Pith. sign in

REVIEW 3 major objections 5 minor 33 references

Measurement-Based Loss Tolerance in Graph-GKP Codes through Syndrome-Resolved Pauli-Frame Decoding

T0 review · 3 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This paper argues that graph–GKP loss tolerance requires a syndrome-resolved decoder in which low-confidence local GKP blocks become located erasures and rejected syndromes still contribute to the logical Pauli-frame posterior.

desk verdict A genuinely new decoder interface for graph-GKP, but the quantitative loss-tolerance claims outrun the provided numerical evidence. read the letter →

arxiv 2608.00830 v1 pith:LYCVHSKE submitted 2026-08-01 quant-ph

Seid Koudia , Symeon Chatzinotas This is my paper
classification quant-ph
keywords GKPcodesgraphlosstolerancePauli-framedecodingsyndrome-resolvedmeasurement-basedquantumcomputationfusion-basedphotonicrepeaters
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 tries to establish a single causal statistical interface between GKP recovery and graph-level control for photonic measurement-based quantum computation. It claims that the availability of a graph node is not an independent random deletion but is generated by thresholding the analog GKP record, so both accepted and rejected local decisions must appear in the global likelihood. The correct output of a graph–GKP module is therefore a posterior over the outgoing logical Pauli frame, not a hard decoded bit. If true, this gives a common interface for logical Pauli measurements, non-Pauli transport, recursive concatenation, and fusion, with finite-size pseudothresholds for square and hexagonal lattices.

What carries the argument

The selected-event closest-coset decoder: a covariance-weighted closest-coset approximation to the exact wrapped logical-coset maximum-likelihood target, evaluated on physical displacement sectors C_{σ,e} = τ_{σ,e} + L_G modulo the invisible sublattice K_b^lat. It carries the argument because the branch/availability event enters the likelihood as a deterministic function of the stored local records, and the resulting normalized sector posterior Π_b(σ,e|D_b) is pushed through the branch Pauli-transfer map τ_b to yield the controller-facing frame posterior Π_b^fr(f|D_b).

What would settle it

Run the same cube and decorated-pentagon benchmarks with the rejected-syndrome factors in Eq. (46) replaced by flag-only factors and with the availability mask drawn independently of the analog records; if accepted-frame error and fusion success do not worsen at the quoted attenuation-reference crossings, the central claim is not supported.

Watch

Extended reading notes

Core claim

The paper's central claim is that the availability mask and graph branch are deterministic functions of the local analog GKP records, expressed by b = B_Γ(Dloc_{1:n}). Consequently, the selected-event likelihood must include the probability of both accepted and rejected local decisions, and the normalized posterior over the outgoing logical Pauli frame, built by pushing the syndrome-resolved sector posterior through the branch's Pauli-transfer map, is the correct object for a controller. The paper derives this selected-event closest-coset decoder, shows that local abstention, accessibility failure, confidence failure, and accepted frame error are distinct events, and applies the same engine

Load-bearing premise

The physical channel after loss and gain compensation is a single additive Gaussian displacement channel, and local GKP recovery is a nondestructive flagged instrument that returns both a refreshed block and a usable syndrome.

Editorial extensions

If this is right

  • A graph–GKP module can export a full Pauli-frame posterior and confidence, allowing soft propagation of frame uncertainty through MBQC instead of a hard decision after every module.
  • Local abstention, accessibility failure, confidence failure, and accepted frame error are separated and not double-counted; raising the confidence threshold trades accepted Pauli error for located erasures that graph redundancy can route around.
  • The same selected-event closest-coset engine applies to logical Pauli measurements, protected A(θ) transport, recursive concatenation, and parity fusion, giving architectures one decoder interface.
  • Finite-size benchmarks show graph topology has a substantial effect on the loss-tolerance boundary, while the hexagonal lattice generally lowers logical failure through its larger shortest logical displacement.
  • Recursive concatenation should propagate posterior channels, not scalar erasure probabilities; scalar accessibility polynomials are only algebraic checks.

