Optimal check weight for two-logical-qubit codes of lengths eleven and twelve
Weight five is necessary and sufficient in both distance-four cases, with an analytic eleven-qubit converse and a complete classification-based twelve-qubit exclusion.
Yuxuan Zhang · Version 1.0, 17 November 2026
Full manuscript (PDF) · Source and executable certificate (ZIP) · Release metadata
A stabilizer code can have many different generating sets. The lightest presentation minimizes the largest Pauli support among the generators. We determine this optimum for codes encoding two logical qubits at distance at least four, using either eleven or twelve physical qubits:
\[W_{\mathrm{opt}}(11,2,d\ge4)=W_{\mathrm{opt}}(12,2,d\ge4)=5.\]The minimization includes all binary stabilizer codes and all generating bases, including degenerate and non-CSS codes. Both entries were left at 4–6 in Wei et al., Theory of low-weight quantum codes, version 2, Table 2. This result closes those two finite intervals.
Why four is not enough
At eleven qubits, an exact shadow-enumerator identity bounds the number of short stabilizer elements. Nine independent generators of weight at most four would saturate that bound. A weight-one generator would shorten to an impossible ten-qubit code; weight-three elements are excluded by equality. All nine generators would therefore have even weight. Weight parity is linear on a commuting Pauli space, so the entire stabilizer would have even weight, contradicting another equality condition in the same certificate. The manuscript gives the two exact identities and the shortening argument.
At twelve qubits, the converse uses the Danielsen–Parker classification. Every possible stabilizer extends to a maximal commuting Pauli space, and each such parent is equivalent to a classified graph code. Inside a fixed parent, all candidate stabilizers are indexed by binary two-dimensional subspaces. Two explicit syndrome tests enforce distance at least four; one of them handles low-weight stabilizers and is needed for degeneracy. A rank test then decides whether any weight-four generating basis exists.
The complete calculation reads 1,323,363 parent representatives, including decomposable codes. After a necessary parent-rank test, it examines 320,777,184 admissible subspaces. None has the required rank ten among its weight-at-most-four elements. The largest rank found is nine. This is an exhaustive finite exclusion using a published classification, rather than a timeout or a search restricted to one family of displayed generators.
Matching constructions
The manuscript lists a weight-five generating basis at each length. Exact enumeration checks commutation, independence, stabilizer signs and the entire symplectic dual. Both codes have distance exactly four: their numbers of weight-four logical Paulis are 60 and 54, respectively.
Appending a fixed qubit to the eleven-qubit construction also proves the twelve-qubit upper bound and supplies a degenerate positive control. A separate twelve-qubit construction without weight-one stabilizers is listed as well.
Reproduction and scope
Extract the source package into an empty directory and run:
python3 reproduce.py
Python 3.10 or later and a C++17 compiler are sufficient. The package includes the classification input with its source and checksum, so the checks need no network connection. The runner recomputes the witnesses, analytic identities and controls, then reruns the complete twelve-qubit exclusion.
The source comparison also checks the two public QEC Challenge distance-four constructions. For each, the span of all stabilizer elements of weight at most five has rank six, while weight six gives rank ten. Their actual optimum is six; changing the displayed basis does not already produce the new upper bound.
The contribution is a pair of exact finite optima. Shadow enumerators, shortening, graph-code classification and rank tests are inherited methods. The result does not settle the general asymptotic question about low-weight quantum codes.
Review status. The full proof received sequential internal review, and three separately implemented classification calculations agreed. This is not independent expert review or formal verification. The twelve-qubit converse depends on the published self-dual classification. The checked primary sources support a distinct contribution; worldwide first priority has not been established. This website release is not a claim of journal or problem-catalog acceptance.
The author directed the research and revision. AI assistance contributed to candidate derivations, literature comparison, implementation, checking and drafting. The manuscript records the same scope and review status.