Reed–Muller Codes: Distance Distribution

• Distance Distribution of Reed–Muller Codes is the enumeration of codewords by Hamming weight, revealing the combinatorial structure and error-correcting capabilities.
• The methodology leverages recursive constructions like the (u, u+v)-technique and derivative methods to precisely characterize weight spectra in various regimes.
• Asymptotic analyses and probabilistic bounds, including binomial approximations, offer practical insights into weight plateaus and performance limits in large finite fields.

A Reed–Muller code is a family of linear codes with broad significance in both theoretical and applied coding theory, combinatorics, and computer science. The distance distribution—also known as the weight spectrum—of a Reed–Muller code refers to the enumeration of codewords by their Hamming weights, providing a precise profile of the code's combinatorial structure and its error-correcting performance. The study of distance distributions includes exact characterizations for certain parameter regimes, asymptotic error bounds, the structure of small-weight codewords, and unified frameworks for codes over various finite fields.

1. Fundamental Definitions and Notation

Let $\mathbb{F}_q$ denote the finite field with $q$ elements. For integers $0 \leq r \leq m$, the $q$-ary Reed–Muller code $\mathrm{RM}_q(r,m)$ consists of evaluation vectors of all $m$-variate polynomials of total degree at most $r$, with entries in $\mathbb{F}_q$, evaluated over the points of $\mathbb{F}_q^m$: $$\mathrm{RM}_q(r,m) = \left\{ \left(f(\alpha)\right)_{\alpha \in \mathbb{F}_q^m} \ \big| \ f \in \mathbb{F}_q[x_1,\ldots,x_m], \ \deg(f) \leq r \right\}$$ The parameters are:

• Length $n = q^m$
• Dimension $k = \sum_{i=0}^r \binom{m+i-1}{i}$
• Minimum distance $d = q^{m-r}$

For binary codes ($q=2$), the weight (distance) spectrum $S(r,m)$ is the set of Hamming weights of all codewords in $\mathrm{RM}(r,m) \subset \mathbb{F}_2^{2^m}$, and the weight enumerator $A_w$ counts the number of codewords of weight $w$.

2. Exact Weight Spectra: Families $RM(m-3,m)$ and $RM(m-4,m)$

Recent work establishes explicit formulas for the weight spectra of two infinite families: $RM(m-3,m)$ for $m \geq 6$ and $RM(m-4,m)$ for $m \geq 8$ (Carlet et al., 2023). The determination proceeds via induction on $m$ utilizing the $(u,u+v)$-construction:

• $S(r+1,m+1) \supseteq S(r,m)+S(r,m)$, the setwise sum of all possible weights.

Weight spectrum for $RM(m-3,m)$ ($m \ge 6$):

$$S(m-3,m) = \{0, 2^m\} \cup \{8,12,14,16,18,\ldots,2^m-14,2^m-12,2^m-8\}$$

Equivalently,

$$S(m-3,m) = \{0,2^m\} \cup \{8 + 2i \mid 0 \leq i \leq 2^{m-1}-8\} \cup \{2^m-8\}$$

Weight spectrum for $RM(m-4,m)$ ($m \ge 8$):

$$S(m-4,m) = \{0,2^m\} \cup \{16,24,28,30,32,34,\ldots,2^m-30,2^m-28,2^m-24,2^m-16\}$$

Or equivalently,

$$S(m-4,m) = \{0,2^m\} \cup \{16 + 2i \mid 0 \leq i \leq 2^{m-1} - 16\} \cup \{2^m-16\}$$

The proofs combine induction with exclusion of "forbidden holes" in the possible weight intervals, rigorously constrained by the Kasami–Tokura characterization:

• For weights in $[2^{m-r}, 2^{m-r+1})$, the only allowable weights take the form $2^{m-r+1} - 2^{m-r+1-p}$ for integer $1 \leq p \leq \max\{ r, m-r-1\}$.

Explicit computations confirm this structure for specific small codes, such as $RM(3,8)$ and $RM(4,9)$, whose complete spectra are tabulated (Carlet et al., 2023).

3. Structured Descriptions: Kasami–Tokura Bound and Forbidden Gaps

The interval $[2^{m-r}, 2^{m-r+1})$ is governed by the Kasami–Tokura theorem: only weights $w = 2^{m-r+1} - 2^{m-r+1-p}$, for suitable $p$, appear. As a result, the weight spectra of $RM(m-c,m)$ for fixed $c$ and large $m$ comprise:

• Isolated "small" weights fully prescribed by Kasami–Tokura
• Further isolated weights in $[2^{c+1}, 2^{c+1}+2^c)$, governed by the Kasami–Tokura–Azumi (KT–A) classification for weights $<2.5d$
• A contiguous sequence of even weights ("central interval") in the middle
• Complements to $2^m$ of the isolated weights

For $c = 1,2,3,4$, the above structure is completely determined. For $c > 4$, this remains conjectural.

