REVIEW 4 major objections 3 minor 4 cited by
Fast quantum measurement tomography with optimal error bounds
T0 review · 4 major / 3 minor · reviewed 2026-08-06 · deepseek-v4-flash
Pith's one-line read A two-step estimator makes POVM tomography sample-optimal in dimension, with matching lower bounds for any non-adaptive protocol.
desk verdict The worst-case QMT result is likely correct and valuable; the average-case claims are not proven as written and the abstract oversells them. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The carrying object is the closed-form least-squares estimator built from a 2-design probe ensemble. For a global 2-design, the dual operators are $\nu_i = d(d+1)|\psi_i\rangle\langle\psi_i| - d\mathbb{1}$, giving $\hat{E}_j = \sum_i \hat{f}_{ij}\nu_i$; for a tensor product of single-qubit 2-designs they factor as $\bigotimes_k(6|\psi_{i_k}\rangle\langle\psi_{i_k}| - 2\mathbb{1})$. The 2-design property makes the frame operator proportional to $X + \mathrm{tr}(X)\mathbb{1}$, so the least-squares inversion is analytic and the estimator is unbiased. A matrix concentration bound on the grouped subset differences $\sum_{j\in x}(\hat{E}_j - E_j)$ is combined with a union bound over all $2^L$ outcome subsets; the projection onto the convex set of POVMs then yields a physical estimate whose operational distance to the true POVM is at most twice the raw error. For the lower bound, the machinery is an $\epsilon/8$-separated packing of $e^{\Omega(d^2)}$ POVMs used to encode a random message, with an information-theoretic bound on the mutual information per use of the measurement.
What would settle it
Run the protocol many times on a known POVM with $N$ set by the theorem and tally how often $d_{op}(E,E^{*}) > \epsilon$; the guarantee says this should happen with frequency at most $\delta$, so a statistically clear violation would refute the bound. A second check is to re-run the hardware experiment with probe calibrations perturbed by the device's measured single-qubit infidelities and see whether the reconstruction shifts by more than the certified $\epsilon$.
Extended reading notes
Core claim
The paper's central claim is Theorem 3: with $N \geq 8(d^3+d^2(1+\epsilon/6))\epsilon^{-2}\ln(2^{L+1}d/\delta)$ uses of the POVM on probe states from a unitary 2-design, the projected least-squares estimate $E^{*}$ satisfies $d_{op}(E,E^{*}) \leq \epsilon$ with probability at least $1-\delta$; the local-2-design variant replaces $d^3$ by $10^n$ and $d^2$ by $4^n$, giving $O(d^{3.33}L/\epsilon^2)$ on $n$-qubit systems. The raw least-squares estimator is unbiased, matrix concentration controls its operator-norm error, and the projection step preserves a physical POVM while at most doubling the distance to the truth. The matching lower bound (Theorem 10) says any non-adaptive, single-copy protocol needs at least $\Omega(d^3/\epsilon^2)$ uses for $\epsilon/16$ accuracy in operational distance, with a comparable $\Omega(d^2L/\epsilon^2)$ bound for average-case distance. The paper therefore establishes that dimension-optimal QMT with finite-sample guarantees is possible at modest classical cost. It explicitly acknowledges that the factor of $L$ is not tight between its upper and lower bounds, and leaves that gap open.
Load-bearing premise
The guarantees require that the probe states are exactly known and exactly form the assumed 2-design; any preparation error acts as an unmodeled bias that can push the reconstruction beyond the certified $\epsilon$.
Editorial extensions
If this is right
- Any non-adaptive single-copy QMT protocol needs at least $\Omega(d^3/\epsilon^2)$ uses for worst-case precision, so the $O(d^3L/\epsilon^2)$ scaling of this protocol is dimension-optimal up to logarithmic factors.
- In average-case distance the protocol needs only $O(d^2L^2/\epsilon^2)$ samples, a factor $d$ better than worst-case, and no non-adaptive protocol can beat the dimension scaling $\Omega(d^2L/\epsilon^2)$.
- Finite-sample error bars are rigorous, so users can choose sample sizes that meet a target accuracy with stated confidence without relying on asymptotic central-limit reasoning.
