Doubly Even Binary Codes

• Doubly even binary codes are linear subspaces of F2^n where every codeword has a Hamming weight divisible by 4, underpinning key concepts in coding theory and algebra.
• These codes, particularly in their self-dual form for lengths divisible by 8, satisfy the Mallows–Sloane bound and enhance error detection and correction capabilities.
• Explicit construction techniques such as four-circulant and composite-matrix methods enable the formation, classification, and application of extremal doubly even self-dual codes in combinatorial designs and code loops.

A doubly even binary code is a linear subspace $C \subseteq \mathbb{F}_2^n$ in which every codeword's Hamming weight is divisible by 4. These codes play a fundamental role in both coding theory and algebraic structures such as code loops. Doubly even self-dual codes—those which are equal to their dual and admit only weights divisible by 4—exist only for lengths $n \equiv 0 \pmod{8}$ and satisfy strong combinatorial and algebraic constraints. The construction and classification of extremal doubly even self-dual codes, which attain the Mallows–Sloane bound on minimum distance, constitute a central theme in contemporary research.

1. Definition and Basic Properties

A binary linear code of length $n$ is a subspace $C \subseteq \mathbb{F}_2^n$. The Hamming weight $\mathrm{wt}(v)$ of $v \in C$ counts positions with value 1. A code $C$ is called doubly even if all $v \in C$ satisfy $\mathrm{wt}(v) \equiv 0 \pmod{4}$ (Pires et al., 2019, Pires et al., 5 Jan 2026). If additionally $C = C^\perp$, the code is self-dual, necessarily of dimension $n/2$ and admitting only lengths divisible by 8.

The extremal minimum distance for a doubly even self-dual code satisfies the Mallows–Sloane bound: $$d \le 4 \left\lfloor \frac{n}{24} \right\rfloor + 4$$ with equality for extremal codes (Harada, 2019). These codes maximize both error-detecting and error-correcting capability among all such codes.

2. Algebraic Constructions and Structural Theorems

The algebraic structure of doubly even binary codes is governed by symmetries of the underlying space and the weight properties. Gleason’s theorem and subsequent extensions dictate the permissible weight enumerators. For any doubly even self-dual code of length $n$,

$$W_C(y) = \sum_{j=0}^{\left\lfloor n/24 \right\rfloor} b_j\, (1 + 14y^4 + y^8)^{n/8-3j}(y^4(1-y^4)^4)^j$$

with integer coefficients $b_j$ (Harada, 2019).

Doubly evenness requires that the block intersection numbers in associated support designs are always even, ensuring heavy combinatorial structure.

3. Constructions: Four-Circulant and Composite-Matrix Methods

Several explicit algebraic constructions exist for doubly even codes, many leveraging automorphism and symmetry.

Four-circulant construction: Let $A, B$ be $n \times n$ binary circulant matrices. The block generator matrix

$$G = \begin{pmatrix} I_n & A & B \ 0 & B^T & A^T \end{pmatrix}$$

defines a self-dual code iff $AA^T + BB^T = I_n$, and doubly evenness if all codewords have weight divisible by 4 (Harada, 2019, Kaya et al., 2014).

Composite matrix/group ring techniques: Codes can be constructed over rings such as $\mathbb{F}_2 + u\mathbb{F}_2$ or $\mathbb{F}_4 + u\mathbb{F}_4$, employing Gray maps to ensure preservation of Lee-weight and duality. For group $G$, generator matrices constructed via composite block-matrices $\Omega(v)$ satisfying orthogonality constraints yield new families of extremal doubly even self-dual codes (Gildea et al., 2021).

See Table 1 for construction types and their parameters:

Construction Type	Typical Lengths	Key Criterion
Four-circulant	72, 80, 96, 112, 120, 128	$AA^T + BB^T = I_n$; weights divisible by 4
Composite matrix	96	Block-matrix satisfies $\Omega(v)\,\Omega(v)^T = I_n$
Duadic extension	$2^m$–lengths extended	Achieves self-dual, doubly even via parity extension

4. Classification and Enumeration of Extremal Codes

Classification of extremal doubly even self-dual codes focuses on inequivalence (typically determined by weight enumerators or automorphism group properties) and explicit enumeration.