Reading between the lines

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

  • Editorial inference: a direct decoder ablation, replacing the accepted/rejected local likelihood factors with hard erasure flags and treating the availability mask as independent, would quantify the practical gain of keeping rejected syndromes; the paper identifies this as a next step but does not run it.
  • Editorial inference: the omitted multiplicity factor R_b(σ,e;D_b) in the closest-coset approximation could be used to construct certified bounds on how far the max-log posterior is from the exact wrapped posterior, which would make small-module threshold claims stronger.
  • Editorial inference: the posterior-interface view suggests network-level confidence-aware routing in all-photonic repeaters, where a graph fragment's local GKP records inform whether to fuse, reroute, store, or discard it.
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

3 major / 5 minor

Summary. The paper develops a causal framework for measurement-based loss tolerance in graph–GKP codes. Local GKP recovery is modeled as a flagged instrument that outputs a refreshed block, a continuous syndrome record, and a confidence score; low-confidence outcomes are converted into located erasures. The availability mask and graph branch are treated as deterministic functions of the same analog records used for error inference, so both accepted and rejected local decisions enter the selected-event likelihood (Eq. 43). The paper derives branch-compatible lattice compilation (Theorem III.1), syndrome-resolved closest-coset decoding, Pauli-frame pushforwards, recursive concatenation, and logical parity fusion (Theorem VII.1). Numerical simulations report finite-size pseudothresholds for square and hexagonal lattices and for adaptive/transversal fusion, with the caveat that all executable results use the closest-coset approximation rather than the exact wrapped likelihood.

Significance. If the framework is correct, it provides a useful unified statistical interface between continuous-variable GKP information and discrete graph-code control, which is relevant for MBQC, fusion-based computation, and photonic repeaters. The algebraic core is carefully presented, with Theorem III.1 and Theorem VII.1 proved and the accessibility polynomials (65) and (69) explicitly enumerated and checked at their fixed points. The paper is transparent about what is and is not included in the benchmarks. However, the quantitative loss-tolerance claims currently rest on an unvalidated closest-coset approximation and thin numerical support: no error bars, no Monte Carlo sample counts, and no code/data release. These gaps do not undermine the formal framework, but they prevent acceptance of the reported numerical thresholds as established results.

major comments (3)
  1. [Sec. V C, Eq. (51)] All executable loss-tolerance results (Figs. 4–9, 12–13 and the quoted intervals in Secs. V E, VI C–D, VII B) use the closest-coset score, not the exact wrapped likelihood. Equation (51) shows that the exact score differs from the CC score by a sector-dependent factor R_b(σ,e;D_b)≥1, and the text states that exact-vs-CC comparison is future work. Since thresholds, erasure decisions, and frame posteriors are calibrated within the CC model, the quantitative pseudothresholds are not yet supported. A small-module exact-vs-CC comparison, or at least a bound on the variation of R_b and the resulting posterior bias, is required before these numbers can be accepted.
  2. [Sec. V E, Eqs. (74)–(76)] The benchmark curves have no error bars, no Monte Carlo sample counts, and no code/data release. The statement near Eq. (54) that frequentist accepted-frame errors are used to calibrate the confidence threshold is not backed by any validation data. Without these, the reported ℓ_ref and ℓ_br intervals and the square/hexagonal differences cannot be distinguished from statistical or systematic noise. Please provide reproducible code/data and convergence diagnostics, including the number of samples per point and confidence intervals.
  3. [Sec. VII B, Figs. 12–13] The adaptive-versus-transversal fusion comparison is not normalized by expected Bell-attempt count or other resource consumption. The paper acknowledges this, but the central practical message of an 'adaptive advantage' is based solely on the half-success attenuation ℓ_1/2 under different attempt schedules. Because the adaptive policy can consume a variable number of Bell attempts and interface blocks, the comparison should at least report the expected number of attempts/resources at each operating point, or be explicitly labeled as a policy-level illustration rather than an architectural advantage.
minor comments (5)
  1. [Sec. VI C c] Typo: 'speudothresholds' should be 'pseudothresholds'.
  2. [Eqs. (75)–(76)] The text uses 'pseudothreshold' for an attenuation-reference crossing. Since ℓ is a transmissivity deficit rather than a Pauli-error probability, 'reference crossing' is the more precise term; consider using it consistently to avoid confusion with fault-tolerance thresholds.
  3. [Figs. 6 and 7] Many curves are plotted without distinct markers, making them hard to distinguish at print size. Adding markers or separating into per-graph panels would improve readability.
  4. [Sec. II C] The notation for records D and subspaces C_i is occasionally left implicit. A short table of symbols for D_loc, D_out, C_b, R_b, and the branches would help the reader navigate the causal structure.
  5. [General] There is no data-availability or code-repository statement. Given the prominent numerical claims, a reproducibility statement is strongly recommended.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the derivation chain is self-contained and the benchmark markers are explicitly defined reference crossings, not fitted predictions.

