Bravyi et al. conjectured that their depth-8 syndrome cycle for the [[144,12,12]] gross code has circuit distance 10. We prove it. In their noise model and for every number of noisy cycles, no set of nine or fewer faults causes an undetected logical error of either Pauli type, and we give explicit ten-fault logical errors of both types. When we released our data on 15 October 2026, the best reported bounds for X-type faults were 9 ≤ d_circ ≤ 12, at 12 rounds; since then, Crocker and Pan have reported the value 10 in both bases, for 12 noisy cycles and for every number of cycles up to 12, respectively. For the same family of circuits we show that the [[90,8,10]] and [[108,8,10]] cycles have circuit distance 8 for every number of cycles, extending values reported for fixed numbers of rounds; for [[72,12,6]] we confirm the known value 6. We also find that, for three of these codes, Google's Tesseract test circuits have the same detector-error-model columns as the depth-8 cycle once restricted to the detectors of the memory basis, for every number of rounds we checked, so circuit-distance results carry over between the two at those round counts. The lower bounds combine 607 SAT refutations, each with a DRAT proof, with exhaustive enumerations of extended-code logicals that carry no certificate. Finally, we independently confirm two results that others reported first: the [[360,12]] bivariate bicycle code has distance 24, and the left-right gross-code circuit of Strikis, Browne and Beverland, as reconstructed from their figures, has circuit distance 11. All code, formulas, proofs and fault sets are available at doi:10.5281/zenodo.23394575.
Bravyi et al. conjectured that their depth-8 syndrome cycle for the [[144,12,12]] gross code has circuit distance 10. We prove it. In their noise model and for every number of noisy cycles, no set of nine or fewer faults causes an undetected logical error of either Pauli type, and we give explicit ten-fault logical errors of both types. When we released our data on 15 October 2026, the best reported bounds for X-type faults were 9 ≤ d_circ ≤ 12, at 12 rounds; since then, Crocker and Pan have reported the value 10 in both bases, for 12 noisy cycles and for every number of cycles up to 12, respectively. For the same family of circuits we show that the [[90,8,10]] and [[108,8,10]] cycles have circuit distance 8 for every number of cycles, extending values reported for fixed numbers of rounds; for [[72,12,6]] we confirm the known value 6. We also find that, for three of these codes, Google's Tesseract test circuits have the same detector-error-model columns as the depth-8 cycle once restricted to the detectors of the memory basis, for every number of rounds we checked, so circuit-distance results carry over between the two at those round counts. The lower bounds combine 607 SAT refutations, each with a DRAT proof, with exhaustive enumerations of extended-code logicals that carry no certificate. Finally, we independently confirm two results that others reported first: the [[360,12]] bivariate bicycle code has distance 24, and the left-right gross-code circuit of Strikis, Browne and Beverland, as reconstructed from their figures, has circuit distance 11. All code, formulas, proofs and fault sets are available at doi:10.5281/zenodo.23394575.
For the bivariate bicycle code [[288,12,18]] ("two-gross"), with the depth-8 syndrome-extraction cycle of Bravyi et al. (arXiv:2308.07915), this deposit gives explicit sets of 16 circuit faults that trigger no detector and apply a nontrivial logical operator. They exist for Z-type and X-type faults and under both readings of the CNOT term labels. One of them lies entirely inside the noise model and memory layout of Beverland et al., "Fail fast" (arXiv:2511.15177: Z-type noise, 18 noisy cycles and one fault-free cycle), whose search reported a lowest weight of 18. So the circuit distance of this cycle is at most 16. No lower bound is claimed, and no claim is made about logical error rates. Contents. A two-page note; the four fault sets listed gate by gate (cycle, timestep, gate, qubit coordinates, Pauli) so they can be checked by hand; a standalone checker (Python 3 + numpy) that builds the code and cycle from the polynomials and replays each set; tamper controls; logs. Checks. Two separately written circuit builds (one written from the papers only) reproduce the matrix sizes in Fail fast, Table 1, and agree on all four sets at cycles 1, 9 and 18; the gate order and indexing were checked against the text of the public BivariateBicycleCodes repository (inspected, not executed). A third propagator, written from the fault table and the papers only, is included. Removing any single fault makes detectors fire in every case tested. Use of AI. The author is not a professional researcher. The search, checks and drafting were carried out by AI models (Anthropic's Claude via Claude Code; OpenAI's ChatGPT as a reviewer) under the author's direction. The author is responsible for the content. No human expert has reviewed the work to date. Corrections are welcome.
