Pith. sign in

REVIEW 2 major objections 3 minor 65 references

Enhancing Quantum State Reconstruction with Structured Classical Shadows

T0 review · 2 major / 3 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Projected classical shadows reconstruct full quantum states with provably optimal copy counts

desk verdict General-state PCS is solid and worth knowing; the low-rank and MPO theorems have a restricted-norm gap that fails under the paper's own projection, so the structured-state claims need proof repair. read the letter →

arxiv 2501.03144 v2 pith:5WHG2ITE submitted 2025-01-06 quant-ph math.OC

Zhen Qin , Joseph M. Lukens , Brian T. Kirby , Zhihui Zhu This is my paper · ORCID
classification quant-phmath.OC
keywords quantumstatetomographyclassicalshadowsHaar-randommeasurementslow-rankstatesmatrixproductoperatorscoveringnumbersFrobeniusnormrecoveryprojectedestimator
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that the classical shadow estimator—normally used to predict individual properties of a quantum state—can be converted into a full state-reconstruction method by adding a projection step. After projecting the unphysical shadow estimate onto the set of allowed states, the recovery error is controlled by the covering-number complexity of that state class. For general $n$-qubit states this needs $O(4^n)$ copies; for rank-$r$ states it drops to $O(2^n r)$, matching information-theoretic optimal bounds in both cases; and for matrix-product states it needs $O(n^2)$ copies, improving on the previous Haar-random $O(n^3)$. If the results hold, one measurement primitive suffices for both property estimation and tomography, with copy counts that adapt to the structure of the target state.

What carries the argument

The load-bearing object is the projection operator $\mathcal{P}_{\mathcal{X}}$ applied to the Haar-random classical shadow estimator, together with the covering-number complexity $\log \mathcal{N}_{1/2}(\mathcal{X})$ of the difference set of the state class $\mathcal{X}$. This covering number controls the sample complexity through a Bernstein concentration inequality over an $\epsilon$-net: for general physical states it is $O(4^n)$, for rank-$r$ states $O(2^n r)$, and for MPO states $O(nD^2\log n)$-style bounds on the log-cover. The actual projections used in practice are the simplex projection onto physical density matrices, rank-$r$ truncation for low-rank states, and tensor-train SVD for MPO states, with quasi-optimality properties that let the analysis go through without solving the exact (possibly hard) projection problem.

What would settle it

Run LR-PCS on a known rank-1 state for a range of measurement counts $M$, and after the simplex projection record the rank of $\rho_{\mathrm{LR-PCS}}-\rho^\star$ and the Frobenius error. If the rank systematically exceeds $2$, or if the error decays as $\sqrt{4^n/M}$ rather than $\sqrt{2^n r/M}$, the rank-restricted argument underlying Theorem 3 is not capturing the actual estimator.

Watch

Extended reading notes

Core claim

The paper's central claim is that the raw classical shadow matrix $\rho_{\mathrm{CS}}$, though generally not a physical state, is a good starting point for full tomography once projected onto the relevant state class $\mathcal{X}$ via $\rho_{\mathrm{PCS}} = \arg\min_{\rho\in\mathcal{X}}\|\rho - \rho_{\mathrm{CS}}\|_F$. The general recovery guarantee (Theorem 1) bounds $\|\rho_{\mathrm{PCS}}-\rho^\star\|_F$ by the square root of the log-covering number of the difference set for $\mathcal{X}$ divided by the number of measurements $M$. Specializing this bound gives $O(\sqrt{4^n/M})$ for all physical states, $O(\sqrt{2^n r/M})$ for rank-$r$ states (Theorem 3), and $O(\sqrt{n^2 D^2\log n/M})$ for matrix-product states with bond dimension $D$ (Theorem 4). The low-rank and MPO guarantees are achieved by practical two-step projections: eigenvalue truncation followed by simplex projection, and tensor-train SVD followed by simplex projection, respectively.

