Pith. sign in

REVIEW 3 major objections 5 minor 35 references

Tomography by Design: An Algebraic Approach to Low-Rank Quantum States

T0 review · 3 major / 5 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read A rank-R density matrix is uniquely fixed by a few of its principal submatrices, and the full state then follows from least squares.

desk verdict A clean idea with a real proof gap: the recovery guarantee fails for a simple non-generic state, and the error bound drops cross terms. read the letter →

arxiv 2602.15202 v2 pith:PM3MKINW submitted 2026-02-16 quant-ph cs.AIcs.NAeess.SPmath.NAstat.CO

Shakir Showkat Sofi , Charlotte Vermeylen , Lieven De Lathauwer This is my paper · ORCID
classification quant-phcs.AIcs.NAeess.SPmath.NAstat.CO
keywords quantumstatetomographylow-rankdensitymatrixcompletionsubspaceestimationisorankalgebraicreconstructionclassicalshadowsdeterministicrecovery
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

The paper aims to show that full quantum state tomography for low-rank states can be done deterministically by measuring a structured set of principal submatrices, not by solving a costly convex program. If the chosen submatrices each have the same rank as the full state, their padded column spaces contain the state's column space, and when their intersection has exactly R dimensions the column space is recovered exactly. An orthonormal basis for that intersection is obtained as the top-R left singular vectors of a concatenated matrix, after which every unknown entry is filled by least-squares regression. This yields O(RD) measurement settings and a reconstruction based only on standard linear algebra, with an explicit bound on how noise in local measurements propagates to the global subspace estimate.

What carries the argument

The load-bearing object is the isorank principal submatrix: a square submatrix on an index set r_l whose rank equals the rank R of the full density matrix. Because ρ_obs^(l) = U_l Σ_l U_l^H, padding its top-R eigenvectors back into the D-dimensional space gives a subspace Q_l that contains col(ρ). Intersecting these local subspaces across blocks recovers the global column space exactly when the intersection has dimension R; the intersection is computed by the closed-form solution of minimizing the sum of squared chordal distances to the Q_l, namely the top-R left singular vectors of the concatenated basis matrix Q_tot. This machinery converts a tomographic inversion problem into two standard

What would settle it

Take the noiseless rank-2 state ρ = 1/2(|00⟩⟨00| + |01⟩⟨01|) on two qubits and use the overlapping block pattern with blocks {00,01,10} and {01,10,11} (R=2, overlap two indices). The second block has matrix diag(0.5,0,0), rank 1, violating isorank, so the subspace intersection computed from the two blocks fails to recover the |00⟩ component and the algorithm's output does not equal ρ despite zero noise.

Watch

Extended reading notes

Core claim

The central claim is that for a rank-R density matrix ρ, full recovery is guaranteed from fully observed isorank principal submatrices whose padded eigen-subspaces Q_l satisfy dim(∩_l Q_l) = R. Under this condition, col(ρ) equals the intersection ∩_l Q_l, and a basis can be computed in closed form as the top-R left singular vectors of Q_tot = [Q^(1), ..., Q^(L)]. The paper proves the global subspace error obeys d_c(Ũ,U) ≤ ε sqrt(2 Σ_l |r_l|)/(δ σ_min(P̃_tot)) under a spectral-gap assumption, and shows that a simple overlapping block-diagonal pattern with pairwise overlap at least R and full coordinate coverage satisfies the required generic conditions. From the recovered column space, each c

Load-bearing premise

The entire recovery guarantee rests on every measured principal submatrix having rank exactly R (the isorank property) and on the generic overlap conditions holding; the paper offers no practical test to confirm these properties for the unknown state.

Editorial extensions

If this is right

  • Selective QST with the proposed overlapping block pattern needs O(RD) measurement settings, achieving deterministic (non-probabilistic) recovery for generic low-rank mixed states.
  • The reconstruction is purely algebraic — EVD, SVD, and least squares — so it avoids the computational cost of nuclear-norm SDPs and non-convex factorization methods.
  • The error bound d_c(Ũ,U) ≤ ε√(2Σ|r_l|)/(δσ_min(P̃_tot)) shows that measurement noise is amplified by a factor controlled by the local spectral gap δ and the conditioning of the aggregate projector; larger overlaps reduce the error.
  • The same structured-entry data can be supplied by classical shadow estimation instead of per-entry measurements, combining low-rank recovery with shadow tomography's logarithmic measurement complexity per observable.
  • For rank-1 states the measurement count reduces to 3D−2, matching known directly-complete pure-state measurements.

