Search-based and learning-based pipelines now publish catalogs of quantum error-correcting codes and of optimised hardware designs. The reported code distances and design equivalences are usually obtained by bounded or heuristic methods. We re-examine two such releases with exact methods: a catalog of 467 bivariate-bicycle-type codes, and a set of ShinkaEvolve-optimised RTLLM Verilog designs. The great majority of claims are confirmed. We identify a small number of cases where the exact value differs from the reported one, give certified replacement codes, and report exact distances for several codes whose published distances were lower bounds. We also report a null result for a selection-concentration hypothesis.
Search-based and learning-based pipelines now publish catalogs of quantum error-correcting codes and of optimised hardware designs. The reported code distances and design equivalences are usually obtained by bounded or heuristic methods. We re-examine two such releases with exact methods: a catalog of 467 bivariate-bicycle-type codes, and a set of ShinkaEvolve-optimised RTLLM Verilog designs. The great majority of claims are confirmed. We identify a small number of cases where the exact value differs from the reported one, give certified replacement codes, and report exact distances for several codes whose published distances were lower bounds. We also report a null result for a selection-concentration hypothesis.
The distance of a quantum code is a fundamental metric that dictates its error-correction capability. While quantum low-density parity-check (qLDPC) families such as Bivariate Bicycle (BB) codes offer a promising pathway to low-overhead fault-tolerant quantum computing, computing their distance is challenging. Current exact solvers and even search heuristics that provide a distance upper bound suffer from severe computational bottlenecks, restricting the discovery of optimal codes. In this paper, we construct a dataset of BB codes with corresponding distance upper bounds obtained through high-budget heuristic searches. We introduce BB-Net, an attention-based neural network architecture for predicting the distance of BB codes. It achieves 96.39% accuracy on the reference distances in the test set, compared with 94.86% for state-of-the-art heuristics with a budget of up to 104 times more computational runtime. On a subset with certified exact distances, BB-Net achieves 98.94%. By relieving the distance-calculation bottleneck, BB-Net enables faster screening of large code spaces, particularly in ML-driven search pipelines, and thereby can accelerate the discovery of novel quantum error-correcting codes, an important capability for advancing scalable fault-tolerant quantum computing. Our code is publicly available: https://github.com/NadavSc/BBNet
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.
Quantum error-correcting codes (QECCs) sit between noisy quantum hardware and reliable computation, so the code parameters used in practice must be trustworthy. The single number that summarizes a code's strength is its distance, yet certifying a distance lower bound is NP-hard in general, placing it beyond the reach of pen-and-paper proofs as well as direct proof-assistant scripting. As a result, distance values in the literature come either from non-scaling hand proofs, or from unverified solvers that leave a trust gap exactly where the code is supposed to provide a guarantee. We present Lean-QEC, the first Lean 4 formalization of stabilizer-code theory that delivers end-to-end, machine-checked distance certificates at industrial code sizes. Lean-QEC formalizes the linear algebra of qubit states, the Pauli group, stabilizer codes, the binary symplectic representation, classical coding theory, and the CSS and Bivariate Bicycle families. To break the combinatorial barrier, Lean-QEC translates the distance condition into a Boolean satisfiability formula through a verified reduction. The pipeline scales through a BitVec-flattened encoding that replaces Lean's Matrix representation, and an error-location encoding that reduces the variable count from n to k⌈log2n⌉. With these, we obtain automatically-generated Lean-checked distance proofs for a large range of industrially viable qLDPC codes within the Bivariate Bicycle and Generalized Bicycle families, including [[90, 8, 10]] and [[70, 6, 9]] BB codes, with the formulation scaling up to 144 qubits when performed outside the Lean kernel. The resulting library is reusable and is designed to plug into broader Lean-based efforts toward end-to-end verification of fault-tolerant quantum computation.
Quantum error correction protects quantum information by encoding it in a code described by three numbers: how many physical qubits it uses, how many logical qubits it protects, and the smallest error it cannot detect. The last is an optimum over exponentially many candidates, so it is read off a solver rather than checked. Here we show that code parameters can be certified by a short certificate, a list or a pairing or a symbolic instance, whose correctness the kernel of a proof assistant decides by computation, and whose size is itself a theorem rather than an estimate. Two published code families have distances settled here: the bivariate bicycle code's [[144,12,12]], certified, and the affine-permutation codes printed at [[1152,580,≤12]] and [[2304,1156,≤14]], settled at twelve and fourteen. Eleven code families and thirty-nine parameter sets follow, with nothing beyond three standard axioms trusted and the pipeline's specification step preset to twenty columns.
