BB codes are unusually easy to misname because the same matrices sit inside several broader constructions. This dictionary records what each view makes transparent, which data it preserves, and which inference it does not license.
| view | basic object | best question | preserved invariant | caveat |
|---|---|---|---|---|
| bivariate polynomial | \(A(x,y),B(x,y)\) in a finite quotient ring | How do four parameters build the checks? | exact matrix in the chosen shift convention | a compact polynomial does not certify distance or circuit layout |
| group algebra | two elements of \(\mathbb F_2[\mathbb Z_l\times\mathbb Z_m]\) | Why do the blocks commute, and what is the parent family? | regular-action algebra and two-block structure | nonabelian 2BGA codes need not be bivariate |
| lifted-product / chain complex | \(R\to R^2\to R\) | Where do logical quotients live? | boundary maps and middle homology | “lifted product,” HGP, and BB are not synonyms |
| Tanner graph | check-variable incidence | Which couplers and degrees are required? | row/column support | incidence omits CNOT order, ancillas, detectors, and fault propagation |
| quotient module / roots | \(R/(A,B)\) and common factors | Why does rank drop, and what happens at a boundary? | image ideal and quotient dimension | common roots alone do not give distance |
| anyon / Koszul | excitation modules and mobility relations | What topological sectors and constraints exist? | ideal/complex data translated into excitations | finite degeneracy need not be a trivial toric stack |
| toric cylinders | symmetry-reduced decoding problems | When can matching replace a generic qLDPC decoder? | selected syndrome/logical sectors under the reduction | operational decoder equivalence is not global local-Clifford equivalence |
The monomial \(x^a y^b\) labels the group element \((a,b)\in\mathbb Z_l\times\mathbb Z_m\). Acting by the regular representation turns group multiplication into matrix multiplication. Since the group is abelian, the regular matrices for \(A\) and \(B\) commute.
This translation explains the relation row:
wang2023abelian develops the abelian/nonabelian two-block setting, while lin2023twoblock places 2BGA codes inside lifted products. The containments are not equalities:
The maps \([B;A]\) and \([A|B]\) are \(R\)-linear. Their composition vanishes, so they define a length-two complex. Forgetting \(R\)-linearity gives the binary CSS matrices; retaining it exposes translations and quotient modules.
The two descriptions preserve the exact code when the same bases and regular representation are fixed. Changing bases can preserve a code space while changing row/qubit order; that distinction matters for the Atlas replay contract, which requires exact matrices rather than equivalence.
The image of \([A|B]\) is the ideal \((A,B)\). The quotient \(R/(A,B)\) therefore measures one rank defect. Factorization and common-root methods can predict when this quotient is nonzero and how its dimension changes with periods or boundaries. postema2025existence uses this structure to characterize existence and asymptotic limitations; wang2026bbs extends the root/boundary analysis.
Rank is still not distance. A quotient class can have a low-weight representative, or every representative may be spread across the torus. Distance asks for the lightest nontrivial class after all stabilizer reductions.
Every nonzero entry of \(A\) or \(B\) is an edge between one check and one data qubit. The gross three-plus-three terms give weight-six checks and three X plus three Z incidences per data qubit. bravyi2024bicycle shows how this graph splits into two planar degree-three coupling layers.
The graph is necessary but insufficient for a fault circuit. A Tanner edge says two systems interact; it does not specify:
shaw2025morphing changes the instantaneous code and graph, illustrating why connectivity reduction cannot be described as a mere redrawing of the same static matrix.
chen2025anyon interprets polynomial/Koszul data as excitation types, mobility constraints, and topological frustration. This picture asks which defects can move under local operators and how global torus constraints create logical sectors.
The caution is essential: the existence of \(k/2\) quotient dimensions does not mean the Hamiltonian is simply \(k/2\) decoupled toric codes. Logical degeneracy, local-Clifford phase equivalence, and decoder decomposition are different equivalence relations.
sahay2026matching exploits symmetries to expose toric-like cylinders and apply MWPM, then combines the result with BP and over-matching. The preserved object is the selected decoding problem: syndromes and logical ambiguity under the reduction.
This does not contradict topological frustration. A decoder may map parts of an inference problem to familiar cylinders without producing a finite-depth local unitary that decomposes the entire BB Hamiltonian into toric copies.
Use:
Calling all of them “the BB representation” erases the question each translation was built to answer.