Load-bearing premise

The low-rank guarantee rests on the assumption, stated as the equality in Eq. (C3), that the difference between the projected estimate and the true state has rank at most $2r$; the proof does not derive this, and the simplex projection can add small positive eigenvalues that break the rank bound.

Editorial extensions

If this is right

  • Classical shadows can serve as a tomographic primitive: the same Haar-random single-copy measurements that predict observables also reconstruct the density matrix with control on the Frobenius error.
  • For general and low-rank states the copy counts match information-theoretic lower bounds, so no independent single-copy measurement strategy can do asymptotically better.
  • For matrix-product states the copy count drops from the previous Haar-random $O(n^3)$ to $O(n^2)$, making structure-aware tomography feasible at larger qubit numbers.
  • The projection step returns a physical density matrix, so downstream uses that require valid states—fidelities, entropies, process benchmarks—can consume the estimate directly.

Reading between the lines

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

  • The same covering-number-plus-projection argument could be replayed for other measurement ensembles such as Clifford, Pauli, or 3-design measurements, yielding analogous guarantees with different shadow-channel constants; the paper itself analyzes only Haar-random measurements.
  • The $O(n^2)$ MPO bound is not degrees-of-freedom optimal: the paper notes that an exact MPO-preserving projection would remove one factor of $n$. Testing alternating projections between the simplex and the MPO set is a natural next step.
  • Because the simplex projection can assign positive weight to zero eigenvalues when the truncated matrix has trace below one, LR-PCS may systematically overestimate rank in the low-copy regime; error metrics that track the eigenvalue spectrum would reveal this behavior.
  • The formalism suggests that other structured state families, such as projected entangled pair operators or MERA states, could inherit similar guarantees once a quasi-optimal projection onto those families is available.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 3 minor

Summary. This paper introduces projected classical shadows (PCS) for full quantum state tomography. The estimator is the standard Haar-random classical shadow rho_CS, followed by a projection onto a structured physical set X: all physical states, low-rank states, or matrix product operator (MPO) states. The paper claims that PCS requires O(4^n) copies for general states (Cor. 1), that the two-step estimator rho_LR-PCS = P_simplex(P_rank-r(rho_CS)) achieves O(2^n r) copies for rank-r states (Thm. 3), and that the two-step estimator rho_MPO-PCS = P_simplex(SVD^tt_D(rho_CS)) achieves O(n^2) copies for MPO states (Thm. 4). The proofs combine a covering-number concentration argument (Thm. 1) with restricted Frobenius norms over difference sets. Numerical experiments on n = 4 and n = 7 qubits illustrate the claimed scalings.

Significance. Section II's exact mean-squared-error formula, Eq. (4), is clean and the general-state guarantee in Cor. 1 follows from convexity, nonexpansiveness, and standard covering bounds; that part is a solid contribution. If the structured-state guarantees could be rigorously justified, PCS would be an attractive and simple method matching information-theoretic sample complexities and improving the prior Haar-random MPO bound. The numerical experiments are consistent with the claimed scalings. However, as detailed below, the low-rank theorem is not proven as written, and the MPO theorem has an unstated Hermiticity assumption, so the central structured-state claims require substantial repair.