Quantum low-density parity-check (qLDPC) codes reduce the qubit overhead of fault-tolerant quantum computation by an order of magnitude, but their decoding is harder than its classical counterpart: because many physical errors are equivalent up to stabilizers, the degenerate maximum-likelihood (ML) decoder must compare the probabilities of entire equivalence classes of errors, that is, partition functions, rather than single errors. The workhorse decoder BP+OSD sidesteps degeneracy heuristically and offers no guarantees. We treat degenerate decoding as probabilistic inference in an undirected graphical model: the probability of each logical class is the partition function of an unconstrained, strictly positive Markov random field over the code's check variables, a construction that generalizes the random-bond Ising mapping of the surface code to arbitrary CSS codes and to spacetime decoding with measurement errors and circuit-level noise. On this model we build two decoders. The first estimates all class partition functions by annealed importance sampling with common random numbers and attaches to every decision a certificate of optimality: a paired bootstrap test, or, composed with constant-factor estimators such as WISH, an exact optimality proof. The second is region-based: the Bethe free energy, whose bias cancels between classes, reproduces exact ML decoding on every tested surface-code instance at millisecond cost, and enlarging the regions to elimination clusters makes exact degenerate ML decoding of the [[72,12,6]] bivariate bicycle code feasible. Across surface codes and the bivariate bicycle codes [[72,12,6]] and [[144,12,12]], under code-capacity, phenomenological, and circuit-level noise, the sampling decoder matches or exceeds BP+OSD while certifying the bulk of its decisions, and the certificate flags exactly the syndromes on which any fast decoder should be distrusted.
Computing the minimum distance of a quantum LDPC code is easy in one direction and expensive in the other: exhibiting a light logical operator is a search, while proving that no lighter one exists is what actually decides the parameter. This note proves an orbit reduction for the second half. If a group of code automorphisms acts on the qubits with r orbits, the minimum distance is the smallest of r restricted problems, one per orbit representative, so the number of solver calls is the number of orbits and not one. We give the a priori condition under which restricting to a single qubit is legitimate, namely transitivity on all qubits, and a counterexample showing what happens without it: on a hypergraph product code with n = 27 and d = 3, the widely used tool dist-m4ri prints a pair of equal bounds, which its own documentation calls a confirmed exact distance, and that value is 6. The reduction is coupled to a proof producing pipeline in which every nonexistence claim is re-verified by an independent checker, drat-trim or, for the largest closure, cake_lpr, whose soundness is proved inside CakeML. This closes with replayable certificates the distances of the five smaller bivariate bicycle codes of Bravyi et al., up to and including the code with n = 288 and d = 18, whose exactness was published with no replayable artifact and is established here at both ends. For the two larger codes it yields the first lower bounds a third party can re-check from the deposited files: d at least 18 for the code with n = 360, against a published upper bound of 24 and against d at least 11 for the best previously re-executable method, and a first lower bound of any kind for the code with n = 756. A separate and much cheaper ingredient, a sweep over logical operators supported on unions of cosets of a subgroup, tightens two upper bounds printed in a recent table of two block group algebra codes. The code printed there as [[140, 8, at most 16]] carries a logical operator of weight 14; the code printed as [[144, 10, at most 16]] has distance exactly 12, established at both ends and so closed four below the published bound. Both witnesses are re-checked outside any solver by two rank computations over the field with two elements, so neither of these two upper bounds rests on a solver at all. Both entries are printed with the inequality sign in that reference, which states no method for the tables they sit in; nothing it proves is contradicted here, and no error of its authors is corrected. What is improved are two bounds they offered as bounds. Finally we show that the two instance form of the reduction is a consequence of commutativity and not of the two block structure. On two block group algebra codes over non-abelian groups, taken from the published catalogue, the translation subgroup has two orbits for one code and twelve for another, so the number of solver calls has to be computed from the group before any search rather than assumed. On three such codes we certify the minimum distance. The argument that makes one orientation enough in the abelian case does not apply there, and we check that it does not, so both orientations are certified separately rather than assumed equal; on these three codes the two turn out to coincide, which is a measured fact about each code and not a consequence of that argument. The published values for these codes come from randomised search, and a certificate of nonexistence a reader can replay is, to our knowledge, new for non-abelian codes of this family. What this work does not claim is stated explicitly in the note: the restriction itself is not our idea, it is an option of dist-m4ri present since 2020; a machine checkable chain covering a symmetry breaking step is not a first; our proofs are checked by an independent checker and not by the kernel of a proof assistant; and the encoding step is not formally verified. A public question asking whether the orbit reduction is already in the literature was posted on MathOverflow on 28 August 2026 and carried no answer at the time of deposit. Every table of numbers in the note is generated from the deposited artifacts by a script that refuses to emit a row it cannot source. 27 pages. Verification scripts and certificate digests included as separate archives. MSC 2020: 81P73, 94B05, 68Q25, 20C05.