---
title: 'Tricycle Codes: Classical & Quantum'
url: https://www.emergentmind.com/topics/tricycle-codes
type: topic
---

# Tricycle Codes: Classical & Quantum

Searching arXiv for papers on tricycle codes, including both classical triple cyclic codes and recent quantum qLDPC tricycle code work.
“Tricycle codes” is used in two distinct coding-theoretic senses in the arXiv literature. In classical algebraic coding theory, the term appears as a synonym for triple cyclic codes over $\mathbb{Z}_2$, namely binary linear codes of block length $(r,s,t)$ whose coordinates are partitioned into three parts and are invariant under cyclic shifts within those parts [1509.05351, 1601.04884]. In recent quantum error-correction literature, “tricycle codes” denotes several families of CSS quantum LDPC codes that extend bicycle-type constructions to three homological dimensions or three-factor algebraic products, with applications to transversal $CCZ$, magic-state generation, and single-shot decoding [2508.10714, 2508.08191]. The common thread is a three-part algebraic structure, but the underlying objects, parameters, and intended applications differ substantially.

## 1. Classical triple cyclic codes over $\mathbb{Z}_2$

A triple cyclic code of length $(r,s,t)$ over $\mathbb{Z}_2$ is a binary linear code of length $r+s+t$ such that the set of coordinates can be partitioned into three parts and any cyclic shift of the coordinates of the parts leaves invariant the code [1509.05351]. Equivalently, for a codeword
\[
c=(c_{1,0},\ldots,c_{1,r-1}\mid c_{2,0},\ldots,c_{2,s-1}\mid c_{3,0},\ldots,c_{3,t-1}),
\]
the blockwise cyclic shift
\[
T(c)=(c_{1,r-1},c_{1,0},\ldots,c_{1,r-2}\mid c_{2,s-1},c_{2,0},\ldots,c_{2,s-2}\mid c_{3,t-1},c_{3,0},\ldots,c_{3,t-2})
\]
is again in the code [1509.05351]. The closely related paper “Z2-Triple cyclic codes and their duals” defines a $\mathbb{Z}_2$-triple cyclic code of block length $(r,s,t)$ as a binary code of length $r+s+t$ such that the code is partitioned into three parts of lengths $r,s$ and $t$ such that each of the three parts is invariant under the cyclic shifts of the coordinates [1601.04884].

These codes are modeled as $\mathbb{Z}_2[x]$-submodules of
\[
R_{r,s,t}=\frac{\mathbb{Z}_2[x]}{\langle x^r-1\rangle}\times\frac{\mathbb{Z}_2[x]}{\langle x^s-1\rangle}\times\frac{\mathbb{Z}_2[x]}{\langle x^t-1\rangle},
\]
with external multiplication
\[
f(x)*(u_1(x),u_2(x),u_3(x))=(f(x)u_1(x),f(x)u_2(x),f(x)u_3(x))
\]
for $f(x)\in\mathbb{Z}_2[x]$ [1509.05351]. This polynomial-module representation is the basic algebraic device for expressing generators, block projections, and duality. The terminology “tricycle code” in this classical setting therefore refers to a three-block cyclic invariance condition rather than to any quantum or homological construction.

## 2. Generator structure, minimal generating sets, and separability

For triple cyclic codes over $\mathbb{Z}_2$, one structural description is
\[
C=\langle (F_1(x)\mid 0\mid 0),\ (0\mid F_2(x)\mid 0),\ (G_1(x)\mid G_2(x)\mid G_3(x))\rangle
\]
with $F_1(x)\mid x^r-1$, $F_2(x)\mid x^s-1$, and $G_3(x)\mid x^t-1$ [1509.05351]. A closely related form given for $\mathbb{Z}_2$-triple cyclic codes is
\[
C=\langle (g_1(x),0,0), (l_1(x),g_2(x),0), (l_2(x),l_3(x),g_3(x))\rangle,
\]
where $g_1(x)\mid x^r-1$, $g_2(x)\mid x^s-1$, and $g_3(x)\mid x^t-1$ [1601.04884]. Both descriptions express the code as a finitely generated $\mathbb{Z}_2[x]$-submodule, and both encode inter-block coupling through nonzero off-diagonal polynomial components.

