Pith. sign in

REVIEW 3 major objections 5 minor 43 references

For any declared set of measurement contexts and covariances, the minimum shot cost of an unbiased linear stratified estimator is exactly computable as a second-order cone program, and its dual supplies a machine-checkable certificate.

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 →

T0 review · deepseek-v4-flash

2026-08-01 19:45 UTC pith:K76UBMAH

load-bearing objection A sound convex-core result with honest scoping; the production numbers are model-certified, not state-certified, but the authors say so themselves, and the certificate machinery is the real contribution. the 3 major comments →

arxiv 2607.16866 v1 pith:K76UBMAH submitted 2026-07-18 quant-ph physics.chem-ph

Certified Optimal Measurement Reduction over Quantum Context Landscapes

classification quant-ph physics.chem-ph
keywords quantum measurement reductionsecond-order cone programshot allocationconic dualitycertified optimalitystratified estimatorscolumn generationcovariance model
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.

This paper establishes that the inner layer of quantum measurement reduction—splitting an observable into context fragments and allocating shots—has a global convex closure. For any declared dictionary of contexts, per-shot costs, score functions, and covariance model, the minimum leading shot cost among unbiased linear stratified estimators is exactly the value of a second-order cone program. The conic dual provides a witness that lower-bounds this minimum, so any schedule's optimality or its gap to optimality becomes a verified statement rather than a heuristic claim. The authors use this certificate to grade existing heuristics and to show that adding fully commuting contexts can save up to 56% of shots on small molecules and 31–70% on production f-element Hamiltonians under the declared covariance model. The framework also supplies a pricing oracle that tells whether a missing measurement context would help.

Core claim

The central discovery is that coefficient splitting and shot allocation over a fixed measurement dictionary—previously treated with iterative heuristics—are exactly solvable by one convex program. Given a finite set of contexts, each with a visible subspace, per-shot cost, and covariance matrix, the optimal unbiased linear stratified estimator has leading shot constant Φ equal to the minimum of Σ √cℓ ∥Fℓ∥_ρ subject to the fragments summing to the target observable. The conic dual of this program yields a witness y with value yᵀh that lower-bounds Φ, while any feasible primal schedule gives an upper bound U, so an independent verifier can recompute L ≤ Φ ≤ U from stored data without trusting

What carries the argument

The central object is the fragment gauge Φ, the optimum of a second-order cone program that jointly optimizes the decomposition of the Hamiltonian into context-visible fragments and the allocation of shots across contexts. The argument is carried by conic duality: the dual problem max yᵀh subject to ∥A_ℓᵀy∥_{Γ_ℓ,*} ≤ √cℓ for every context produces a lower-bound witness, with the dual seminorm built from the Moore–Penrose pseudoinverse and an explicit range condition so singular covariance matrices do not break the certificate. The same dual witness doubles as a pricing oracle for column generation: a missing context whose reduced cost exceeds one provably improves the schedule, and if none e

Load-bearing premise

The production-scale savings (31–70%) rest on two things a reader cannot inspect: the f-element Hamiltonians come from an unpublished companion pipeline, and the certificates use a Hartree–Fock-proxy covariance model that the paper itself calls 'a design model rather than the prepared state'—if either misrepresents the real systems, the headline numbers change.

What would settle it

Take any certified-optimal schedule, add a single ghost Pauli direction (a zero-target coordinate) to a context's score basis, re-solve the SOCP, and check whether Φ falls below the certified L; if it does, the original certificate was not global over the physically readable algebra. Alternatively, on one f-element system such as CeO, replace the Hartree–Fock-proxy covariance with the available ADAPT-VQE statevector covariance and compare the certified Φ and shot-savings percentage.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • For any declared dictionary, the optimal coefficient splitting and shot allocation are computed exactly by one convex program, replacing iterative alternation heuristics.
  • Every schedule in the declared landscape can be graded by a certified gap U/L−1; a gap of zero proves global optimality in that landscape.
  • On the tested molecules, standard grouped-measurement heuristics are exactly optimal for H2 but leave factors of 2.1–7.7 in shots within their own settings for H2O.
  • Adding fully commuting, Clifford-accessible contexts lowers the certified optimum by up to 56% on BeH2 and by 31–70% on production f-element Hamiltonians, under the declared covariance model.
  • The dual witness enables column generation, either finding a missing context that provably helps or certifying optimality over a landscape larger than the solved dictionary.

