Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Degeneracy cutting, a local post-processing step that removes one qubit per stabilizer generator before rerunning belief propagation, matches or beats the accuracy of ordered statistics decoding for quantum LDPC codes while reducing cost fr

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

T0 review · deepseek-v4-flash

2026-08-04 10:41 UTC pith:QHNNKPR6

load-bearing objection A genuinely useful, local linear-time post-processor for BP decoding that earns a serious referee, with one unproven rank-preservation heuristic and a few empirical gaps that should be addressed rather than blocking publication. the 3 major comments →

arxiv 2510.08695 v2 pith:QHNNKPR6 submitted 2025-10-09 quant-ph

Degeneracy Cutting: A Local and Efficient Post-Processing for Belief Propagation Decoding of Quantum Low-Density Parity-Check Codes

classification quant-ph MSC 81P7094B05 PACS 03.67.Pp
keywords degeneracy cuttingbelief propagation decodingquantum LDPC codespost-processing decoderordered statistics decodingdetector degeneracy matrixsurface codesbivariate bicycle codes
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper proposes a post-processing step for belief-propagation (BP) decoding of quantum LDPC codes. The step, called degeneracy cutting, looks at each stabilizer generator and permanently sets to zero the error probability of the qubit BP considers least likely to have flipped within that generator's support, then reruns BP. The authors argue this breaks the local degeneracies that stall BP, at only linear computational cost and using only local comparisons. Numerically, they show BP plus degeneracy cutting approaches or beats BP plus ordered statistics decoding for surface and bivariate bicycle codes under code-capacity noise, and extends to phenomenological and circuit-level noise via a newly introduced detector degeneracy matrix. If correct, this offers a real-time, parallel-friendly decoding route for fault-tolerant quantum computers.

Core claim

The central claim is that degeneracy cutting works by, for every X-type stabilizer generator, deleting from the Tanner graph the qubit with the smallest BP-estimated error probability, then rerunning BP once; the deletion removes one representative of each locally degenerate error class, steering BP to a valid recovery. The paper claims this single local pruning step is enough to make BP plus degeneracy cutting match or exceed the accuracy of BP plus ordered statistics decoding for both surface codes and BB codes under code-capacity noise, at O(n) rather than O(n^3) cost. The same principle extends to phenomenological and circuit-level noise once one replaces stabilizer rows with the detecto

What carries the argument

The central mechanism is degeneracy cutting: for each row of the X-type parity-check matrix (or of the generalized detector degeneracy matrix), identify the variable node with the lowest BP posterior error probability within that row's support and set its probability to zero (equivalently delete its column), then rerun BP on the modified Tanner graph, the bipartite graph linking qubits to stabilizer checks. This removes one representative of each local degenerate error class and breaks short loops. The detector degeneracy matrix generalizes the stabilizer generators to settings where new trivial errors arise from measurement errors or gate-induced error propagation, allowing the same local p

Load-bearing premise

The load-bearing premise is the heuristic that removing one qubit per stabilizer generator leaves the reduced Z-parity-check matrix with full row rank 'almost always', so that every measured syndrome still has a solution in the pruned graph; the paper gives no proof or worst-case bound for this, only numerical evidence.