A minimal generating set over $\mathbb{Z}_2$ is obtained by taking appropriate shifts of the module generators. In the formulation of [1509.05351],
\[
S_1=\{x^i*(F_1(x)\mid 0\mid 0):0\le i<r-\deg F_1(x)\},
\]
\[
S_2=\{x^i*(0\mid F_2(x)\mid 0):0\le i<s-\deg F_2(x)\},
\]
\[
S_3=\{x^i*(G_1(x)\mid G_2(x)\mid G_3(x)):0\le i<t-\deg G_3(x)\},
\]
and
\[
C=\operatorname{Span}_{\mathbb{Z}_2}(S_1\cup S_2\cup S_3).
\]
The same source states that
\[
|C|=2^{r+s+t-\deg F_1(x)-\deg F_2(x)-\deg G_3(x)}.
\]

The notion of separability isolates the case in which the three blocks decouple. A triple cyclic code is called separable if
\[
C=C_r\times C_s\times C_t,
\]
where $C_r,C_s,C_t$ are individual cyclic codes for each block [1509.05351]. This is equivalent to the conditions $F_1(x)\mid G_1(x)$ and $F_2(x)\mid G_2(x)$, in which case the generator $(G_1(x),G_2(x),G_3(x))$ can be replaced by $(0,0,G_3(x))$ [1509.05351]. The companion discussion in [1601.04884] states that a triple cyclic code is the direct sum of three cyclic codes if all $l_i(x)=0$, whereas nonzero interconnection polynomials induce dependencies between the blocks. This suggests that separability is the decisive algebraic distinction between purely blockwise cyclic structure and genuinely coupled three-block codes.

## 3. Projections, distance bounds, and duality in the classical setting

The block projections of a triple cyclic code are themselves cyclic codes. Specifically,
\[
C_r=(\gcd(F_1(x),G_1(x))),\qquad
C_s=(\gcd(F_2(x),G_2(x))),\qquad
C_t=(G_3(x)),
\]
with sizes determined by the degrees of these generator polynomials [1509.05351]. The same source gives the minimum-distance inequality
\[
d(C)\ge \min\{d(C_r),d(C_s),d(C_t)\},
\]
and states that equality holds if the code is separable.

Duality preserves the triple cyclic structure. The dual $C^\perp$ of a triple cyclic code is again a triple cyclic code of the same block structure [1509.05351]. A bilinear form is introduced on $R_{r,s,t}$:
\[
o(u(x),v(x)):=\sum_{i=0}^{m-1}x^{m-1-i}\left[u_1(x)v_1^*(x)+u_2(x)v_2^*(x)+u_3(x)v_3^*(x)\right]\bmod (x^m-1),
\]
where $m=\operatorname{lcm}(r,s,t)$ and $^*$ denotes reciprocal polynomial [1509.05351]. Then
\[
C^\perp=\{u\in R_{r,s,t}:o(u(x),c(x))=0 \pmod{x^m-1},\ \forall c(x)\in C\}.
\]

Explicit relations between code generators and dual generators are also given. For example,
\[
F_1^\perp(x)=\frac{x^r-1}{\gcd(F_1(x),G_1(x))^*},\qquad
F_2^\perp(x)=\frac{x^s-1}{\gcd(F_2(x),G_2(x))^*},
\]
and
\[
G_3^\perp(x)=\frac{x^t-1}{G_3^*(x)}\cdot \frac{1}{h(x)},
\]
where
\[
h(x)=\gcd(F_1(x)F_2(x),F_1(x)G_2(x),F_2(x)G_1(x))^*
\]
[1509.05351]. The paper [1601.04884] likewise states that dual generators are derived from parity-check polynomials, in analogy with the dual of an ordinary cyclic code. In the classical literature, then, “tricycle” primarily denotes a structured generalization of cyclic and double cyclic codes, organized by module-theoretic generators, reciprocal polynomials, and gcd relations.