full rationale

The paper's central claim is a formal statistical interface: local GKP recovery records determine the availability mask and branch, and the selected-event likelihood in Eq. (43) correctly conditions on both accepted and rejected local decisions. This is a causal modeling identity derived from the declared Gaussian displacement channel, not a prediction obtained from fitted parameters. The closest-coset approximation in Eq. (50) is explicitly identified as the sole sector-score approximation, with the exact wrapped likelihood retained as the target; the paper states that reported posterior odds are probabilities within the declared closest-coset model and are not asserted to equal exact maximum-likelihood posteriors. No load-bearing step relies on a self-citation: the cited works [11,18,19,31,32] are background for repeater, bosonic, and multimode contexts, and the channel composition, GKP decoding targets, and graph-code facts are standard results from external references. The benchmark markers in Eqs. (75)-(76) are self-consistency definitions, but the paper explicitly disclaims them: Eq. (75) is called 'a reference crossing, not a fault-tolerance threshold,' and Eq. (76) 'compares only the propagation-induced increments.' These are conventions for interpreting finite-size numerics, not a disguised reuse of the input as the output. The main numerical limitation, the unvalidated closest-coset approximation, is disclosed rather than concealed, and the paper lists an exact-versus-CC comparison as future work. Therefore the derivation chain is self-contained and no circular step was found.

Assumptions & free parameters 5 free parameters · 5 assumptions · 1 invented entities

The paper introduces no free parameters fitted to data to make the central framework work; the physical and decoder parameters (squeezing, loss, confidence thresholds) are declared inputs of the model. The main postulates are the Gaussian displacement channel, the flagged GKP recovery instrument, the static-lattice periodicity, and the closest-coset approximation. The conceptual erasure flags are operational, not physical. The thresholds Gamma_loc and Gamma_fr_th are the nearest thing to hand-chosen knobs, and they are not reported as values, which weakens reproducibility.

