OctoInformBeta

OpenAI preprint, September 2026 · §2. The matrix construction target · §2

§2 (displayed): enlarging the polar scale: For 0<c≤d0 < c \le d, cK∘⊆dK∘cK^\circ \subseteq dK^\circ, and when L∘L^\circ is finitely covered by translates of cK∘cK^\circ, N(L∘,dK∘)≤N(L∘,cK∘)N(L^\circ, dK^\circ) \le N(L^\circ, cK^\circ)

Larger dilates of a polar body need no more translates.

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 ι\iota be a finite index set. On the dual side of the duality conjecture for metric entropy one covers the polar L∘L^\circ of the cube L=[−1,1]ιL = [-1,1]^\iota, the ℓ1\ell_1 unit ball, by translates of a dilate tK∘tK^\circ, with scale t>0t > 0, of the polar of a set KK, and counts the translates. The question here is how that count moves with the scale: can covering L∘L^\circ by translates of a larger dilate of K∘K^\circ ever take more of them?

The library states the answer for every subset KK of ℝι\mathbb{R}^\iota, convex or not, symmetric or not, and for every pair of scales 0<c≤d0 < c \le d.

Polars, Cubes and Covers

The real space ℝι\mathbb{R}^\iota over an index set ι\iota is the vector space of all functions from ι\iota to ℝ\mathbb{R}, a point xx being the family (xi)i∈ι(x_i)_{i \in \iota} of its coordinates.✦ For a finite ι\iota, the pairing of two points xx and yy of ℝι\mathbb{R}^\iota is their standard inner product ⟨x,y⟩=∑i∈ιxiyi.\langle x, y \rangle = \sum_{i \in \iota} x_i y_i .(1)✦

The polar K∘K^\circ of a subset KK of ℝι\mathbb{R}^\iota is the set of points whose pairing with every point of KK is at most 11, K∘={y∈ℝι:⟨x,y⟩≤1 for every x∈K},K^\circ = \{\, y \in \mathbb{R}^\iota : \langle x, y \rangle \le 1 \text{ for every } x \in K \,\},(2)✦ and for a real number cc, the dilate cK∘cK^\circ of K∘K^\circ is the set cK∘={cy:y∈K∘}cK^\circ = \{\, cy : y \in K^\circ \,\}.

The cube LL of ℝι\mathbb{R}^\iota is the unit ball of the sup norm, L={x∈ℝι:|xi|≤1 for every i∈ι}=[−1,1]ι.L = \{\, x \in \mathbb{R}^\iota : |x_i| \le 1 \text{ for every } i \in \iota \,\} = [-1,1]^\iota .(3)✦

For subsets AA and BB of ℝι\mathbb{R}^\iota and points z1,…,zMz_1, \dots, z_M of ℝι\mathbb{R}^\iota, the translates z1+B,…,zM+Bz_1 + B, \dots, z_M + B cover AA when every point xx of AA has x−zj∈Bx - z_j \in B for some jj.✦ A pair (A,B)(A, B) of subsets of ℝι\mathbb{R}^\iota is finitely coverable when finitely many translates of BB cover AA.✦

The covering number N(A,B)N(A, B) of a subset AA of ℝι\mathbb{R}^\iota by a subset BB is the least number of translates of BB, with centers anywhere in ℝι\mathbb{R}^\iota, that cover AA, N(A,B)=min⁡{M∈ℕ:A⊆⋃j=1M(zj+B) for some z1,…,zM∈ℝι},N(A, B) = \min \Bigl\{\, M \in \mathbb{N} : A \subseteq \bigcup_{j=1}^{M} (z_j + B) \text{ for some } z_1, \dots, z_M \in \mathbb{R}^\iota \,\Bigr\},(4)✦ and it is 00 when no finite family of translates of BB covers AA.

The Claim

Theorem 1. Let ι\iota be a finite set, KK a subset of ℝι\mathbb{R}^\iota, and cc and dd real numbers with 0<c≤d0 < c \le d. Then cK∘⊆dK∘.cK^\circ \subseteq dK^\circ .(5)✦ If moreover the pair (L∘,cK∘)(L^\circ, cK^\circ) is finitely coverable, then the pair (L∘,dK∘)(L^\circ, dK^\circ) is finitely coverable and N(L∘,dK∘)≤N(L∘,cK∘).N(L^\circ, dK^\circ) \le N(L^\circ, cK^\circ) .(6)✦

What Is Checked

The claim is stated over the real space ℝι\mathbb{R}^\iota of a finite index set with the pairing (1), the polar (2) of an arbitrary subset KK and its dilates, the cube (3), and the covering number (4) together with finite coverability by translates.