What would settle it

Search for a qLDPC code (e.g., row weight 4 or 6) and an error pattern for which the pruned parity-check matrix is rank-deficient, making the syndrome unsatisfiable after cutting and forcing the second BP run to fail; a concrete check is to compute, for increasing code distances, the fraction of all syndromes for which the cut set destroys full row rank of H_Z and to see whether that fraction vanishes as n grows or remains bounded away from zero.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • If correct, BP plus degeneracy cutting provides a decoder for qLDPC codes whose total cost is linear in the number of qubits, the same scaling as BP alone, while matching the accuracy of BP plus ordered statistics decoding, which costs O(n^3).
  • For bivariate bicycle codes under code-capacity noise, the authors report that BP plus degeneracy cutting achieves a lower decoding failure probability than BP plus ordered statistics decoding, in addition to the cost reduction.
  • The detector degeneracy matrix extends stabilizer-based post-processing to phenomenological and circuit-level noise models, where new trivial errors arise from the syndrome extraction process itself; with it, BP plus degeneracy cutting stays within roughly an order of magnitude of BP plus ordered statistics decoding.
  • Because degeneracy cutting uses only local comparisons within each stabilizer generator and runs BP at most twice, it preserves BP's amenability to parallel hardware implementation, unlike post-processors that sort or compare across all qubits.
  • From BP plus degeneracy cutting plus ordered statistics decoding matching plain BP plus ordered statistics decoding, the pruned Tanner graph still contains valid solutions even when degeneracy cutting fails, indicating the accuracy gap lies in BP's search rather than in lost syndrome consistency.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Because degeneracy cutting only compares probabilities within each stabilizer support and touches each qubit once, the pruning step maps directly onto the same parallel message-passing hardware used by BP; this makes BP plus degeneracy cutting a plausible building block for on-chip, real-time decoders, an implication the paper notes but does not develop.
  • The detector degeneracy matrix is defined generically, so it could be extended to higher-weight trivial errors (weight 4 and beyond); the paper's circuit-level results degrade with code distance partly because only weight-3 degeneracies are included, so filling in higher-weight rows is a concrete way to test whether the gap can be closed.
  • Combining degeneracy cutting with recently proposed modifications of BP itself, such as automorphism ensembles, memory effects, or oscillating-bit dynamics, is left open by the paper; since degeneracy cutting is orthogonal to those changes, such combinations might push failure rates below BP plus ordered statistics decoding while keeping linear cost.
  • If the rank-preservation heuristic is confirmed or guaranteed by a reweighting rule, the resulting linear-cost, local decoder would be especially attractive for fault-tolerant protocols targeting constant or polylog overhead, where classical decoding latency is often a bottleneck.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes degeneracy cutting (DC), a post-processing step for belief-propagation (BP) decoding of qLDPC codes. For each X-stabilizer generator, DC removes (sets to zero) the variable node with the lowest BP-estimated error probability, then reruns BP on the modified Tanner graph. The authors argue this breaks local degeneracies while preserving syndrome solvability, and that the procedure is O(n) and uses only local comparisons. They report numerical results for rotated surface codes and bivariate bicycle (BB) codes under code-capacity noise, claiming that BP+DC approaches BP+OSD accuracy, and for BB codes outperforms BP+OSD while reducing cost from O(n^3) to O(n). They also extend DC to phenomenological and circuit-level noise by introducing a detector degeneracy matrix (H_DDM) built from identified low-weight trivial errors, with numerical results showing performance within an order of magnitude of BP+OSD.

Significance. If the reported performance is robust, BP+DC offers a practical lightweight alternative to BP+OSD for real-time decoding of qLDPC codes: it retains BP's linear scaling and locality, making it amenable to parallel hardware implementation. The paper is careful in its numerical methodology: error bars are reported, the circuit-level noise model is specified in detail, and for BB codes the construction of H_DDM is verified to include all weight-3 trivial errors (with an explicit exception for [[108,8,10]]). The introduction of the detector degeneracy matrix is a useful conceptual step that generalizes stabilizer-based post-processing to realistic noise models. However, the core correctness argument is heuristic, and two load-bearing choices (the rank-preservation assumption and the choice of second-BP input) are not formalized; these gaps limit the strength of the claims as they currently stand.

major comments (3)
  1. [Sec. III A, Algorithm 1, and Fig. 2] The only correctness argument for DC is the statement: "Because we delete one qubit per stabilizer generator, codes whose row weight is four or larger will almost always retain full row rank in H_Z..." (Sec. III A). This is not a proof, and the relevant condition is not row rank but whether the surviving columns of H_Z still span the observed syndrome space. For BB codes, n=2m and there are m_X=m_Z potential cuts, so the reduced parity-check submatrix can be square or nearly square; a square submatrix of a rank-deficient matrix that spans the original column space is a delicate property. If the cuts destroy syndrome solvability, Algorithm 1 returns a syndrome-inconsistent estimate or declares failure, so the failure probability includes artifacts of the cut rather than logical errors. BP+DC+OSD in Fig. 2 provides an audit for surface codes, but no equivalent audit is reported for BB code
  2. [Sec. III B, Algorithm 1 line 9] Algorithm 1 specifies the second BP run with input p̂ (the posterior from the first BP). However, for BB codes the authors write: "When applying BP+DC to BB codes, we find that accuracy improves if, in the second BP decoding step (line 9 of Algorithm 1), we replace the estimated error probabilities p̂ with the prior error probabilities p as input." For surface codes, using the prior degrades performance. Thus the headline claim that BP+DC outperforms BP+OSD for BB codes depends on a code-family-dependent choice not present in Algorithm 1 and justified only by empirical observation. The paper states the formal analysis is left to future work. As written, the method is underspecified: a decoder whose input distribution is selected post hoc per code family cannot be evaluated as a single algorithm. Please report results for both input choices for both families, or specify a data-independent
  3. [Sec. IV B-D and Eq. (A4)] The detector degeneracy matrix H_DDM is constructed from numerically identified low-weight trivial errors. The completeness verification in Appendix A3 covers only weight-3 trivial errors for the listed BB codes (with an appended row block for [[108,8,10]]), and no analogous verification is reported for surface codes or for the phenomenological model of Sec. IV C. Since H_DDM determines which variable nodes are cut, an incomplete H_DDM could remove a qubit needed by all valid corrections, causing unsolvable syndrome problems. The authors acknowledge higher-weight degeneracies as future work, which is honest, but the claim that BP+DC is "comparable" to BP+OSD in these settings needs the same audit as in the code-capacity surface-code case: report BP+DC+OSD for all failing samples and state whether the reduced detector check matrix always has a solution. This is directly testable with the
minor comments (5)
  1. [Abstract and Fig. 5 caption] The abstract states BP+DC achieves "comparable" performance to BP+OSD, while Fig. 5's caption says "within an order of magnitude." These are different quantitative claims; please align the wording.
  2. [Eq. (A5)] The notation in the appended row block for the [[108,8,10]] code is unclear: "A_2^1 A_3 + A_1 A_2^3 + I" — specify whether superscripts denote powers of the matrices A_2 and A_2 or some other operation; this is important for reproducibility.
  3. [Sec. II C 2] The description of OSD says "the columns of a parity-check matrix with the smallest p̂_i values are temporarily removed so that the reduced parity-check matrix becomes invertible." For a general rectangular parity-check matrix, "invertible" should be read as "full column rank" or similar; consider rephrasing for precision.
  4. [Sec. III A, Algorithm 2] Algorithm 2 is presented as an alternative to line 7, but it is not explicitly stated that it must be executed before line 9; if the probabilities are set to zero after the first BP, this is fine, but the ordering should be clarified in the main text for readers implementing from the pseudo-code.
  5. [References] The paper cites several relevant recent decoders (Refs. [35]-[44]) but misses the ordered Tanner forest method in the comparison table if any; also consider citing the source of the surface-code syndrome measurement schedule [63] in the main text, not only in Sec. IV D.