Reading between the lines

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

  • If the isorank condition is not verified in practice, a state whose support is aligned with the measurement blocks (e.g., a superposition supported on only one block) can silently violate rank(ρ_obs^(l)) = R; one could detect this by comparing the block ranks to the assumed R and adaptively enlarging blocks.
  • The subspace-recovery formulation suggests an incremental or streaming tomography protocol: blocks can be measured one at a time and the global column space updated by adjusting the SVD of Q_tot, which may be useful when measurement settings are scheduled adaptively.
  • The same intersection-of-subspaces principle extends naturally to tensor-train or matrix-product-state completions from fiberwise observations, where the local subspaces come from matricized blocks rather than principal submatrices.
  • The error bound invites a design heuristic: choose the overlap d that minimizes the noise-amplification term σ_min(P̃_tot) for a fixed measurement budget, rather than the minimal d=1 pattern.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes an algebraic method for low-rank quantum state tomography. Given a selection pattern of principal submatrices of the density matrix, the algorithm computes the column space of ρ as the intersection of subspaces induced by isorank submatrices: each submatrix is diagonalized, its top-R eigenvectors are padded to the full space, and the top-R left singular vectors of the concatenated padded basis are taken as an estimate of col(ρ). The state is then reconstructed by least squares. The authors claim deterministic recovery guarantees for generic low-rank states, a noisy subspace error bound, O(RD) measurement complexity, and numerical superiority over nuclear-norm minimization and Burer-Monteiro factorization on a five-qubit Ginibre-ensemble example.

Significance. If the recovery guarantee and error analysis were rigorous, the algebraic approach would be a useful contribution: it replaces convex optimization by standard numerical linear algebra, has a simple closed-form reconstruction, and its structured measurement count is competitive with compressed-sensing QST. The paper also sensibly draws on the existing algebraic matrix-completion literature rather than re-deriving it. However, the central recovery condition is state-dependent and is only asserted as 'generic', and the noisy perturbation bound contains a false norm identity and an unquantified spectral quantity. As it stands, the significance of the claimed deterministic guarantee is not established to the standard required for publication.