major comments (2)
  1. [Appendix C, Eq. (C3) / Theorem 3] The proof of Theorem 3 hinges on the equality ||rho_LR-PCS - rho*||_F = ||rho_LR-PCS - rho*||_{F,2r}, which requires rank(rho_LR-PCS - rho*) <= 2r. This is not guaranteed: P_simplex applied after P_rank-r can assign positive eigenvalues to the zero eigenvalues whenever the truncated matrix has trace below one, so rho_LR-PCS can be full rank even for r = 1. A concrete Hermitian trace-one example is d = 4, rho_CS with eigenvalues (0.4, 0.4, 0.4, -0.2) and rho* = diag(1, 0, 0, 0); P_rank-1 followed by the simplex projection gives diag(0.55, 0.15, 0.15, 0.15), so the error matrix has eigenvalues (-0.45, 0.15, 0.15, 0.15), i.e., rank 4. Its Frobenius norm is approximately 0.520, while the maximum inner product against rank-2 traceless unit-Frobenius matrices is approximately 0.490, so Eq. (C3)'s equality fails and the subsequent covering argument over bX_{2r} does not control the actual error. The O(2^n r) bound of Theorem 3 is therefore not established as written.
  2. [Appendix D, Eq. (D3) / Theorem 4] The proof of Theorem 4 requires that SVD^tt_D(rho_CS) be Hermitian so that P_simplex in Eq. (17) is defined and so that the restricted-norm manipulation in Eq. (D3) is meaningful. TT-SVD of a Hermitian tensor is not automatically Hermitian, and the manuscript neither proves a symmetry-preserving property of its TT-SVD nor symmetrizes the output. If SVD^tt_D(rho_CS) is not Hermitian, the eigenvalue projection P_simplex is not defined, and the equality ||P_trace(SVD^tt_D(rho_CS)) - rho*||_F = ||.||_{F,2D} in Eq. (D3) requires the matrix inside to be Hermitian. This is a load-bearing gap in the MPO guarantee. For completeness, the separate concern that P_simplex raises the bond dimension of rho_MPO-PCS does not by itself invalidate Eq. (D3), because the restricted norm there is applied to P_trace(SVD^tt_D(rho_CS)) - rho* rather than to rho_MPO-PCS - rho*; the real issue is the unstated Hermiticity assumption.
minor comments (3)
  1. [Appendix D, Eq. (D5)] The covering-number bound is stated as ((4n + epsilon)/epsilon)^{4n D^2} even though the net is over bX_{2D}; replacing D by 2D in Lemma 3 changes only the constant in the exponent. The asymptotic statement is unaffected, but the displayed formula should be corrected or the notation clarified.
  2. [Section III, MPO subsection] The text says the bond dimension of rho_MPO-PCS 'may differ slightly from D due to the simplex projection, but the recovery error still depends on D.' This claim is not justified in the proof and should be either proved or removed, since the numerical and theoretical dependence on D are the main quantitative claims for MPO states.
  3. [Section IV, Fig. 2 caption] The caption states that the figures span M = 250 to M = 10000, but the r = 1 and r = 4 panels appear to begin at M = 1000; the caption should match the plotted ranges.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: PCS bounds are derived via generic covering arguments, not by fitting or by assuming the conclusions.

full rationale

The central claims—O(4^n) copies for general states, O(2^n r) for rank-r states, and O(n^2 D^2 log n) for MPO states—are obtained from a standard covering-number framework (Theorem 1) applied to Haar-random classical shadows, with the projection analyzed through nonexpansiveness and quasi-optimality lemmas. The general-state MSE in Eq. (4) is computed directly from the shadow estimator and is not reused as the projection guarantee. The low-rank and MPO bounds use covering numbers for the respective difference sets; these covering bounds are either proved in the appendices (Lemma 3) or cited from prior work, and they do not assume the target sample-complexity bound. The authors' own prior results ([30,31,52,59]) supply technical covering and tomography lemmas and the O(n^3) baseline being improved, but the PCS theorems do not reduce to those citations: the projection argument and restricted-norm concentration are carried out in the paper. The only substantive concern is the asserted restricted-norm equalities in Eqs. (C3) and (D3), which require that P_simplex after rank/bond-dimension truncation leaves the error in the low-rank/MPO set; this premise is unproven and possibly false, but it is a proof soundness gap rather than a circular definition, because the restricted norm is not defined in terms of the theorem's conclusion and no fitted parameter or self-citation forces the bound. Accordingly, no circularity step meets the evidentiary bar.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The central claims rest on standard tools such as covering numbers, moment bounds, and TT-SVD quasi-optimality, all drawn from cited literature. The two ad hoc assumptions listed above are load-bearing for the low-rank and MPO theorems and are not established in the paper.