Circularity Check

0 steps flagged

No material circularity: DC is a heuristic post-processor validated by external benchmarks; its rank-preservation rationale is unproven but not circular, and self-citations are contextual only.

full rationale

The paper's central claims are empirical, not derived from the target result: BP+DC is benchmarked against BP+OSD and other external decoders on surface and BB codes under code-capacity, phenomenological, and circuit-level noise. The DC rule—remove the lowest-probability variable node in the support of each H_X row and rerun BP—is not constructed so that success is guaranteed by definition; its performance is checked by simulation. The detector degeneracy matrix H_DDM is built from code/circuit structure: stabilizer generators and numerically enumerated weight-3 trivial errors satisfying e H_DCM^T = 0 and e O^T = 0, which is independent of the decoder's output. The only load-bearing analytical assertion, Sec. III A: 'Because we delete one qubit per stabilizer generator, codes whose row weight is four or larger will almost always retain full row rank in H_Z, so the reduced Tanner graph still supports a solution for every measured syndrome,' is an unproven heuristic; if it fails, DC can create unsolvable syndromes, but that is a correctness/soundness risk, not a circular reduction. The paper also honestly reports a post-hoc choice for BB codes: 'we find that accuracy improves if, in the second BP decoding step..., we replace the estimated error probabilities p_hat with the prior error probabilities p as input,' and notes it is not always beneficial and defers formal analysis to future work. This is a tuning choice on the same benchmark data, which may inflate reported performance, but it is not a prediction forced by construction. Self-citations (Refs. [7,8,24,48,49]) appear in contextual statements about fault-tolerant computation and hardware, not as load-bearing support for the DC decoder. No uniqueness theorem, ansatz, or known result is repackaged via self-citation. Score 1 reflects the absence of circularity while noting minor contextual self-citation.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 1 invented entities

The central claims rest on few numeric parameters; the method is heuristic and validated numerically. The main free choices are the second-BP input distribution (per code family) and the composition of the detector degeneracy matrix (per code). Standard linear algebra over F2 and the detector error model are background assumptions.

free parameters (3)
  • second_BP_input_distribution = prior p for BB codes; posterior \hat p for surface codes
    Sec. III B: the BB-code outperformance over BP+OSD requires resetting the second BP input to the prior p, while this is not beneficial for surface codes. The paper states this and leaves formal analysis to future work; the choice is post hoc per code family.
  • detector_degeneracy_matrix_completeness = all weight-3 trivial errors; extra row block for [[108,8,10]]
    Sec. IV D and App. A3: H_DDM is built from numerically identified weight-3 trivial errors; for the [[108,8,10]] BB code an additional row block is appended to ensure completeness. Which degeneracies are included affects DC decisions and is adjusted per code.
  • dc_iteration_count = 1
    Sec. III A: 'we numerically observe that additional iterations do not enhance performance'—the graph is modified only once.
axioms (4)
  • domain assumption BP posterior marginals are accurate enough that the lowest-probability qubit in each stabilizer support is the least likely true error
    The entire DC heuristic rests on this; the paper justifies it only via the degeneracy example in Fig. 1, not by proof or systematic analysis.
  • ad hoc to paper Deleting one variable node per stabilizer generator preserves existence of a syndrome-consistent solution
    Sec. III A: 'codes whose row weight is four or larger will almost always retain full row rank in H_Z' is a heuristic claim with no proof or worst-case analysis.
  • ad hoc to paper The numerically identified weight-3 trivial errors are exactly the minimal non-trivial degeneracies; higher-weight degeneracies are negligible
    Sec. IV D and App. A3: completeness is numerically verified only for listed BB codes, not surface codes; the growing gap with distance for surface codes suggests incompleteness may matter.
  • domain assumption Standard detector error model with independent error mechanisms
    Sec. IV A: decoding under circuit-level noise is modeled via detector check matrix and independent mechanisms from Stim; correctness of this reduction is assumed from prior literature.
invented entities (1)
  • detector degeneracy matrix H_DDM independent evidence
    purpose: Generalizes X-type stabilizer rows to detector-error-model settings so DC can be applied under phenomenological and circuit-level noise
    It is constructed from the code and circuit (H_X, H_Z, H_DCM) and satisfies H_DDM H_DCM^T = 0; an independent group can verify the construction. It is a mathematical object, not a physical entity.