major comments (3)
  1. [Sec. III-B and III-D] The central recovery equality col(ρ)=∩_l Q_l is asserted under 'generic conditions' cited from [25,27,28], but the two state-dependent hypotheses—(i) every measured principal submatrix is isorank and (ii) dim ∩_l Q_l = R—are neither derived from rank(ρ)=R nor checked by Algorithm 1. They can fail for physically meaningful states even when all observed blocks have rank R. Example: D=3, R=1, ρ=|v><v| with v=(1,0,1)^T/√2, and blocks r1={1,2}, r2={2,3}. Both 2×2 principal submatrices have rank 1, the blocks share one row and cover [3], so the stated pattern conditions are met, yet Q1=Q2=span{e1,e3}, the intersection has dimension 2, and the top singular vector of [Q^(1),Q^(2)] is not col(ρ). The wording in III-D ('This condition is generically satisfied when ...') is too weak to support the subsequent 'deterministic' recovery claim; the hypotheses need to be stated as explicit assumptions an
  2. [Sec. III-B, Error Analysis] The displayed identity ∥P̃_tot − P_tot∥_F^2 = Σ_l ∥P̃_Ql − P_Ql∥_F^2 is incorrect. From the definition P_tot = Σ_l Q^(l)Q^(l)H, the difference is a sum, so the squared Frobenius norm contains cross terms Σ_{l≠k} Tr((P̃_Ql − P_Ql)(P̃_Qk − P_Qk)); these are not generally zero. Consequently the stated noise accumulation bound is unproved. This is load-bearing because the paper's deterministic subspace-error guarantee is one of its central claims. The authors should replace this step with a valid perturbation analysis, for example by bounding the cross terms directly or by treating the sum as a single Hermitian matrix and using a sin-Θ theorem for the top-R invariant subspace of P̃_tot.
  3. [Sec. III-B, final bound] The final chordal-distance bound divides by σ_min(P̃_tot), but no lower bound on this quantity is provided or related to the local gap δ, L, or the overlap pattern. Moreover, P_tot = Σ_l Q^(l)Q^(l)H is a sum of projectors and is not itself an orthogonal projection (unless L=1), so the invocation of [33, Thm. 1.1]—a perturbation bound for orthogonal projections—does not apply as written. Without an explicit spectral-gap assumption on P̃_tot and a statement of the theorem's actual hypotheses, the displayed guarantee can be vacuous.
minor comments (5)
  1. [Algorithm 1, line 4] The typesetting `[S_rl U^(l) S_rc_l]` should read `[S_{r_l}U^{(l)}, S_{r_l^c}]`; as printed it looks like a matrix product rather than a concatenation.
  2. [Notation] Introduce σ_min(·) explicitly, and clarify that P_tot is called an 'aggregate projector' but is not an orthogonal projection matrix in general.
  3. [Sec. IV] The experiments report medians over 20 trials without error bars, and only one system size, one rank, and one SNR are tested. The Ginibre ensemble satisfies the generic assumptions almost surely, so the experiment cannot detect the failure mode illustrated in the major comment above; a structured-state example would be informative.
  4. [Sec. III-E] The shadow-tomography discussion is a short remark with no analysis or experiments. If it is intended as a contribution, more detail is needed; otherwise it should be labeled as outlook.
  5. [Sec. III-D] The measurement-complexity formula M = (R+d)^2 + (L−1)((R+d)^2−R^2) counts real parameters of the union of Hermitian blocks; the text should state this explicitly and define L in terms of D, R, and d for the pattern in Fig. 1.

Circularity Check

0 steps flagged · score 2.0 of 10

Minor self-citation in recovery conditions, but no circular reduction in the derivation.

full rationale

The paper's central recovery claim—col(ρ)=∩_l Q_l and the closed-form computation via the top-R singular vectors of Q_tot—is derived in Section III-B from the isorank assumption and the dimension condition dim(∩_l Q_l)=R. The containment col(ρ)⊆Q_l follows from the EVD of each observed principal submatrix; equality then follows from equal dimensions. This is a genuine derivation, not an assumption renamed as a result. The main unproved premise is the generic intersection condition, stated in Section III-D: "This condition is generically satisfied when some submatrices share at least R overlapping rows ... see [25], [27], [28]." Two of those references ([27], [28]) are prior works by the same group (De Lathauwer and Sofi), which is a self-citation; however, [25] is an independent source, and the paper does not reduce its central claim to that citation alone. The noisy error bound is an application of external perturbation theorems [32], [33], and the measurement-complexity calculation is derived from the proposed structured pattern rather than from a fitted constant. No fitted parameter is later reported as a prediction, and no equation is equal to its input by construction. The skeptic's counterexample (e.g., v∝(1,0,1) with blocks {1,2} and {2,3}) shows that the isorank/generic-intersection conditions can fail for non-generic states; that is a conditional-guarantee robustness issue, not circularity, because the paper explicitly frames the guarantee as generic and does not claim to certify the condition from data. Overall, the derivation is self-contained apart from the cited generic-condition lemma, so the circularity score is low.

Assumptions & free parameters 1 free parameters · 6 assumptions · 0 invented entities

No new physical entities are introduced. The only hand-chosen design parameter is the block-size parameter d that controls the measurement pattern. The main load-bearing assumptions are the known-rank and generic-isorank requirements, plus unquantified spectral-gap/conditioning assumptions in the error analysis.