- Readout-error mitigation can use the certified POVM estimate directly: the operational-distance bound controls the residual post-correction error in the triangle-inequality analysis of Section V.
- For $n$-qubit local 2-designs, the sample count grows as $O(10^n L/\epsilon^2)$, i.e. as $d^{\log_2 10}\approx d^{3.32}$, trading a mild loss in dimension scaling for easier probe preparation.
Reading between the lines
- The same analytic-frame plus projection template should apply to estimating more general objects—quantum instruments or channels with continuous outcomes—where the projection set changes but the unbiased linear estimator remains closed form.
- The ideal-probe assumption could be tested by deliberately perturbing the assumed probe states: the theorem predicts the reconstruction error should stay within the certified $\epsilon$, while the hardware demonstration suggests preparation and readout noise will push it out; quantifying this crossover would be a direct stress test.
- Closing the $L$-gap likely needs a packing argument whose separation or information cost depends on the number of outcomes, rather than the current construction which is sensitive mainly to dimension.
- A practical rule-of-thumb emerges: use global 2-designs whenever they are cheap to prepare (dimension-optimal $d^3$ scaling), and local 2-designs when only single-qubit randomized measurements are available.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript proposes a two-step projected least-squares protocol for quantum measurement tomography (QMT). For an L-outcome POVM on a d-dimensional system, the protocol first computes an analytical least-squares estimate from measurements on known probe states drawn from a global or local 2-design, then projects the estimate onto the set of physical POVMs. The main claims are: Theorem 3, O(d^3 L ln(d)/epsilon^2) samples suffice for operational distance with global 2-designs, and O(d^{3.33} L/epsilon^2) for local 2-designs; Theorem 4, O(d^2 L^2 ln(dL)/epsilon^2) for average-case distance; Theorems 10 and 15, lower bounds Omega(d^3/epsilon^2) and Omega(d^2 L/epsilon^2) for any non-adaptive, single-copy protocol; and an experimental demonstration on superconducting transmon qubits. The upper-bound proofs use matrix Bernstein inequalities with explicit K and sigma^2 parameters, a triangle-inequality projection step, and Hoeffding bounds for trace terms; the lower bounds use packing constructions and Fano's inequality.
Significance. If the operational-distance results survive review, the paper delivers a valuable and elegant result: a computationally light QMT protocol with explicit non-asymptotic error bounds and dimension-optimal scaling, together with a concrete application to readout-error mitigation. The analytic form of the least-squares estimator for 2-design probe ensembles and the explicit Bernstein parameters are useful and largely check out. However, the average-case upper bound is not proven as stated because it relies on a false inequality, and the average-case lower bound contains normalization inconsistencies. In addition, the arXiv abstract overstates the L-scaling and the claimed optimality. The central advertised claim of simultaneous optimality for both operational and average-case distances is therefore currently unsupported.
major comments (4)
- [Appendix C, Eq. (C3) and Theorem 4] The inequality d_av(E, \hat E) <= (1/(2d)) \sum_i (||E_i - \hat E_i||_F + |tr(E_i - \hat E_i)|) is false. For traceless errors with ||\Delta_i||_F = s for all i, d = 2 and L = 3, the left-hand side is s sqrt(3/4) while the right-hand side is 3s/4. The subsequent proof bounds each Frobenius and trace term by epsilon/(4L) and then uses Eq. (C3) to conclude d_av(E, \hat E) <= epsilon/2; without a valid sum-to-square-root bound, those individual bounds do not control d_av. Consequently the sample complexity stated in Theorem 4 and Eq. (C1) is not established by the given proof. A correct proof would need to control the squared Frobenius and trace terms collectively, which changes the required per-element accuracy and hence the resulting sample complexity.
