OctoInformBeta

OpenAI preprint, September 2026 · §7. Compact obstructions in infinite-dimensional Banach spaces

Finite witnesses: For each kk and A≥1A \ge 1, a finite subset of SS forces distortion above AA, and finite subsets govern embeddings into proper targets

Finite subsets of SS forcing large distortion in ℝk\mathbb{R}^k..

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 set SS is a subset of the Hilbert space ℓ2\ell^2, built over the plane from colored strips at a sequence of decreasing scales, with points of the plane lifted in directions orthogonal to it.

Given a dimension kk and a bound A≥1A \ge 1, the question is whether finitely many points of SS already rule out every map of distortion at most AA into Euclidean space ℝk\mathbb{R}^k: is there a finite subset of SS that admits no such map?

A normed abelian group EE is proper when its closed balls are compact. For a nonempty metric space XX and a proper target EE, the question also runs the other way: when every finite subset of XX maps into EE with no distance shrunk and none stretched by more than the factor AA, does XX itself?

The Set SS

The base plane ℝ2\mathbb{R}^2 is the set of pairs p=(x,y)p = (x, y) of real numbers, xx the horizontal and yy the vertical coordinate, with the sup distance max⁡(|x−x′|,|y−y′|)\max(|x - x'|, |y - y'|).✦ The Hilbert space ℓ2\ell^2 is the space of real sequences (cn)n≥0(c_n)_{n \ge 0} with ∑ncn2<∞\sum_n c_n^2 < \infty, with the norm ∥c∥=(∑ncn2)1/2\|c\| = (\sum_n c_n^2)^{1/2}.✦ The base vector of a point p=(x,y)p = (x, y) of the plane is the vector (x,y,0,0,…)(x, y, 0, 0, \dots) of ℓ2\ell^2, with xx at index 00 and yy at index 11.✦

The levels are j=0,1,2,…j = 0, 1, 2, \dots, and the scale of level jj is rj=1000−(j+1)r_j = 1000^{-(j+1)}.✦ The number of colors NjN_j and the width factor WjW_j of level jj are the positive integers Nj=u+1N_j = u + 1 and Wj=v+1W_j = v + 1, where ⟨⋅,⋅⟩:ℕ×ℕ→ℕ\langle \cdot, \cdot \rangle \colon \mathbb{N} \times \mathbb{N} \to \mathbb{N} is the pairing bijection, (m,s)(m, s) is the pair with ⟨m,s⟩=j\langle m, s \rangle = j, and (u,v)(u, v) is the pair with ⟨u,v⟩=m\langle u, v \rangle = m.

For a width hh and NN colors, the line is cut into the open strips (qh,(q+1)h)(qh, (q+1)h), q∈ℤq \in \mathbb{Z}, the strip of index qq having color qmod⁡Nq \bmod N; a real number tt lies in a strip of color ii when qh<t<(q+1)hqh < t < (q+1)h for an integer qq with qmod⁡N=iq \bmod N = i.✦ At level jj the horizontal strips qrj<y<(q+1)rjq r_j < y < (q+1) r_j and the vertical strips qWjrj<x<(q+1)Wjrjq W_j r_j < x < (q+1) W_j r_j are colored with NjN_j colors, and the region Uj,iU_{j,i} of level jj and color ii is the union of the strips of color ii of both kinds: Uj,i=⋃qmod⁡Nj=i(ℝ×(qrj,(q+1)rj)∪(qWjrj,(q+1)Wjrj)×ℝ).U_{j,i} = \bigcup_{q \bmod N_j = i} \Bigl( \mathbb{R} \times \bigl(q r_j, (q+1) r_j\bigr) \;\cup\; \bigl(q W_j r_j, (q+1) W_j r_j\bigr) \times \mathbb{R} \Bigr).(1) A point pp of the base plane is allowed at level jj for color ii when p∈Uj,ip \in U_{j,i}.✦

The coordinate index of level jj and color ii is 2+⟨j,i⟩2 + \langle j, i \rangle, and ej,ie_{j,i} is the unit vector of ℓ2\ell^2 at that index; distinct pairs (j,i)(j, i) receive distinct indices, all different from the indices 00 and 11 of the base vectors.✦ The level vector of level jj and color ii is rjej,ir_j e_{j,i}.✦

A finitely supported tuple is a sequence w=(wj)j≥0w = (w_j)_{j \ge 0} of natural numbers with wj=0w_j = 0 for all but finitely many jj; the entry wj=0w_j = 0 stands for no displacement at level jj, the entry wj=i+1w_j = i + 1 for a displacement of color ii.✦ A finitely supported tuple ww is admissible at a point pp of the base plane when wj≤Njw_j \le N_j and p∈Uj,wj−1p \in U_{j, w_j - 1} at every level jj with wj≠0w_j \neq 0.✦ The displacement vector of ww is the sum of the level vectors rjej,wj−1r_j e_{j, w_j - 1} over the levels jj with wj≠0w_j \neq 0✦, and the point (p,w)(p, w) is the vector of ℓ2\ell^2 equal to the base vector of pp plus the displacement vector of ww: (p,w)=(x,y,0,0,…)+∑j:wj≠0rjej,wj−1.(p, w) = (x, y, 0, 0, \dots) + \sum_{j \,:\, w_j \neq 0} r_j\, e_{j, w_j - 1}.(2)✦ The set SS is the subset of ℓ2\ell^2 of the points (p,w)(p, w) with pp in the base plane and ww a finitely supported tuple admissible at pp; its distance is the distance ∥z−z′∥\|z - z'\| of ℓ2\ell^2 between two points z,z′z, z' of SS.

The Claim

Theorem 1. Let kk be a natural number and A≥1A \ge 1. There is a finite subset FF of SS such that no map ff from FF to Euclidean space ℝk\mathbb{R}^k satisfies, for some constant a>0a > 0 and all z,z′∈Fz, z' \in F, a∥z−z′∥≤∥f(z)−f(z′)∥≤Aa∥z−z′∥;a \|z - z'\| \le \|f(z) - f(z')\| \le A a \|z - z'\|;(3)✦ such a subset FF exists with |F|≥2.|F| \ge 2.(4)✦

Theorem 2. Let XX be a nonempty metric space with distance dd, let EE be a proper normed abelian group and let A≥0A \ge 0. If every finite subset FF of XX admits a map f:F→Ef \colon F \to E satisfying, for all x,y∈Fx, y \in F, d(x,y)≤∥f(x)−f(y)∥≤Ad(x,y),d(x, y) \le \|f(x) - f(y)\| \le A\, d(x, y),(5)✦ then some map f:X→Ef \colon X \to E satisfies (5) for all x,y∈Xx, y \in X. If EE is moreover a real normed space and no map f:X→Ef \colon X \to E satisfies (5) for all x,y∈Xx, y \in X, then XX has a finite subset FF such that no map f:F→Ef \colon F \to E satisfies, for some constant a>0a > 0 and all x,y∈Fx, y \in F, ad(x,y)≤∥f(x)−f(y)∥≤Aad(x,y).a\, d(x, y) \le \|f(x) - f(y)\| \le A a\, d(x, y).(6)✦

What Is Checked

The first theorem is stated over the set SS of the points (2) with ww admissible at pp, under the distance of ℓ2\ell^2, and over Euclidean space ℝk\mathbb{R}^k; the second is stated over an arbitrary nonempty metric space and an arbitrary proper normed abelian group, a proper real normed space in its second statement.