## 4. Quantum tricycle codes as finite-blocklength qLDPC constructions

In quantum error correction, “tricycle codes” refers to CSS quantum LDPC codes obtained by extending bicycle-type constructions to three homological dimensions or to three-factor balanced products. “Magic tricycles: efficient magic state generation with finite block-length quantum LDPC codes” introduces tricycle codes as a class of finite block-length quantum LDPC codes generalizing bicycle codes to three homological dimensions [2508.10714]. In that work, the codes are defined using three group-algebra elements $a,b,c$ from a group algebra $F=\mathbb{F}_2[G]$, with parity-check matrices
\[
H_X=\begin{bmatrix}\mathbf{A}^T & \mathbf{B}^T & \mathbf{C}^T\end{bmatrix},
\]
\[
H_Z=
\begin{bmatrix}
\mathbf{C} & \mathbf{0} & \mathbf{A}\\
\mathbf{0} & \mathbf{C} & \mathbf{B}\\
\mathbf{B} & \mathbf{A} & \mathbf{0}
\end{bmatrix},
\]
acting on $N=3n_G$ qubits grouped into three blocks or sectors [2508.10714].

A related but distinct family is the trivariate tricycle (TT) code family, introduced as CSS codes based on a length-3 chain complex and defined from three trivariate polynomials [2508.08191]. In that formulation,
\[
\partial_3=\begin{bmatrix}a&b&c\end{bmatrix},\qquad
\partial_2=
\begin{bmatrix}
0&c&b\\
c&0&a\\
b&a&0
\end{bmatrix},\qquad
\partial_1=\begin{bmatrix}a&b&c\end{bmatrix},
\]
and the CSS check matrices are
\[
H_X\equiv \partial_1=[A\;\;B\;\;C],\qquad
H_Z=\partial_2^\top=
\begin{bmatrix}
0&C^\top&B^\top\\
C^\top&0&A^\top\\
B^\top&A^\top&0
\end{bmatrix},
\]
with meta-checks
\[
M_Z=\partial_3^\top=[A^\top\;\;B^\top\;\;C^\top]
\]
[2508.08191]. The 3D toric code belongs to this construction via the choice $A=1+x$, $B=1+y$, $C=1+z$ [2508.08191].

The more general perspective of “Multivariate Multicycle Codes for Complete Single-Shot Decoding” places tricycle or TT codes at $t=3$ within a Koszul-complex framework. Over
\[
S=\mathbb{F}_2[x,y,z]/\langle x^\ell-1,y^m-1,z^p-1\rangle,
\]
the length-3 Koszul complex yields explicit boundary maps
\[
\partial_3=\begin{bmatrix}F&G&H\end{bmatrix},\qquad
\partial_2=
\begin{bmatrix}
G&H&0\\
F&0&H\\
0&F&G
\end{bmatrix},\qquad
\partial_1=\begin{bmatrix}H\\G\\F\end{bmatrix},
\]
with TT codes supporting only $Z$-metachecks, hence only partial single-shot decoding [2601.18879]. A plausible implication is that current quantum usage reserves “tricycle” for three-factor CSS/qLDPC constructions, while different papers emphasize either balanced-product language, chain-complex language, or group-algebra language.

## 5. Fault-tolerant gates, single-shot features, and decoding

A major motivation for quantum tricycle codes is the coexistence of LDPC structure with logical non-Clifford gates. The finite-blocklength tricycle codes of [2508.10714] support constant-depth physical circuits that implement logical $CCZ$ gates between three code blocks. For the standard $(w_X=12,w_Z=8)$ codes, a depth-18 circuit is always possible, and by imposing further constraints, depths as low as $4$ are attainable at some cost in code rate and/or distance [2508.10714]. The same paper states that tricycle codes are single-shot in the $Z$ basis, with meta-checks
\[
H_{meta}=\begin{bmatrix}\mathbf{B}&\mathbf{A}&\mathbf{C}\end{bmatrix}
\]
enabling single-shot state preparation and error correction in constant depth [2508.10714].

The TT-code paper presents a different but related set of fault-tolerant features. It states that TT codes combine partial single-shot decodability, a large set of transversal Clifford gates and automorphisms within and between code blocks, and, for several sub-constructions, constant-depth implementations of a non-Clifford $CCZ$ gate [2508.08191]. All TT codes possess several transversal $CZ$ gates that can partially address logical qubits between two code blocks, and shift automorphisms induced by monomial translations can implement Clifford gates within a code block [2508.08191]. For codes where $A,B,C$ are all weight-2, or more generally for polynomials admitting a cup-product structure, a depth-2 circuit of physical $CCZ$ gates on triples of qubits within each cube can implement logical $CCZ$ gates between code blocks [2508.08191].

