OctoInformBeta

OpenAI preprint, September 2026 · §4. Spreading Fourier coefficients

The binary recursion: For each 0<δ<10<\delta<1, some pnp_n of the binary recursion is a trigonometric polynomial with non-proportional positive frequencies of coefficient square sum at least (1−δ)/2(1-\delta)/2

Functions on tori bounded by one, with spread Fourier coefficients.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

Take a real function ff on the torus 𝕋n=(ℝ/ℤ)n\mathbb{T}^n = (\mathbb{R}/\mathbb{Z})^n with values in [−1,1][-1,1], and expand it in the characters e(s⋅x)e(s\cdot x), s∈ℤns \in \mathbb{Z}^n, where e(t)=exp⁡(2πit)e(t) = \exp(2\pi i t) and s⋅x=∑k<nskxks\cdot x = \sum_{k<n} s_k x_k. How close to 11 can the square sum of its Fourier coefficients come when no two of its frequencies are proportional, except a frequency ss and its negative −s-s? The binary recursion is an explicit construction for this question: a sequence of functions pnp_n on 𝕋n\mathbb{T}^n, each obtained from the one before by adding a coordinate.

The Binary Recursion

A point of 𝕋n+1\mathbb{T}^{n+1} is written (x0,x′)(x_0, x'), with x0∈𝕋x_0 \in \mathbb{T} its first coordinate and x′=(x1,…,xn)∈𝕋nx' = (x_1, \dots, x_n) \in \mathbb{T}^n the remaining ones. The binary recursion is the sequence of real functions pnp_n on 𝕋n\mathbb{T}^n given by p0=0p_0 = 0 on the one-point torus 𝕋0\mathbb{T}^0 and pn+1(x0,x′)=pn(x′)+1−pn(x′)22cos⁡2πx0.p_{n+1}(x_0, x') = p_n(x') + \frac{1 - p_n(x')^2}{2}\,\cos 2\pi x_0 .(1)✦

The torus 𝕋n\mathbb{T}^n carries its Haar probability measure, the product of the Haar probability measures of the circle in its coordinates, and integrals over 𝕋n\mathbb{T}^n are taken against it.✦ The energy MnM_n of pnp_n is its mean square over 𝕋n\mathbb{T}^n: Mn=∫𝕋npn(x)2dx.M_n = \int_{\mathbb{T}^n} p_n(x)^2 \, dx .(2)✦

Positive Frequencies

An integer vector s∈ℤns \in \mathbb{Z}^n is positive if it is nonzero and its first nonzero coordinate equals 11.✦ The linear form of an integer vector s∈ℤns \in \mathbb{Z}^n is the linear functional ℓs\ell_s on ℝn\mathbb{R}^n given by ℓs(x)=s⋅x\ell_s(x) = s\cdot x.✦

The Claim

Theorem 1. For every n≥0n \ge 0 and every x∈𝕋nx \in \mathbb{T}^n, |pn(x)|≤1,|p_n(x)| \le 1 ,(3)✦ and the energies of consecutive steps satisfy Mn+1=Mn+18∫𝕋n(1−pn(y)2)2dy.M_{n+1} = M_n + \frac{1}{8} \int_{\mathbb{T}^n} \bigl(1 - p_n(y)^2\bigr)^2 \, dy .(4)✦ Let 0<δ<10 < \delta < 1. For some nn there are a nonempty finite set SS of positive vectors in ℤn\mathbb{Z}^n and nonzero complex numbers csc_s, s∈Ss \in S, such that pn(x)=∑s∈Scse(s⋅x)+∑s∈Scse(s⋅x)¯(x∈𝕋n),1−δ≤2∑s∈S|cs|2≤1.p_n(x) = \sum_{s \in S} c_s\, e(s\cdot x) + \overline{\sum_{s \in S} c_s\, e(s\cdot x)} \quad (x \in \mathbb{T}^n), \qquad 1 - \delta \le 2 \sum_{s \in S} |c_s|^2 \le 1 .(5)✦ If ss and tt are distinct positive vectors in ℤn\mathbb{Z}^n, then for every real rr ℓs≠rℓt.\ell_s \ne r\, \ell_t .(6)✦

What Is Checked

The claims are stated over the binary recursion pnp_n of (1), its energy MnM_n of (2) for the Haar probability measure on 𝕋n\mathbb{T}^n, the positive integer vectors and the linear forms ℓs\ell_s of integer vectors.