OctoInformBeta

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

Compact assembly and its covering bound: Finite clusters within aj/100a_j/100 of ajva_j v, aj=2−(j+1)a_j = 2^{-(j+1)}, ∥v∥=1\|v\| = 1, each doubling at most LL, form with 00 a compact set doubling at most 1+5L21+5L^2

Doubling clusters at dyadic scales assemble into a compact doubling set.

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

Compact subsets of an infinite-dimensional real Banach space can be assembled from finite pieces, one at each of a sequence of scales tending to zero. Take a unit vector vv of a real normed space BB and the points v/2,v/4,v/8,…v/2, v/4, v/8, \dots of the ray through vv, which converge to the origin. Around each of them place a finite set, small against its distance from the origin, and add the origin itself. Is the resulting set compact, and if every piece is doubling with a common constant, is it doubling, and with what constant?

The configuration raises two further questions: how far apart points of two different pieces are, and how many pieces have a scale comparable with a given radius.

Doubling Sets and Clusters

A subset KK of a metric space XX is doubling at most Λ\Lambda, for a natural number Λ\Lambda, if for every point xx of KK and every r>0r > 0 there are points z1,…,zmz_1, \dots, z_m of KK, with m≤Λm \le \Lambda, such that {y∈K:d(y,x)<r}⊆⋃k=1m{y∈K:d(y,zk)<r/2}.\{\, y \in K : d(y, x) < r \,\} \subseteq \bigcup_{k=1}^{m} \{\, y \in K : d(y, z_k) < r/2 \,\}.(1)✦

A subset KK of a metric space XX is ambient doubling at most Λ\Lambda in XX, for a natural number Λ\Lambda, if for every point xx of KK and every r>0r > 0 there are points z1,…,zmz_1, \dots, z_m of XX that lie in KK, with m≤Λm \le \Lambda, such that the inclusion (1) holds.✦

The cluster scale of index j≥0j \ge 0 is the number aj=2−(j+1).a_j = 2^{-(j+1)}.(2)✦ For a unit vector vv of a real normed space BB, a cluster of index jj along vv is a finite subset CjC_j of BB whose points all lie within distance aj/100a_j/100 of the point ajva_j v: ∥z−ajv∥≤aj/100for every z∈Cj.\| z - a_j v \| \le a_j/100 \qquad \text{for every } z \in C_j.(3)

The Claim

Theorem 1. Let BB be a real normed space, vv a unit vector of BB and LL a natural number, and let C0,C1,C2,…C_0, C_1, C_2, \dots be clusters along vv as in (3), CjC_j of index jj.

(a) If every CjC_j is doubling at most LL, then {0}∪⋃j≥0Cjis compact and doubling at most 1+5L2.\{0\} \cup \bigcup_{j \ge 0} C_j \quad \text{is compact and doubling at most } 1 + 5L^2.(4)✦

(b) If every CjC_j is ambient doubling at most LL in BB, then {0}∪⋃j≥0Cjis ambient doubling at most 1+5L2 in B.\{0\} \cup \bigcup_{j \ge 0} C_j \quad \text{is ambient doubling at most } 1 + 5L^2 \text{ in } B.(5)✦

(c) If oo is a point of a metric space XX and E0,E1,E2,…E_0, E_1, E_2, \dots are finite subsets of XX shrinking to oo, in the sense that for every ε>0\varepsilon > 0 there is an NN such that every point of every EjE_j with j≥Nj \ge N lies at distance less than ε\varepsilon from oo, then {o}∪⋃j≥0Ejis compact.\{o\} \cup \bigcup_{j \ge 0} E_j \quad \text{is compact}.(6)✦

(d) If i<ji < j are indices and pp, qq are points of BB with ∥p−aiv∥≤ai/100\|p - a_i v\| \le a_i/100 and ∥q−ajv∥≤aj/100\|q - a_j v\| \le a_j/100, then ∥p−q∥≥97200ai.\|p - q\| \ge \tfrac{97}{200}\, a_i.(7)✦

(e) For every r>0r > 0, #{j≥0:r/4<aj≤40097r}≤5.\#\{\, j \ge 0 : r/4 < a_j \le \tfrac{400}{97}\, r \,\} \le 5.(8)✦

What Is Checked

The claim is stated over the doubling condition (1) and its ambient form, and over clusters (3) at the cluster scales (2) along a unit vector of a real normed space; part (c) is stated over finite subsets of an arbitrary metric space.