Pith. sign in

REVIEW 2 major objections 6 minor 17 references

(2,2)-GB Codes: Classification and Comparison with weight-4 Surface Codes

T0 review · 2 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper constructs three infinite families of (2,2)-generalized bicycle codes that reach optimal surface-code parameters, including the first optimal even-distance family previously claimed impossible.

desk verdict New even-distance (2,2)-GB family closes a real gap; the main distance proof has a fixable hole at Corollary III.3. read the letter →

arxiv 2507.21237 v1 pith:3COJBQP5 submitted 2025-07-28 cs.IT math.ITquant-ph

François Arnault , Philippe Gaborit , Nicolas Saussay This is my paper · ORCID
classification cs.ITmath.ITquant-ph MSC 81P7005C2594B65
keywords generalizedbicyclecodesquantumerrorcorrectionCSSCayleygraphssurfaceminimumdistancelatticelowerboundcodeequivalence
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 establishes that carefully chosen (2,2)-generalized bicycle (GB) codes, quantum CSS codes built from two circulant matrices with two ones per row, achieve the same optimal parameters as the best 2D weight-4 surface codes. The key tool is a lower bound showing that the minimum distance of $\mathrm{GB}(1+X,1+X^{\alpha},n)$ is at least the shortest Manhattan norm of a non-zero vector in an associated 2D lattice. Using that bound, the authors build three infinite families with parameters $[[2n^2,2,n]]$, $[[4r^2,2,2r]]$, and $[[(2t+1)^2+1,2,2t+1]]$. The even-distance family $[[4r^2,2,2r]]$ closes a gap that earlier work had declared impossible for GB codes. They also introduce a CSS-preserving equivalence relation to prove that two of the families are genuinely new relative to known toric and rotated surface codes, while the third is an alternative realization of the optimal odd-distance surface code.

What carries the argument

The load-bearing object is the lattice $L = \mathbb{Z}(n,0)+\mathbb{Z}(\alpha,-1)$ associated to $\mathrm{GB}(1+X,1+X^{\alpha},n)$, together with $\lambda(L)$, the minimal Manhattan norm over non-zero lattice vectors. Theorem II.7 states $d \geq \lambda(L)$; the proof lifts simple cycles of the quotient Cayley graph to $\mathbb{Z}^2$-walks and shows that each nontrivial cycle gives a non-zero lattice endpoint. This turns code design into a lattice-geometry problem: choose $(n,\alpha)$ so that every short $\mathbb{Z}^2$-path fails to close modulo $(n,\alpha)$. A second piece of machinery is the CSS graph-preserving (CGP) equivalence relation, a permutation-only equivalence that preserves both the CSS form and the underlying Cayley graph, used with a classical 2-isomorphism-to-isomorphism theorem to prove structural distinctness.

What would settle it

Take a concrete pair $(n,\alpha)$, for example $n=8,\alpha=3$, exhaustively compute the minimum distance of $\mathrm{GB}(1+X,1+X^{\alpha},n)$, and compare it to $\lambda(\mathbb{Z}(n,0)+\mathbb{Z}(\alpha,-1))$; any instance with $d < \lambda$ would falsify Theorem II.7. A more targeted test is to enumerate simple cycles of $(\mathbb{Z}/n\mathbb{Z},1,\alpha)$ with zero net displacement and check whether any is not a sum of faces, which would directly falsify Corollary III.3.

Watch

Extended reading notes

Core claim

The central discovery is that the minimum distance of a (2,2)-GB code can be read off from a lattice: for the code $\mathrm{GB}(1+X,1+X^{\alpha},n)$, with $n\geq 6$ and $1\leq \alpha\leq n-1$, the distance is at least $\lambda(L)$, the smallest Manhattan length of a non-zero vector in $L = \mathbb{Z}(n,0) + \mathbb{Z}(\alpha,-1)$. The proof maps every cycle in the Cayley graph $(\mathbb{Z}/n\mathbb{Z},1,\alpha)$ to a walk in $\mathbb{Z}^2$ whose endpoint lies in $L$ and whose Manhattan norm is no larger than the cycle length; a cycle that is not a sum of faces must have non-zero endpoint. With this bound in hand, the authors select lattices with large Manhattan minimum to obtain explicit optimal codes. In particular $\mathrm{GB}(1+X,1+X^{2r-1},2r^2)$ has parameters $[[4r^2,2,2r]]$, contradicting the earlier claim that optimal even-distance GB codes could not exist, and the families are distinguished from known surface codes by a graph-based CSS-preserving equivalence relation.