- [Appendix D, Lemma 12 and Theorem 15] Lemma 12 is stated as a packing of POVMs separated by epsilon/4 in d_av, but the proof's conclusion in Eq. (D11) is sqrt(d) d_av(E,F) >= epsilon/4, which is a weaker separation by a factor sqrt(d). Theorem 15 then assumes that an algorithm accurate to epsilon/8 in d_av can decode such a packing; this only follows if the packing is separated in d_av, not in sqrt(d) d_av. The step from the Fano lower bound for sqrt(d) d_av to the claimed Omega(d^2 L/epsilon^2) lower bound for d_av is therefore unjustified. Additionally, the function F defined in Eq. (D3) has the wrong prefactor: each of the L/2 unitary pairs contributes two POVM elements with coefficient 2epsilon/L, so F = (2epsilon/(dL)) \sum_j f_j, not (epsilon/(dL)) \sum_j f_j. Moreover, F is not generally a lower bound for d_av; for equal f_j = s, one has sqrt(d) d_av = (2epsilon/sqrt(L)) s while F written as in Eq. (D3) equals epsilon s/(2d), so the comparison depends on L and d.
- [Abstract and Section III.C] The arXiv abstract states that the average-case sample complexity is O(d^2 L/epsilon^2) and that the protocol is sample-optimal in both the dimension and the number of outcomes. The actual full-text abstract and Theorem 4 state O(d^2 L^2 ln(dL)/epsilon^2) for the average case, and Section VII explicitly acknowledges an unresolved gap in the L-dependence. The abstract therefore overclaims the L-scaling and the optimality in the number of outcomes; it should be corrected to match what is proven, namely dimension-optimality up to logarithmic factors.
- [Appendix D, Eqs. (D28)-(D29)] The final translation of the lower bound from sqrt(d) d_av to d_av is not valid as written. The paper obtains N >= Omega(d^3 L/epsilon^2) for precision epsilon in sqrt(d) d_av, equivalently precision epsilon/sqrt(d) in d_av, and then concludes N >= Omega(d^2 L/epsilon^2) for precision epsilon in d_av. Since epsilon is a weaker accuracy requirement than epsilon/sqrt(d), this monotonicity goes in the wrong direction: a lower bound for the harder accuracy does not imply the claimed lower bound for the easier one. This step needs to be repaired or removed from the proof.
minor comments (3)
- [Section IV.B, Lemma 9] Lemma 9 states I(X:Y) <= (N/(2 ln 2)) epsilon^2/(d+1), but the proof and Eq. (40) use I(X:Y) <= (N/ln 2) epsilon^2/(d+1), missing a factor of 2 in the lemma statement.
- [Section VI and Appendix E] The theoretical guarantees in Theorems 3 and 4 assume that the probe states are exactly known and exactly form a 2-design, whereas the hardware demonstration uses gates with single-qubit infidelities of order 10^-3 and readout fidelities near 97%. The demonstrated reconstructions therefore do not inherit the stated error bounds; this limitation should be stated explicitly in Section VI.
- [Throughout] There are several typos and minor wording errors: 'Infomationally' should be 'Informationally', 'conrrect' should be 'correct', 'nate interactions' should be 'native interactions', 'showvases' should be 'shows', 'Frobenious' should be 'Frobenius', and 'i.d.d' should be 'i.i.d.'.
Circularity Check
No circular derivation found: upper bounds use matrix Bernstein on the least-squares estimator and lower bounds use independent packing/Fano arguments; the few self-citations are lineage and not load-bearing.
full rationale
Theorem 3's sample bound is obtained by applying matrix Bernstein (Theorem 1, external Ref. 40) to the random matrices (1/N)sum(X_k^(x) - F^(x)) and union-bounding over the 2^L subsets; the estimator in Eq. (7) is a linear function of measured frequencies, so no fitted parameter is later renamed as a prediction. Theorem 4 uses the same estimator and the same Bernstein/Hoeffding machinery; even if Appendix C's inequality (C3) is mathematically questionable, that is a proof-validity defect, not a reduction of the result to its inputs. The lower bounds in Theorems 10 and 15 construct explicit epsilon-packings and invoke Fano inequality and mutual-information bounds from Refs. 32 and 45-47; they do not assume the desired sample complexity. The only self-citations are Ref. 29 (frame-operator identity for 2-designs, a parameter-free algebraic fact that does not contain the target bound), Ref. 30 (projected least-squares lineage), and Ref. 44 (choice of projection metric); none carries the sample-complexity argument. The paper also explicitly flags its own limitation in Section VII: 'Still, a gap remains in the dependence on the number of POVM outcomes L... we leave the problem of establishing tight L-scaling bounds as an open question for future work' - an honest admission, not a circular step. The experimental sections assume exactly known probe states; the quoted gate and readout imperfections are a validity caveat for the demonstration, not circularity. Overall, the central derivations are self-contained against external concentration and information-theoretic tools; the score 2 reflects only minor non-load-bearing self-citations.