assumptions (6)
  • standard math Bernstein-type moment bound E[|s_m|^a] <= 6 * 2^{a-2} a! for Haar random rank-one measurements.
    Invoked in Appendix B after Eq. (B7); justified by unitary p-design properties and matrix moment bounds from Refs [55,56]. Appears correct.
  • standard math Covering number of the MPO difference set is at most ((4n + epsilon)/epsilon)^{4nD^2}.
    Proved in Lemma 3 using spectral-norm nets of tensor cores; this is a standard covering argument.
  • standard math Projection onto a closed convex set is nonexpansive in Frobenius norm.
    Used in Appendix B first inequality, Appendix C first inequality, and Appendix D first inequality.
  • standard math TT-SVD is quasi-optimal with factor 1 + sqrt(n-1) in Frobenius norm for tensor-train rank D.
    Used in Appendix D third inequality; it is a standard property of tensor-train SVD from Ref [40].
  • ad hoc to paper The difference rho_LR-PCS - rho* has rank at most 2r.
    Unstated premise of the equality in Eq. (C3). The simplex projection after rank-r eigenvalue truncation can assign positive values to zero eigenvalues, potentially making rho_LR-PCS full rank; no argument for rank <= 2r is given.
  • ad hoc to paper For A = SVD^tt_D(rho_CS), ||P_simplex(A) - rho*||_F <= ||P_trace(A) - rho*||_F.
    Used in the first inequality of Eq. (D3) but not derived. P_simplex and P_trace are different projections and this ordering is not generally obvious.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing Quantum State Reconstruction with Structured Classical Shadows." pith.science (2026). https://pith.science/paper/5WHG2ITE

@misc{pith2026250103144,
  author       = {Pith},
  title        = {Pith review of: Enhancing Quantum State Reconstruction with Structured Classical Shadows},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5WHG2ITE}},
  note         = {Machine review of arXiv:2501.03144}
}
abstract

Quantum state tomography (QST) remains the prevailing method for benchmarking and verifying quantum devices; however, its application to large quantum systems is rendered impractical due to the exponential growth in both the required number of total state copies and classical computational resources. Recently, the classical shadow (CS) method has been introduced as a more computationally efficient alternative, capable of accurately predicting key quantum state properties. Despite its advantages, a critical question remains as to whether the CS method can be extended to perform QST with guaranteed performance. In this paper, we address this challenge by introducing a projected classical shadow (PCS) method with guaranteed performance for QST based on Haar-random projective measurements. PCS extends the standard CS method by incorporating a projection step onto the target subspace. For a general quantum state consisting of $n$ qubits, our method requires a minimum of $O(4^n)$ total state copies to achieve a bounded recovery error in the Frobenius norm between the reconstructed and true density matrices, reducing to $O(2^n r)$ for states of rank $r<2^n$ -- meeting information-theoretic optimal bounds in both cases. For matrix product operator states, we demonstrate that the PCS method can recover the ground-truth state with $O(n^2)$ total state copies, improving upon the previously established Haar-random bound of $O(n^3)$. Simulation results further validate the effectiveness of the proposed PCS method.

Figures

Figures reproduced from arXiv: 2501.03144 by the authors.

Figure 1
Figure 1. FIG. 1. Illustration of proposed PCS method. Given an initial CS estimate [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Mean squared error as a function of state copies [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Mean squared error as a function of state copies [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Mean square error as a function of the number of state copies [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. Mean squared error as a function of the total qubit number with [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

65 extracted references · 53 canonical work pages

  1. [1]

    CS yields an unbiased estimate (E[ρCS] = ρ⋆), whereas the solution from QST is often biased [43]

    work page

  2. [2]

    While the CS estimator is typically unphysical (not PSD), leading QST methods like MLE [5], projected least squares [44], and Bayesian inference [9] enforce physicality by construction

    work page

  3. [3]

    CS boasts significantly lower computational complexity compared to QST

    work page

  4. [4]

    For M ≪ 2n, CS outperforms QST in predicting cer- tain linear observables, not in predicting the entire state [17, 45]

    work page

  5. [5]