pith-pipeline@v1.3.0-alltime-deepseek · 21691 in / 11392 out tokens · 97446 ms · 2026-08-04T10:41:42.461118+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Degeneracy Cutting: A Local and Efficient Post-Processing for Belief Propagation Decoding of Quantum Low-Density Parity-Check Codes." pith.science (2026). https://pith.science/paper/QHNNKPR6

@misc{pith2026251008695,
  author       = {Pith},
  title        = {Pith review of: Degeneracy Cutting: A Local and Efficient Post-Processing for Belief Propagation Decoding of Quantum Low-Density Parity-Check Codes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QHNNKPR6}},
  note         = {Machine review of arXiv:2510.08695}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Quantum low-density parity-check (qLDPC) codes are promising for realizing scalable fault-tolerant quantum computation due to their potential for low-overhead protocols. A common approach to decoding qLDPC codes is to use the belief propagation (BP) decoder, followed by a post-processing step to enhance decoding accuracy. For fast decoding, the post-processing algorithm is desirable to have a small computational cost and rely only on local operations on the Tanner graph to facilitate parallel implementation. To address this requirement, we propose degeneracy cutting (DC), an efficient post-processing technique for the BP decoder that operates on information restricted to the support of each stabilizer generator. DC selectively removes one variable node with the lowest error probability for each stabilizer generator, significantly improving decoding performance while retaining the favorable computational scaling and structure amenable to parallelization inherent to BP. We further extend our method to realistic noise models, including phenomenological and circuit-level noise models, by introducing the detector degeneracy matrix, which generalizes the notion of stabilizer-induced degeneracy to these settings. Numerical simulations demonstrate that BP+DC achieves decoding performance approaching that of BP followed by ordered statistics decoding (BP+OSD) in several settings, while requiring significantly less computational cost. Our results present BP+DC as a promising decoder for fault-tolerant quantum computing, offering a valuable balance of accuracy, efficiency, and suitability for parallel implementation.

Figures

Figures reproduced from arXiv: 2510.08695 by Hayata Yamasaki, Kento Tsubouchi, Shiro Tamiya.

Figure 1
Figure 1. Figure 1: FIG. 1. An illustration of degeneracy in belief propagation decoding and degeneracy cutting (DC). Circles represent variable [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Performance of the BP decoder, BP+OSD decoder, BP+DC decoder, and BP+DC+OSD decoder for surface codes [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Schematic illustration of degeneracy arising from [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Performance of the BP decoder, BP+OSD decoder, BP+DC decoder, and BP+DC+OSD decoder for surface codes [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. Performance of the BP decoder, BP+OSD decoder, BP+DC decoder, and BP+DC+OSD decoder for surface [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Impulse Decoding of Quantum LDPC Codes: Equivalence of Degeneracy and Code-Shortening

    quant-ph 2026-06 unverdicted novelty 6.0

    Degeneracy in quantum LDPC codes equals classical code shortening at the decoder, enabling impulse decoding that beats BP+OSD plus a residual-error follow-up step.

Reference graph

Works this paper leans on

80 extracted references · 17 linked inside Pith · cited by 1 Pith paper

  1. [1]

    In this work, we consider therotated surface code of code distanced[52, 53], which is defined on ad×dsquare lattice and has parameters [[d 2,1, d]]

    Surface codes Thesurface codeis a planar realization of Kitaev’s toric code [45–47]. In this work, we consider therotated surface code of code distanced[52, 53], which is defined on ad×dsquare lattice and has parameters [[d 2,1, d]]. Data qubits are on vertices, auxiliary qubits for syndrome measurements are placed at the centers of the plaquettes, and bo...

  2. [2]

    Let Il ∈F l×l 2 denote the identity matrix andS l ∈F l×l 2 the cyclic shift matrix, defined by (S l)i,j = 1 ifj= i+ 1 modl, and 0 otherwise

    Bivariate bicycle codes Bivariate bicycle (BB) codes [16] are a class of qLDPC codes implemented on a two-layer qubit architecture. Let Il ∈F l×l 2 denote the identity matrix andS l ∈F l×l 2 the cyclic shift matrix, defined by (S l)i,j = 1 ifj= i+ 1 modl, and 0 otherwise. The construction uses two commuting matricesx=S l ⊗I m andy=I l ⊗S m, which 4 [[n, k...