4. Asymptotic and Probabilistic Bounds: Character-Sum Framework

A general asymptotic analysis of $\mathrm{RM}_q(r,m)$ over arbitrary finite fields is achieved via the character-sum method (Kolekar, 26 Jan 2026). For any received word $u$ and $0 \leq w \leq n$, the coset-weight distribution $A_w(u)$ admits a binomial-approximation: $\mathcal{C}_w = \binom{n}{w} (q-1)^w q^{k-n}$ with an explicit error bound: $$|A_w(u) - \mathcal{C}_w| \leq C(r,m,q) \binom{n}{w-1} (q-1)^{w-1} q^{k-n+1-\delta(r,m,q)}$$ where $$\delta(r,m,q) = \binom{\lfloor (r+1)/2 \rfloor + m}{m}$$ and $C(r,m,q)$ is polynomial in $m$ and $q$. For fixed $(r,m)$ and $q \to \infty$, the ratio $A_w(u)/\mathcal{C}_w \to 1$ uniformly in $w$.

This character-sum framework generalizes previous results for Reed–Solomon codes (the $m=1$ case) and reveals that, for large $q$, the Reed–Muller distance distribution is sharply concentrated around the binomial estimate.

5. Global Weight Distribution: Plateaus and Combinatorial Structure

The cumulative weight distribution $W_{n,d}(\alpha)$ and the multiplicities at given weights exhibit a "plateau" phenomenon (0811.2356):

• For each $k$, the distribution $W_{n,d}(\alpha)$ remains essentially constant in $\alpha \in [2^{-k}, 2^{-k+1})$, rising exponentially at the cutoff points $\alpha = 2^{-k}$.
• For $\alpha \geq 1/2$, $W_{n,d}(\alpha) = 2^{\Theta(n^d)}$.
• The asymptotics in each plateau obey $\exp\bigl(\Theta(n^{d-k})\bigr)$ as $n \to \infty$ with $d$ fixed.

Upper and lower bounds for $W_{n,d}(\alpha)$ are established: $$W_{n,d}\left(2^{-k}(1-\epsilon)\right) \leq \exp\left( C \cdot n^{d-k} \right)$$

$W_{n,d}(2^{-k}) \geq 2^{c_k n^{d-k}}$

for constants $C = C(d, \epsilon)$, $c_k = c_k(d)$.

6. Techniques: $(u,u+v)$-Construction and Derivative Methods

The recursive $(u,u+v)$-construction underpins the inductive computation of spectra. For $RM(r+1,m+1)$, each codeword can be written as $(u, u+v)$ with $u \in RM(r+1,m)$, $v \in RM(r,m)$. This implies $S(r+1,m+1) \supseteq S(r,m) + S(r,m)$, tightly constraining possible weights.

For asymptotic upper bounds and plateaus, the discrete derivative method is central (0811.2356). Mapping Boolean codewords into $\{\pm1\}$, repeated application of directional differences $\partial_a f(x) = f(x+a)f(x)$ uncovers bias and allows for representations of low-weight words via a controlled number of derivatives, bounding the number of possible codewords at each weight level.

On the enumeration side, the character-sum approach utilizes:

• Lagrange-indicator polynomials for zero-set specification
• Additive and multiplicative characters on the quotient algebra of polynomials to enforce coefficient constraints
• Evaluation of Gauss sums and Möbius-inversion to control error terms, combined with the Li–Wan permutation sieve for distinctness in summations (Kolekar, 26 Jan 2026)

7. Open Problems and Conjectural Spectra

For the general family $RM(m-c,m)$ with $c$ fixed and $m \gg c$, it is conjectured that the weight spectrum consists precisely of:

• Isolated gaps at small weights as predicted by Kasami–Tokura and KT–A results
• A single run of consecutive even weights ("central interval") in the middle
• Complements to $2^m$ of the exceptional weights

This conjecture remains open for $c \ge 5$ (Carlet et al., 2023). The rigorous study of the coset-weight distribution for codes over large finite fields is now addressed systematically, but extensions to higher $r$, detailed spectra for nonbinary codes, and explicit combinatorial characterizations in the intermediate regime remain active topics.

---

Key References:

• "The weight spectrum of two families of Reed–Muller codes" (Carlet et al., 2023)
• "On the Distance Distribution of Reed–Muller Codes" (Kolekar, 26 Jan 2026)
• "The List-Decoding Size of Reed-Muller Codes" (0811.2356)