Where Pith is reading between the lines

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

  • The certificate's dependence on a declared score basis points to a concrete next problem: derive a reduced-cost test for admitting 'ghost' directions (zero-target Pauli products) so that certificates automatically cover the full physically readable algebra of each context.
  • The paper's empirical calibration of the covariance radius (exponent ≈0.8 instead of worst-case 1) suggests that a structure-aware concentration inequality for chemical covariances could tighten data-driven certificates by an order of magnitude; proving such a bound is an open, high-value extension.
  • The same stratified-estimator certificate could be applied to other sampling targets, such as Hadamard-test estimators for dynamical correlation functions, whose probe observables define a natural context dictionary.
  • If the benchmark community adopts this reporting standard, future measurement-reduction claims will need to distinguish 'better point in the same landscape' from 'enlarged landscape'—shifting research effort toward dictionary design and context pricing.

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

3 major / 5 minor

Summary. The paper develops a certificate framework for quantum measurement reduction. For any declared finite dictionary of measurement contexts, per-shot costs, score functions, and covariance model, it formulates the minimum leading shot cost among unbiased linear stratified estimators as a second-order cone program (Theorem 1, Corollary 1), whose value is Φ²/ε². The conic dual supplies an independently checkable lower-bound witness (Theorem 2), pilot measurements yield finite-sample simultaneous covariance brackets (Theorem 3), and these give high-confidence robust certificates under covariance uncertainty (Theorem 4). A dual pricing oracle is used for column generation over omitted contexts (Proposition 1), and the nonconvex outer-layer tasks — dictionary design, covariance-radius calibration, context pricing — are handled by the RANGE global optimizer. Numerical studies range from synthetic landscapes to molecular benchmarks, pilot-data robust certificates, dictionary compression, and production 29–35-qubit f-element Hamiltonians, for which the paper reports 31–70% certified shot savings under a declared Hartree–Fock-proxy covariance model.

Significance. If the main theorem holds, the paper provides a global convex closure of the coefficient-splitting and shot-allocation freedom in existing measurement-reduction heuristics, together with a machine-checkable certificate that does not require trusting the optimizer. The explicit dual feasibility-repair procedure (Appendix D), the finite-sample covariance brackets, the column-generation pricing oracle, and the clear division between the convex core and the nonconvex outer layer are genuine contributions. The central conditional statement is sound and useful as a common benchmark standard. The production-scale numbers, however, rest on inputs that are not inspectable in the submitted manuscript — an unpublished companion pipeline and a proxy covariance model that the authors themselves call “a design model rather than the prepared state” — so the durable scientific value of the paper is its certificate machinery rather than the specific f-element savings.

