OctoInformBeta

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

Proposition 4.1: For 0<δ<1/200<\delta<1/20, a real trigonometric polynomial bounded by one on a torus, with coefficients controlled by the curvatures of its modes

An auxiliary frequency packet exists for each small parameter.

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

The item is an auxiliary object for the spreading of Fourier coefficients: a real trigonometric polynomial on a finite-dimensional torus, bounded by one and close to one in L2L^2 norm, together with a velocity, a real vector vv that gives each frequency aa of the polynomial a curvature a⋅va\cdot v, none of them zero. A parameter δ\delta with 0<δ<1/200<\delta<1/20 sets the tolerances: the curvatures are small in total, and they dominate the squared Fourier coefficients up to a factor close to one.

Let 𝕋=ℝ/ℤ\mathbb T=\mathbb R/\mathbb Z be the circle. For a finite set JJ, 𝕋J\mathbb T^J is the torus of points y=(yj)j∈Jy=(y_j)_{j\in J} with each yj∈𝕋y_j\in\mathbb T. Write e(x)=exp⁡(2πix)e(x)=\exp(2\pi i x) for x∈ℝx\in\mathbb R. For an integer vector a∈ℤJa\in\mathbb Z^J, the character of frequency aa, also called the mode aa, is y↦e(a⋅y)=∏j∈Je(ajyj)y\mapsto e(a\cdot y)=\prod_{j\in J}e(a_jy_j), and for a velocity v∈ℝJv\in\mathbb R^J the curvature of the mode aa is λa=a⋅v=∑j∈Jajvj\lambda_a=a\cdot v=\sum_{j\in J}a_jv_j.

Frequency Packets

Let II and JJ be finite sets and δ\delta a real number. A frequency packet of parameter δ\delta over II and JJ is a triple (a,c,v)(a,c,v) of nonzero, pairwise nonparallel integer vectors ai∈ℤJa_i\in\mathbb Z^J and complex coefficients ci∈ℂc_i\in\mathbb C, one of each for every i∈Ii\in I, and a velocity v∈ℝJv\in\mathbb R^J, such that the real trigonometric polynomial F(y)=∑i∈Icie(ai⋅y)+∑i∈Icie(ai⋅y)¯,y∈𝕋J,F(y)=\sum_{i\in I}c_i\,e(a_i\cdot y)+\overline{\sum_{i\in I}c_i\,e(a_i\cdot y)}, \qquad y\in\mathbb T^J,(1) and the curvatures λi=ai⋅v\lambda_i=a_i\cdot v of its modes satisfy supy∈𝕋J|F(y)|≤1,(1−3δ)2≤2∑i∈I|ci|2,2∑i∈I|λi|<1,λi≠0and|ci|2≤Kδ2|λi|for every i∈I,\begin{aligned} &\sup_{y\in\mathbb T^J}|F(y)|\le 1,\qquad (1-3\delta)^2\le 2\sum_{i\in I}|c_i|^2, \qquad 2\sum_{i\in I}|\lambda_i|<1,\\ &\lambda_i\neq 0\quad\text{and}\quad |c_i|^2\le K_\delta^2\,|\lambda_i| \quad\text{for every } i\in I, \end{aligned}(2)✦ where Kδ2=(1+δ)3/(1−δ)K_\delta^2=(1+\delta)^3/(1-\delta).

A finite frequency packet of parameter δ\delta is a frequency packet of parameter δ\delta over a nonempty finite set II of modes and a finite set JJ of coordinates.✦ Write 𝒫δ\mathcal P_\delta for the collection of finite frequency packets of parameter δ\delta.

The Claim

Theorem 1. Let δ\delta be a real number with 0<δ<1/200<\delta<1/20. Then 𝒫δ≠⌀.\mathcal P_\delta\neq\varnothing.(3)✦

In the paper, the result reads: for every δ\delta with 0<δ<1/200<\delta<1/20 there are integers m≥1m\ge 1 and r≥2r\ge 2, nonzero pairwise nonparallel vectors a1,…,ar∈ℤma_1,\dots,a_r\in\mathbb Z^m, a real trigonometric polynomial FF on 𝕋m\mathbb T^m with Fourier support contained in A={±a1,…,±ar}A=\{\pm a_1,\dots,\pm a_r\}, and a vector v∈ℝmv\in\mathbb R^m such that ∥F∥∞≤1\|F\|_\infty\le 1, ∥F∥2≥1−3δ\|F\|_2\ge 1-3\delta and ∑a∈A|λa|<1\sum_{a\in A}|\lambda_a|<1, where λa=a⋅v\lambda_a=a\cdot v, while λa≠0\lambda_a\neq 0 and |Fˆ(a)|/|λa|≤Kδ|\widehat F(a)|/\sqrt{|\lambda_a|}\le K_\delta for every a∈Aa\in A.

What Is Checked

The claim (3) is stated over the finite frequency packets of parameter δ\delta, built on the frequency packets of (2) and their polynomials (1).