Harada’s work (Harada, 2019) produced numerous new extremal codes of lengths 112, 120, 128. For length 112, three inequivalent codes (H₁₁₂, D₁₁₂, E₁₁₂) were identified, distinguished by their distinct multisets of inner-product counts among weight-20 codewords. In length 120, at least 526 inequivalent codes were constructed, distinguished by the parameter $a$ in their unique weight enumerator forms.

For length 96, both four-circulant and composite-matrix constructions have yielded more than 200 inequivalent doubly even [96,48,16] codes, substantially enriching the space of possible codes (Kaya et al., 2014, Gildea et al., 2021).

5. Code Loops and Algebraic Applications

A fundamental application of doubly even codes is the construction of code loops—Moufang loops with a central subgroup $\{ \pm 1 \}$, where the binary code's structure induces loop multiplication via a cocycle $\phi$ (Pires et al., 2019, Pires et al., 5 Jan 2026). For $V \subseteq \mathbb{F}_2^n$ doubly even,

$L(V) = \{ \pm 1 \} \times V,$

with product $$(\varepsilon, u)\cdot(\delta, v) = (\varepsilon\delta\phi(u, v), u+v)$$, and the factor set $\phi$ determined by codeword weights and pairwise intersections.

Minimal and reduced representations of code loops are classified by the degree and the “type” (sorted tuple of equivalence-class sizes among coordinates). For ranks 3 and 4, the types determine the loop up to isomorphism, and minimal code representations have been explicitly constructed (Pires et al., 5 Jan 2026).

6. Designs and Weight Distribution

Doubly even codes are tightly linked with combinatorial designs through their support structures. The Assmus–Mattson theorem ensures that, for extremal doubly even self-dual codes, the supports of minimum-weight codewords form t-designs, typically with $t=5$ (Horiguchi et al., 2013). It is shown that higher strength (e.g., $t=6,7,8$) is extremely rare, only potentially realized in certain sporadic lengths and never for $t=8$.

The weight enumerator and block intersection statistics impose strong integrality conditions on possible support designs, forbidding the existence of many hypothetical high-t designs attached to extremal doubly even codes.

7. Extensions, Generalizations, and Open Problems

Recent work has generalized extension theorems to more general rings, leading to new families of doubly even self-dual codes (notably via F₂+uF₂ and F₄+uF₄-lifts) and new weight enumerators (Kaya et al., 2014). For cyclic and duadic codes, parity extension produces doubly even self-dual codes of arbitrary length $n+1$ where $n$ is odd, with root-like lower bounds on minimum distance (Wu et al., 2024).

An open direction remains the full classification of code loop representations in higher rank, as well as determining necessary and sufficient conditions for the existence of extremal doubly even codes of specific length and automorphism structure (Harada et al., 2015). The congruence properties among weight distributions for small lengths (notably 24) suggest deeper modular connections (Nagaoka et al., 2023).

• (Harada, 2019): Harada, “New doubly even self-dual codes having minimum weight 20”
• (Kaya et al., 2014): Kaya & Yıldız, “Extension theorems for self-dual codes over rings and new binary self-dual codes”
• (Gildea et al., 2021): Yildiz & Kaya, “New binary self-dual codes of lengths 80, 84 and 96 from composite matrices”
• (Pires et al., 5 Jan 2026): Pires, Grishkov, Rasskazova, “Representations of code loops by binary codes”
• (Horiguchi et al., 2013): Harada, “On the support designs of extremal binary doubly even self-dual codes”
• (Wu et al., 2024): Li, Tang, Ding, “Binary duadic codes and their related codes with a square-root-like lower bound”
• (Nagaoka et al., 2023): Nagaoka & Oura, “Note on the Type II codes of length $24$”
• (Harada et al., 2015): Harada & Munemasa, “On $s$-extremal singly even self-dual $[24k+8,12k+4,4k+2]$ codes”