major comments (3)
  1. [Sec. XIH, Tables VIII–X] The production-scale claims — 31–70% QWC+FC shot savings, the QWC versus QWC+FC comparison, and the block-encoding-versus-sampling “two currencies” observation — are computed from Hamiltonians produced by an unpublished companion pipeline (Zahariev, Glezakou et al., in preparation) and under an HF-proxy covariance model that the authors themselves describe as “a design model rather than the prepared state” and which “cannot testify about correlation strength.” Since the certificate is exact only for the declared covariance model, these headline numbers are not independently checkable from the submitted manuscript; a reader cannot determine whether the savings would persist with correlated-state covariances or with the true CASSCF/ADAPT-VQE operators. The manuscript should either include the operators and covariance data (e.g., in the artifact or a supplementary file) or explicitly demote
  2. [Sec. III Remark; Table IV] The certificate is exact only over the declared score basis of each context. The paper notes that ghost directions readable in the same physical setting are excluded and that a reduced-cost test for ghost admission remains to be derived. Table IV itself shows ICS entries at 0.99 and 0.93 below the authors’ certified Φ², which the text correctly attributes to a richer declared dictionary. This scoping is not a mathematical flaw in the theorem, but it is load-bearing for the paper’s framing: “Certified Optimal Measurement Reduction” is certified only relative to the declared score bases and context dictionary. The abstract and conclusion should state this qualification more prominently, and the authors should discuss how certificates can be extended to ghost-augmented bases or explain why that extension is out of scope.
  3. [Sec. V, Theorem 3 proof, Step 3] The displayed calculation gives the coefficient 5√2 (≈7.07) for the m√(κ/N) term, which cannot be absorbed by the stated radius r(δ)=4m√(κ/N)+3/2 mκ/N. The coefficient should be 5/√2 (≈3.54). As printed, the absorption step is invalid, and since this theorem underpins the data-driven robust certificates, the proof must be corrected. The intended argument is clear, but the written inequality is false and a reader cannot verify the theorem as stated.
minor comments (5)
  1. [Sec. XIF, Table VII] The text says the searched LiH dictionary “beats greedy given 74% more contexts (23 contexts at +0.2% versus 40 at +1.4%)”, but Table VII lists the greedy result as 19 contexts / +1.5% and the search result as 23 contexts / +0.2%. The 40-context greedy number is unexplained; please reconcile.
  2. [Table VI caption] The caption correctly notes that the six rows are separate simulated experiments rather than one jointly certified family, but the table body could make this even clearer by labeling each row as an independent realization.
  3. [Sec. III Remark] The phrase “ghost or shared Pauli products of Ref. [16]” would benefit from a one-sentence definition of a ghost direction in the remark itself, rather than relying on the reader to fetch the reference.
  4. [Sec. VII, Eq. (30)] The notation ⟨Fℓ⟩²/pℓ in the expression for Vrand should be defined explicitly as the squared true expectation divided by the context-selection probability; the distinction between estimator variance and one-shot score variance is central to the argument.
  5. [Appendix D, Proposition 2] The verifier is described as “trusting nothing but linear algebra” but operates in floating point; the appendix is explicit about this, which is good. A short remark that the repaired bounds are exact in real arithmetic and that floating-point residuals are reported, not directed rounding, would help avoid overclaiming in later sections.

Circularity Check

0 steps flagged

Central SOCP/duality derivation is self-contained; no circular reduction found. Production-scale tables rest on an unpublished companion source and a declared HF-proxy covariance model, a material-access limitation, not circularity.

full rationale

No step in the derivation chain reduces a predicted quantity to a fitted input or to a self-citation. Theorem 1 defines Φ as the value of a convex program over observable fragments and shot counts, and Appendix A derives the optimal cost by minimizing Σ c_ℓ N_ℓ subject to Σ v_ℓ/N_ℓ ≤ ε², giving (Σ sqrt(c_ℓ v_ℓ))²/ε²; this is an optimization result, not an identity with an input. The dual certificate (Theorem 2, Proposition 2) is constructed via Lagrangian weak duality and a feasibility-repair map, so the bracket L ≤ Φ ≤ U is recomputed from stored data and declared tolerances rather than trusted from solver output. The calibration of the radius shape (c,a)=(0.67,0.81) in Sec. XIE is explicitly excluded from formal certificates: 'we use the proved radius for every certificate reported in this paper.' The ghost-direction and declared-landscape caveats (Sec. III Remark; Sec. XIC) narrow the scope of the certificates but do not make a conclusion equal to an input. The genuine burden is material access: production f-element Hamiltonians are supplied by 'companion resource-estimates paper: Zahariev, Glezakou et al., in preparation,' and production certificates use 'a design model rather than the prepared state' whose covariance 'cannot testify about correlation strength' (Sec. XIH). These are explicit, disclosed input assumptions for Tables VIII–X; they affect the empirical conclusions but not the logical derivation of the framework. Score 2 reflects the minor self-citations and the unavailable companion source, not a circular step.