Syndrome extraction for a CSS code has depth at least the Tanner-graph degree Δ, attained by edge colouring if gate order is ignored; correctness adds a parity constraint. We prove three general obstructions to depth-Δ extraction and show that, for every two-block group-algebra code with weight-three polynomials whose checks overlap in at most two qubits, no correct single-round circuit with one ancilla per check and one CNOT per Tanner edge has depth six: the minimum is seven, so the seven-layer cycle of Bravyi \emph{et al.} is optimal for their seven bivariate bicycle codes. This is the first analytic, family-wide impossibility result, extending prior instance-specific solver-based infeasibility results. The hypothesis cannot be dropped: a single orbit of four-qubit overlaps with a collision element of even order and 2-height at most eight makes depth six attainable, and a [[72,8,6]] code of this kind has a verified six-layer circuit. The refutations carry DRAT certificates. As a further application of the same scheduling scheme, we construct Stab-QRAM, a quantum random access memory for affine Boolean data that uses only CNOT and X gates and has the minimum depth possible for this gate set.
We construct three new weight-6 bivariate-bicycle (BB) codes—two [[336,20,12]] codes, inequivalent under the BB presentation symmetries, over Z₂ × Z₈₄ and a [[360,24,12]] code over Z₆ × Z₃₀—and certify exact distance 12 in both the X and Z sectors. They encode 25% and 50% more logical qubits than the best distance-12, weight-6 code reported by IBM's evolutionary search (k = 16) [Cruz-Benito et al., arXiv:2606.02418]. They are also the only ones: a symmetry-reduced enumeration of every connected weight-6 BB code with three-term polynomials, n ≤ 360 and k ≥ 20 eliminates 1,448 of the 1,584 candidates with checked logical operators of weight at most 11, and the remaining 136 presentations fall into three classes under group automorphisms, renormalisation and A ↔ B, represented by the new codes; within the polynomial ansatz of [Bravyi et al., Nature 627, 778 (2024)], every one of the 432 such candidates has distance at most 10. The same certificates settle the distances of the IBM ladder [Bravyi et al., Nature 627, 778 (2024)]—[[72,12,6]], [[90,8,10]], [[108,8,10]], the gross code [[144,12,12]] and the two-gross code [[288,12,18]], both sectors—and certify 19 ≤ d ≤ 24 for [[360,12]], whose distance we establish to be exactly 24 by exhaustive enumeration, resolving the upper bound reported in [Bravyi et al., Nature 627, 778 (2024)]. For the depth-8 syndrome-extraction circuit they certify circuit-level distance exactly 6 for [[72,12,6]] in both memory bases and at least 6 for every other code tested. Certified lower bounds are DRAT refutations for n ≤ 144 and replayable search trees above, whose single pruning lemma is proved in Lean 4; a reader replays any certificate with a standard-library Python verifier plus drat-trim or a 200-line C replayer, without the software that produced it. In a 1.1-million-shot circuit-level benchmark under one frozen BP-OSD decoder, the new codes' block failure probabilities at p = 0.004 are 3.6–6.2 times lower than the gross code's and 3.4–5.6 times higher than the two-gross code's, in both memory bases. Every computation—search, certification, replay and benchmark—ran on one laptop (Apple M3 Pro, 18 GB) with no GPU and no cluster. Contents of the release paper/ — the manuscript (LaTeX source, bibliography and PDF). certificates/ — one manifest per code and sector (<code_id>_<sector>.json), its standard-library Python verifier (verify_qec_<code_id>_<sector>.py), the proofs in proofs/ (DRAT proofs and gzipped proof trees), the problem files and C source of the tree replayer, the Lean 4 minimality lemma (qec_minimality.lean), the weight-6 envelope witnesses with their checkers, the distance table, FORMAT.md (the certificate format) and REPLAY.md (a clean replay of every certificate: 44 of 44 pass). simulation/ — the projected detector error models, the BP-OSD and BP-LSD evaluation scripts, every 2,000-shot chunk record with its seed, the decoder-tuning grid and the joint-model controls. SHA256SUMS — the SHA-256 of every file. Verification In certificates/: python3 verify_qec_<code_id>_<sector>.py rebuilds the code from its polynomials, checks the logical basis by rank, regenerates every CNF or problem file byte for byte and replays the proof; its last line is the verdict. DRAT rows need drat-trim (set DRAT_TRIM); tree rows need a C compiler. The Lean lemma checks with lake env lean qec_minimality.lean in a project with the pinned Lean 4 v4.33.1 and Mathlib 0df444a3 (exit status 0). Integrity: shasum -a 256 -c SHA256SUMS. Hardware Every computation — search, certificate production, replay and the benchmark — ran on one MacBook Pro (Apple M3 Pro, 11 CPU cores, 18 GB memory, macOS 15.5), with no GPU and no cluster. License: Licensed by SmartLedger Solutions under CC BY-NC 4.0 (LICENSE.md). Commercial use requires a written license from SmartLedger Solutions (labs@smartledger.solutions). The certificate producer is proprietary and not distributed; verification does not need it.
