OctoInformBeta

OpenAI preprint, September 2026 · §3. A measure on the continuum gives separable quotients

Lemma 3.3: Under continuum additivity and a continuum lower bound on the norm density of the evaluation image

Uniformly nonzero functionals with null evaluations.

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 issue is the relation between the norm density of an evaluation image and almost everywhere vanishing for families of bidual functionals. Let XX be a real normed space and let E⊆X*E\subseteq X^* be a norm separable linear subspace, with the norm and real vector space operations inherited from X*X^*.

The evaluation operator RE:X→E*R_E\colon X\to E^* is the continuous linear map defined by (1). REx(f)=f(x)(x∈X,f∈E).R_E x(f)=f(x)\qquad (x\in X,\ f\in E).(1)✦ The closed unit ball BXB_X and its evaluation image Y⊆E*Y\subseteq E^* are the sets in (2); YY carries the norm topology inherited from E*E^*. BX={x∈X:∥x∥≤1},Y=RE(BX).B_X=\{x\in X:\|x\|\leq 1\},\qquad Y=R_E(B_X).(2)

Write 𝔠=2ℵ0\mathfrak c=2^{\aleph_0} and let κ\kappa be the initial ordinal of cardinality 𝔠\mathfrak c. A measure μ\mu on the full power set 𝒫(κ)\mathcal P(\kappa) is 𝔠\mathfrak c-additive if every pairwise disjoint family (Ai)i∈I(A_i)_{i\in I} of subsets of κ\kappa, indexed by a set II of cardinality less than 𝔠\mathfrak c, satisfies (3). μ(⋃i∈IAi)=∑i∈Iμ(Ai).\mu\!\left(\bigcup_{i\in I}A_i\right)=\sum_{i\in I}\mu(A_i).(3)

The Claim

Theorem 1. Let μ\mu be a 𝔠\mathfrak c-additive probability measure on 𝒫(κ)\mathcal P(\kappa) that vanishes on singletons. Suppose that every norm dense subset of YY has cardinality at least 𝔠\mathfrak c. There are ε>0\varepsilon>0 and a family (hα)α<κ(h_\alpha)_{\alpha<\kappa} in the unit sphere SE**S_{E^{**}} whose pullbacks fα=hα∘RE∈X*f_\alpha=h_\alpha\circ R_E\in X^* satisfy (4). ε<∥fα∥≤1(α<κ),μ{α<κ:hα(y)≠0}=0(y∈Y).\begin{aligned} \varepsilon&<\|f_\alpha\|\leq 1 &&(\alpha<\kappa),\\ \mu\{\alpha<\kappa:h_\alpha(y)\ne 0\}&=0 &&(y\in Y). \end{aligned}(4)✦

In the paper, the result reads: if XX and EE satisfy the hypotheses of Lemma 3.2, including the measure hypothesis and the absence of an infinite-dimensional norm separable quotient of XX, then there are ε>0\varepsilon>0 and hα∈SE**h_\alpha\in S_{E^{**}} for α<κ\alpha<\kappa such that fα=hα∘REf_\alpha=h_\alpha\circ R_E satisfies ε<∥fα∥≤1\varepsilon<\|f_\alpha\|\leq 1 and μ{α<κ:hα(y)≠0}=0\mu\{\alpha<\kappa:h_\alpha(y)\ne 0\}=0 for every y∈RE(BX)y\in R_E(B_X).

What Is Checked

The claim in (4) is stated over evaluation on a norm separable dual subspace, a probability measure on the full power set of κ\kappa satisfying (3) and vanishing on singletons, and an explicit continuum lower bound on the norm density of the image in (2). Under these measure hypotheses, the continuum is a regular cardinal, every union of fewer than 𝔠\mathfrak c null sets is null, and every subset of κ\kappa of cardinality less than 𝔠\mathfrak c is null.