Self-Dual Bicycle Codes on 2D Lattices

• The paper introduces self-dual bicycle codes defined by Laurent-polynomial data on a 2D lattice, achieving improved encoding rates and transversal Clifford operations.
• Self-duality is enforced through inversion symmetry, which aligns X- and Z-check supports to enable efficient transversal CNOT, Hadamard, and S gates.
• The construction leverages weight-8 stabilizers on twisted tori to optimize the rate-distance tradeoff and enhance local syndrome extraction.

Self-dual bivariate bicycle codes are translation-invariant CSS quantum low-density parity-check stabilizer codes defined through Laurent-polynomial data on a two-dimensional lattice and constrained by an inversion symmetry that makes the $X$- and $Z$-check structures coincide. Introduced as a broad family of self-dual bivariate bicycle (BB) codes with transversal Clifford gates, they combine the bivariate translational framework of BB codes with the self-duality classically associated with color codes, while the main constructions emphasize weight-8 stabilizers on twisted tori, higher encoding rates than surface and color codes, and transversal CNOT, Hadamard, and $S$ gates (Liang et al., 6 Oct 2025).

1. Algebraic definition and lattice realization

Self-dual bivariate bicycle codes are formulated over the Laurent-polynomial ring

$R=\mathbb{Z}_2[x^{\pm 1},y^{\pm 1}],$

with $x$ and $y$ representing lattice translations in two spatial directions. Qubits live on the vertices of a $2$D lattice, described in the main construction as honeycomb but explicitly noted as generalizable. A translation-invariant CSS code is specified by two polynomials $f(x,y)$ and $g(x,y)$ through

$S_X= \begin{bmatrix} f(x,y) & g(x,y)\ \hline 0 & 0 \end{bmatrix}, \qquad S_Z= \begin{bmatrix} 0 & 0\ \hline \overline{g(x,y)} & \overline{f(x,y)} \end{bmatrix},$

where

$Z$0

The defining self-duality condition is

$Z$1

so every stabilizer has an inversion, or mirror, symmetry. In this setting the $Z$2- and $Z$3-stabilizers have the same support. The paper identifies this equality of support as the structural feature that is crucial for logical Clifford gates and that also simplifies code analysis.

Boundary conditions are imposed by placing the code on a twisted torus determined by two basis vectors $Z$4, which need not be orthogonal. For $Z$5 and $Z$6, the number of physical qubits is

$Z$7

The number of logical qubits is given by

$Z$8

where $Z$9 and $S$0 specify the twisted torus (Liang et al., 6 Oct 2025).

2. Self-duality and transversal Clifford structure

The defining operational significance of self-duality is that it turns a BB code into a self-dual CSS code with identical $S$1/$S$2 stabilizer geometry. In the main family, this is paired with doubly even stabilizer weights, specifically weights divisible by $S$3. The paper’s principal constructions focus on weight-$S$4 checks, so the relevant generators are doubly even.

Under these conditions, the codes admit the full set of transversal Clifford gates available in the paper’s framework. Transversal CNOT acts between code blocks, Hadamard acts transversally within a block because the code is self-dual CSS, and transversal $S$5 is available because the $S$6-type stabilizers are doubly even. In a canonical dual basis satisfying

$S$7

the logical action is

$S$8

This places self-dual BB codes in direct conceptual proximity to color codes. The paper explicitly notes that the full set of transversal Clifford gates is usually available only in self-dual CSS codes, notably color codes, and positions the new BB family as retaining that gate property while improving rate and distance at comparable overhead. The same work also states that weight-$S$9 self-dual codes reduce to the color code or its multiples, making weight $R=\mathbb{Z}_2[x^{\pm 1},y^{\pm 1}],$0 the minimal stabilizer weight for new, nontrivial self-dual BB codes with improved rate-distance tradeoffs (Liang et al., 6 Oct 2025).

3. Parameters, rate, and representative constructions

Code parameters are reported in the standard form $R=\mathbb{Z}_2[x^{\pm 1},y^{\pm 1}],$1, with $R=\mathbb{Z}_2[x^{\pm 1},y^{\pm 1}],$2 the number of physical qubits, $R=\mathbb{Z}_2[x^{\pm 1},y^{\pm 1}],$3 the number of logical qubits, and $R=\mathbb{Z}_2[x^{\pm 1},y^{\pm 1}],$4 the code distance. The principal search in the defining paper enumerates weight-$R=\mathbb{Z}_2[x^{\pm 1},y^{\pm 1}],$5 self-dual BB codes with up to $R=\mathbb{Z}_2[x^{\pm 1},y^{\pm 1}],$6 and selects optimal examples by maximizing the Bravyi–Poulin–Terhal metric $R=\mathbb{Z}_2[x^{\pm 1},y^{\pm 1}],$7.