    Currently no known method for similarly reducing CS scaling exists

    Including prior information about state structure allows for a reduction in scaling in QST (see Tables I and II). Currently no known method for similarly reducing CS scaling exists. In other words, CS requires O(4n) mea- surements for estimating the full state, as demonstrated in Eq. (5). In the next section we investigate methods for incorporating prior ...

    work page 2000

  6. [6]

    J. Haah, A. Harrow, Z. Ji, X. Wu, and N. Yu, IEEE Transactions on Information Theory 63, 5628 (2017)

    work page 2017

  7. [7]

    Bertrand and P

    J. Bertrand and P. Bertrand, Foundations of Physics 17, 397 (1987)

    work page 1987

  8. [8]

    V ogel and H

    K. V ogel and H. Risken, Physical Review A40, 2847 (1989)

    work page 1989

Show all 65 references
  1. [9]

    Leonhardt, Physical review letters 74, 4101 (1995)

    U. Leonhardt, Physical review letters 74, 4101 (1995)

    1995

  2. [10]

    Hradil, Physical Review A 55, R1561 (1997)

    Z. Hradil, Physical Review A 55, R1561 (1997)

    1997

  3. [11]

    D. F. V . James, P. G. Kwiat, W. J. Munro, and A. G. White, Phys. Rev. A 64, 052312 (2001)

    2001

  4. [12]

    Blume-Kohout, arXiv:1202.5270 (2012)

    R. Blume-Kohout, arXiv:1202.5270 (2012)

    2012 arXiv

  5. [13]

    Fano, Reviews of modern physics 29, 74 (1957)

    U. Fano, Reviews of modern physics 29, 74 (1957)

    1957

  6. [14]

    ˇReh´aˇcek, Z

    J. ˇReh´aˇcek, Z. Hradil, and M. Je ˇzek, Physical Review A 63, 040303 (2001)

    2001

  7. [15]

    Blume-Kohout, New Journal of Physics 12, 043034 (2010)

    R. Blume-Kohout, New Journal of Physics 12, 043034 (2010)

    2010

  8. [16]

    Granade, J

    C. Granade, J. Combes, and D. Cory, new Journal of Physics 18, 033024 (2016)

    2016

  9. [17]

    J. M. Lukens, K. J. Law, A. Jasra, and P. Lougovski, New Jour- nal of Physics 22, 063038 (2020)

    2020

  10. [18]

    P. Sen, A. S. Bhatia, K. S. Bhangu, and A. Elbeltagi, Plos one 17, e0262346 (2022)

    2022

  11. [19]

    Faist and R

    P. Faist and R. Renner, Physical review letters 117, 010404 (2016)

    2016

  12. [20]

    Lohani, B

    S. Lohani, B. T. Kirby, M. Brodsky, O. Danaci, and R. T. Glasser, Machine Learning: Science and Technology1, 035007 (2020)

    2020

  13. [21]

    Kyrillidis, A

    A. Kyrillidis, A. Kalev, D. Park, S. Bhojanapalli, C. Caramanis, and S. Sanghavi, npj Quantum Information 4, 1 (2018)

    2018

  14. [23]

    Z. Zhu, J. M. Lukens, and B. T. Kirby, Quantum 8, 1455 (2024)

    2024

  15. [24]

    F. G. Brand ˜ao, R. Kueng, and D. S. Franc ¸a, arXiv preprint arXiv:2009.08216 (2020)

    2020 arXiv

  16. [25]

    Y . Liu, D. Wang, S. Xue, A. Huang, X. Fu, X. Qiang, P. Xu, H.-L. Huang, M. Deng, C. Guo, et al., Physical Review A 101, 052316 (2020)

    2020

  17. [26]

    Lloyd, M

    S. Lloyd, M. Mohseni, and P. Rebentrost, Nature physics 10, 631 (2014)

    2014

  18. [27]

    Machine learning assisted quantum state to- mography,

    M. K. Kurmapu, “Machine learning assisted quantum state to- mography,” (2020)

    2020

  19. [28]

    Kueng, H

    R. Kueng, H. Rauhut, and U. Terstiege, Appl. Comput. Har- mon. Anal. 41, 88 (2017)

    2017

  20. [29]