Assumptions & free parameters
assumptions (6)
- standard math Matrix Bernstein inequality (Theorem 1) for sums of bounded zero-mean Hermitian matrices.
- standard math Fano inequality consequence (Lemma 2): decoding a uniform message over a set of size |X| requires I(X:Y) = Omega(log|X|).
- domain assumption Packing of exp(Omega(d^2)) unitaries with (1/d)||Ui P Ui† - Uj P Uj†||_1 >= 1/4 for a fixed rank-d/2 projector P (Lemma 5).
- standard math Concentration of Lipschitz functions on products of unitary groups (Theorem 11).
- domain assumption Probe ensemble is an exact unitary 2-design and probe states are prepared perfectly.
- domain assumption The average-case distance d_av (Eq. 4) equals the average total-variation distance over a 4-design ensemble.
Cite this review
Pith. "Pith review of Fast quantum measurement tomography with optimal error bounds." pith.science (2026). https://pith.science/paper/KIMYPAI3
@misc{pith2026250704500,
author = {Pith},
title = {Pith review of: Fast quantum measurement tomography with optimal error bounds},
year = {2026},
howpublished = {\url{https://pith.science/paper/KIMYPAI3}},
note = {Machine review of arXiv:2507.04500}
}
abstract
We present a two-step protocol for quantum measurement tomography that is light on classical co-processing cost and still achieves optimal sample complexity. Given measurement data from a known probe state ensemble, we first apply least-squares estimation to produce an unconstrained approximation of the POVM, and then project this estimate onto the set of valid quantum measurements. For a POVM with $L$ outcomes acting on a $d$-dimensional system, we show that the protocol requires $\mathcal{O}\left((d^3+d^2L)/\epsilon^2\right)$ samples to achieve error $\epsilon$ in worst-case distance, and $\mathcal{O}(d^2 L/\epsilon^2)$ samples in average-case distance. We further establish two matching sample complexity lower bounds of $\Omega((d^3 + d^2 L) /\epsilon^2)$ and $\Omega(d^2 L/\epsilon^2)$ for any non-adaptive, single-copy POVM tomography protocol. Hence, our projected least squares POVM tomography is sample-optimal in both the dimension and the number of outcomes for both distances. Our method admits an analytic form when using global or local 2-designs as probe ensembles and enables rigorous non-asymptotic error guarantees. Finally, we also complement our findings with empirical performance studies carried out on a noisy superconducting quantum computer with flux-tunable transmon qubits.
Figures
Figures from the paper (3 more)
Forward citations
Cited by 4 Pith papers
-
Near-Optimal Learning of Local Lindbladians
Near-optimal algorithm learns local Lindbladians via finite-time probes and classical shadows with Õ(Λ²/ε²) channel uses and matching lower bounds showing dissipative terms block Heisenberg-limited scaling.
-
Quantum memory advantage for quantum process tomography
Learning a channel with d_in input and d_out output dimensions without quantum memory requires Θ((d_in d_out)^3/ε²) queries even with classical adaptivity, while quantum memory gives Θ((d_in d_out)^2/ε²).
-
Sketch Tomography: Hybridizing Classical Shadow and Matrix Product State
Sketch tomography reconstructs a matrix-product-state density matrix from classical Pauli-shadow data via sketched tensor-train equations, with a claimed O(n^2) sample guarantee.
-
Quantum tomography for non-iid sources
Projected least-squares tomography reconstructs the time-averaged quantum state or channel with the same sample complexity as in the i.i.d. case, even when the source is fully adaptive.