free parameters (5)
  • local confidence threshold Gamma_loc
    Decoder parameter in the flagged GKP recovery instrument (Eq. 21); chosen by hand to trade accepted Pauli error against located erasure. All numerical panels depend on it, but its value is not reported as tuned.
  • frame confidence threshold Gamma_fr_th
    Threshold for declaring bottom_conf in Algorithm 3; selected by hand and not reported.
  • fusion parity confidence threshold Delta_th = 0.50
    Joint acceptance rule for adaptive fusion in Sec. VII B; chosen as 0.50 for the simulations.
  • closest-coset sector normalizer Z_CC_{b,sigma,e}
    Sector-dependent normalizer for non-translation-covariant instruments, mentioned in Sec. V C but never computed; any numerical use would require calibration.
  • Tie-breaking rule and Monte Carlo estimator details
    The closest-vector solver uses a deterministic tie rule and the frame error is measured in Monte Carlo (Sec. V C), but neither the tie rule nor sample counts are specified, leaving a hidden degree of freedom in the reported curves.
assumptions (5)
  • domain assumption The physical channel is an additive Gaussian displacement channel with covariance Sigma_eps = Sigma_prep + Sigma_LA + ... (Eq. 16), where pure loss followed by quantum-limited amplification gives Sigma_LA(i) = (1-eta_i)/(2*pi*eta_i) I.
    Sec. II C. Load-bearing model for all simulations and likelihoods; real loss or finite-squeezing preparation may deviate.
  • domain assumption Local GKP recovery is a trace-preserving flagged instrument (Eq. 22) producing an accepted refreshed block or a deliberate erasure, with the rejected syndrome retained in the likelihood.
    Sec. II C. The framework's 'causal' availability pattern depends on this instrument; no specific recovery circuit is designed or simulated.
  • domain assumption Physical errors remain periodic under the static compiled lattice L_G; the branch lattice L_G,b is used only for observable reconstruction.
    Sec. III C after Eq. (36). This protects the sector decomposition Eq. (46) from branch-dependent changes; if the recovery backaction changed the periodicity, the quotient decoding would need revision.
  • domain assumption Closest-coset (max-log) approximation is a faithful proxy for the exact wrapped maximum-likelihood decoder; all numerical results use it.
    Sec. V C, Eq. (50). The paper notes the approximation is not exact and calls for ablation, but every benchmark rests on it.
  • standard math Standard graph-code progenitor facts: the fiber construction gives Q_G isotropic, logical classes are affine cosets, and the displacement embedding Phi is faithful.
    Sec. III A, Appendix A. Classical stabilizer code theory; correctly applied.
invented entities (1)
  • Located-decoder-erasure flags (bottom_loc, bottom_acc, bottom_conf)
    purpose: Operational labels for the three causal failure modes in the selected-event decoder; they are classical bookkeeping devices, not new physical objects.
    These are introduced as conceptual shorthand throughout Secs. III-VIII. They have no falsifiable handle outside the paper's own decoder model, but they are not claimed to be physical entities.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Measurement-Based Loss Tolerance in Graph-GKP Codes through Syndrome-Resolved Pauli-Frame Decoding." pith.science (2026). https://pith.science/paper/LYCVHSKE

@misc{pith2026260800830,
  author       = {Pith},
  title        = {Pith review of: Measurement-Based Loss Tolerance in Graph-GKP Codes through Syndrome-Resolved Pauli-Frame Decoding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LYCVHSKE}},
  note         = {Machine review of arXiv:2608.00830}
}
read the original abstract

Graph codes offer multiple physical representatives of logical observables, while Gottesman-Kitaev-Preskill (GKP) codes retain analog information about bosonic displacement noise. We develop a causal framework that unifies these mechanisms for measurement-based loss tolerance under pure loss followed by quantum-limited amplification. In this framework, each local GKP recovery produces a refreshed logical block, a continuous syndrome record, and a confidence score for the inferred Pauli class. Low-confidence outcomes are deliberately converted into located erasures, so the availability pattern is generated directly from the bosonic data rather than sampled independently. Both accepted and rejected syndromes contribute to a syndrome-resolved posterior over the graph branch, which determines accessible logical representatives and the outgoing logical Pauli frame. We derive decoder-conditioned branch restriction, signed-outcome reconstruction, Pauli-frame updating, recursive concatenation of graph-GKP modules, and syndrome-resolved logical fusion. Numerical simulations across several squeezing levels identify task-dependent loss-tolerance behavior and finite-depth pseudothresholds for square and hexagonal GKP lattices. The resulting graph-GKP interface provides a unified causal control layer for fault-tolerant MBQC, fusion-based computation, and all-photonic repeaters.

Figures

Figures reproduced from arXiv: 2608.00830 by the authors.

