and P. Zhang, Efficient classical shadow tomography through many-body localization dynamics, Quantum8, 1467 (2024)

2024
[19]

S. Chen, W. Yu, P. Zeng, and S. T. Flammia, Robust shadow estimation, PRX Quantum2, 030348 (2021)

2021
[20]

Aliakbarpour, V

M. Aliakbarpour, V. Braverman, N.-H. Chia, C.-Y. Lin, Y. Liu, A. Oufkir, and Y.-C. Shen, Shadow to- mography against adversaries (2025), arXiv preprint, arXiv:2512.05451 [quant-ph]

work page arXiv 2025
[21]

M. West, F. Sauvage, A. Sen, R. Forestano, D. Wierichs, N. Killoran, D. Grinko, M. Cerezo, and M. Larocca, Clas- sical shadows with arbitrary group representations (2026), arXiv preprint, arXiv:2604.01429 [quant-ph]

work page arXiv 2026
[22]

Gandhari, V

S. Gandhari, V. V. Albert, T. Gerrits, J. M. Taylor, and M. J. Gullans, Precision bounds on continuous-variable state tomography using classical shadows, PRX Quantum 5, 010346 (2024)

2024
[23]

Boyd and B

G. Boyd and B. Koczor, Training variational quantum circuits with CoVaR: Covariance root finding with clas- sical shadows, Physical Review X12, 041022 (2022), arXiv:2204.08494 [quant-ph]

work page arXiv 2022
[24]

Hoeffding, Probability inequalities for sums of bounded random variables, Journal of the American Sta- tistical Association58, 13 (1963)

W. Hoeffding, Probability inequalities for sums of bounded random variables, Journal of the American Sta- tistical Association58, 13 (1963)

1963

[1] [1]

This remaining issue is done in the next subsection. F. Empirical mean control and sample-centering transfer The centering identity in Lemma 2 reduces the pas- sage from true-centered to sample-centered covariance to a bound on∥¯ˆzN∥2

[2] [2]

We now bound the selected empirical mean

We control this quantity using the selected centered-radius bound(72), together with coordinatewise Hoeffding inequalities and a union bound over thelselected coordinates. We now bound the selected empirical mean. The rank- one correction is controlled by its only nonzero eigenvalue, ∥¯ˆzN ¯ˆz⊤ N∥op =∥¯ˆzN∥2 2.The coordinates of ˆzi need not be independen...

[3] [3]

T. W. Anderson,An Introduction to Multivariate Statis- tical Analysis, 3rd ed. (Wiley, 2003)

2003

[4] [4]

K. V. Mardia, J. T. Kent, and J. M. Bibby,Multivariate Analysis(Academic Press, 1979)

1979

[5] [5]

I. T. Jolliffe,Principal Component Analysis, 2nd ed. (Springer, 2002)

2002

[6] [6]

J. A. Tropp, User-friendly tail bounds for sums of random matrices, Foundations of Computational Mathematics12, 389 (2012)

2012

[7] [7]

R. A. Horn and C. R. Johnson,Matrix Analysis, 2nd ed. (Cambridge University Press, 2012)

2012

[8] [8]

Davis and W

C. Davis and W. M. Kahan, The rotation of eigenvec- tors by a perturbation. III, SIAM Journal on Numerical Analysis7, 1 (1970)

1970

[9] [9]

G. W. Stewart and J.-G. Sun,Matrix Perturbation Theory (Academic Press, 1990)

1990

[10] [10]

Y. Yu, T. Wang, and R. J. Samworth, A useful variant of the Davis–Kahan theorem for statisticians, Biometrika 102, 315 (2015), arXiv:1405.0680 [math.ST]

work page internal anchor Pith review Pith/arXiv arXiv 2015

[11] [11]

Vershynin,High-Dimensional Probability: An Intro- duction with Applications in Data Science(Cambridge University Press, 2018)

R. Vershynin,High-Dimensional Probability: An Intro- duction with Applications in Data Science(Cambridge University Press, 2018)

2018

[12] [12]

Hadfield, S

C. Hadfield, S. Bravyi, R. Raymond, and A. Mezzacapo, Measurements of quantum hamiltonians with locally- biased classical shadows, Communications in Mathemati- cal Physics391, 951 (2022). 30

2022

[13] [13]

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)

2020

[14] [14]

Elben, S

A. Elben, S. T. Flammia, H.-Y. Huang, R. Kueng, J. Preskill, B. Vermersch, and P. Zoller, The random- ized measurement toolbox, Nature Reviews Physics5, 9 (2023), arXiv:2203.11374 [quant-ph]

work page arXiv 2023

[15] [15]

K. Bu, D. E. Koh, R. J. Garcia, and A. Jaffe, Classi- cal shadows with pauli-invariant unitary ensembles, npj Quantum Information10, 6 (2024)

2024

[16] [16]

Innocenti, S

L. Innocenti, S. Lorenzo, I. Palmisano, F. Albarelli, A. Fer- raro, M. Paternostro, and G. M. Palma, Shadow tomog- raphy on general measurement frames, PRX Quantum4, 040328 (2023), arXiv:2301.13229 [quant-ph]

work page arXiv 2023

[17] [17]

Ippoliti, Y

M. Ippoliti, Y. Li, T. Rakovszky, and V. Khemani, Opera- tor relaxation and the optimal depth of classical shadows, Physical Review Letters130, 230403 (2023)

2023

[18] [18]

Zhou and P

T.-G. Zhou and P. Zhang, Efficient classical shadow tomography through many-body localization dynamics, Quantum8, 1467 (2024)

2024

[19] [19]

S. Chen, W. Yu, P. Zeng, and S. T. Flammia, Robust shadow estimation, PRX Quantum2, 030348 (2021)

2021

[20] [20]

Aliakbarpour, V

M. Aliakbarpour, V. Braverman, N.-H. Chia, C.-Y. Lin, Y. Liu, A. Oufkir, and Y.-C. Shen, Shadow to- mography against adversaries (2025), arXiv preprint, arXiv:2512.05451 [quant-ph]

work page arXiv 2025

[21] [21]

M. West, F. Sauvage, A. Sen, R. Forestano, D. Wierichs, N. Killoran, D. Grinko, M. Cerezo, and M. Larocca, Clas- sical shadows with arbitrary group representations (2026), arXiv preprint, arXiv:2604.01429 [quant-ph]

work page arXiv 2026

[22] [22]

Gandhari, V

S. Gandhari, V. V. Albert, T. Gerrits, J. M. Taylor, and M. J. Gullans, Precision bounds on continuous-variable state tomography using classical shadows, PRX Quantum 5, 010346 (2024)

2024

[23] [23]

Boyd and B

G. Boyd and B. Koczor, Training variational quantum circuits with CoVaR: Covariance root finding with clas- sical shadows, Physical Review X12, 041022 (2022), arXiv:2204.08494 [quant-ph]

work page arXiv 2022

[24] [24]

Hoeffding, Probability inequalities for sums of bounded random variables, Journal of the American Sta- tistical Association58, 13 (1963)

W. Hoeffding, Probability inequalities for sums of bounded random variables, Journal of the American Sta- tistical Association58, 13 (1963)

1963