Load-bearing premise

The argument assumes that a simple loop in the cyclic graph with equal counts of forward and backward steps of each type unwraps to a simple loop in the integer grid; the paper states this without proof, and the lower bound depends on it.

Editorial extensions

If this is right

  • The family $\mathrm{GB}(1+X,1+X^{2r-1},2r^2)$ realizes optimal even-distance parameters $[[4r^2,2,2r]]$ for every $r$, so even distances are no longer a gap for (2,2)-GB codes.
  • The family $\mathrm{GB}(1+X,1+X^n,n^2)$ matches the toric code parameters $[[2n^2,2,n]]$ while being inequivalent to the standard toric code under the paper's CGP-equivalence.
  • The odd-distance family $\mathrm{GB}(1+X,1+X^{2t+1},t^2+(t+1)^2)$ is CGP-equivalent to the best-known odd-distance 2D surface code, giving an alternative construction of the same code.
  • The lattice criterion $d \geq \lambda(L)$ gives a concrete design rule: any pair $(n,\alpha)$ whose lattice $\mathbb{Z}(n,0)+\mathbb{Z}(\alpha,-1)$ has large Manhattan minimum yields a high-distance (2,2)-GB code.
  • The classification tables list all extremal non-equivalent (2,2)-GB codes of length below 200, with representatives and counts, providing a benchmark for future decoder and implementation studies.

Reading between the lines

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

  • If Theorem II.7 extends to parameters outside the stated range, the search for optimal (2,2)-GB codes becomes a purely number-theoretic optimization over $(n,\alpha)$; checking this computationally for all lengths below 200 would be a direct test.
  • The same cycle-lifting argument may generalize to (a,b)-GB codes with $a,b>2$ by passing to higher-dimensional lattices, potentially producing optimal families beyond weight 4.
  • The CGP-equivalence framework suggests that the relevant invariant for comparing Cayley-graph CSS codes is the isomorphism class of the underlying abelian group; if true, classification of equivalence classes reduces to a group-theory problem.
  • The even-distance result may prompt re-examination of other 'impossible' parameter claims for GB codes, since the mechanism that broke the even-distance barrier is a lattice choice rather than a change of code family.
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 / 6 minor

Summary. This paper studies (2,2)-Generalized Bicycle (GB) codes, CSS codes built from pairs of binary circulant matrices with two nonzero entries per row and viewed as Cayley graphs (Z/nZ,a,b). The main theoretical result is Theorem II.7, a lower bound d_min ≥ λ(L) on the minimum distance of GB(1+X,1+X^α,n), where L is the lattice Z(n,0)+Z(α,-1) and λ(L) is the shortest Manhattan norm of a nonzero lattice vector. Using this bound, the authors construct three infinite families with optimal parameters [[2n^2,2,n]], [[4r^2,2,2r]], and [[(2t+1)^2+1,2,2t+1]]; the even-distance family is claimed to be the first optimal even-distance (2,2)-GB construction. They then introduce a CSS-preserving 'CGP-equivalence' relation for comparing Cayley-graph-based CSS codes, prove non-CGP-equivalence of the first two families to the standard Kitaev and rotated even-distance surface codes, prove equivalence of the third family to the optimal odd-distance surface code, and provide a computational classification of extremal (2,2)-GB codes with length below 200.

