ResearchQuantum error correctionGSI-R-0011 · Note · Exact value

Two non-interleaved schedules and the period-11 schedule of G. Pan (to 40 cycles) give [[288, 12, 18]] memory circuits of circuit distance 18.

Circuit distance is the fewest faults that flip a logical observable undetected. Two non-interleaved schedules of ours, C1 and C2, and G. Pan’s period-11 gross-code schedule (up to 40 noisy cycles) give [[288, 12, 18]] memory circuits of distance 18 in both bases. A tesseract-decoder [[360, 12, 24]] memory circuit (24 rounds, SI1000 noise) has X-basis distance at least 19.

Before this work
Bravyi et al. [1] print dcirc ≤ 18 for [[288, 12, 18]] and list [[360, 12, ≤ 24]]. Cheng, Guo and F. Pan [2] report schedules reaching 18; dist-m4ri [3] lists [12, 23] for the 360-qubit X-basis circuit, nothing for Z (read 15 November 2026).
Evidence
One engine. Exhaustive computations by programs of ours. A second program, written separately, reproduces C1 in full and part of the period-11 result, and nothing of C2 or the 360-qubit circuit. The two share a method family and its machine-unchecked correctness argument, our circuit builders and Stim. No review outside the lab has taken place.
What you can check
One half: C1 and C2. Their extended codes build in under a second with extended_code.py; an exact solver of your own must show that no 17 or fewer of the 720 columns sum to zero syndrome and nonzero observable action.
Not established
That every undetected logical fault set has at least 18 faults (19 for the 360-qubit circuit): the exclusions are unreleased exhaustive computations of ours, not certificates others can check.
Method
Not published.

What was known

Strikis, Browne and Beverland [4] prove that a non-interleaved single-ancilla CNOT circuit has circuit distance equal to that of its extended code: the code’s check and observable matrices with one column added per hook fault, a single fault on a check’s ancilla that spreads to the data qubits of the check’s later CNOTs. For the gross code [[144, 12, 12]] they bound all non-interleaved circuits by 11 (Proposition 4).

SourcePrior resultTimesteps per cycle
Bravyi et al. [1]Standard schedule of [[288, 12, 18]]: dcirc ≤ 18. Distance 16 is certified in [2] and in the two-gross note.8
Cheng, Guo and F. Pan [2], 5 November 2026Of 8,496 uniform schedules of seven CNOT layers per cycle (one gate order for all checks of a type), 84 have distance 18 at one noisy cycle; for one of them 17 or fewer faults are excluded at two cycles and 16 or fewer at three.9
G. Pan [5], release 2.0, 24 October 2026Period-11 circuits of circuit distance 12 for the gross code and 10 for [[90, 8, 10]], their code distances; no result for [[288, 12, 18]].11
Timesteps per cycle count reset and measurement; the eight of the standard schedule and the nine of [2] are from [2].

To our knowledge (searched 15 November 2026), [2] is the earliest dated public report of schedules reaching 18 for this code.

The result

C1 and C2

For both error types the extended code of C1 and of C2 has distance 18: no set of 17 or fewer of its 720 columns has zero syndrome and nonzero observable action. By Theorem 1 of [4], in its noise model, their non-interleaved memory circuits have circuit distance 18 in both bases for every number of cycles.

terms (ell = m = 12):   A1 = x^3   A2 = y^2   A3 = y^7   B1 = y^3   B2 = x   B3 = x^2
P  = A1 A3 A2 B2 B1 B3
P' = A1 A3 A2 B3 B1 B2
Q  = A1 A2 A3 B1 B2 B3
C1: X check (a, b) and Z check (a, b) use order P if a + b is even, else Q
C2: X check (a, b) and Z check (a, b) use order P' if a is even, else Q
e(x^i y^j) = (i, j)
X check (a, b): left qubit (a, b) + e(A_k), right qubit (a, b) + e(B_k)
Z check (a, b): left qubit (a, b) − e(B_k), right qubit (a, b) − e(A_k)
Orders run first CNOT to last. Single-ancilla: one ancilla per check. Non-interleaved: in every cycle all X-check CNOTs precede all Z-check CNOTs. Interleaved layerings of these orders have circuit distance at least 18 by inequality (12) of [4], at most 18 by the code distance.
Period-11 schedule