free parameters (1)
  • Block-size parameter d = d = 1..6 in experiments
    The size of the overlapping measured blocks is chosen by hand as R+d. It is not fitted to make the derivation work, but it controls the measurement-versus-accuracy trade-off and is swept in the experiments.
assumptions (6)
  • domain assumption Density matrices are Hermitian, PSD, unit-trace, and the E^Re/E^Im observables access individual matrix entries.
    Standard QST framework used in Sec. II; no derivation is given.
  • domain assumption Rank R is known a priori.
    Algorithm 1 takes R as input; no rank estimation or model-selection procedure is provided.
  • domain assumption Observed principal submatrices are isorank and the overlap/coverage conditions hold generically.
    Sec. III-B and III-D; for non-generic states a block may drop rank and the algebraic recovery fails.
  • standard math The generic recovery theorem for submatrix completion (dimension of intersection = R) holds as stated in [25,27,28].
    The paper cites these works rather than proving the identifiability condition itself.
  • standard math Perturbation bounds of Wedin [32] and Li et al. [33] apply to the noisy local EVDs.
    Used in Sec. III-B error analysis.
  • domain assumption Positive spectral gap: λ_R(ρ̃_(l)_obs) ≥ δ > ε for all l, and σ_min(P̃_tot) > 0.
    Assumed in the error analysis; no quantitative evidence or verification is provided in the experiments.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Tomography by Design: An Algebraic Approach to Low-Rank Quantum States." pith.science (2026). https://pith.science/paper/PM3MKINW

@misc{pith2026260215202,
  author       = {Pith},
  title        = {Pith review of: Tomography by Design: An Algebraic Approach to Low-Rank Quantum States},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PM3MKINW}},
  note         = {Machine review of arXiv:2602.15202}
}
read the original abstract

We present an algebraic algorithm for quantum state tomography that leverages measurements of certain observables to estimate structured entries of the underlying density matrix. Under low-rank assumptions, the remaining entries can be obtained solely using standard numerical linear algebra operations. The proposed algebraic matrix completion framework applies to a broad class of generic, low-rank mixed quantum states and, compared with state-of-the-art methods, is computationally efficient while providing deterministic recovery guarantees.

Figures

Figures reproduced from arXiv: 2602.15202 by the authors.

Figure 1
Figure 1. Overlapping block diagonal pattern C. From Subspace Identification to Full State Reconstruction After computing a basis matrix U ∈ C D×R, we can recover ρ column-wise. Each column ρ:c can be expressed as ρ:c = Ux. If at least R entries of this column are known, we can generically obtain a unique solution xˆ ∈ C R×1 by solving S ⊤ rl ρ:c = [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. The proposed method yields higher accuracy than both references [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 2 linked inside Pith

  1. [3]

    Quantum state tomography via compressed sensing,

    D. Gross, Y .-K. Liu, S. T. Flammia, S. Becker, and J. Eisert, “Quantum state tomography via compressed sensing,” Phys. Rev. Lett., vol. 105, p. 150 401, 15 Oct. 2010

    2010

  2. [1]

    M. A. Nielsen and I. L. Chuang, Quantum computation and quantum information. Cambridge university press, 2010

    2010

  3. [2]

    Efficient quantum state tomography,

    M. Cramer, M. B. Plenio, S. T. Flammia, et al. , “Efficient quantum state tomography,” Nat. Commun. , vol. 1, no. 1, p. 149, 2010

    2010

  4. [4]

    Experimental quantum state tomography via compressed sampling,

    W.-T. Liu, T. Zhang, J.-Y . Liu, P.-X. Chen, and J.-M. Yuan, “Experimental quantum state tomography via compressed sampling,” Phys. Rev. Lett., vol. 108, p. 170 403, 17 2012

    2012

  5. [5]

    Colloquium: Area laws for the entanglement entropy,

    J. Eisert, M. Cramer, and M. B. Plenio, “Colloquium: Area laws for the entanglement entropy,” Rev. Mod. Phys., vol. 82, no. 1, pp. 277–306, 2010

    2010

  6. [6]

    Universal low-rank matrix recovery from pauli measurements,

    Y .-K. Liu, “Universal low-rank matrix recovery from pauli measurements,” Adv. Neural Inf. Process. Syst., vol. 24, 2011

    2011

  7. [7]

    A nonlinear programming algorithm for solving semidefinite programs via low-rank factorization,

