--- title: Generalised Bicycle Codes in Quantum Error Correction url: https://www.emergentmind.com/topics/generalised-bicycle-codes-gb-codes type: topic --- # Generalised Bicycle Codes in Quantum Error Correction Generalized Bicycle Codes (GB Codes) are a large and flexible family of quantum low-density parity-check (qLDPC) codes constructed from pairs of commuting binary circulant matrices, whose rich algebraic structure supports highly efficient code design, scalable decoding, and substantial hardware adaptability. GB codes unify and generalize surface codes, quantum toric codes, and classical bicycle codes within a single algebraic framework, enabling high code rates, sublinear-to-linear distance scaling, and practical implementation in diverse quantum computing architectures. ## 1. Algebraic Definition and Code Construction A generalized bicycle (GB) code is a Calderbank–Shor–Steane (CSS) quantum code defined from two $\ell \times \ell$ binary circulant matrices $A$ and $B$ (or, equivalently, two binary polynomials $a(x), b(x) \in \mathbb F_2[x]/(x^\ell-1)$). The commutation condition $AB = BA$ is automatically satisfied in the circulant case. The CSS parity-check matrices are: \[ H_X = [ A \mid B ] , \qquad H_Z = [ B^T \mid A^T ] , \] each of size $\ell \times 2\ell$. The code parameters are as follows: - Block length: $n = 2\ell$ - Dimension: $k = n - \operatorname{rank} H_X - \operatorname{rank} H_Z = 2\deg h(x), \quad h(x) = \gcd(a(x), b(x), x^\ell-1)$ - Minimum distance: $d$ (typically determined numerically for finite length; see Section 3). In group-algebraic generalizations (2BGA codes), $A$ and $B$ are constructed via left and right multiplication actions of elements $a, b$ from the group algebra $\mathbb F_2[G]$ of a (possibly non-abelian) finite group $G$ of order $\ell$. The code length becomes $n = 2\ell$, and the commutation still holds due to the algebraic structure [2305.06890, 2306.16400]. Key algebraic and combinatorial properties: - The code is specified by two generating polynomials or group algebra elements, drastically reducing storage and design complexity compared to explicit matrices. - Row weight $w = \operatorname{wt}(a) + \operatorname{wt}(b)$ controls the LDPC property and maps to the check locality. - The construction unifies surface codes, toric codes, and classical bicycle codes as special cases [1212.6703, 2203.17216]. ## 2. Parameters, Distance Bounds, and Infinite Families The critical code parameters—length $n$, dimension $k$, distance $d$, and row weight $w$—are determined algebraically. Although no closed-form general formula for the minimum distance exists, several tight bounds and infinite code families have been established: - For weight-4 generators $(w = 4)$, exhaustive enumerations and explicit constructions show that up to canonical equivalence, GB codes achieve $d = \Theta(\sqrt{n})$ [2203.17216, 2507.21237]. - Three infinite families of $(2,2)$-GB codes (with $w=4$) have been constructed with parameters: - $[[2n^2,2,n]]$ (square toric codes) - $[[4r^2,2,2r]]$ (optimal even distance, previously thought impossible for GB codes) - $[[(2t+1)^2+1,2,2t+1]]$ (optimal odd distance) [2507.21237] - For higher row weights ($w=6,8$), the minimum distance empirically follows: - $d \approx A(w)n^{1/2}+B(w)$, with $A(w)$ increasing in $w$ and $B(w)$ a small constant (e.g., $A(4)\approx1.02$, $A(6)\approx1.66$, $A(8)\approx2.13$) [2203.17216]. Lower/upper distance bounds: - Upper: $d = O(n^{1-1/(w-1)})$ for check row weight $w$ (from geometric locality in $D\leq w-1$ dimensions). - Lower: $d = \Omega(n^{1/2})$ from explicit algebraic or GV-type constructions, as well as mappings to hypergraph-product codes [2203.17216, 1212.6703, 2306.16400]. ## 3. Structural Features, LDPC and Decoding Properties GB codes are quantum LDPC codes, with each check of bounded weight $w$: - Any cyclic permutation of a circulant matrix row is also a valid check, yielding a large and highly overcomplete set of minimum-weight generators [2203.17216, 2502.19406]. - Overcompleteness enables syndrome redundancy, which significantly enhances fault tolerance against syndrome measurement errors, as redundant syndrome bits enable single-shot or two-shot decoding with practical error suppression [2502.19406]. Efficient decoders: - Belief-propagation (BP) combined with ordered-statistics decoding (OSD) provides near-optimal finite-length decoding in practice (MBP$_4$+ADOSD$_4$) [2505.06157]. - Sliding-window and single-/two-shot sequential decoding exploit the redundancy; $T=2$ measurement rounds are sufficient to recover nearly full code distance in logical error rates, outperforming schemes with redundant checks dropped [2502.19406]. - Redundant stabilizer generators yield syndrome codes of distance $d_S = \min(\operatorname{wt}a, \operatorname{wt}b)$, enhancing error-detection and rapid convergence under BP+OSD [2502.19406]. ## 4. Generalizations, Classification, and Graph-Theoretic Equivalence The group-algebraic extension (2BGA codes) replaces cyclic groups with arbitrary finite groups $G$, yielding quantum