OctoInformBeta

OpenAI preprint, September 2026 · §3. An anchor for successive choices

Lemma 3.2 [Finite simultaneous approximation]: Simultaneously in norm on finite input sets, with nonexpansiveness preserved when assumed

Convex families approximate weak closure points.

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

Let XX be a real normed space and K⊆XK\subseteq X. Finite simultaneous approximation seeks a single selfmap whose values are close to those of a given selfmap at every input in a prescribed finite set. All errors and nonexpansiveness estimates use the ambient norm of XX.

Write XwX_{\mathrm w} for the weakly topologized copy of XX. The weak topological carrier KwK_{\mathrm w} of KK is the subset of XwX_{\mathrm w} consisting of the weak representatives of points of KK, equipped with the induced weak topology.✦ The norm-to-weak identification ι:K→Kw\iota:K\to K_{\mathrm w} is the bijection sending each point to its weak representative.✦

The space 𝒲(K)=KwK\mathcal W(K)=K_{\mathrm w}^{K} of weak-valued selfmaps of KK consists of all functions from KK to KwK_{\mathrm w}, with the topology of pointwise weak convergence.✦ For B⊆𝒲(K)B\subseteq\mathcal W(K), let B¯\overline B denote its closure in this topology.

The ambient weak-valued map Af:K→XwA_f:K\to X_{\mathrm w} associated with f∈𝒲(K)f\in\mathcal W(K) is the function ff with its values regarded as points of XwX_{\mathrm w}.✦ The underlying norm selfmap f̃:K→K\widetilde f:K\to K of ff is specified by (1): f̃(a)=ι−1(f(a))(a∈K).\widetilde f(a)=\iota^{-1}(f(a))\qquad(a\in K).(1)✦

A convex family BB of weak-valued selfmaps of KK is a subset of 𝒲(K)\mathcal W(K) whose ambient maps form a convex subset of the real vector space XwKX_{\mathrm w}^{K}, as specified by (2): (1−t)Af+tAg∈{Au:u∈B}(f,g∈B,0≤t≤1).(1-t)A_f+tA_g\in\{A_u:u\in B\} \qquad(f,g\in B,\ 0\leq t\leq1).(2)✦

The Claim

Theorem 1. Let B⊆𝒲(K)B\subseteq\mathcal W(K) be a convex family and let h∈B¯h\in\overline B. For a finite index set II, a tuple (ai)i∈I(a_i)_{i\in I} of points of KK, and ε>0\varepsilon>0, there is g∈Bg\in B satisfying (3): ∥g̃(ai)−h̃(ai)∥<ε(i∈I).\|\widetilde g(a_i)-\widetilde h(a_i)\|<\varepsilon \qquad(i\in I).(3)✦ For a finite subset F⊆KF\subseteq K and ε>0\varepsilon>0, there is g∈Bg\in B satisfying (4): ∥g̃(a)−h̃(a)∥<ε(a∈F).\|\widetilde g(a)-\widetilde h(a)\|<\varepsilon \qquad(a\in F).(4)✦ If every g̃\widetilde g with g∈Bg\in B is nonexpansive, then (5) holds: ∥h̃(a)−h̃(b)∥≤∥a−b∥(a,b∈K).\|\widetilde h(a)-\widetilde h(b)\|\leq\|a-b\| \qquad(a,b\in K).(5)✦

In the paper, the result reads: if B⊆KKB\subseteq K^K is a nonempty convex family of nonexpansive maps and hh belongs to its pointwise weak closure, then hh is nonexpansive and, for every integer q≥1q\geq1, points a1,…,aq∈Ka_1,\ldots,a_q\in K and δ>0\delta>0, some g∈Bg\in B satisfies max⁡1≤i≤q∥g(ai)−h(ai)∥<δ\max_{1\leq i\leq q}\|g(a_i)-h(a_i)\|<\delta.

What Is Checked

The statements (3) and (4) are over a convex family of weak-valued selfmaps of an arbitrary subset KK of a real normed space, and (5) adds the nonexpansiveness hypothesis.

The norm-to-weak identification satisfies the inverse relations (6): ι−1(ι(a))=a(a∈K),ι(ι−1(z))=z(z∈Kw).\iota^{-1}(\iota(a))=a\quad(a\in K),\qquad \iota(\iota^{-1}(z))=z\quad(z\in K_{\mathrm w}).(6)