Figure 1
Figure 1. FIG. 1. Graph–GKP decoding pipeline. Pure loss followed by ideal gain compensation produces one additive Gaussian [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Decoder-conditioned branch restriction and non-Pauli terminal interface. (a) A locally ambiguous block is converted to [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Observable compatibility is not physical-error compatibility. The branch word [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Fixed-progenitor closest-coset benchmarks for square GKP blocks. Panels (a)–(d) use the seven-block cube code for [PITH_FULL_IMAGE:figures/full_fig_p019_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. Representative graph-code families used in the graph-family simulations of Figs. [PITH_FULL_IMAGE:figures/full_fig_p019_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6. Graph-family excess-failure benchmarks for hexagonal-lattice GKP blocks decoded with the closest-coset decoder. The [PITH_FULL_IMAGE:figures/full_fig_p020_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7. Square-lattice counterpart of Fig. [PITH_FULL_IMAGE:figures/full_fig_p020_7.png]
Figure 8
Figure 8. Figure 8: FIG. 8. Finite-depth recursive closest-coset loss sweeps for hexagonal GKP blocks. Columns use [PITH_FULL_IMAGE:figures/full_fig_p022_8.png]
Figure 9
Figure 9. Figure 9: FIG. 9. Square-lattice counterpart of the finite-depth recursion study in Fig. [PITH_FULL_IMAGE:figures/full_fig_p022_9.png]
Figure 10
Figure 10. Figure 10: FIG. 10. Fusion-sector logic. The commuting measured parity generators define the Bell parity subgroup. Outcome sectors are [PITH_FULL_IMAGE:figures/full_fig_p024_10.png]
Figure 11
Figure 11. Figure 11: FIG. 11. Transversal and adaptive logical-fusion policies. All [PITH_FULL_IMAGE:figures/full_fig_p025_11.png]
Figure 12
Figure 12. Figure 12: FIG. 12. Logical CV/GKP fusion success with closest-coset parity decoding for hexagonal GKP blocks at parity-confidence [PITH_FULL_IMAGE:figures/full_fig_p026_12.png]
Figure 13
Figure 13. Figure 13: FIG. 13. Square-lattice counterpart of the logical-fusion benchmark in Fig. [PITH_FULL_IMAGE:figures/full_fig_p026_13.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

33 extracted references · 26 canonical work pages

  1. [1]

    It remains to verify consistency of the phase assignment

    The decomposition is therefore unique. It remains to verify consistency of the phase assignment. For two graph labelss, t∈QG, define A= Φ(s), B= Φ(t), w=s⊕t, W= Φ(w), and let c=A+B−W∈L loc. Closure of the signed graph stabilizer, together with Eq. (B4), gives πϵ(w) =πϵ(s) +πϵ(t)−πA T JlocB +πW T Jlocc−ϕ loc(c) (mod 2π).(B11) UsingA+B=c+Win Eq. (B9), ϕG(A+...

    work page

  2. [2]

    Vischi, G

    M. Vischi, G. Di Bartolomeo, M. Proietti, S. Koudia, F. Cerocchi, M. Dispenza,et al., Physical Review Re- search6, 033337 (2024)

    work page 2024

  3. [3]

    F. Hoch, E. Caruccio, G. Rodari, T. Francalanci, A. Suprano, T. Giordani,et al., Nature Communications 16, 902 (2025)

    work page 2025

  4. [4]

    Eisert and J

    J. Eisert and J. Preskill, Mind the gaps: The fraught road to quantum advantage (2025), arXiv:2510.19928

    arXiv 2025

  5. [5]

    P. Zeng, G. Zheng, Q. Xu, and L. Jiang, Error-structure- tailored early fault-tolerant quantum computing (2025), arXiv:2511.19983

    arXiv 2025

  6. [6]

    Raussendorf and H

    R. Raussendorf and H. J. Briegel, Physical Review Letters 86, 5188 (2001)

    work page 2001

  7. [7]

    Raussendorf, D

    R. Raussendorf, D. E. Browne, and H. J. Briegel, Physical Review A68, 022312 (2003)

    work page 2003

  8. [8]

    Varnava, D

    M. Varnava, D. E. Browne, and T. Rudolph, Physical Review Letters97, 120501 (2006)

    work page 2006

Show all 33 references
  1. [9]

    Morley-Short, M

    S. Morley-Short, M. Gimeno-Segovia, T. Rudolph, and H. Cable, Quantum Science and Technology4, 025014 (2019)

    2019

  2. [10]

    T. J. Bell, L. A. Pettersson, and S. Paesani, PRX Quan- tum4, 020328 (2023)

    2023

  3. [11]

    B. Li, K. Goodenough, F. Rozpędek, and L. Jiang, Physi- cal Review Letters134, 190801 (2025)

    2025

  4. [12]