Axiom & Free-Parameter Ledger

4 free parameters · 6 axioms · 0 invented entities

No new physical entities are postulated; the certificate, dual witness, and discrete RANGE mode are software/math constructs. The framework's certificates treat the dictionary, covariance model, and costs as declared inputs, so they are not fitted degrees of freedom; the only genuinely fitted parameters are the calibration radius shape (c,a), which is explicitly excluded from all formal certificates, plus the declared caps and p, which the authors flag. Production claims additionally lean on unpublished operators from the companion pipeline.

free parameters (4)
  • depolarization strength p = 0.02 (sweeps at 1e-3, 1e-4; p=0 limit analyzed)
    Chosen to model preparation error and to keep covariances non-singular so certificates retain guarantees; the p=0 limit is reported to degrade certificates (Sec. XIB).
  • calibrated covariance-radius shape (c, a) = 0.67, 0.81
    Sec. XIE: fit by RANGE on simulated pilot data for LiH, minimizing radius subject to empirical simultaneous coverage ≥ 95%; explicitly labeled empirical and excluded from formal certificates.
  • declared dictionary caps = visibility cap 200, FC pool cap 1500 (f-element); visibility cap 80 (CH4)
    Hand-chosen caps define the declared landscape; all production savings numbers (31-70%) are certified only over these capped dictionaries, with class-level claims explicitly disclaimed (Sec. XIH).
  • covariance model choice (HF-proxy) = HF product-state Pauli expectations (0 or ±1) plus p=0.02 depolarization
    Sec. XIH: production certificates use the declared HF-proxy because exact covariances are unavailable at 29-35 qubits; authors note a correlated-state model could change results.
axioms (6)
  • standard math Matrix Bernstein inequality with variance bound E[X_i^2] ⪯ (m^2/4)I for ±1 outcome vectors
    Theorem 3 proof, Ref [17].
  • standard math Conic strong duality holds when the target lies in the relative interior of the dictionary span after quotienting zero-variance directions
    Theorem 2 / Appendix B; certificates remain valid lower bounds without it by weak duality.
  • standard math PSD order reverses under the Moore-Penrose pseudoinverse on the common range: Γ ⪯ Γ̄ implies z^T Γ^† z ≥ z^T Γ̄^† z on range(Γ̄)
    Appendix C, used for Theorem 4.
  • domain assumption Declared context score bases produce unbiased single-shot estimators for the target observable algebra
    Def. 1 and Eq. (5); the certificate is exact only over the declared score basis (Sec. III Remark on ghost directions).
  • ad hoc to paper Production f-element Hamiltonians from the companion CASSCF/compression pipeline are the correct operators for the target chemistry
    Sec. XIH; the operators are not included in the preprint and cannot be checked by a reader.
  • domain assumption Pilot measurements are i.i.d. over ±1 outcome vectors within each context
    Theorem 3; needed for the finite-sample simultaneous coverage statement.

pith-pipeline@v1.3.0-alltime-deepseek · 24327 in / 34105 out tokens · 329283 ms · 2026-08-01T19:45:01.694318+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Certified Optimal Measurement Reduction over Quantum Context Landscapes." pith.science (2026). https://pith.science/paper/K76UBMAH