codes from Cayley graphs (including nonabelian cases) [2305.06890, 2306.16400, 2409.09830]. Parameters depend on subgroup structure, with code dimension given by formulas involving idempotent projectors and group algebra ideals. Classification of $(2,2)$-GB codes reveals: - Three infinite families described above achieve distance matching optimal weight-4 surface codes, including previously elusive even-distance optimal GB instances [2507.21237]. - Classification under CSS-graph-preserving (CGP) equivalence distinguishes code families, identifying which are genuinely inequivalent to known surface code constructions. - Extensive tables of inequivalent codes for $2n<200$ document all extremal codes, including new codes surpassing prior distance records for specified block lengths [2507.21237]. Generalization to quantum Margulis codes (using $SL(2,p)$ and Cayley graph expansion) yields quantum LDPC codes with logarithmic girth and numerically observed thresholds near $14\%$ under BP-OSD decoding [2409.09830]. ## 5. Practical Implementations, Hardware Adaptation, and Performance GB codes support hardware-adaptable and scalable architectures: - The bivariate bicycle subclass (BB codes) is compatible with two-dimensional qubit layouts (e.g., superconducting and neutral atom architectures), enabling efficient syndrome extraction and high code rate [2508.20858, 2311.16980, 2506.03094]. - The Louvre routing framework reduces ancilla qubit degree (from 6 to 4.5 or 4) with CXSWAP and SWAP-based schedules while preserving or marginally degrading logical error rates. These protocols eliminate long-range couplers and are robust to defective or absent sites [2508.20858]. - Neutral-atom implementations exploit collective movement primitives (AOD), bundling checks as parallel circuits, and achieve up to $10\times$ lower qubit overhead and $2\!-\!3\times$ faster cycle times versus surface codes, supporting quantum memory hierarchies combining high-density GB code memory with surface code compute regions [2311.16980]. - Modular quantum computing architectures integrate high-rate BB codes with universal logical instruction sets and lattice-surgery-style operations, yielding substantial qubit savings and lower logical error rates in end-to-end resource estimates compared to surface codes [2506.03094]. Performance and threshold behavior: - GB and 2BGA codes achieve thresholds comparable to or exceeding classical toric/surface codes (phenomenological thresholds $\sim14\%$ for small quantum codes; see also [2401.07583]). - Quantum Margulis codes (generalized 2BGA) show threshold-like crossovers at $p \approx 14\%$ [2409.09830]. - Finite-length performance of GB codes is competitive or superior to quantum Tanner and single parity-check product codes in block error rates at moderate code lengths, particularly when row weight is increased to $w\geq 8$ [2505.06157]. ## 6. Design Guidelines, Extensions, and Open Problems Guidelines for code synthesis: - For moderate to high rates: use polynomial-based construction with small gcd divisor degrees; increase row weight to raise code distance [2505.06157, 2401.07583]. - For hardware-constrained layouts: leverage code extension variants (modular, triple-block, circulant-symmetric) to balance locality and scalability [2401.07583]. - Redundant minimum-weight stabilizer measurement is essential for high-fidelity, low-latency fault tolerance on near-term devices [2502.19406]. Research directions: - Circuit-level noise analysis, hardware-specific optimization of qubit layouts, and interface design for hybrid memory architectures remain largely open questions. - The role of GB codes in quantum memory hierarchies, particularly for high-T-count circuits where load/store overhead becomes negligible, is a topic of active architectural study [2311.16980]. - Further analytic progress on rate/distance tradeoffs, decoding thresholds, and generator locality for non-cyclic cases is ongoing [2507.21237, 2306.16400]. ## 7. Summary Table: Key Families and Parameters | Family/Class | Parameters | Generating Polynomials | Distance Scaling | Reference | |-------------------------------|----------------------------------|-----------------------------|------------------------|-------------------| | (2,2)-GB (Family I) | $[[2n^2,2,n]]$ | $1+X,\ 1+X^{n}$ | $d=n$ | [2507.21237] | | (2,2)-GB (Family II) | $[[4r^2,2,2r]]$ | $1+X,\ 1+X^{2r-1}$ | $d=2r$ | [2507.21237] | | (2,2)-GB (Family III) | $[[(2t+1)^2+1,2,2t+1]]$ | $1+X,\ 1+X^{2t+1}$ | $d=2t+1$ | [2507.21237] | | BB (bivariate) | $\llbracket 2\ell m, k, d \rrbracket$| $a(x,y),\ b(x,y)$ | $d=O(\min\{\ell,m\})$ | [2311.16980] | | General GB, $w=4,6,8$ | $[[2\ell,2,d]]$ for $b(x)=1+x$ | $a(x),\ b(x)$ (wt 2,3,4) | $d \approx A(w) n^{1/2}$| [2203.17216] | Parameters and code families illustrate the flexibility of the GB code construction, encompassing optimal surface-code distances, high rates, and practical finite-length constructions superior to known surface and product codes in several regimes. --- Source: https://www.emergentmind.com/topics/generalised-bicycle-codes-gb-codes