Significance. If the proof gaps identified below are closed, the paper is a solid and useful contribution. The lattice lower bound is elegant and parameter-free, the three distance computations are explicit and checkable, and the non-equivalence arguments via Whitney's theorem are substantial. The construction of optimal even-distance (2,2)-GB codes addresses a real gap in the literature relative to [13]. The paper also ships a classification repository [17], which supports reproducibility of the tables. The constructions involve no fitted constants and the lower-bound lattice vectors are computed directly, so I see no circularity.

major comments (2)
  1. [III.C-4 / Corollary III.3] The proof of Corollary III.3 asserts without argument that if C is a simple cycle of (Z/nZ,1,α) with n_1(C)=n_{-1}(C) and n_α(C)=n_{-α}(C), then the associated Z^2-walk γ_C starting at the origin is also a simple cycle. This is the load-bearing step of the proof of Theorem II.7: without simplicity, Lemma III.2 cannot be applied, and the endpoint P_r could be zero without forcing C to be a sum of faces, so the lower bound d≥λ(L) would not follow. The assertion is very likely correct (if P_i=P_j, then C_i=C_0+Φ(P_i)=C_j, contradicting simplicity of C), but the argument is absent. Please add this proof explicitly.
  2. [III.D / Lemma III.2] The induction proving Lemma III.2 is only a sketch. The step 'within Int(Γ), at least one of these two scenarios is true' does not formally establish the existence of a square S whose removal leaves a simple Z^2-cycle γ through the origin surrounding exactly q squares; the cases of S sharing one or two edges with Γ, and the handling of squares incident with the origin, need a rigorous case analysis. Because Lemma III.2 is used in Corollary III.3 and hence in Theorem II.7, this gap should be closed (e.g., by a standard cell-decomposition argument) before the lower-bound result is considered proved.
minor comments (6)
  1. [III.C] The step-type list contains a typo: 'k+α → α' should be 'k+α → k' (or 'k → k−α').
  2. [IV.B / Lemma IV.3] The displayed inequality '2r|rx+y| − |y| |+|y|' has a typo and should read '2r|rx+y|-|y|+|y|'.
  3. [Abstract and Section VI] The claim that the first two families are inequivalent to 'all previously known optimal weight-4 2D surface codes' is broader than what is proved; the proofs cover the periodic-lattice surface codes of [4] and the Kitaev toric code. Please qualify the claim to the class actually treated.
  4. [III, proof of Theorem II.7] The reduction to 1≤α≤n/2 via Proposition II.6 is stated but not demonstrated; a short derivation using the invariance under X→X^{-1} and column shifts should be included.
  5. [IV.A and IV.C] The small cases n=2 and t=1 in Lemmas IV.2 and IV.4 are verified 'by computer simulations'; please specify the exact method (e.g., exhaustive enumeration of codewords or the specific SageMath/Magma function) so the reader can reproduce those checks.
  6. [Throughout] There are numerous formatting artifacts in the equations (e.g., 'X n2', '[[4r2,2,2r|', and missing superscripts in the introduction); a careful proofreading pass is needed.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the minimum-distance bound and the three optimal families are derived parameter-free with explicit upper-bound codewords; the only questionable step (Corollary III.3) is an unproved lemma, not a circular reduction.