@misc{pith2026260716866,
  author       = {Pith},
  title        = {Pith review of: Certified Optimal Measurement Reduction over Quantum Context Landscapes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/K76UBMAH}},
  note         = {Machine review of arXiv:2607.16866}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Quantum-measurement reduction contains two distinct global-optimization layers: a continuous problem of splitting an observable and allocating shots within a fixed measurement dictionary, and a nonconvex outer problem of designing the dictionary and calibrating its data-driven uncertainty model. We solve the inner layer globally and certifiably as a second-order cone program (SOCP), and use RANGE, a robust adaptive nature-inspired global optimizer, for the combinatorial and statistical outer layer. For any declared set of contexts, per-shot costs, score functions, and covariance model, the SOCP returns the minimum leading shot cost among unbiased linear stratified estimators. The conic dual supplies an independently checkable lower-bound witness; after feasibility repair, an external verifier recomputes $L \le \Phi \le U$ from stored data without trusting the optimizer. Pilot measurements yield simultaneous finite-sample covariance brackets, and the dual becomes a pricing oracle for omitted contexts. Discrete RANGE searches covering sub-dictionaries, Pareto compression fronts, and candidate contexts; continuous RANGE performs an explicitly empirical, coverage-constrained calibration of covariance-radius models, while rigorous certificates retain the proved finite-sample radius. RANGE compresses molecular context dictionaries by 4.3-6.1x at 0.2-2.1% certified-frontier excess. Standard strategies are exactly optimal for H2 yet leave factors of 2.1-7.7 in shots within their own settings by H2O. Adding fully commuting contexts lowers the certified optimum by up to 56%; on 29-35-qubit production f-element Hamiltonians under a declared Hartree-Fock-proxy covariance model, the capped-dictionary enlargement saves 31-70% of the shots, and transformations reducing block-encoding cost need not reduce sampling cost.

Figures

Figures reproduced from arXiv: 2607.16866 by Federico Zahariev, Vanda Glezakou.

Figure 1
Figure 1. Figure 1: FIG. 1. Measurement reduction as geometry plus optimization. The allowed contexts define visible subspaces. A covariance [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] 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

43 extracted references · 5 linked inside Pith

  1. [1]

    Each context contributes a visible matrixAℓ and a costcℓ

    Dictionary generation.Generate candidate contexts from Pauli grouping, Clifford/stabilizer transforma- tions, fermionic fragments, hardware-native bases, or any mixture of these. Each context contributes a visible matrixAℓ and a costcℓ

  2. [2]

    Pilot data should return covariance setsKℓ or PSD sandwichesΓ ℓ⪯Γ ℓ⪯ Γℓ

    Covariance acquisition.Use exact simulation for diagnostics, aclassicalproxystateforpre-experiment design, or pilot measurements for robust certification. Pilot data should return covariance setsKℓ or PSD sandwichesΓ ℓ⪯Γ ℓ⪯ Γℓ

  3. [3]

    (13) or its robust counterpart

    Certified optimization.Solve Eq. (13) or its robust counterpart. Export the declared model and raw primal–dual pair. The verifier repairs feasibility where possible, recomputes the two objective bounds, and reports the certified gap

  4. [4]

    which heuristic won?

    Landscape expansion.Use the dual witness as a pricing signal. If a missing context violates Eq.(25), add it and reoptimize. Stop when the dual gap and pricing residual are below declared tolerances. The benchmark suite required for a high-impact submis- sion should report not only root-mean-square error versus shots, but also the certificate numbersU,L,U/...

  5. [5]

    The same schema works for robust certificates by usingΓℓ in the primal andΓℓ in the dual

    reportU cert/Lcert−1, the corresponding shot overhead, and the achieved floating-point residuals. The same schema works for robust certificates by usingΓℓ in the primal andΓℓ in the dual. Appendix F: Benchmark protocol for a submission-level study A decisive benchmark should include both performance and certificates. For each molecule or many-body instanc...

  6. [6]

    Production-scale benchmarks(Sec. XIH). Model- certified measurement costs forf-element Hamilto- nians from an exascale pipeline, together with, to our knowledge, the first dual-certified stage-by-stage study of how block-encoding-motivated compression reshapes sampling cost. Fix a finite dictionary of allowed measurement contexts. A context may be a Pauli...