The period-11 gross-code schedule of G. Pan [5], applied by term name to [[288, 12, 18]], has circuit distance 18 in both bases for 1 to 40 noisy cycles, and for 41 to 45 given the time identity that the detector error model of R + 1 cycles is that of R cycles with one cycle repeated (checked for R ≤ 44).

{"P": 11,
 "X": {"reset": 0,  "order": "A2A1A3B3B2B1", "steps": [1, 2, 3, 4, 5, 9], "meas": 10},
 "Z": {"reset": 10, "order": "B1B2B3A3A1A2", "steps": [0, 4, 5, 6, 7, 8], "meas": 9}}
Transcription of u11_spec.json of [5]. Per check type: step modulo 11 of the reset, the six CNOTs in the order given, the measurement; a step below the reset step falls in the next period. Z round j starts one step before X round j.

Period-11 noise: two-qubit depolarising faults after every CNOT, single-qubit faults on every idle qubit, flipped resets and measurements, noise-free data preparation and readout. Our builder is not released; on the gross code it gives the circuit distance 12 listed in [5].

360-qubit circuit

The X-basis memory circuit of [[360, 12, 24]] in the test data of tesseract-decoder [6] (24 rounds, SI1000 noise [7]) has circuit distance at least 19, so with [12, 23] listed for it in [3] it lies in 19 to 23. Z basis: at least 16. We know of no circuit-level result for this code published by IBM.

r=24,d=24,p=0.001,noise=si1000,c=bivariate_bicycle_X,nkd=[[360,12,24]],q=720,iscolored=True,A_poly=x^9+y+y^2,B_poly=y^3+x^25+x^26.stim
File name in testdata/bivariatebicyclecodes. The benchmark [3] circuit is this file at p = 0.0001, differing in noise strengths only; circuit distance does not depend on p. The Z-basis file has c=bivariate_bicycle_Z. Code distance: at most 24 in [1], certified 24 in [2].
ResultChecked by
C1, C2: no 17 or fewer columns, both error typesFirst program: all four extended codes. Second program: C1 only.
Period-11: distance 18 to 40 cycles, to 45 given the time identityFirst program: to 40 cycles. Second program: to 20 cycles. A third solver: to 17 cycles. Both on the first program’s inputs. Upper bound replayed at 1, 2, 5, 17 and 30 cycles.
360-qubit: X at least 19, Z at least 16X: first program, weights to 18. Z: a reduced problem only.
Evidence by result. Period-11 upper bound 18: a weight-18 data logical operator fires no detector at any cycle count.
ProgramGrowth per unit of weightMeasured as
Ours, 2 programs sharing a method4.9 to 5.0search nodes, weights 14 to 15, both bases
Search code of [5], release 2.0, run by us10 to 11CPU time on one thread, Z basis, weights 7 to 10, at machine load 60 to 82
Two-gross memory circuit of GSI-R-0003, measured by us (Internal level). One program of ours decided distance 16 in 85.0 s (Z; 108 s on repeat) and 78.7 s (X), excluding 5.6 s preparation, on 16 threads of a loaded shared machine. Not compared with [2].

How to check it

python3 extended_code.py C1.design
X-type errors: 144 detector rows, 12 observables, 720 columns (288 data, 432 residual)
Z-type errors: 144 detector rows, 12 observables, 720 columns (288 data, 432 residual)
C2.design: same sizes. Residual: per check, hook errors on 2, 3 and 4 data qubits; those on 1 and 5 equal data-qubit columns modulo the check. A second argument writes the columns, which our programs’ extended codes match up to renumbering of detectors and observables.

