OctoInformBeta

OpenAI preprint, September 2026 · §2. Weak compactness and a minimal invariant set

Lemma 2.1 [Weak compactness and separation facts]: Reflexive weak compactness and simultaneous approximation from pointwise weak closure

Weak topology and finite norm approximation.

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

Weak closedness concerns subsets of a normed space, whereas pointwise weak closure concerns families of vectors. Finite simultaneous approximation asks for one member of a convex family that approximates a given member at every coordinate in a prescribed finite list, using the original norm.

Let XX be a real normed space, let X*X^* be its continuous dual and X**X^{**} its bidual, and write XwX_{\mathrm w} for the same vector space with its weak topology. Let ι:X→Xw\iota:X\to X_{\mathrm w} be the canonical linear identification. A subset KK of XX is weakly closed when it satisfies (1), and weakly compact when it satisfies (2): ι(K) is closed in Xw.\iota(K)\text{ is closed in }X_{\mathrm w}.(1)✦ ι(K) is compact in Xw.\iota(K)\text{ is compact in }X_{\mathrm w}.(2)✦

The closed ball B¯(x,r)\overline B(x,r), for x∈Xx\in X and r∈ℝr\in\mathbb R, is the subset defined by (3): B¯(x,r)={y∈X:∥y−x∥≤r}.\overline B(x,r)=\{y\in X:\|y-x\|\leq r\}.(3)

A canonically reflexive real normed space XX is one for which the canonical map J:X→X**J:X\to X^{**} is surjective, with JJ given by (4): J(x)(ℓ)=ℓ(x)(x∈X,ℓ∈X*).J(x)(\ell)=\ell(x)\qquad(x\in X,\ \ell\in X^*).(4)✦

Let AA be an arbitrary index set and II a finite index set. Give XwAX_{\mathrm w}^{A} its product topology, so that closure in this space is pointwise weak closure. For a list a:I→Aa:I\to A, the finite evaluation map Ea:XwA→XIE_a:X_{\mathrm w}^{A}\to X^I is the map in (5): Ea(f)i=ι−1(f(a(i)))(i∈I).E_a(f)_i=\iota^{-1}(f(a(i)))\qquad(i\in I).(5)✦ Here XIX^I has its finite product norm, and ιI:XI→(XI)w\iota_I:X^I\to (X^I)_{\mathrm w} denotes the canonical identification with its weakly topologized counterpart.

The Claim

Theorem 1. Let XX be a real normed space. Its weak topology satisfies Xw is Hausdorff.X_{\mathrm w}\text{ is Hausdorff}.(6)✦ If K⊂XK\subset X is norm-closed and convex, then ι(K) is closed in Xw.\iota(K)\text{ is closed in }X_{\mathrm w}.(7)✦ For every x∈Xx\in X and r∈ℝr\in\mathbb R, ι(B¯(x,r)) is closed in Xw.\iota(\overline B(x,r))\text{ is closed in }X_{\mathrm w}.(8)✦ If XX is canonically reflexive and K⊂XK\subset X is norm-closed, bounded and convex, then ι(K) is compact in Xw.\iota(K)\text{ is compact in }X_{\mathrm w}.(9)✦ For every finite index set II and every a:I→Aa:I\to A, ιI∘Ea:XwA→(XI)w is continuous.\iota_I\circ E_a:X_{\mathrm w}^{A}\longrightarrow (X^I)_{\mathrm w} \text{ is continuous}.(10)✦ Let S⊂XwAS\subset X_{\mathrm w}^{A} be convex, let ff belong to its pointwise weak closure, and let a:I→Aa:I\to A be a finite list. For every ε>0\varepsilon>0 there is g∈Sg\in S such that ∥ι−1(g(a(i)))−ι−1(f(a(i)))∥<ε(i∈I).\bigl\|\iota^{-1}(g(a(i)))-\iota^{-1}(f(a(i)))\bigr\|<\varepsilon \qquad(i\in I).(11)✦

In the paper, the result reads: for a real normed space XX, every v∈Xv\in X admits ℓ∈X*\ell\in X^* with ∥ℓ∥≤1\|\ell\|\leq1 and ℓ(v)=∥v∥\ell(v)=\|v\|; norm-closed convex sets are weakly closed and the norm is weakly lower semicontinuous; on every finite Cartesian power XIX^I, the weak topology is the product weak topology and the weak and norm closures of convex sets coincide; if XX is reflexive and complete, every norm-closed bounded convex subset is weakly compact; arbitrary products of compact Hausdorff spaces are compact.

What Is Checked

The assertions (6), (7) and (8) concern the normed space XX and its weak counterpart XwX_{\mathrm w}; (9) additionally assumes canonical reflexivity through the map (4). The assertions (10) and (11) use the product weak topology on XwAX_{\mathrm w}^{A} and finite coordinate lists, with convexity required for the approximation assertion. The finite evaluation map in (5) is linear.