    Gut ¸˘a, J

    M. Gut ¸˘a, J. Kahn, R. Kueng, and J. A. Tropp, Journal of Physics A: Mathematical and Theoretical 53, 204001 (2020)

    2020

  21. [30]

    Z. Qin, C. Jameson, Z. Gong, M. B. Wakin, and Z. Zhu, IEEE Transactions on Information Theory 70, 5030 (2024)

    2024

  22. [31]

    V oroninski, arXiv preprint arXiv:1309.7669 (2013)

    V . V oroninski, arXiv preprint arXiv:1309.7669 (2013)

    2013 arXiv

  23. [32]

    Liu, Advances in Neural Information Processing Systems 24 (2011)

    Y .-K. Liu, Advances in Neural Information Processing Systems 24 (2011)

    2011

  24. [33]

    Eisert, M

    J. Eisert, M. Cramer, and M. B. Plenio, Rev. Mod. Phys. 82, 277 (2010)

    2010

  25. [34]

    Pirvu, V

    B. Pirvu, V . Murg, J. I. Cirac, and F. Verstraete, New Journal of Physics 12, 025012 (2010)

    2010

  26. [35]

    K. Noh, L. Jiang, and B. Fefferman, Quantum 4, 318 (2020)

    2020

  27. [36]

    Grier, H

    D. Grier, H. Pashayan, and L. Schaeffer, Quantum 8, 1373 (2024)

    2024

  28. [37]

    Z. Qin, C. Jameson, A. Goldar, M. B. Wakin, Z. Gong, and Z. Zhu, arXiv preprint arXiv:2410.02583 (2024)

    2024

  29. [38]

    Huang, R

    H.-Y . Huang, R. Kueng, and J. Preskill, Nature Physics 16, 1050 (2020)

    2020

  30. [39]

    Acharya, S

    A. Acharya, S. Saha, and A. M. Sengupta, Physical Review A 104, 052418 (2021)

    2021

  31. [40]

    to truncate it to an MPS, and then normalize the re- sult to unit length. As a result, entry ρ⋆(i1 · · ·i7, j1 · · ·j7) can be expressed as ρ⋆(i1 · · ·i7, j1 · · ·j7) = ( U ⋆ 1 i1 ⊗ U ⋆ 1 j1 † ) · · ·(U ⋆ 7 i7 ⊗ U ⋆ 7 j7 † ) = X ⋆ 1 i1,j1 · · ·X ⋆ 7 i7,j7, where ⊗ denotes the ...

  32. [41]

    Struchalin, Y

    G. Struchalin, Y . A. Zagorovskii, E. Kovlakov, S. Straupe, and S. Kulik, PRX Quantum 2, 010307 (2021)

    2021

  33. [42]

    A. A. Akhtar, H.-Y . Hu, and Y .-Z. You, Quantum 7, 1026 (2023)

    2023

  34. [43]

    Ippoliti, Quantum 8, 1293 (2024)

    M. Ippoliti, Quantum 8, 1293 (2024)

    2024

  35. [44]

    Becker, N

    S. Becker, N. Datta, L. Lami, and C. Rouz ´e, IEEE Transactions on Information Theory (2024)

    2024

  36. [45]

    Chen and X

    Y . Chen and X. Ye, arXiv preprint arXiv:1101.6081 (2011)

    2011 arXiv

  37. [46]

    Oseledets, SIAM Journal on Scientific Computing 33, 2295 (2011)

    I. Oseledets, SIAM Journal on Scientific Computing 33, 2295 (2011)

    2011

  38. [47]

    M. A. Nielsen and I. L. Chuang, Cambridge (2000)

    2000

  39. [48]

    ,AK } ∈ C2n×2n , s

    Specifically, a POVM is characterized as a set of PSD matri- ces: {A1, . . . ,AK } ∈ C2n×2n , s. t. PK k=1 Ak = I2n. Each POVM element Ak corresponds to a potential outcome of a quantum measurement with the special case of projec- tive measurements corresponding to the case wh...

