OctoInformBeta

OpenAI preprint, September 2026 · Section 2. Smooth passages and spectral densities

Lemma 2.3: The cutoff law lies below the prescribed probability density and loses arbitrarily little mass

A smooth jitter approximates a positive density.

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

A smooth real function on the circle has a distribution of values on ℝ\mathbb R. The approximation problem is to prescribe that distribution after restricting it by a smooth cutoff supported on an arc where the function is increasing.

Write 𝕋=ℝ/ℤ\mathbb T=\mathbb R/\mathbb Z, with normalized Haar measure dada represented by Lebesgue measure on (0,1)(0,1). A strictly positive smooth probability density ρ\rho on ℝ\mathbb R is an integrable C∞C^\infty real function satisfying ρ(y)>0(y∈ℝ),∫ℝρ(y)dy=1.\rho(y)>0\quad(y\in\mathbb R), \qquad \int_{\mathbb R}\rho(y)\,dy=1.(1) Here dydy is Lebesgue measure on ℝ\mathbb R.

A smooth real circle jitter J:𝕋→ℝJ:\mathbb T\to\mathbb R is a function represented on ℝ\mathbb R by a smooth function satisfying J(a+1)=J(a)J(a+1)=J(a). For 0<δ<1/80<\delta<1/8, let Iδ=(δ,1−δ)I_\delta=(\delta,1-\delta), also viewed as an arc of 𝕋\mathbb T. A cutoff ψ:ℝ→[0,1]\psi:\mathbb R\to[0,1] on IδI_\delta is a smooth function whose support is a compact subset of IδI_\delta; the support is the closure of the set where ψ\psi is nonzero. The cutoff law μJ,ψ\mu_{J,\psi} is the measure on ℝ\mathbb R defined by μJ,ψ=J*(ψ(a)2𝟏(0,1)(a)da).\mu_{J,\psi} =J_*\bigl(\psi(a)^2\mathbf 1_{(0,1)}(a)\,da\bigr).(2)

The Claim

Theorem 1 (Lemma 2.3). Let ρ\rho be a strictly positive smooth probability density as in (1), and let ε>0\varepsilon>0. There exist δ∈ℝ\delta\in\mathbb R and smooth real functions J,ψ,ρ̃J,\psi,\widetilde\rho on ℝ\mathbb R, with ψ\psi and ρ̃\widetilde\rho compactly supported, satisfying (3), where μJ,ψ\mu_{J,\psi} is the cutoff law (2): 0<δ<18,2δ<ε,J(a+1)=J(a)(a∈ℝ),J′(a)>0(a∈Iδ),supp⁡ψ⊂Iδ,0≤ψ(a)≤1(a∈ℝ),0≤ρ̃(y)≤ρ(y)(y∈ℝ),ρ̃∈L1(ℝ),μJ,ψ=ρ̃(y)dy,∫ℝ|ρ(y)−ρ̃(y)|dy<ε,∫01(1−ψ(a)2)da<ε.\begin{aligned} &0<\delta<\tfrac18,\qquad 2\delta<\varepsilon,\\ &J(a+1)=J(a)\quad(a\in\mathbb R), \qquad J'(a)>0\quad(a\in I_\delta),\\ &\operatorname{supp}\psi\subset I_\delta, \qquad 0\leq\psi(a)\leq1\quad(a\in\mathbb R),\\ &0\leq\widetilde\rho(y)\leq\rho(y)\quad(y\in\mathbb R), \qquad \widetilde\rho\in L^1(\mathbb R),\\ &\mu_{J,\psi}=\widetilde\rho(y)\,dy,\\ &\int_{\mathbb R}|\rho(y)-\widetilde\rho(y)|\,dy<\varepsilon, \qquad \int_0^1\bigl(1-\psi(a)^2\bigr)\,da<\varepsilon. \end{aligned}(3)✦

In the paper, the result reads: for every strictly positive smooth probability density ρ\rho on ℝ\mathbb R and every ε>0\varepsilon>0, there are a smooth real circle function JJ, an open arc I′⊂𝕋I'\subset\mathbb T with |𝕋\I′|<ε|\mathbb T\setminus I'|<\varepsilon and J′>0J'>0 on I′I', and a smooth function 0≤ψ≤10\leq\psi\leq1 compactly supported in I′I', such that J*(ψ2da)=ρ̃(y)dyJ_*(\psi^2\,da)=\widetilde\rho(y)\,dy for a smooth compactly supported density ρ̃\widetilde\rho with ∫ℝ|ρ−ρ̃|dy<ε\int_{\mathbb R}|\rho-\widetilde\rho|\,dy<\varepsilon and ∫𝕋(1−ψ2)da<ε\int_{\mathbb T}(1-\psi^2)\,da<\varepsilon; the cutoff law is a subprobability and is not renormalized.

What Is Checked

The claim is stated over a strictly positive smooth probability density on ℝ\mathbb R and a positive tolerance, with the cutoff law understood as the pushforward of Lebesgue measure weighted by the squared cutoff on (0,1)(0,1).