    S. Burer and R. D. C. Monteiro, “A nonlinear programming algorithm for solving semidefinite programs via low-rank factorization,” Math. Program., vol. 95, pp. 329–357, 2003

    2003

  8. [8]

    Provable compressed sensing quantum state tomography via non-convex methods,

    A. Kyrillidis, A. Kalev, D. Park, S. Bhojanapalli, C. Carama- nis, and S. Sanghavi, “Provable compressed sensing quantum state tomography via non-convex methods,” npj Quantum Inf., vol. 4, no. 1, p. 36, 2018

    2018

Show all 35 references
  1. [9]

    Matrix product state representations,

    D. Perez-Garcia, F. Verstraete, M. M. Wolf, and J. I. Cirac, “Matrix product state representations,” Quantum Inf. Comput., vol. 7, pp. 401–430, 2007

    2007

  2. [10]

    Tensor-train decomposition,

    I. Oseledets, “Tensor-train decomposition,” SIAM J. Sci. Com- put., vol. 33, pp. 2295–2317, 2011

    2011

  3. [11]

    Efficient tomography of a quantum many-body system,

    B. P. Lanyon, C. Maier, M. Holz ¨apfel, et al. , “Efficient tomography of a quantum many-body system,” Nat. Phys. , vol. 13, no. 12, pp. 1158–1162, 2017

    2017

  4. [12]

    Tensor train quantum state tomography using compressed sensing,

    S. S. Sofi, C. Vermeylen, and L. De Lathauwer, “Tensor train quantum state tomography using compressed sensing,” in 2025 33rd European Signal Processing Conference (EU- SIPCO), 2025, pp. 1332–1336

    2025

  5. [13]

    Permutationally invariant quantum tomography,

    G. T ´oth, W. Wieczorek, D. Gross, R. Krischek, C. Schwem- mer, and H. Weinfurter, “Permutationally invariant quantum tomography,” Phys. Rev. Lett. , vol. 105, no. 25, p. 250 403, 2010

    2010

  6. [14]

    Permutationally invariant state reconstruction,

    T. Moroder, P. Hyllus, G. T ´oth, et al. , “Permutationally invariant state reconstruction,” New J. Phys. , vol. 14, no. 10, p. 105 001, 2012

    2012

  7. [15]

    Shadow tomography of quantum states,

    S. Aaronson, “Shadow tomography of quantum states,” in Proceedings of the 50th annual ACM SIGACT symposium on theory of computing , 2018, pp. 325–338

    2018

  8. [16]

    Predicting many properties of a quantum system from very few measurements,

    H.-Y . Huang, R. Kueng, and J. Preskill, “Predicting many properties of a quantum system from very few measurements,” Nat. Phys., vol. 16, no. 10, pp. 1050–1057, 2020

    2020

  9. [17]

    Strictly- complete measurements for bounded-rank quantum-state to- mography,

    C. H. Baldwin, I. H. Deutsch, and A. Kalev, “Strictly- complete measurements for bounded-rank quantum-state to- mography,” Phys. Rev. A, vol. 93, no. 5, p. 052 105, 2016

    2016

  10. [18]

    Direct reconstruction of the quantum density matrix by strong measurements,

    L. Calderaro, G. Foletto, D. Dequal, P. Villoresi, and G. Vallone, “Direct reconstruction of the quantum density matrix by strong measurements,” Phys. Rev. Lett. , vol. 121, no. 23, p. 230 501, 2018

    2018

  11. [19]

    Direct measurement of density-matrix elements using a phase-shifting technique,

    T. Feng, C. Ren, and X. Zhou, “Direct measurement of density-matrix elements using a phase-shifting technique,” Phys. Rev. A, vol. 104, no. 4, p. 042 403, 2021

    2021

  12. [20]

    Selective quantum state tomography,

    J. Morris and B. Daki ´c, “Selective quantum state tomography,” arXiv preprint arXiv:1909.05880 , 2019

    1909 arXiv

  13. [21]

    Direct reconstruction of the quantum density matrix elements with classical shadow tomography,

    Y . Wang, “Direct reconstruction of the quantum density matrix elements with classical shadow tomography,” arXiv preprint arXiv:2505.15243, 2025