  3. [3]

    Given the prior bit-flip probabili- tiesp= (p 1,

    Belief propagation decoder Thebelief propagation (BP) decoderis an iterative message-passing algorithm whose application to error correction was pioneered by Gallager for decoding classi- cal LDPC codes [25]. Given the prior bit-flip probabili- tiesp= (p 1, . . . , pn), theZ-type parity-check matrixH Z, and an observed syndromes Z, the decoder estimates t...

  4. [4]

    !"#="′!"#

    Post-processing for BP decoder Although the BP decoder alone is not effective for de- coding qLDPC codes, the estimated marginal error prob- abilitiesˆ pobtained from BP can be post-processed to produce a more accurate correction ˆeX . A prominent example of such post-processing isordered statistics de- coding(OSD): the columns of a parity-check matrix wi...

  5. [5]

    P. W. Shor, Fault-tolerant quantum computation, inPro- ceedings of 37th conference on foundations of computer science(IEEE, 1996) pp. 56–65

  6. [6]

    Aharonov and M

    D. Aharonov and M. Ben-Or, Fault-tolerant quantum computation with constant error, inProceedings of the twenty-ninth annual ACM symposium on Theory of com- puting(1997) pp. 176–188

  7. [7]

    Aliferis, D

    P. Aliferis, D. Gottesman, and J. Preskill, Quantum accu- racy threshold for concatenated distance-3 codes, Quan- tum Info. Comput.6, 97–165 (2006)

  8. [8]

    Gottesman, An introduction to quantum error correction and fault-tolerant quantum computation, arXiv:0904.2557 (2009)

    D. Gottesman, An introduction to quantum error correction and fault-tolerant quantum computation, arXiv:0904.2557 (2009)

  9. [9]

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

  10. [10]

    Gottesman, Fault-tolerant quantum computation with constant overhead, Quantum Inf

    D. Gottesman, Fault-tolerant quantum computation with constant overhead, Quantum Inf. Comput.14, 1338 (2014)

  11. [11]

    Tamiya, M

    S. Tamiya, M. Koashi, and H. Yamasaki, Polylog- time-and constant-space-overhead fault-tolerant quan- tum computation with quantum low-density parity-check codes, arXiv:2411.03683 (2024)

  12. [12]

    Yamasaki and M

    H. Yamasaki and M. Koashi, Time-efficient constant- space-overhead fault-tolerant quantum computation, Na- ture Physics20, 247–253 (2024)

  13. [13]

    Tillich and G

    J.-P. Tillich and G. Zemor, Quantum ldpc codes with positive rate and minimum distance proportional to the square root of the blocklength, IEEE Trans. Inf. Theory 60, 1193–1202 (2014)

  14. [14]

    Leverrier, J.-P

    A. Leverrier, J.-P. Tillich, and G. Zemor, Quantum ex- pander codes, in2015 IEEE 56th Annual Symposium on Foundations of Computer Science(IEEE, 2015) p. 810–824

  15. [15]

    M. B. Hastings, J. Haah, and R. O’Donnell, Fiber bundle codes: breaking the n 1/2 polylog( n ) barrier for quan- tum ldpc codes, inProceedings of the 53rd Annual ACM SIGACT Symposium on Theory of Computing, STOC ’21 (ACM, 2021) p. 1276–1288

  16. [16]

    N. P. Breuckmann and J. N. Eberhardt, Quantum low- density parity-check codes, PRX Quantum2, 040101 (2021)

  17. [17]

    Panteleev and G

    P. Panteleev and G. Kalachev, Quantum ldpc codes with almost linear minimum distance, IEEE Trans. Inf. The- ory68, 213–229 (2022)

  18. [18]

    Panteleev and G

    P. Panteleev and G. Kalachev, Asymptotically good quantum and locally testable classical ldpc codes, inPro- ceedings of the 54th Annual ACM SIGACT Symposium on Theory of Computing, STOC 2022 (Association for Computing Machinery, New York, NY, USA, 2022) p. 375–388

  19. [19]

    Dinur, M.-H

    I. Dinur, M.-H. Hsieh, T.-C. Lin, and T. Vidick, Good quantum ldpc codes with linear time decoders, inPro- ceedings of the 55th Annual ACM Symposium on Theory of Computing, STOC 2023 (Association for Computing Machinery, New York, NY, USA, 2023) p. 905–918

  20. [20]

    Bravyi, A

    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, Nature627, 778 (2024)

  21. [21]

    Fawzi, A

    O. Fawzi, A. Grospellier, and A. Leverrier, Constant overhead quantum fault tolerance with quantum ex- pander codes, Commun. ACM64, 106 (2020)

  22. [22]