full rationale

The paper's central derivation is self-contained. Theorem II.7 lower-bounds the minimum distance by lambda(L), the shortest Manhattan norm in the explicitly defined lattice L = Z(n,0) + Z(alpha,-1); no fitted constant or externally imported distance value is used. The three families in Proposition IV.1 are each verified from both sides: the lower bound is obtained by computing lambda(L) directly, and the upper bound is an explicit low-weight codeword, such as the weight-n vector (0, sum_k X^{nk}) in Lemma IV.2, the weight-2r vector in Lemma IV.3, and the weight-(2t+1) pair (U,V) in Lemma IV.4. Thus the equalities d=n, d=2r, and d=2t+1 do not reduce to their own inputs; they are exact matches between a proven bound and an exhibited codeword. The claim that even-distance optimal GB codes were previously considered impossible is attributed to Pryadko and Wang [13] and contradicted by an explicit construction, so it is not a self-citation chain. The authors cite their own GitHub repository [17], but only as a supplement to the classification tables, not as evidence for any theorem. The one genuinely delicate step is Corollary III.3, which asserts without proof that a balanced simple quotient cycle lifts to a simple grid cycle; this is a proof gap that could invalidate Theorem II.7 if false, but it is not a circular step: the asserted implication is not equivalent to the theorem's conclusion, and the paper does not define the conclusion in terms of the premise. For the same reason, the skeptical observation about the missing lattice-lift proof is a correctness risk to be resolved by an explicit argument, not evidence that the derivation is circular.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The paper relies on standard algebraic and graph-theoretic facts (circulant polynomial algebra, Cayley graph incidence matrices, Whitney's theorem). One load-bearing assumption is that balanced simple torus cycles lift to simple lattice cycles, asserted in Corollary III.3 without proof. No fitted parameters or invented entities are introduced.

assumptions (3)
  • standard math Whitney's theorem: a 2-isomorphism between 3-connected graphs implies graph isomorphism (Theorem VI.3).
    Used in Section VI.B to reduce CGP-equivalence to graph isomorphism.
  • domain assumption The quotient map from Z^2 lattice walks to the (Z/nZ,1,alpha) Cayley graph is a covering map; simple cycles lift to simple lattice cycles (Corollary III.3).
    The claim that balanced step counts imply a simple Z^2-cycle is asserted without proof in Section III-D; it is load-bearing for Theorem II.7.
  • domain assumption CSS code minimum distance equals the shortest non-trivial cycle not in the face span for these Cayley graph codes (Section III.A).
    Standard for homological CSS codes; used implicitly to translate the graph-theoretic cycle bound to d_min.

how reviews work

0 comments
Cite this review

Pith. "Pith review of (2,2)-GB Codes: Classification and Comparison with weight-4 Surface Codes." pith.science (2026). https://pith.science/paper/3COJBQP5

@misc{pith2026250721237,
  author       = {Pith},
  title        = {Pith review of: (2,2)-GB Codes: Classification and Comparison with weight-4 Surface Codes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3COJBQP5}},
  note         = {Machine review of arXiv:2507.21237}
}
read the original abstract

Generalized Bicycle (GB) codes offer a compelling alternative to surface codes for quantum error correction. This paper focuses on (2,2)-Generalized Bicycle codes, constructed from pairs of binary circulant matrices with two non-zero elements per row. Leveraging a lower bound on their minimum distance, we construct three novel infinite families of optimal (2,2)-GB codes with parameters [[ 2n^2, 2, n ]], [[ 4r^2, 2, 2r ]], and [[(2t + 1)^2 + 1, 2, 2t + 1 ]]. These families match the performance of Kitaev's toric code and the best 2D weight-4 surface codes, reaching known theoretical limits. In particular, the second family breaks a long-held belief by providing optimal even-distance GB codes, previously deemed impossible. All are CSS codes derived from Cayley graphs. Recognizing that standard equivalence relations do not preserve their CSS structure, we introduce a CSS-preserving equivalence relation for rigorous comparison of Cayley graph-based CSS codes. Under this framework, the first two families are inequivalent to all previously known optimal weight-4 2D surface codes, while the third family is equivalent to the best-known odd-distance 2D surface code. Finally, we classify all extremal, non-equivalent (2,2)-GB codes with length below 200 and present a comparison table with existing notable 2D weight-4 surface codes.

Figures

Figures reproduced from arXiv: 2507.21237 by the authors.