    Koudia, Physica Scripta99, 015115 (2024)

    S. Koudia, Physica Scripta99, 015115 (2024)

    2024

  5. [13]

    Gottesman, A

    D. Gottesman, A. Kitaev, and J. Preskill, Physical Review A64, 012310 (2001)

    2001

  6. [14]

    M. V. Larsen, J. E. Bourassa, S. Kocsis, J. F. Tasker, R. S. Chadwick, C. González-Arciniegas, J. Hastrup, C. E. Lopetegui-González, F. M. Miatto, A. Motamedi,et al., Nature642, 587 (2025)

    2025

  7. [15]

    Conrad, J

    J. Conrad, J. Eisert, and F. Arzani, Quantum6, 648 (2022)

    2022

  8. [16]

    M. Lin, C. Chamberland, and K. Noh, PRX Quantum4, 040334 (2023)

    2023

  9. [17]

    Zheng, W

    G. Zheng, W. He, G. Lee, K. Noh, and L. Jiang, PRX Quantum6, 030314 (2025)

    2025

  10. [18]

    Wang and L

    Z. Wang and L. Jiang, Physical Review X15, 021003 (2025)

    2025

  11. [19]

    Koudia and A

    S. Koudia and A. Gharbi, International Journal of Quan- tum Information19, 2150026 (2021)

    2021

  12. [20]

    Koudia and A

    S. Koudia and A. Gharbi, International Journal of Quan- tum Information17, 1950055 (2019)

    2019

  13. [21]

    K. Noh, V. V. Albert, and L. Jiang, IEEE Transactions on Information Theory65, 2563 (2019)

    2019

  14. [22]

    Hastrup and U

    J. Hastrup and U. L. Andersen, Physical Review A108, 052413 (2023)

    2023

  15. [24]

    Gottesman,Stabilizer Codes and Quantum Error Cor- rection, Ph.D

    D. Gottesman,Stabilizer Codes and Quantum Error Cor- rection, Ph.D. thesis, California Institute of Technology (1997), arXiv:quant-ph/9705052

    1997 arXiv

  16. [25]

    Schlingemann and R

    D. Schlingemann and R. F. Werner, Physical Review A 65, 012308 (2001)

    2001

  17. [26]

    Enrico Fermi

    M. Hein, W. Dür, J. Eisert, R. Raussendorf, M. Van den Nest, and H. J. Briegel, inQuantum Computers, Algo- rithms and Chaos, Proceedings of the International School of Physics “Enrico Fermi”, Vol. 162 (IOS Press, 2006) pp. 115–218

    2006

  18. [27]

    Bartolucci, P

    S. Bartolucci, P. Birchall, H. Bombin, H. Cable, C. Daw- son, M. Gimeno-Segovia, E. Johnston, K. Kieling, N. Nick- erson, M. Pant,et al., Nature Communications14, 912 (2023)

    2023

  19. [28]

    W. Song, N. Kang, Y.-S. Kim, and S.-W. Lee, Physical Review Letters133, 050605 (2024)

    2024

  20. [29]

    Fukui, R

    K. Fukui, R. N. Alexander, and P. van Loock, Physical Review Research3, 033118 (2021)

    2021

  21. [30]

    Rozpędek, K

    F. Rozpędek, K. P. Seshadreesan, P. Polakos, L. Jiang, and S. Guha, Physical Review Research5, 043056 (2023)

    2023

  22. [31]

    S. N. Swain and T. C. Ralph, arXiv preprint arXiv:2604.09002 (2026)

    2026 arXiv

  23. [32]

    Koudia, L

    S. Koudia, L. Oleynik, J. Ur Rehman, and S. Chatzinotas, Communications Physics8, 351 (2025)

    2025

  24. [33]

    Koudia and S

    S. Koudia and S. Chatzinotas, npj Quantum Information 11, 162 (2025)

    2025

  25. [34]

    Dehaene and B

    J. Dehaene and B. De Moor, Physical Review A68, 042318 (2003)

    2003

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.