    L. Z. Cohen, I. H. Kim, S. D. Bartlett, and B. J. Brown, Low-overhead fault-tolerant quantum computing using long-range connectivity, Sci. Adv.8, eabn1717 (2022)

  23. [23]

    Q. Xu, J. P. Bonilla Ataides, C. A. Pattison, N. Raveen- dran, D. Bluvstein, J. Wurtz, B. Vasi´ c, M. D. Lukin, L. Jiang, and H. Zhou, Constant-overhead fault-tolerant 14 quantum computation with reconfigurable atom arrays, Nat. Phys.20, 1084 (2024)

  24. [24]

    T. J. Yoder, E. Schoute, P. Rall, E. Pritchett, J. M. Gam- betta, A. W. Cross, M. Carroll, and M. E. Beverland, Tour de gross: A modular quantum computer based on bivariate bicycle codes, arXiv:2506.03094 (2025)

  25. [25]

    Q. T. Nguyen and C. A. Pattison, Quantum fault tol- erance with constant-space and logarithmic-time over- heads, inProceedings of the 57th Annual ACM Sympo- sium on Theory of Computing(2025) pp. 730–737

  26. [26]

    B. M. Terhal, Quantum error correction for quantum memories, Rev. Mod. Phys.87, 307 (2015)

  27. [27]

    Barber, K

    B. Barber, K. M. Barnes, T. Bialas, O. Bu˘ gdaycı, E. T. Campbell, N. I. Gillespie, K. Johar, R. Rajan, A. W. Richardson, L. Skoric, C. Topal, M. L. Turner, and A. B. Ziad, A real-time, scalable, fast and resource-efficient de- coder for a quantum computer, Nat. Electron.8, 84–91 (2025)

  28. [28]

    Takada and H

    Y. Takada and H. Yamasaki, Doubly-polylog-time- overhead fault-tolerant quantum computation by a polylog-time parallel minimum-weight perfect matching decoder, arXiv:2503.13601 (2025)

  29. [29]

    Gallager, Low-density parity-check codes, IRE Trans

    R. Gallager, Low-density parity-check codes, IRE Trans. Inf. Theory8, 21 (1962)

  30. [30]

    D. J. MacKay and R. M. Neal, Near shannon limit perfor- mance of low density parity check codes, Electron. Lett. 32, 1645 (1996)

  31. [31]

    T. J. Richardson, M. A. Shokrollahi, and R. L. Ur- banke, Design of capacity-approaching irregular low- density parity-check codes, IEEE Trans. Inf. Theory47, 619 (2001)

  32. [32]

    F. R. Kschischang, B. J. Frey, and H.-A. Loeliger, Factor graphs and the sum-product algorithm, IEEE Trans. Inf. Theory47, 498 (2001)

  33. [33]

    Tanner, A recursive approach to low complexity codes, IEEE Trans

    R. Tanner, A recursive approach to low complexity codes, IEEE Trans. Inf. Theory27, 533 (1981)

  34. [34]

    Poulin and Y

    D. Poulin and Y. Chung, On the iterative decoding of sparse quantum codes, Quantum Inf. Comput.8, 987 (2008)

  35. [35]

    Raveendran and B

    N. Raveendran and B. Vasi´ c, Trapping sets of quantum ldpc codes, Quantum5, 562 (2021)

  36. [36]

    Fuentes, J

    P. Fuentes, J. E. Martinez, P. M. Crespo, and J. Garcia- Fr ´ ıas, Degeneracy and its impact on the decoding of sparse quantum codes, IEEE Access9, 89093 (2021)

  37. [37]

    Panteleev and G

    P. Panteleev and G. Kalachev, Degenerate quantum ldpc codes with good finite length performance, Quantum5, 585 (2021)

  38. [38]

    Roffe, D

    J. Roffe, D. R. White, S. Burton, and E. Campbell, De- coding across the quantum low-density parity-check code landscape, Phys. Rev. Res.2, 043423 (2020)

  39. [39]

    Du Crest, M

    J. Du Crest, M. Mhalla, and V. Savin, Stabilizer inactiva- tion for message-passing decoding of quantum ldpc codes, in2022 IEEE Information Theory Workshop (ITW) (IEEE, 2022) pp. 488–493

  40. [40]

    Du Crest, F

    J. Du Crest, F. Garcia-Herrero, M. Mhalla, V. Savin, and J. Valls, Check-agnosia based post-processor for message- passing decoding of quantum ldpc codes, Quantum8, 1334 (2024)

  41. [41]

    H. Yao, W. A. Laban, C. H¨ ager, A. G. i Amat, and H. D. Pfister, Belief propagation decoding of quantum ldpc codes with guided decimation, in2024 IEEE In- ternational Symposium on Information Theory (ISIT) (IEEE, 2024) pp. 2478–2483

  42. [42]

    A. Gong, S. Cammerer, and J. M. Renes, Toward low- latency iterative decoding of qldpc codes under circuit- level noise, arXiv:2403.18901 (2024)