The distance of a classical or quantum code is a key figure of merit which reflects its capacity to detect errors. Quantum LDPC code families have considerable promise in reducing the overhead required for fault-tolerant quantum computation, but calculating their distance is challenging with existing methods. We generally assess the performance of a quantum code under circuit level error models, and for such scenarios the circuit distance is an important consideration. Calculating circuit distance is in general more difficult than finding the distance of the corresponding code as the detector error matrix of the circuit is usually much larger than the code's check matrix. In this work, we benchmark a wide range of distance-finding methods for various classical and quantum code families, as well as syndrome-extraction circuits. We consider both exact methods (such as Brouwer-Zimmermann, connected cluster, SAT and mixed integer programming) and heuristic methods which have lower run-time but can only give a bound on distance (examples include random information set, syndrome decoder algorithms, and Stim undetectable error methods). We further develop the QDistEvol algorithm and show that it performs well for the quantum LDPC codes in our benchmark. The algorithms and test data have been made available to the community in the codeDistance Python package.
Code and circuit distances characterize how strongly quantum error correction suppresses logical errors and guide code comparisons, schedule choices and resource estimates. Yet many published distances are only upper bounds from heuristic searches. Here, we express both distances as tropical partition functions and evaluate them with rank-based dynamic programming and weight-bounded search. Circuit reductions and relaxations shrink the problems, and proved lower bounds with explicit logical errors of matching weight certify exact values. On two benchmarks, our method certifies more instances than all other tested exact solvers combined. We certify distances for codes with up to 31,200 qubits, such as Google's modular hyperbolic codes, and for memory circuits with up to 8×105 fault mechanisms, such as those of high-rate Cornucopia codes. Our audit of 26 recent papers identifies overestimated distances and non-tight upper bounds in 10 papers, with a maximum discrepancy of at least eight. Certified distances also guide circuit design and show which defective chips remain usable. In particular, we certify circuit distance 16 for IBM's two-gross memory circuits with the standard schedule, not the assumed 18, whereas redesigned schedules reach 18 at one noisy cycle, at the cost of one more timestep per cycle. At 99.98% junction yield, keeping gross-code memory blocks with dead checks that are certified at distance 10 or more raises the estimated usable share from 53% for perfect blocks alone to 94% with an improved schedule.
We present an analytical theory of greedy peeling decoding for bivariate bicycle (BB) codes under circuit-level noise. The deferred greedy decoder achieves 330x latency reduction over belief propagation (BP) at p = 10^{-3} while maintaining identical logical error rate. Our main theoretical contribution is a closed-form collision resolution factor A_0 = |true collisions| / |birthday collisions|, derived from XOR syndrome analysis with no free parameters, that quantifies the fraction of detector-sharing fault pairs genuinely blocking iterative peeling. For the [[144,12,12]] Gross code, A_0 = 0.8685 (within 0.5% of the empirical value), with shared-2 pairs (4-cycles) always resolving under peeling. We show A_0 depends on the mean fault-graph degree d-bar rather than code size: A_0 = 0.87 for d-bar = 52 (Gross family) versus A_0 = 0.76 for d-bar = 17 ([[32,8,6]]). We establish a syndrome code stopping distance d_S = n/4.5 for the Gross family and demonstrate that [[32,8,6]] (d_S = 4) enables two-shot streaming decoding: T = 2 rounds achieve 89% peeling success with 1.29 +/- 0.03 LER ratio versus T = 12, at estimated latency ~50 ns. The full formula P_peel = exp(-A_0 * gamma_analytic * exp(-BTp) * n * p^2) is validated across five BB codes, four noise levels, and four values of T with R^2 = 0.86. Cross-platform reproduction of the Kunlun [[18,4,4]] experiment matches their hardware LER within 0.73 percentage points.