  7. [7]

    verify thath∈range (A), reconstruct or validate the declared repair map throughABA =A, and repair the raw primal residual

  8. [8]

    recompute the feasible primal valueUcert; 19

  9. [9]

    test AT ℓy∈range (Γℓ)for every ℓ, rescale a range-valid witness to dual feasibility, and returnLcert = 0if any range test fails

  10. [10]

    recompute the dual valueLcert =h Tycert

  11. [11]

    M. Li, M. Lin, and M. J. S. Beach, Resource-optimized grouping shadow for efficient energy estimation, Quantum9, 1694 (2025)

  12. [12]

    Verteletskyi, T.-C

    V. Verteletskyi, T.-C. Yen, and A. F. Izmaylov, Measurement optimization in the variational quantum eigensolver using a minimum clique cover, The Journal of Chemical Physics152, 124114 (2020)

  13. [13]

    and whose equivalence theorems [25, 26] are pre- cisely optimality certificates in the sense used here. For multiresponse experiments with correlated outcomes, the 8 reduction ofc-optimal design to second-order cone pro- gramming was established by Sagnol [15], with an equiva- lent geometric characterization by Dette and Holland-Letz [14]; under the mappi...

  14. [14]

    T.-C. Yen, A. Ganeshram, and A. F. Izmaylov, Deterministic improvements of quantum measurements with grouping of compatible operators, non-local transformations, and covariance estimates, npj Quantum Information9, 14 (2023)

  15. [15]

    B. Wu, J. Sun, Q. Huang, and X. Yuan, Overlapped grouping measurement: A unified framework for measuring quantum states, Quantum7, 896 (2023)

  16. [16]

    Huang, R

    H.-Y. Huang, R. Kueng, and J. Preskill, Predicting many properties of a quantum system from very few measurements, Nature Physics16, 1050 (2020)

  17. [17]

    Huang, R

    H.-Y. Huang, R. Kueng, and J. Preskill, Efficient estimation of pauli observables by derandomization, Physical Review Letters127, 030503 (2021)

  18. [18]

    Hadfield, S

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

  19. [19]

    K. Wan, W. J. Huggins, J. Lee, and R. Babbush, Matchgate shadows for fermionic quantum simulation, Communications in Mathematical Physics404, 629 (2023)

  20. [20]

    S. Choi, I. Loaiza, and A. F. Izmaylov, Fluid fermionic fragments for optimizing quantum measurements of electronic hamiltonians in the variational quantum eigensolver, Quantum7, 889 (2023)

  21. [21]

    L. E. Fischer, T. Dao, I. Tavernelli, and F. Tacchino, Dual-frame optimization for informationally complete quantum measurements, Physical Review A109, 062415 (2024)

  22. [22]

    Gresch and M

    A. Gresch and M. Kliesch, Guaranteed efficient energy estimation of quantum many-body hamiltonians using Shadow- Grouping, Nature Communications16, 689 (2025)

  23. [23]

    Zahariev and G

    F. Zahariev and G. Ortiz, Multiplicative cost factorization and necessary conditions for quantum advantage in dynamical simulation (2026), manuscript in preparation

  24. [24]

    Elfving, Optimum allocation in linear regression theory, The Annals of Mathematical Statistics23, 255 (1952)

    G. Elfving, Optimum allocation in linear regression theory, The Annals of Mathematical Statistics23, 255 (1952)

  25. [25]

    Dette and T

    H. Dette and T. Holland-Letz, A geometric characterization of c-optimal designs for heteroscedastic regression, The Annals of Statistics37, 4088 (2009)

  26. [26]

    Sagnol, Computing optimal designs of multiresponse experiments reduces to second-order cone programming, Journal of Statistical Planning and Inference141, 1684 (2011)

    G. Sagnol, Computing optimal designs of multiresponse experiments reduces to second-order cone programming, Journal of Statistical Planning and Inference141, 1684 (2011)

  27. [27]