The paper “Transversal dimension jump for product qLDPC codes” uses “bivariate tricycle codes” for small 3D qLDPC examples embedded in a broader lifted-product framework [2510.07269]. It identifies explicit 3D–2D code pairs, including the $[[81,3,5]]$–$[[54,2,6]]$ pair, where the 3D tricycle codes admit depth-2 $CCZ$, weight-6 stabilizers, and pseudo-thresholds $\gtrsim 0.4\%$ [2510.07269]. That work also describes a code-switching protocol between a 3D tricycle code and a 2D component code using one-way transversal CNOTs and teleportation logic [2510.07269]. In this line of research, tricycle codes are not merely storage codes; they are positioned as ingredients in universal fault-tolerant computation.

## 6. Concrete code instances, layouts, thresholds, and generalizations

Several explicit quantum code families and parameter points appear in the recent literature. The finite-blocklength tricycle paper reports examples such as $[[375,15,\leq 15]]$, $[[648,18,\leq 18]]$, $[[192,27,8]]$, and $[[72,9,6]]$, states a circuit-noise threshold of $>0.4\%$ with a Belief-Propagation + Ordered-Statistics Decoder (BPOSD), and gives an optimal CNOT-depth-$12$ syndrome measurement schedule for the standard instances [2508.10714]. The TT-code paper reports circuit-level thresholds of $0.3\%$ in the $Z$ error channel and $1\%$ in the $X$ error channel with single-shot decoding, and states that numerical searches found candidates with improved parameters relative to the 3D toric code, using up to $48\times$ fewer data qubits as equivalent 3DTC encodings [2508.08191]. It also gives examples such as $[[72,6,6]]$, $[[48,3,4]]$, and $[[90,3,5]]$ [2508.08191].

The resource-oriented paper on transversal dimension jump emphasizes a small bivariate tricycle code $[[81,3,5]]$ as an example with depth-2 $CCZ$ and highly efficient magic-state preparation: a single round of stabilizer measurements followed by depth-2 $CCZ$ and postselection produces states with error $<10^{-9}$ and success probability $\sim 35\%$ [2510.07269]. By contrast, “Magic tricycles” emphasizes deterministic low-overhead distillation without requiring post-selection, using single-shot state preparation and error correction [2508.10714]. These are different protocols in different constructions, not interchangeable statements about a single universal tricycle family.

A neighboring line of work, “Multivariate Bicycle Codes,” focuses on trivariate bicycle (TB) codes rather than tricycle codes, but it is closely adjacent conceptually because it also exploits a three-variable algebraic structure in QLDPC design [2406.19151]. TB-QLDPC codes are introduced as an extension of multivariate bicycle constructions and include explicit weight-5 examples with bi-planar structure, toric layouts, and syndrome circuits of depth $7$ [2406.19151]. The paper reports that one example can encode $4$ logical qubits with distance $5$ into $60$ physical qubits using weight-5 check measurements of circuit depth $7$, whereas a comparable surface code requires $200$ physical qubits [2406.19151]. This does not make TB codes identical to tricycle codes, but it suggests a broader three-variable design trend in finite-length qLDPC research.

The generalization beyond tricycle codes is made explicit in “Multivariate Multicycle Codes for Complete Single-Shot Decoding,” which states that multivariate multicycle (MM) codes unify and generalize bivariate bicycle codes, multivariate bicycle codes, abelian two-block group algebra codes, generalized bicycle codes, trivariate tricycle codes, and $n$-dimensional toric codes [2601.18879]. In that framework, tricycle codes correspond to length-$3$ chain complexes, while MM codes with $t\ge 4$ possess both $X$- and $Z$-metachecks and therefore complete single-shot decoding [2601.18879]. This places tricycle codes in a larger algebraic hierarchy: they are sufficiently structured to support partial single-shot decoding and logical non-Clifford gates, but not the full metacheck symmetry available in longer multicycle constructions.

Source: https://www.emergentmind.com/topics/tricycle-codes