  43. [43]

    A. d. iOlius and J. E. Martinez, The closed-branch de- coder for quantum ldpc codes, arXiv:2402.01532 (2024)

  44. [44]

    Hillmann, L

    T. Hillmann, L. Berent, A. O. Quintavalle, J. Eisert, R. Wille, and J. Roffe, Localized statistics decoding: A parallel decoding algorithm for quantum low-density parity-check codes, arXiv:2406.18655 (2024)

  45. [45]

    Wolanski and B

    S. Wolanski and B. Barber, Ambiguity clustering: an accurate and efficient decoder for qldpc codes, arXiv:2406.14527 (2024)

  46. [46]

    A. d. iOlius, I. E. Martinez, J. Roffe, and J. E. Martinez, An almost-linear time decoding algorithm for quantum ldpc codes under circuit-level noise, arXiv:2409.01440 (2024)

  47. [47]

    K. Yin, X. Fang, J. Ruan, H. Zhang, D. Tullsen, A. Sorn- borger, C. Liu, A. Li, T. Humble, and Y. Ding, Symbreak: Mitigating quantum degeneracy issues in qldpc code de- coders by breaking symmetry, arXiv:2412.02885 (2024)

  48. [48]

    K. R. Ott, B. Het´ enyi, and M. E. Beverland, Decision-tree decoders for general quantum ldpc codes, arXiv:2502.16408 (2025)

  49. [49]

    Dennis, A

    E. Dennis, A. Kitaev, A. Landahl, and J. Preskill, Topo- logical quantum memory, J. Math. Phys.43, 4452 (2002)

  50. [50]

    A. Y. Kitaev, Fault-tolerant quantum computation by anyons, Ann. Phys. (N. Y.)303, 2 (2003)

  51. [51]

    A. G. Fowler, M. Mariantoni, J. M. Martinis, and A. N. Cleland, Surface codes: Towards practical large-scale quantum computation, Phys. Rev. A86, 032324 (2012)

  52. [52]

    Sunami, A

    S. Sunami, A. Goban, and H. Yamasaki, Transver- sal surface-code game powered by neutral atoms, arXiv:2506.18979 (2025)

  53. [53]

    Sunami, S

    S. Sunami, S. Tamiya, R. Inoue, H. Yamasaki, and A. Goban, Scalable networking of neutral-atom qubits: Nanofiber-based approach for multiprocessor fault-tolerant quantum computers, PRX Quantum6, 010101 (2025)

  54. [54]

    A. R. Calderbank and P. W. Shor, Good quantum error- correcting codes exist, Phys. Rev. A54, 1098 (1996)

  55. [55]

    Steane, Multiple particle interference and quantum er- ror correction, Proceedings of the Royal Society A: Math- ematical, Physical and Engineering Sciences452(1996)

    A. Steane, Multiple particle interference and quantum er- ror correction, Proceedings of the Royal Society A: Math- ematical, Physical and Engineering Sciences452(1996)

  56. [56]

    Bombin and M

    H. Bombin and M. A. Martin-Delgado, Optimal resources for topological two-dimensional stabilizer codes: Com- parative study, Phys. Rev. A76, 012305 (2007)

  57. [57]

    Tomita and K

    Y. Tomita and K. M. Svore, Low-distance surface codes under realistic quantum noise, Phys. Rev. A90, 062320 (2014)

  58. [58]

    A. J. Landahl, J. T. Anderson, and P. R. Rice, Fault-tolerant quantum computing with color codes, arXiv:1108.5738 (2011)

  59. [59]

    Mansour and N

    M. Mansour and N. Shanbhag, High-throughput ldpc de- coders, IEEE Transactions on Very Large Scale Integra- tion (VLSI) Systems11, 976 (2003)

  60. [60]

    Chytas, N

    D. Chytas, N. Raveendran, and B. Vasic, Enhanced min- sum decoding of quantum codes using previous iteration dynamics, arXiv:2501.05021 (2025)

  61. [61]

    Koutsioumpas, H

    S. Koutsioumpas, H. Sayginel, M. Webster, and D. E. Browne, Automorphism ensemble decoding of quantum ldpc codes, arXiv:2503.01738 (2025)

  62. [62]

    M¨ uller, T

    T. M¨ uller, T. Alexander, M. E. Beverland, M. B¨ uhler, B. R. Johnson, T. Maurer, and D. Vandeth, Improved belief propagation is sufficient for real-time decoding of 15 quantum memory, arXiv:2506.01779 (2025)

  63. [63]

    M. Wang, A. Li, and F. Mueller, Fully parallelized bp decoding for quantum ldpc codes can outperform bp-osd, arXiv:2507.00254 (2025)

  64. [64]

    P.-J. H. Derks, A. Townsend-Teague, A. G. Burchards, and J. Eisert, Designing fault-tolerant circuits using de- tector error models, arXiv:2407.13826 (2024)

  65. [65]