An exact solver of your own must find no set of 17 or fewer columns with zero syndrome and nonzero observable action in either extended code; one such set refutes the schedule. The other two results need circuits rebuilt in Stim from the files below.

What it changes

Non-interleaved circuits reach the code distance of [[288, 12, 18]]; for the gross code they cannot [4].

Limits

  • The exclusions. Unreleased exhaustive computations by our own programs. A certificate or a rerun by others would close it.
  • Period-11. Beyond 45 cycles the result is assumed; a machine-checked time identity would close that. The value is for the round pairing above, as in schedule.py of [5]; the unshifted pairing has undetected logical sets of 16 faults at the ends.
  • Z basis, 360-qubit circuit. The bound 16 comes from a reduced problem whose own distance is 16; a direct search shows at least 14. A direct exclusion of 15 faults would close it.
  • Cost. As non-interleaved circuits C1 and C2 need at least 12 CNOT layers per cycle (7 in [1]); the depth of interleaved layerings is not recorded.
  • Reduction. Theorem 1 of [4] is not re-proved; a direct search of C1 at 2 cycles excludes 12 or fewer X-type faults. A direct search at 17 faults would close it.

Files

  • C1.designCNOT order of each of the 288 checks of C1
  • C2.designCNOT order of each of the 288 checks of C2
  • extended_code.pybuilds the 720-column extended code of each error type from a design file; Python 3, no packages
  • u11_spec.jsonperiod-11 gross-code schedule of [5], transcribed above; public file of another author
  • 360-qubit X-basis circuitStim circuit of the third result; public data of another group

References

  1. [1]S. Bravyi, A. W. Cross, J. M. Gambetta, D. Maslov, P. Rall, T. J. Yoder, High-threshold and low-overhead fault-tolerant quantum memory, Nature 627, 778–782 (2024), arXiv:2308.07915v2: Tables 1 and 3. Link
  2. [2]B. Cheng, N. Guo, F. Pan, Certifying quantum code and circuit distances at scale, arXiv:2611.08247v1 (5 November 2026): main text, Supplementary Tables 13 and 14. Link
  3. [3]dist-m4ri (GitHub repository QEC-pages/dist-m4ri), benchmark/BENCHMARK.md at commit e9cac19 (5 November 2026, the last change to the file; read 15 November 2026). Link
  4. [4]A. Strikis, D. E. Browne, M. E. Beverland, High-performance syndrome extraction circuits for quantum codes, arXiv:2603.05481v2 (14 September 2026): Definitions 2 and 3, Theorem 1, Proposition 4. Link
  5. [5]G. Pan (GitHub account GavinNull; the repository page names Guancheng Pan as author), gross-code-distance-12 (GitHub repository): release 2.0 (24 October 2026) has period-11 circuits; release 3.0 (8 November 2026) supersedes it and gives period 10 for the gross code; the README and release notes do not mention [[288, 12, 18]] (read 15 November 2026). Link
  6. [6]quantumlib/tesseract-decoder (GitHub repository), directory testdata/bivariatebicyclecodes, unchanged since commit 98c663d (1 March 2025): Stim circuits of bivariate bicycle codes with noise=si1000, including nkd=[[360,12,24]] with r=24 (listing read 15 November 2026). Link
  7. [7]C. Gidney, M. Newman, A. Fowler, M. Broughton, A fault-tolerant honeycomb memory, Quantum 5, 605 (2021), arXiv:2108.10457: definition of the SI1000 noise model. Link

Cite asKishan Kumar, Circuit distances of bivariate bicycle syndrome schedules, GSI-R-0011, https://generalised.si/research/bivariate-bicycle-circuit-distances, 2026.

Published16 November 2026

History
  • G. Pan publishes release 2.0 of the gross-code repository with period-11 circuits [5].
  • Cheng, Guo and F. Pan post a preprint with schedules of distance 18 at one noisy cycle for the two-gross code [2].
  • G. Pan publishes release 3.0, which supersedes release 2.0 and gives period 10 for the gross code [5].
  • This account published.

All results