Reference graph
Works this paper leans on
-
[1]
Temme, S
K. Temme, S. Bravyi, and J. M. Gambetta, Error mitiga- tion for short-depth quantum circuits, Phys. Rev. Lett. 119, 180509 (2017)
2017
-
[2]
Preskill, Quantum Computing in the NISQ era and beyond, Quantum2, 79 (2018)
J. Preskill, Quantum Computing in the NISQ era and beyond, Quantum2, 79 (2018)
2018
-
[3]
S. Endo, Z. Cai, S. C. Benjamin, and X. Yuan, Hybrid quantum-classical algorithms and quantum error mitiga- tion, J. Phys. Soc. Jpn.90, 032001 (2021)
work page 2021
-
[4]
Y. Kim, A. Eddins, S. Anand, K. X. Wei, E. Van Den Berg, S. Rosenblatt, H. Nayfeh, Y. Wu, M. Zale- tel, K. Temme,et al., Evidence for the utility of quan- tum computing before fault tolerance, Nature618, 500 (2023)
2023
-
[5]
Li and S
Y. Li and S. C. Benjamin, Efficient variational quantum simulator incorporating active error minimization, Phys. Rev. X7, 021050 (2017)
2017
-
[6]
Y. Chen, M. Farahzad, S. Yoo, and T.-C. Wei, Detector tomography on IBM quantum computers and mitigation of an imperfect measurement, Phys. Rev. A100, 052315 (2019)
work page 2019
-
[7]
F. B. Maciejewski, Z. Zimbor´ as, and M. Oszmaniec, Mit- igation of readout noise in near-term quantum devices by classical post-processing based on detector tomography, Quantum4, 257 (2020)
work page 2020
- [8]
Show all 63 references
-
[9]
van den Berg, Z
E. van den Berg, Z. K. Minev, and K. Temme, Model-free readout-error mitigation for quantum expectation values, Phys. Rev. A105, 032620 (2022)
2022
-
[10]
Funcke, T
L. Funcke, T. Hartung, K. Jansen, S. K¨ uhn, P. Stornati, and X. Wang, Measurement error mitigation in quan- tum computers through classical bit-flip correction, Phys. Rev. A105, 062404 (2022)
2022
-
[11]
Korhonen, H
K. Korhonen, H. Vappula, A. Glos, M. Cattaneo, Z. Zim- bor´ as, E.-M. Borrelli, M. A. Rossi, G. Garc ´ ıa-P´ erez, and D. Cavalcanti, Practical techniques for high-precision measurements on near-term quantum hardware and ap- plications in molecular energy estimation, npj Quant...
2025
-
[12]
Of course, when imple- menting this POVM measurements on a noisy device, the resulting channel will not be the expected one
relies on the inversion of the measurement channel of a spherical two design, which is proportional to the effective depolarizing channel. Of course, when imple- menting this POVM measurements on a noisy device, the resulting channel will not be the expected one. There- fore, ...
-
[13]
Huang, R
H.-Y. Huang, R. Kueng, and J. Preskill, Predicting many properties of a quantum system from very few measure- ments, Nat. Phys.16, 1050 (2020)
2020
-
[14]
S. Chen, W. Yu, P. Zeng, and S. T. Flammia, Robust shadow estimation, PRX Quantum2, 030348 (2021)
2021
-
[15]
D. E. Koh and S. Grewal, Classical Shadows With Noise, Quantum6, 776 (2022)
2022
-
[16]
Brieger, I
R. Brieger, I. Roth, and M. Kliesch, Compressive gate set tomography, PRX Quantum4, 010325 (2023)
2023
-
[17]
Fiur´ aˇ sek, Maximum-likelihood estimation of quantum measurement, Phys
J. Fiur´ aˇ sek, Maximum-likelihood estimation of quantum measurement, Phys. Rev. A64, 024102 (2001)
2001
-
[18]
G. M. D’Ariano, L. Maccone, and P. L. Presti, Quantum calibration of measurement instrumentation, Phys. Rev. Lett.93, 250407 (2004)
2004
-
[19]
J. S. Lundeen, A. Feito, H. Coldenstrodt-Ronge, K. L. Pregnell, C. Silberhorn, T. C. Ralph, J. Eisert, M. B. Plenio, and I. A. Walmsley, Tomography of quantum de- tectors, Nat. Phys.5, 27 (2009)
