Self-Dual BB Codes

• Self-dual BB codes are self-dual error-correcting codes that merge quantum CSS constructions with classical LDPC features to support transversal gate operations.
• They employ circulant matrices and stacking methodologies to ensure commutation and self-orthogonality, enabling efficient fault-tolerant syndrome extraction.
• The codes achieve high logical rates and competitive pseudo-thresholds, making them promising for scalable quantum architectures and robust classical coding.

Self-dual BB (Bivariate Bicycle) codes are a family of quantum and classical error-correcting codes that combine the structure of low-density parity-check (LDPC) codes with robust topological and transversal gate properties. Developed in the context of CSS (Calderbank-Shor-Steane) codes, these codes exhibit self-duality, high code rates, low stabilizer weights, and support transversal Clifford operations. Self-dual BB codes play a pivotal role both in quantum information processing—enabling fault-tolerant architectures with efficient logical gates—and in classical coding theory as extremal self-dual binary codes for various lengths.

1. Formal Structure and Self-Duality

Self-dual BB codes are constructed as translation-invariant CSS codes on two-dimensional tori. The defining data consist of two binary matrices $A,B\in\mathbb{F}_2^{N\times N}$ ($N = l\cdot m$), usually built from products of commuting circulant matrices. These generate stabilizer groups by shifting a polynomial "seed" over all lattice translations. In the standard CSS formalism, the code is self-dual if the $X$ and $Z$ stabilizer structures become identical up to relabeling: $H_X = H_Z$ This property enables transversal logical Hadamard and phase ($S$) gates, as each stabilizer is mapped onto itself (or its dual) under these Clifford operations (Liang et al., 6 Oct 2025).

For quantum codes, the self-duality requirement is reflected in the polynomial language by $g(x, y) = \overline{f(x, y)}$, where $\overline{(\cdot)}$ denotes variable inversion. This ensures that $H_XH_Z^T=0$ and that the stabilizer generators are simultaneously doubly-even and self-orthogonal (Liang et al., 6 Oct 2025). In classical settings, self-duality is defined by the property $C = C^\perp$ under standard dot product, where $N = l\cdot m$0 is the code subspace.

2. Construction Methodologies

The canonical bivariate bicycle (BB) construction starts with circulant blocks: $N = l\cdot m$1 where $N = l\cdot m$2 denotes the circulant matrix for the Laurent polynomial $N = l\cdot m$3. Self-duality mandates $N = l\cdot m$4.

A significant recent development is the "stacking" procedure for constructing self-dual quantum LDPC (qLDPC) codes from non-self-dual BB codes. Given matrices $N = l\cdot m$5, one builds a $N = l\cdot m$6 block check matrix: $N = l\cdot m$7 This guarantees that $N = l\cdot m$8 (self-duality) and $N = l\cdot m$9 (commutation), as required for CSS codes. The corresponding polynomial seed formalism utilizes a formal layer variable $X$0 to express the stacked stabilizers: $X$1 The code parameters—length $X$2, number of logical qubits $X$3, and distance $X$4—are then computed via polynomial module quotients and distance searches (Liu et al., 17 Feb 2026).

The construction is highly modular:

• Classical circulant-product codes can be used as seeds.
• Non-self-dual LDPC codes can be stacked by matrix or polynomial methods.
• Lattice embeddings (including twisted torus boundary conditions) expand the family of possible code parameters and locality features (Liang et al., 6 Oct 2025, Liu et al., 17 Feb 2026).

3. Code Parameters and Representative Families

Self-dual BB codes are characterized by $X$5 (quantum $X$6 or classical $X$7). Notable instances include:

• $X$8, $X$9, $Z$0, $Z$1 (see exhaustive enumeration for $Z$2) (Liang et al., 6 Oct 2025).
• Double-layer (stacked) BB codes such as $Z$3, $Z$4, $Z$5, generated by stacking particular pairs of commuting circulant-product matrices (Liu et al., 17 Feb 2026).