Figure 1
Figure 1. (Z/4Z, 1, 2) graph Definition II.2 (Cocycles and faces of (G, g1, g2)). Let (G, g1, g2) be a graph constructed as in Definition II.1 • For any vertex g ∈ G, we define its associated cocycle (in the context of this graph’s construction) as the multiset consisting in the edges connecting g to its four neighbours: {g, g + g1}, {g, g − g1}, {g, g + g2}, {g, g − g2}. Note: This definition treats edges as distinct even if… view at source ↗
Figure 2
Figure 2. Mapping a walk in the Cayley graph (Z/8Z, 1, 3) (left) to a path in the grid Z 2 (right) [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Two possible configurations for the boundary of [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Connectivity in the Cayley graph (Z/2r 2Z, 1, 2r − 1) with r = 3 is preserved after removing vertices a = 1 and a + 1 − (2r − 1) = 5 thanks to the edge between a − 2 = 17 and a − 2 + (2r − 1) = 4, represented with the dotted blue line. The edge a − 2 → a − 2 + (2r − 1)…
Figure 5
Figure 5. Figure 5: Connectivity in the Cayley graph (Z/2r 2Z, 1, 2r − 1) with r = 3 is preserved after removing vertices a = 1 and b = 7 ̸= a + 1 − (2r − 1) thanks to the edge between b + 1 = 8 and b + 1 − (2r − 1) = 3, represented with the dotted blue line. The edge b + 1 → b + 1 − (2r …
Figure 6
Figure 6. Figure 6: Three independent paths from  0 0  to R =  a b  on the torus Z 2/L where L = Z  2r 0 LZ  r r  and parameters satisfying 0 < a ≤ r and 0 ≤ b < r − 1. Here r = 6, a = 3 and b = 2. Case 1.2: b = r − 1 (see [PITH_FULL_IMAGE:figures/full_fig_p018_6.png]
Figure 7
Figure 7. Figure 7: Three independent paths from  0 0  to R =  a b  on the torus Z 2/L where L = Z  2r 0 LZ  r r  and parameters satisfying r − 1 ≤ a ≤ r and b = r − 1. Here r = 6 and a = b = r − 1. Case 2: r < a < 2r • Path 1 (green path in [PITH_FULL_IMAGE:figures/full_fig_p018…
Figure 8
Figure 8. Figure 8: Three independent paths from  0 0  to R =  a b  on the torus Z 2/L where L = Z  2r 0 LZ  r r  and parameters satisfying r < a < 2r and 0 ≤ b < r. Here r = 6, a = 9 and b = 3. Case 3: 2r ≤ a ≤ 3r − 2 If a can be equal to 2r, we have to take different shape of pa…
Figure 9
Figure 9. Figure 9: Three independent paths from  0 0  to R =  a b  on the torus Z 2/L where L = Z  2r 0 LZ  r r  and parameters satisfying 2r ≤ a ≤ 3r − 2 and 0 < b < r − 1. Here r = 6, a = 13 and b = 3 [PITH_FULL_IMAGE:figures/full_fig_p019_9.png]
Figure 10
Figure 10. Figure 10: Three independent paths from  0 0  to R =  a b  on the torus Z 2/L where L = Z  2r 0 LZ  r r  and parameters satisfying 2r ≤ a ≤ 3r − 2 and b = r − 1. Here r = 6, a = 13 and b = 5. In every case, using vertices of Z 2 ∩ P, we have constructed 3 paths between …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

17 extracted references · 10 canonical work pages

  1. [13]

    Distance bounds for generalized bicycle codes,

    R. Wang and L. P. Pryadko, “Distance bounds for generalized bicycle codes,” arXiv:2203.17216, 2022

    arXiv 2022

  2. [17]

    Classification of (2,2)-Generalized Bicycle (GB) Codes,

    F. Arnault, P. Gaborit, N. Saussay, “Classification of (2,2)-Generalized Bicycle (GB) Codes,” GitHub repository, 2025. Accessed on: Jul. 13, 2025. [Online]. Available: https://github.com/NicolasSaussay/weight-4_GB-Codes_Classification

    work page 2025

  3. [1]

    Polynomial-time algorithms for prime factorization and discrete logarithms on a quantum computer,

    P. W. Shor, “Polynomial-time algorithms for prime factorization and discrete logarithms on a quantum computer,”SIAM Review, vol. 41, no. 2, pp. 303–332, 1999