    Higgott and C

    O. Higgott and C. Gidney, Sparse blossom: correcting a million errors per core second with minimum-weight matching, Quantum9, 1600 (2025)

  66. [66]

    Gidney, Stim: a fast stabilizer circuit simulator, Quan- tum5, 497 (2021)

    C. Gidney, Stim: a fast stabilizer circuit simulator, Quan- tum5, 497 (2021)

  67. [67]

    A. R. O’Rourke and S. Devitt, Compare the pair: Ro- tated vs. unrotated surface codes at equal logical error rates, arXiv:2409.14765 (2024). Appendix A: Circuit-level noise model for BB codes In this appendix, we provide details of the circuit-level noise model used for BB codes. Section A 1 introduces the syndrome measurement circuit proposed in Ref. [16...

  68. [68]

    Syndrome measurement circuit In this section, we describe the syndrome measurement circuit for BB codes as introduced in Ref. [16]. Consider a 2n-qubit quantum circuit, wherenof the qubits are data qubits,m X =n/2 qubits are auxiliary qubits forX-type syndrome measurements, and the remainingm Z =n/2 qubits are auxiliary qubits forZ-type syndrome measure- ...

  69. [69]

    Then, the following eight steps are repeated forTrounds:

    fori= 1 ton/2 do •Idle(q(X, i)) •InitZ(q(Z, i)) •Idle(q(L, i)) •Idle(q(R, i)) Here,Idledenotes the identity operation (i.e., doing nothing), andInitZinitializes the qubit to|0⟩. Then, the following eight steps are repeated forTrounds:

  70. [70]

    fori= 1 ton/2 do •InitX(q(X, i)) •CNOT(q(R, A⊤ 1 (i)), q(Z, i)) •Idle(q(L, i))

  71. [71]

    fori= 1 ton/2 do •CNOT(q(X, i), q(L, A2(i))) •CNOT(q(R, A⊤ 3 (i)), q(Z, i))

  72. [72]

    fori= 1 ton/2 do •CNOT(q(X, i), q(R, B2(i))) •CNOT(q(L, B⊤ 1 (i)), q(Z, i))

  73. [73]

    fori= 1 ton/2 do •CNOT(q(X, i), q(R, B1(i))) •CNOT(q(L, B⊤ 2 (i)), q(Z, i))

  74. [74]

    fori= 1 ton/2 do •CNOT(q(X, i), q(R, B3(i))) •CNOT(q(L, B⊤ 3 (i)), q(Z, i))

  75. [75]

    fori= 1 ton/2 do •CNOT(q(X, i), q(L, A1(i))) •CNOT(q(R, A⊤ 2 (i)), q(Z, i))

  76. [76]

    fori= 1 ton/2 do •CNOT(q(X, i), q(L, A3(i))) •MeasZ(q(Z, i)) •Idle(q(R, i))

  77. [77]

    In the circuit-level noise model, each operation de- scribed above is followed by a potential error character- ized by a physical error ratep

    fori= 1 ton/2 do •MeasX(q(X, i)) •InitZ(q(Z, i)) •Idle(q(L, i)) •Idle(q(R, i)) Here,InitXinitializes the qubit to|+⟩;CNOTapplies a CNOT gate with the first qubit as control and the sec- ond as target;MeasZandMeasXperformZ- andX-basis measurements, respectively. In the circuit-level noise model, each operation de- scribed above is followed by a potential e...

  78. [78]

    !"#="′!"#

    Detector check matrix Based on the syndrome measurement schedule de- scribed in Sec. A 1, we construct the detector check ma- trixH DCM. As noted in the main text, we focus on de- codingX-type errors usingZ-type syndrome measure- ment outcomes obtained fromMeasZ(q(Z, i)), while dis- carding theX-type syndrome measurement results from MeasX(q(X, i)). Furth...

  79. [79]

    ′!"#"!!#=

    Detector degeneracy matrix Corresponding to the detector check matrixH DCM de- scribed in Sec. A 2, we construct the detector degeneracy matrixH DDM as "′!"#"!!#= "′!"# "& 6% 11%/2 ・・・ %/2%5%)+% %+'&)+'& 5% "′!"# ) . (A3) The matrix is formed by stackingTsubmatricesH ′ DDM ∈ F11n/2×6n 2 , each representing degeneracy introduced by errors in thet-th round....

  80. [80]

    We numerically verify that all weight-3 errorsesat- isfyingeH ⊤ DCM =0andeO ⊤ =0are included as row vectors ofH DDM for all BB codes presented in Ref

    By this construction, the resulting detector degener- acy matrixH DDM satisfiesH DCMH ⊤ DDM = 0, and has row weight 6 and column weight 7. We numerically verify that all weight-3 errorsesat- isfyingeH ⊤ DCM =0andeO ⊤ =0are included as row vectors ofH DDM for all BB codes presented in Ref. [16], except for the [[108,8,10]] BB code. We also confirm that the...