OctoInformBeta

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

Lemma 4.2: For smooth compactly supported gg, ∫g(x)e(Tβx2/2−ux)dx\int g(x)\,e(T\beta x^2/2-ux)\,dx is (T|β|)−1/2(T|\beta|)^{-1/2} times a phase times g(u/(Tβ))+O(1/T)g(u/(T\beta))+O(1/T), uniformly in uu

Uniform stationary phase for quadratic oscillatory integrals.

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 phase e(x)e(x) of a real number xx is the point e2πixe^{2\pi i x} of the unit circle, at angle 2πx2\pi x. For real numbers t≠0t \neq 0 and uu, the quadratic phase of curvature tt and linear frequency uu is the function x↦e(tx2/2−ux)x \mapsto e(tx^2/2 - ux) on the real line. Its stationary point is x=u/tx = u/t, where the derivative tx−utx - u of tx2/2−uxtx^2/2 - ux vanishes.

Take a smooth function g:ℝ→ℂg \colon \mathbb{R} \to \mathbb{C} of compact support, a real β≠0\beta \neq 0 and a large T>0T > 0, and integrate gg against the quadratic phase of curvature TβT\beta and linear frequency uu. Stationary phase approximates this integral through the value of gg at the stationary point u/(Tβ)u/(T\beta), with an error that decays as TT grows. The question is how fast it decays, and whether one bound serves every real uu, wherever the stationary point falls.

The library states the answer together with an exact identity that expresses the integral of a Schwartz function against a quadratic phase of any real curvature t≠0t \neq 0 through the inverse Fourier transform of the function.

The Inverse Fourier Transform

The inverse Fourier transform of an integrable function f:ℝ→ℂf \colon \mathbb{R} \to \mathbb{C} is the function f̌:ℝ→ℂ\check f \colon \mathbb{R} \to \mathbb{C} given by f̌(ξ)=∫ℝf(x)e(xξ)dx,ξ∈ℝ.\check f(\xi) = \int_{\mathbb{R}} f(x)\, e(x\xi)\, dx, \qquad \xi \in \mathbb{R}.(1)

The Claim

Theorem 1. (a) Let f:ℝ→ℂf \colon \mathbb{R} \to \mathbb{C} be a Schwartz function, and let t≠0t \neq 0 and uu be real numbers. Then ∫ℝf(x)e(tx22−ux)dx=(−it)−1/2∫ℝf̌(ξ)e(−(ξ+u)22t)dξ,\int_{\mathbb{R}} f(x)\, e\Big(\frac{t x^2}{2} - u x\Big)\, dx = (-it)^{-1/2} \int_{\mathbb{R}} \check f(\xi)\, e\Big(-\frac{(\xi + u)^2}{2t}\Big)\, d\xi,(2)✦ where (−it)1/2(-it)^{1/2} is the principal square root of −it-it.

(b) Let g:ℝ→ℂg \colon \mathbb{R} \to \mathbb{C} be a smooth function of compact support, and let β≠0\beta \neq 0, T>0T > 0 and uu be real numbers. Then there is a complex number EE with |E|≤(π/(T|β|))∫ℝξ2|ǧ(ξ)|dξ|E| \le \big(\pi/(T|\beta|)\big) \int_{\mathbb{R}} \xi^2\, |\check g(\xi)|\, d\xi such that ∫ℝg(x)e(Tβx22−ux)dx=e(sgn⁡(β)/8−u2/(2Tβ))T|β|(g(uTβ)+E).\int_{\mathbb{R}} g(x)\, e\Big(\frac{T\beta x^2}{2} - u x\Big)\, dx = \frac{e\big(\operatorname{sgn}(\beta)/8 - u^2/(2T\beta)\big)}{\sqrt{T|\beta|}} \Big(g\Big(\frac{u}{T\beta}\Big) + E\Big).(3)✦

In the paper, the result reads: for a smooth function gg of compact support and a fixed real β≠0\beta \neq 0, as T→∞T \to \infty, uniformly in real uu, ∫ℝg(x)e(Tβx2/2−ux)dx=e(sgn⁡(β)/8−u2/(2Tβ))(T|β|)−1/2(g(u/(Tβ))+Og,β(1/T))\int_{\mathbb{R}} g(x)\, e(T\beta x^2/2 - ux)\, dx = e(\operatorname{sgn}(\beta)/8 - u^2/(2T\beta))\, (T|\beta|)^{-1/2} \big(g(u/(T\beta)) + O_{g,\beta}(1/T)\big). In (3) that error term is EE, with an explicit bound that holds for every T>0T > 0 and does not depend on uu.

What Is Checked

Both claims are stated over the phase e(x)=e2πixe(x) = e^{2\pi i x} and the inverse Fourier transform (1): (2) for a Schwartz function ff and real t≠0t \neq 0 and uu, and (3) for a smooth compactly supported gg and real β≠0\beta \neq 0, T>0T > 0 and uu.