    1999

  4. [2]

    M. A. Nielsen and I. L. Chuang,Quantum Computation and Quantum Information: 10th Anniversary Edition. Cambridge University Press, 2010

    2010

  5. [3]

    Fault-tolerant quantum computation by anyons,

    A. Kitaev, “Fault-tolerant quantum computation by anyons,”Annals of Physics, vol. 303, no. 1, pp. 2–30, Jan. 2003

    work page 2003

  6. [4]

    Homological error correction: Classical and quantum codes,

    H. Bombin and M. A. Martin-Delgado, “Homological error correction: Classical and quantum codes,”Journal of Mathematical Physics, vol. 48, no. 5, p. 052105, May 2007

    work page 2007

  7. [5]

    Quantum tanner codes,

    A. Leverrier and G. Zémor, “Quantum tanner codes,” arXiv:2202.13641, 2022

    arXiv 2022

  8. [6]

    Asymptotically good quantum and locally testable classical LDPC codes,

    P. Panteleev and G. Kalachev, “Asymptotically good quantum and locally testable classical LDPC codes,” arXiv:2111.03654, 2022

    arXiv 2022

Show all 17 references
  1. [7]

    High-threshold and low-overhead fault-tolerant quantum memory,

    S. Bravyi, A. W. Cross, J. M. Gambetta, D. Maslov, P. Rall, and T. J. Yoder, “High-threshold and low-overhead fault-tolerant quantum memory,” arXiv:2308.07915, 2024

    2024 arXiv

  2. [8]

    Fiber bundle codes: Breaking then 1/2polylog(n)barrier for quantum LDPC codes,

    M. B. Hastings, J. Haah, and R. O’Donnell, “Fiber bundle codes: Breaking then 1/2polylog(n)barrier for quantum LDPC codes,” inProc. 53rd Annu. ACM SIGACT Symp. Theory Comput. (STOC), 2021, pp. 1276–1288

    2021

  3. [9]

    Good quantum error-correcting codes exist,

    A. R. Calderbank and P. W. Shor, “Good quantum error-correcting codes exist,”Physical Review A, vol. 54, no. 2, pp. 1098–1105, Aug. 1996

    1996

  4. [10]

    Error correcting codes in quantum theory,

    A. M. Steane, “Error correcting codes in quantum theory,”Phys. Rev. Lett., vol. 77, no. 5, pp. 793–797, Jul. 1996

    1996

  5. [11]

    Quantum Kronecker sum-product low-density parity-check codes with finite rate,

    A. A. Kovalev and L. P. Pryadko, “Quantum Kronecker sum-product low-density parity-check codes with finite rate,”Phys. Rev. A, vol. 88, no. 1, p. 012311, Jul. 2013

    2013

  6. [12]

    Sparse-graph codes for quantum error correction,

    D. MacKay, G. Mitchison, and P. McFadden, “Sparse-graph codes for quantum error correction,”IEEE Trans. Inf. Theory, vol. 50, no. 10, pp. 2315–2330, 2004

    2004

  7. [14]

    Analysis of the error-correcting radius of a renormalisation decoder for Kitaev’s toric code,

    W. Rozendaal and G. Zémor, “Analysis of the error-correcting radius of a renormalisation decoder for Kitaev’s toric code,”arXiv, arXiv:2309.12165, Sep. 2023

    2023 arXiv

  8. [15]

    Collection of codes constructed for ‘Distance bounds for generalized bicycle codes’,

    R. Wang and L. P. Pryadko, “Collection of codes constructed for ‘Distance bounds for generalized bicycle codes’,” GitHub repository, 2022. Accessed on: Mar. 30, 2022. [Online]. Available: https://github.com/QEC-pages/GB-codes

    2022

  9. [16]

    Congruent graphs and the connectivity of graphs,

    H. Whitney, “Congruent graphs and the connectivity of graphs,”American Journal of Mathematics, vol. 54, no. 1, pp. 150–168, 1932

    1932

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.