  40. [49]

    Schwemmer, L

    C. Schwemmer, L. Knips, D. Richart, H. Weinfurter, T. Mo- roder, M. Kleinmann, and O. G ¨uhne, Physical review letters 114, 080403 (2015)

    2015

  41. [50]

    J. A. Smolin, J. M. Gambetta, and G. Smith, Physical review letters 108, 070502 (2012)

    2012

  42. [51]

    J. M. Lukens, K. J. Law, and R. S. Bennink, npj Quantum In- formation 7, 113 (2021)

    2021

  43. [52]

    We chose the label “simplex” for this set since the eigenvalues {λk} of all physical states define a standard simplex, i.e., λk ≥ 0 and P k λk = 1. 9

  44. [53]

    Specifically, i1 · · ·in represents the (i1 + Pn ℓ=2 2ℓ−1(iℓ − 1))- th row

  45. [54]

    A. H. Werner, D. Jaschke, P. Silvi, M. Kliesch, T. Calarco, J. Eisert, and S. Montangero, Physical review letters 116, 237201 (2016)

    2016

  46. [55]

    The thermal state is then defined as ρ⋆ = e−H/T trace(e−H/T )

    The thermal state is generated from the 1D quantum Ising model H = Pn−1 j=1 σz j σz j+1 + Pn j=1 σx j with σa j = I2j−1 ⊗ σa ⊗ I2n−j ∈ R2n×2n , a = x, zand σx = 0 1 1 0 , σz = 1 0 0 −1 . The thermal state is then defined as ρ⋆ = e−H/T trace(e−H/T )

  47. [56]

    The GHZ state is constructed as ρ⋆ = gg † where g =h 1√ 2 0 · · ·0 1√ 2 i⊤ ∈ R2n×1

  48. [57]

    Cramer, M

    M. Cramer, M. B. Plenio, S. T. Flammia, R. Somma, D. Gross, S. D. Bartlett, O. Landon-Cardinal, D. Poulin, and Y .-K. Liu, Nature communications 1, 1 (2010)

    2010

  49. [58]

    Jameson, Z

    C. Jameson, Z. Qin, A. Goldar, M. B. Wakin, Z. Zhu, and Z. Gong, arXiv preprint arXiv:2408.07115 (2024)

    2024 arXiv

  50. [59]

    J. I. Cirac, D. Perez-Garcia, N. Schuch, and F. Verstraete, Re- views of Modern Physics 93, 045003 (2021)

    2021

  51. [60]

    Haegeman, T

    J. Haegeman, T. J. Osborne, H. Verschelde, and F. Verstraete, Physical review letters 110, 100402 (2013)

    2013

  52. [61]

    A. A. Mele, arXiv preprint arXiv:2307.08956 (2023)

    2023 arXiv

  53. [62]

    Z. Zhu, Q. Li, G. Tang, and M. B. Wakin, IEEE Transactions on Information Theory 67, 1308 (2021)

    2021

  54. [63]

    E. J. Cand` es, X. Li, Y . Ma, and J. Wright, J. ACM58, 1 (2011)

    2011

  55. [64]

    Holtz, T

    S. Holtz, T. Rohwedder, and R. Schneider, Numerische Math- ematik 120, 701 (2012)

    2012

  56. [65]

    Z. Qin, M. B. Wakin, and Z. Zhu, arXiv preprint arXiv:2401.02592 (2024)

    2024 arXiv

  57. [66]

    Zhang and D

    A. Zhang and D. Xia, IEEE Transactions on Information The- ory 64, 7311 (2018). Appendix A: Proof of Equation (4) Proof. We expand E ∥ρCS − ρ⋆∥2 F as follows: E ∥ρCS − ρ⋆∥2 F = E 1 M MX m=1 ρm − ρ⋆ 2 F = E * 1 M MX m=1 (ρm − ρ⋆), 1 M MX m=1 (ρm − ρ⋆) + = 1 M 2 E MX m=1 ∥ρm − ρ...

    2018

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.