2009
-
[20]
Feito, J
A. Feito, J. Lundeen, H. Coldenstrodt-Ronge, J. Eisert, M. B. Plenio, and I. A. Walmsley, Measuring measure- ment: theory and practice, New J. Phys.11, 093038 (2009). 11
2009
-
[21]
Zhang, A
L. Zhang, A. Datta, H. B. Coldenstrodt-Ronge, X.-M. Jin, J. Eisert, M. B. Plenio, and I. A. Walmsley, Recursive quantum detector tomography, New J. Phys.14, 115005 (2012)
2012
-
[22]
Cattaneo, M
M. Cattaneo, M. A. Rossi, K. Korhonen, E.-M. Bor- relli, G. Garc ´ ıa-P´ erez, Z. Zimbor´ as, and D. Cavalcanti, Self-consistent quantum measurement tomography based on semidefinite programming, Phys. Rev. Res.5, 033154 (2023)
2023
-
[23]
Grandi, A
S. Grandi, A. Zavatta, M. Bellini, and M. G. Paris, Ex- perimental quantum tomography of a homodyne detec- tor, New J. Phys.19, 053015 (2017)
2017
-
[24]
Y. Wang, S. Yokoyama, D. Dong, I. R. Petersen, E. H. Huntington, and H. Yonezawa, Two-stage estimation for quantum detector tomography: Error analysis, numerical and experimental results, IEEE Trans. Inf. Theory67, 2293 (2021)
2021
-
[25]
S. Xiao, Y. Wang, J. Zhang, D. Dong, S. Yokoyama, I. R. Petersen, and H. Yonezawa, On the regularization and optimization in quantum detector tomography, Au- tomatica155, 111124 (2023)
2023
-
[26]
Nielsen, J
E. Nielsen, J. K. Gamble, K. Rudinger, T. Scholten, K. Young, and R. Blume-Kohout, Gate set tomography, Quantum5, 557 (2021)
2021
-
[27]
O’Donnell and J
R. O’Donnell and J. Wright, Efficient quantum tomog- raphy, inProceedings of the Forty-Eighth Annual ACM Symposium on Theory of Computing, STOC ’16 (Asso- ciation for Computing Machinery, New York, NY, USA,
-
[28]
J. Haah, A. W. Harrow, Z. Ji, X. Wu, and N. Yu, Sample- optimal tomography of quantum states, IEEE Trans. Inf. Theory63, 5628 (2017)
2017
-
[29]
Kueng, H
R. Kueng, H. Rauhut, and U. Terstiege, Low rank matrix recovery from rank one measurements, Appl. Comput. Harmon. Anal.42, 88 (2017)
2017
-
[30]
Gut ¸˘ a, J
M. Gut ¸˘ a, J. Kahn, R. Kueng, and J. A. Tropp, Fast state tomography with optimal error bounds, J. Phys. A: Math. Theor53, 204001 (2020)
2020
-
[31]
Surawy-Stepney, J
T. Surawy-Stepney, J. Kahn, R. Kueng, and M. Guta, Projected least-squares quantum process tomography, Quantum6, 844 (2022)
2022
-
[32]
S. Chen, J. Li, B. Huang, and A. Liu, Tight bounds for quantum state certification with incoherent measure- ments, in2022 IEEE 63rd Annual Symposium on Foun- dations of Computer Science (FOCS)(IEEE, 2022) pp. 1205–1213
2022
-
[33]
Lowe and A
A. Lowe and A. Nayak, Lower bounds for learn- ing quantum states with single-copy measurements, arXiv:2207.14438 (2022)
2022 arXiv
-
[34]
Anshu and S
A. Anshu and S. Arunachalam, A survey on the com- plexity of learning quantum states, Nat. Rev. Phys.6, 59 (2024)
2024
-
[35]
Navascu´ es and S
M. Navascu´ es and S. Popescu, How energy conservation limits our measurements, Phys. Rev. Lett.112, 140502 (2014)
2014
-
[36]
Pucha la, L
Z. Pucha la, L. Pawela, A. Krawiec, and R. Kukulski, Strategies for optimal single-shot discrimination of quan- tum measurements, Phys. Rev. A98, 042103 (2018)
2018
-
[37]
F. B. Maciejewski, Z. Pucha la, and M. Oszmaniec, Ex- ploring quantum average-case distances: Proofs, proper- ties, and examples, IEEE Trans. Inf. Theory69, 4600 (2023)
2023
-
[38]
Ahlswede and A
R. Ahlswede and A. Winter, Strong converse for identi- fication via quantum channels, IEEE Trans. Inf. Theory 48, 569 (2002)
