OctoInformBeta

OpenAI preprint, September 2026 · §3. Common mass, smoothing, and score moments

Lemma 3.1: Independent copies of a centered finite random variable with a bounded even moment

Coefficient energy bounds multilinear even moments.

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 finite probability law PP on a finite outcome set α\alpha is a mass function p:α→ℝp:\alpha\to\mathbb R satisfying (1). p(a)≥0(a∈α),∑a∈αp(a)=1.p(a)\ge 0\quad(a\in\alpha),\qquad \sum_{a\in\alpha}p(a)=1.(1)✦ The expectation 𝔼P[f]\mathbb E_P[f] of a real function ff on α\alpha is the weighted sum in (2). 𝔼P[f]=∑a∈αp(a)f(a).\mathbb E_P[f]=\sum_{a\in\alpha}p(a)f(a).(2)✦

For an integer n≥0n\ge0, let I={1,…,n}I=\{1,\ldots,n\}. The product law P⊗nP^{\otimes n} on αI\alpha^I is the probability law of nn independent samples from PP, with masses given by (3). P⊗n(x)=∏j∈Ip(xj),x∈αI.P^{\otimes n}(x)=\prod_{j\in I}p(x_j),\qquad x\in\alpha^I.(3)✦

For a real coefficient array AA on the subsets of II and a function Z:α→ℝZ:\alpha\to\mathbb R, the multilinear polynomial FA:αI→ℝF_A:\alpha^I\to\mathbb R is the function in (4). FA(x)=∑S⊆IA(S)∏j∈SZ(xj).F_A(x)=\sum_{S\subseteq I}A(S)\prod_{j\in S}Z(x_j).(4)✦ The weighted coefficient energy ℰR(A)\mathcal E_R(A), for R∈ℝR\in\mathbb R, is the counting-measure sum in (5). ℰR(A)=∑S⊆IA(S)2R2|S|.\mathcal E_R(A)=\sum_{S\subseteq I}A(S)^2R^{2|S|}.(5)✦

The squared norm ∥f∥L2k(Q)2\|f\|_{L^{2k}(Q)}^2 of a real function ff on the outcome set of a finite probability law QQ, with k≥1k\ge1 an integer, is the probability-weighted moment in (6).✦ For a real function ff on the outcome set of a finite probability law QQ, the even-moment quantity ∥f∥L2k(Q)2\|f\|_{L^{2k}(Q)}^2 at k=0k=0 is the real number 11. ∥f∥L2k(Q)2=(𝔼Q[|f|2k])1/k.\|f\|_{L^{2k}(Q)}^2 =\bigl(\mathbb E_Q[|f|^{2k}]\bigr)^{1/k}.(6)✦

The Claim

Theorem 1. Let k≥1k\ge1 be an integer, and let Z:α→ℝZ:\alpha\to\mathbb R be centered under PP, so that 𝔼P[Z]=0\mathbb E_P[Z]=0. Let n≥0n\ge0 be an integer and AA a real coefficient array on the subsets of II. If 𝔼P[|Z|2k]≤1\mathbb E_P[|Z|^{2k}]\le1, the even moment obeys (7). 𝔼P⊗n[FA2k]≤ℰ12k(A)k.\mathbb E_{P^{\otimes n}}[F_A^{2k}] \le \mathcal E_{12k}(A)^k.(7)✦ If L>0L>0 and 𝔼P[|Z|2k]≤L2k\mathbb E_P[|Z|^{2k}]\le L^{2k}, the squared norm obeys (8). ∥FA∥L2k(P⊗n)2≤ℰ12kL(A).\|F_A\|_{L^{2k}(P^{\otimes n})}^2 \le \mathcal E_{12kL}(A).(8)✦

In the paper, the result reads: for an even integer t≥2t\ge2, h≥0h\ge0, and independent centered real random variables (Zj)j∈I(Z_j)_{j\in I} with ∥Zj∥Lt≤h\|Z_j\|_{L^t}\le h, every integer r≥1r\ge1 and real coefficient array AA on the rr-subsets of II satisfy ∥∑|S|=rA(S)∏j∈SZj∥Lt≤(C1th)r(∑|S|=rA(S)2)1/2\bigl\|\sum_{|S|=r}A(S)\prod_{j\in S}Z_j\bigr\|_{L^t} \le(C_1th)^r\bigl(\sum_{|S|=r}A(S)^2\bigr)^{1/2}, with C1C_1 absolute.

What Is Checked

The claims are stated over finite probability laws and their product laws, arbitrary real coefficient arrays indexed by subsets, and centered real functions with the stated even moment bounds.