    Choi, T.-C

    S. Choi, T.-C. Yen, and A. F. Izmaylov, Improving quantum measurements by introducing “ghost” pauli products, Journal of Chemical Theory and Computation18, 7394 (2022)

  28. [28]

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

  29. [29]

    Zhang, M

    D. Zhang, M. Z. Makoś, R. Rousseau, and V.-A. Glezakou, RANGE: A robust adaptive nature-inspired global explorer of potential energy surfaces, The Journal of Chemical Physics163, 152501 (2025). 20

  30. [30]

    I. L. Huidobro-Meezs and R. A. Vargas-Hernández, Reducing quantum measurements in qubit-based overlapping group- ing methods for quantum energy estimation through better initializations, arXiv preprint arXiv:2607.02794 (2026), arXiv:2607.02794 [quant-ph]

  31. [31]

    Rowland, R

    J. Rowland, R. Sarkar, N. P. D. Sawaya, N. M. Tubman, and R. LaRose, Overlapped groupings for quantum energy estimation: Maximal variance reduction and deterministic algorithms for reducing variance, arXiv preprint arXiv:2604.07156 (2026), arXiv:2604.07156 [quant-ph]

  32. [32]

    Malmi, K

    J. Malmi, K. Korhonen, D. Cavalcanti, and G. García-Pérez, Enhanced observable estimation through classical optimization of informationally over-complete measurement data – beyond classical shadows, arXiv preprint arXiv:2401.18049 (2024), arXiv:2401.18049 [quant-ph]

  33. [33]

    Korhonen, S

    K. Korhonen, S. Mangini, J. Malmi, H. Vappula, and D. Cavalcanti, Improving shadow estimation with locally-optimal dual frames, arXiv preprint arXiv:2511.02555 (2025), arXiv:2511.02555 [quant-ph]

  34. [34]

    Shlosberg, A

    A. Shlosberg, A. J. Jena, P. Mukhopadhyay, J. F. Haase, F. Leditzky, and L. Dellantonio, Adaptive estimation of quantum observables, Quantum7, 906 (2023)

  35. [35]

    Heyraud, H

    V. Heyraud, H. Chomet, and J. Tilly, Unified framework for matchgate classical shadows, arXiv preprint arXiv:2409.03836 (2024), arXiv:2409.03836 [quant-ph]

  36. [36]

    Kiefer and J

    J. Kiefer and J. Wolfowitz, The equivalence of two extremum problems, Canadian Journal of Mathematics12, 363 (1960)

  37. [37]

    Pukelsheim,Optimal Design of Experiments(SIAM, Philadelphia, 2006)

    F. Pukelsheim,Optimal Design of Experiments(SIAM, Philadelphia, 2006)

  38. [38]

    G. M. D’Ariano and P. Perinotti, Optimal data processing for quantum measurements, Physical Review Letters98, 020403 (2007)

  39. [39]

    Zhu, Quantum state estimation with informationally overcomplete measurements, Physical Review A90, 012115 (2014)

    H. Zhu, Quantum state estimation with informationally overcomplete measurements, Physical Review A90, 012115 (2014)

  40. [40]

    Crawford, B

    O. Crawford, B. van Straaten, D. Wang, T. Parks, E. Campbell, and S. Brierley, Efficient quantum measurement of pauli operators in the presence of finite sampling error, Quantum5, 385 (2021)

  41. [41]

    Zahariev and V.-A

    F. Zahariev and V.-A. Glezakou, Beyond Orbital Rotations: Correlation-Rank Limits and Clifford-Accessible Measurement, from Algebra and Global Optimization (2026), companion manuscript

  42. [42]

    Zahariev and G

    F. Zahariev and G. Ortiz, Quantum algorithms for time-dependent open-boundary correlation functions (2026), manuscript in preparation

  43. [43]

    J. A. Tropp,An Introduction to Matrix Concentration Inequalities, Foundations and Trends in Machine Learning, Vol. 8 (Now Publishers, 2015) pp. 1–230