2002
-
[39]
J. A. Tropp, User-friendly tail bounds for sums of random matrices, Found. Comput. Math.12, 389 (2012)
2012
-
[40]
Gross, Recovering low-rank matrices from few coef- ficients in any basis, IEEE Trans
D. Gross, Recovering low-rank matrices from few coef- ficients in any basis, IEEE Trans. Inf. Theory57, 1548 (2011)
2011
-
[41]
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
-
[42]
Dankert, R
C. Dankert, R. Cleve, J. Emerson, and E. Livine, Exact and approximate unitary 2-designs and their application to fidelity estimation, Phys. Rev. A80, 012304 (2009)
2009
-
[43]
J. M. Renes, R. Blume-Kohout, A. J. Scott, and C. M. Caves, Symmetric informationally complete quantum measurements, J. Math. Phys.45, 2171 (2004)
2004
-
[44]
Klappenecker and M
A. Klappenecker and M. Rotteler, Mutually unbiased bases are complex projective 2-designs, inProceedings. International Symposium on Information Theory, 2005. ISIT 2005.(IEEE, 2005) pp. 1740–1744
2005
-
[45]
Barber` a-Rodr ´ ıguez, L
J. Barber` a-Rodr ´ ıguez, L. Zambrano, A. Ac ´ ın, and D. Fa- rina, Boosting projective methods for quantum process and detector tomography, Physical Review Research7, 013208 (2025)
2025
-
[46]
Scarlett and V
J. Scarlett and V. Cevher, An introductory guide to fano’s inequality with applications in statistical estima- tion, arXiv:1901.00555 (2019)
2019 arXiv
-
[47]
S. T. Flammia, D. Gross, Y.-K. Liu, and J. Eisert, Quan- tum tomography via compressed sensing: error bounds, sample complexity and efficient estimators, New J. Phys. 14, 095022 (2012)
2012
-
[48]
I. Roth, R. Kueng, S. Kimmel, Y.-K. Liu, D. Gross, J. Eisert, and M. Kliesch, Recovering quantum gates from few average gate fidelities, Phys. Rev. Lett.121, 170502 (2018)
2018
-
[49]
Efthymiou, A
S. Efthymiou, A. Orgaz-Fuertes, R. Carobene, J. Cereijo, A. Pasquale, S. Ramos-Calderer, S. Bordoni, D. Fuentes- Ruiz, A. Candido, E. Pedicillo,et al., Qibolab: an open- source hybrid quantum operating system, Quantum8, 1247 (2024)
2024
-
[50]
Pasquale, E
A. Pasquale, E. Pedicillo, J. Cereijo, S. Ramos-Calderer, A. Candido, G. Palazzo, R. Carobene, M. Gobbo, S. Efthymiou, Y. P. Tan,et al., Qibocal: an open-source framework for calibration of self-hosted quantum devices, arXiv:2410.00101 (2024)
2024 arXiv
-
[51]
Jiang, A
Z. Jiang, A. Kalev, W. Mruczkiewicz, and H. Neven, Op- timal fermion-to-qubit mapping via ternary trees with applications to reduced quantum states learning, Quan- tum4, 276 (2020)
2020
-
[52]
Z. You, Q. Liu, and Y. Zhou, Circuit optimiza- tion of qubit IC-POVMs for shadow estimation, arXiv:2409.05676 (2024)
2024 arXiv
-
[53]
Oufkir, Sample-optimal quantum process tomog- raphy with non-adaptive incoherent measurements, arXiv:2301.12925 (2023)
A. Oufkir, Sample-optimal quantum process tomog- raphy with non-adaptive incoherent measurements, arXiv:2301.12925 (2023)
2023 arXiv
-
[54]
Meckes and M
E. Meckes and M. Meckes, Spectral measures of powers of random matrices, ECP18, 1 (2013). 12 Appendix A: Estimator using2-designs
2013
-
[55]
The map is defined as [M(X)] i = 1 M ⟨ψi|X|ψ i⟩,(A1) where{|ψ i⟩}M i=1 is a set of states that forms a 2-design
Least-squares estimator using2-designs The measurement step in the protocol defines a lin- ear mapM:M(C d)→R M from POVM elements to probabilities. The map is defined as [M(X)] i = 1 M ⟨ψi|X|ψ i⟩,(A1) where{|ψ i⟩}M i=1 is a set of states that forms a 2-design. In the experimen...