The code distance typically scales as $Z$6, and for best-performing codes, the Bravyi–Poulin–Terhal bound $Z$7 is saturated to within constant factors (Liang et al., 6 Oct 2025). Logical rates are significantly higher than for topological codes like the surface or color code.

A representative table:

$Z$8	Polynomial Seed $Z$9	Boundary Twist
64, 8, 8	$H_X = H_Z$0	$H_X = H_Z$1
120, 8, 12	$H_X = H_Z$2	$H_X = H_Z$3
160, 8, 16	$H_X = H_Z$4	$H_X = H_Z$5

Each code supports exactly weight-8 stabilizer generators embedded locally on a honeycomb lattice.

4. Logical Gate Implementation and Transversal Clifford Operations

Self-dual BB codes support transversal implementation of the entire Clifford group:

• Transversal CNOT: Acts as logical CNOT across all logical qubits pairwise.
• Transversal Hadamard: Swaps $H_X = H_Z$6 and $H_X = H_Z$7 stabilizers and logicals, exploiting the CSS and self-duality structure.
• Transversal $H_X = H_Z$8 (phase): Preserves doubly-even $H_X = H_Z$9-type stabilizers and acts as a logical phase gate, as all weight-8 stabilizers incur phases $S$0.

By virtue of these properties, self-dual BB codes are the only known high-rate qLDPC codes simultaneously admitting all Clifford logicals by weight-preserving transversal gates (besides color codes and minor variants) (Liang et al., 6 Oct 2025).

5. Numerical Performance and Fault Tolerance

Simulations of double-layer self-dual BB codes under full circuit-level depolarizing noise (including data/ancilla gates and readout) indicate pseudo-thresholds in excess of $S$1. For example:

• $S$2: $S$30.6%
• $S$4: $S$50.75%
• $S$6: $S$70.7% The logical failure rate drops sharply with code size and distance, and the codes already outperform classical hypergraph-product qLDPC constructions of similar lengths (Liu et al., 17 Feb 2026). These properties make self-dual BB codes attractive candidates for scalable quantum memory and fault-tolerant Clifford computation.

6. Comparison to Related Constructions and Applications

Self-dual BB codes stand at the intersection of three important code families:

• LDPC/Bicycle codes: Retain high rates, low-weight stabilizers, and translation invariance.
• Color codes: Share self-dual structure and Clifford gate transversality but are restricted to lower rates/higher overhead with growing code distance.
• Hypergraph-product and surface codes: Self-dual BBs achieve better constant rates and distances for similar resource budgets.

The codes also enable advanced operations such as:

• Lattice surgery with flexible boundary manipulation.
• Fast syndrome extraction and efficient matching decoders.
• Specializations to magic-state distillation using triorthogonal subspaces (see the construction of triorthogonal matrices from self-dual codes in (Shi et al., 2024)).

Open problems involve decoder optimization, thresholds under various noise models, and extension to higher stabilizer weights or planar codes (Liang et al., 6 Oct 2025).

7. Generalizations and Future Directions

The stacking procedure underlying self-dual BB codes is widely applicable:

• Any pair of commuting circulant-product matrices can be stacked to produce a self-dual CSS LDPC code with guaranteed commutation and self-orthogonality (Liu et al., 17 Feb 2026).
• Extensions to higher-dimensional and higher-weight codes are conjectured to further improve parameters.
• Modifications to boundary conditions ("twisted" or reflection tori) provide fine-grained control over logical qubit number and code distance (Liang et al., 6 Oct 2025).
• There is strong potential for systematic magic-state distillation via triorthogonal projections and connections to classical extremal self-dual codes (Shi et al., 2024).

Self-dual BB codes thus provide a unifying framework that merges high-rate LDPC architectures, topological protection, and fault-tolerant logical gate sets, motivating continued research in quantum code optimization and implementation.