    2025 arXiv

  14. [22]

    Dropping convexity for faster semi-definite optimization,

    S. Bhojanapalli, A. Kyrillidis, and S. Sanghavi, “Dropping convexity for faster semi-definite optimization,” in Conference on Learning Theory , PMLR, 2016, pp. 530–582

    2016

  15. [23]

    Deterministic symmetric pos- itive semidefinite matrix completion,

    W. E. Bishop and B. M. Yu, “Deterministic symmetric pos- itive semidefinite matrix completion,” in Advances in Neural Information Processing Systems, Z. Ghahramani, M. Welling, C. Cortes, N. Lawrence, and K. Weinberger, Eds., vol. 27, Curran Associates, Inc., 2014

    2014

  16. [24]

    The algebraic combinatorial approach for low-rank matrix completion,

    F. J.Kir ´aly, L. Theran, and R. Tomioka, “The algebraic combinatorial approach for low-rank matrix completion,” J. Mach. Learn. Res. , vol. 16, pp. 1391–1436, 2015

    2015

  17. [25]

    A characterization of deterministic sampling patterns for low- rank matrix completion,

    D. L. Pimentel-Alarc ´on, N. Boston, and R. D. Nowak, “A characterization of deterministic sampling patterns for low- rank matrix completion,” IEEE J. Sel. Top. Signal Process. , vol. 10, pp. 623–636, 2016

    2016

  18. [26]

    Tensor train completion of multi-way data observed along one mode,

    S. S. Sofi, S. Hendrikx, and L. De Lathauwer, “Tensor train completion of multi-way data observed along one mode,” in Proceedings of the 32nd EUSIPCO , 2024, pp. 1067–1071

    2024

  19. [27]

    Multilinear singular value decomposition–based completion with fibers observed in a single mode,

    M. Sørensen, S. Hendrikx, and L. De Lathauwer, “Multilinear singular value decomposition–based completion with fibers observed in a single mode,” SIAM J. Matrix Anal. Appl. , vol. 46, no. 2, pp. 1061–1090, 2025

    2025

  20. [28]

    Tensor train completion from fiberwise observations along a single mode,

    S. S. Sofi and L. De Lathauwer, “Tensor train completion from fiberwise observations along a single mode,” arXiv preprint arXiv:2509.18149, 2025

    2025

  21. [29]

    Positive definite completions of partial hermitian matrices,

    R. Grone, C. R. Johnson, E. M. S ´a, and H. Wolkowicz, “Positive definite completions of partial hermitian matrices,” Linear Algebra Appl. , vol. 58, pp. 109–124, 1984

    1984

  22. [30]

    The positive definite completion problem revis- ited,

    R. L. Smith, “The positive definite completion problem revis- ited,” Linear Algebra Appl. , vol. 429, no. 7, pp. 1442–1452, 2008

    2008

  23. [31]

    Chordal graphs and semidefinite optimization,

    L. Vandenberghe and M. S. Andersen, “Chordal graphs and semidefinite optimization,” Found. Trends Optim. , vol. 1, no. 4, pp. 241–433, 2015

    2015

  24. [32]

    Perturbation bounds in connection with singular value decomposition,

    P. Wedin, “Perturbation bounds in connection with singular value decomposition,” BIT, vol. 12, no. 1, pp. 99–111, Mar. 1972

    1972

  25. [33]

    New bounds for perturbation of the orthogonal projection,

    B. Li, W. Li, and L. Cui, “New bounds for perturbation of the orthogonal projection,” Calcolo, vol. 50, no. 1, pp. 69–78, Mar. 2013

    2013

  26. [34]

    The power of convex relaxation: Near-optimal matrix completion,

    E. J. Cand `es and T. Tao, “The power of convex relaxation: Near-optimal matrix completion,” IEEE Trans. Inf. Theory , vol. 56, no. 5, pp. 2053–2080, 2010

  27. [35]

    Pure-state informationally complete and “re- ally

    J. Finkelstein, “Pure-state informationally complete and “re- ally” complete measurements,” Phys. Rev. A , vol. 70, no. 5, p. 052 107, 2004

    2004

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.