-
[56]
For the matrix Bernstein inequality, Eq
Parameters for matrix Bernstein inequality using 2-designs Now we want to bound the error in the protocol. For the matrix Bernstein inequality, Eq. (5), we need to cal- culate the parametersKandσfor the random matrices 1 N ˆX (x) k −F (x) , whereF (x) is a POVM element that gr...
-
[57]
Then, we haveM=m n elements of the form |ψi⟩=|ψ i1 ψi2
Least-squares estimator using single-qubit 2-designs We now assume we have ann-qubit system and that the set of states{|ψ i⟩}M i=1 we measure is a tensor product ofnsingle-qubit sets{|ψ ik ⟩}m ik=1 that form a 2-design. Then, we haveM=m n elements of the form |ψi⟩=|ψ i1 ψi2 . ...
-
[58]
Parameters for matrix Bernstein inequality using local 2-designs For the matrix Bernstein inequality we need to calculate the parametersKandσfor the random matrices 1 N ˆX (x) k −F (x) , whereF (x) is a POVM element that groups the outcomesj∈xand ˆXk = Nn k=1 (6|ψik ⟩⟨ψik | −2...
-
[59]
To do this, we will use the following [53]: 15 Theorem 11.Letkanddbe positive integers andM= U1(d)× · · · ×Uk(d)a space equipped with theL 2-sum of Hilbert–Schmidt metrics
Construction of anϵ-packing Let{U j}L/2 j=1 be a set of Haar-random unitaries,Pa rank-d/2 projector andE {Uj } a POVM withLelements given by Ej = (1−ϵ) L 1+ 2ϵ L UjP U† j Ej+L/2 = (1 +ϵ) L 1− 2ϵ L UjP U† j (D1) We want to construct a large set of POVMs of this form such that √...
-
[60]
Then, there exists a set ofR∈ 16 exp(Ω(d2L))POVMs of the form of Eq.(D1)which forms anϵ/4packing for √ d dav and satisfies I(X:Y)≤I({U j}:Z),(D12) whereY= (Y 1,
Upper bound on mutual information for the average distance Lemma 13.LetX∼U nif[R],{U j}L/2 j=1 be a set of independent Haar random unitaries and{ρ i}N i=1 a set ofNquantum states. Then, there exists a set ofR∈ 16 exp(Ω(d2L))POVMs of the form of Eq.(D1)which forms anϵ/4packing ...
-
[61]
Lower bound on the sample complexity ofdav Theorem 15.Any procedure for quantum measurement tomography of a POVM on ad-dimensional Hilbert space 17 that isϵ/8accurate in average distance using nonadap- tive, single-copy measurements on known input states re- quires N∈Ω d2L ϵ2 ...
-
[62]
Furthermore, we also characterize this POVM as a two-qubit measurement, instead of a projec- tion into a single qubit system
andϕ= 3π/4 [50] to generate a SIC-POVM [51]. Furthermore, we also characterize this POVM as a two-qubit measurement, instead of a projec- tion into a single qubit system. In Fig. 4 we showcase the reconstruction. The results are similar to the ones in the main text, but in an ...
-
[63]
The choice of a CZ gate as the entangling operation is due to the nate interactions of the experimental device where this protocol is deployed
andϕ= 3π/4, this results in a Symmetrical Informationally Complete measurement. The choice of a CZ gate as the entangling operation is due to the nate interactions of the experimental device where this protocol is deployed. 19 Re(E*j) Re(Ej) Im(E*j) Im(Ej) |E*j−Ej| Re(E*j) Re(...
Reviewed August 6, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.