Code	$R=\mathbb{Z}_2[x^{\pm 1},y^{\pm 1}],$8	$R=\mathbb{Z}_2[x^{\pm 1},y^{\pm 1}],$9	$x$0
$x$1	16	4	4
$x$2	40	6	6
$x$3	56	6	8
$x$4	64	8	8
$x$5	120	8	12
$x$6	152	6	16
$x$7	160	8	16

For these examples, the reported $x$8 values are $x$9, $y$0, $y$1, $y$2, $y$3, $y$4, and $y$5, respectively. The construction singled out in formula form is

$y$6

The comparative claim made in the source is that self-dual BB codes typically have much higher rates than surface and color codes for given $y$7 and $y$8. The paper’s explicit example is $y$9 against the toric code $2$0, together with the observation that standard color-code examples such as $2$1 and $2$2 generally carry lower $2$3 at comparable small sizes. This suggests that the principal architectural gain of the self-dual BB family lies not only in transversality, but in the simultaneous retention of nontrivial encoding rate (Liang et al., 6 Oct 2025).

4. Weight-8 locality, twisted tori, and geometric optimization

The central finite-length constructions are realized on twisted tori rather than on untwisted rectangular periodic lattices. The source attributes two functions to the twist: improvement of code parameters such as $2$4 and $2$5, and better localization of stabilizer checks. The intended effect is that short-range generators can be recovered from a code family that is specified algebraically rather than by a purely Euclidean local tiling.

The representative $2$6 code is realized on a twisted torus with basis vectors $2$7 and $2$8, which imposes the periodicity relations

$2$9

More generally, the twisted-torus description allows one to choose periodicity vectors that improve stabilizer locality while preserving the translation-invariant algebraic description.

The paper states that all checks in the main family have weight $f(x,y)$0 and that appropriate twisting makes them local or effectively local. In the thermodynamic limit, all stabilizers are described as effectively local, a point presented as relevant for practical circuit depth and syndrome extraction. The same summary also attributes topological protection to the design in the sense that all local, commuting operators with the stabilizer are themselves stabilizers. Within the article’s terminology, this identifies the self-dual BB family as a topological $f(x,y)$1D qLDPC construction rather than merely a sparse algebraic code (Liang et al., 6 Oct 2025).

5. Relation to color codes, generic BB codes, and self-correction claims

Self-dual bivariate bicycle codes are best understood against two nearby baselines. The first is the color code, described in the source as the most prominent self-dual CSS family and one whose transversal Clifford gates have been demonstrated experimentally. The second is the broader BB family, which provides the algebraic and geometric framework but is not generically self-dual. Later work makes this distinction explicit: generic bivariate BB codes offer high rates and large distances, but are not self-dual, so they do not directly support transversal Clifford gates; self-dual variants must therefore be specially constructed, for example by stacking or by imposing explicit symmetry conditions (Liu et al., 17 Feb 2026).

The comparison with toric and surface-code families is similarly structural. The toric code is quoted as $f(x,y)$2, so its rate scales as $f(x,y)$3. The self-dual BB family is presented as improving on this tradeoff at finite block length while preserving LDPC sparsity and transversality. The resulting contrast is not only geometric but computational: self-dual BB codes place high-rate encoding and transversal Clifford logic in a single CSS framework (Liang et al., 6 Oct 2025).

A separate clarification concerns passive self-correction. The broader BB literature does not support the claim that BB codes are self-correcting under local decoders. In work on ZSZ codes, BB codes are reported to show no evidence of a sustainable threshold under the same passive decoder, whereas the non-Abelian $f(x,y)$4 generalization does. Self-duality in the BB sense therefore should not be conflated with passive self-correcting behavior; it is a statement about the symmetry of $f(x,y)$5 and $f(x,y)$6 sectors and about transversal Clifford implementation, not a proof of autonomous memory protection under local decoding (Guo et al., 29 Jul 2025).

6. Extensions, stacked variants, and later directions

Subsequent work generalized the self-dual BB idea by constructing self-dual qLDPC codes through stacking non-self-dual base codes. For BB codes this yields double-layer bivariate bicycle codes, double-layer twisted BB codes, and double-layer reflection codes. In the stacking construction, a base CSS code with

$f(x,y)$7

is converted into a self-dual code with

$f(x,y)$8

The physical interpretation is a doubled system in which qubits occupy two layers and stabilizer support is distributed within and across them. That work reports circuit-level-noise simulations with an efficient Tesseract decoder and states that pseudo-thresholds exceed $f(x,y)$9 for codes with logical operators of odd weight (Liu et al., 17 Feb 2026).

Other developments broadened the implementation and analysis landscape of BB-derived constructions. Floquetification of Abelian two-block group algebra codes, including BB codes, leads to Stairway codes. In that framework, self-duality is inherited if and only if the underlying BB code is self-dual and the ZX-calculus decomposition together with the time-tiling are performed symmetrically for both $g(x,y)$0 and $g(x,y)$1 checks; the stated requirement is that the original stabilizer weight be divisible by $g(x,y)$2, matching the weight-$g(x,y)$3 emphasis of the self-dual BB family (Jacoby et al., 27 Feb 2026).

Finally, decoding-oriented work on BB codes identified algebraic constraints linking rate, stabilizer redundancy, and single-shot syndrome robustness. For coprime BB codes, the greatest common divisor polynomial $g(x,y)$4 determines both the logical dimension and the syndrome code, with a strict equality between quantum rate and stabilizer redundancy density. Although this analysis is not restricted to the self-dual subclass, it provides a design principle directly relevant to self-dual BB variants intended for measurement-limited architectures: increasing syndrome distance requires accepting a corresponding structural constraint on rate (Rowshan, 3 Jan 2026).