OctoInformBeta

OpenAI preprint, September 2026 · Section 1. Introduction

Section 1 (setting): real Lipschitz-free spaces and approximation properties: The closed evaluation span, its metric embedding, and compact-uniform approximation of the identity

Lipschitz-free spaces and finite-rank 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

Let (M,d,o)(M,d,o) be a pointed metric space and let XX be a real normed space. Evaluation associates linear functionals to the points of MM. The approximation conditions concern the identity operator on XX and uniform error on norm-compact subsets of XX.

Write ℒ(X)\mathcal L(X) for the space of bounded linear operators from XX to XX, with the operator norm.

The Claim

Definition 1. The space Lip⁡0(M)\operatorname{Lip}_0(M) of basepoint-vanishing real Lipschitz functions is the normed space in (1), equipped with the Lipschitz norm. Lip⁡0(M)={f:M→ℝ:f(o)=0,∃C≥0∀x,y∈M,|f(x)−f(y)|≤Cd(x,y)}.\operatorname{Lip}_0(M) =\left\{f:M\to\mathbb R: f(o)=0,\quad \exists C\geq0\ \forall x,y\in M,\quad |f(x)-f(y)|\leq C d(x,y)\right\}.(1)✦

Write Lip⁡0(M)*\operatorname{Lip}_0(M)^* for the continuous real dual of Lip⁡0(M)\operatorname{Lip}_0(M). For x∈Mx\in M, the evaluation functional δx∈Lip⁡0(M)*\delta_x\in\operatorname{Lip}_0(M)^* is the bounded linear functional given by (2). δx(f)=f(x)(f∈Lip⁡0(M)).\delta_x(f)=f(x)\qquad\bigl(f\in\operatorname{Lip}_0(M)\bigr).(2)✦

The real Lipschitz-free space F(M)F(M) of the pointed metric space (M,d,o)(M,d,o) is the closed linear span of the evaluation functionals in Lip⁡0(M)*\operatorname{Lip}_0(M)^*, with its inherited norm, as in (3). The canonical Dirac map δ:M→F(M)\delta:M\to F(M) sends xx to the evaluation functional δx\delta_x, regarded as an element of this closed span.✦ F(M)=span⁡ℝ{δx:x∈M}¯⊆Lip⁡0(M)*.F(M)=\overline{\operatorname{span}_{\mathbb R} \{\delta_x:x\in M\}} \subseteq\operatorname{Lip}_0(M)^*.(3)✦

A real normed space XX has the approximation property (AP) when it satisfies the finite-rank approximation condition (4). For every norm-compact C⊆X and every ε>0, there exist T∈ℒ(X) and η∈ℝ such thatdim⁡T(X)<∞,η<ε,∥Tu−u∥≤η(u∈C).\begin{gathered} \text{For every norm-compact }C\subseteq X\text{ and every }\varepsilon>0, \\ \text{ there exist }T\in\mathcal L(X)\text{ and }\eta\in\mathbb R\text{ such that} \\ \dim T(X)<\infty,\qquad \eta<\varepsilon,\qquad \|Tu-u\|\leq\eta\quad(u\in C). \end{gathered}(4)✦

For λ∈ℝ\lambda\in\mathbb R, a real normed space XX has the λ\lambda-bounded approximation property (λ\lambda-BAP) when it satisfies (5). For every norm-compact C⊆X and every ε>0, there exist T∈ℒ(X) and η∈ℝ such thatdim⁡T(X)<∞,∥T∥≤λ,η<ε,∥Tu−u∥≤η(u∈C).\begin{gathered} \text{For every norm-compact }C\subseteq X\text{ and every }\varepsilon>0, \\ \text{ there exist }T\in\mathcal L(X)\text{ and }\eta\in\mathbb R\text{ such that} \\ \dim T(X)<\infty,\qquad \|T\|\leq\lambda,\qquad \eta<\varepsilon, \qquad \|Tu-u\|\leq\eta\quad(u\in C). \end{gathered}(5)✦

Theorem 1. For every pointed metric space (M,d,o)(M,d,o) and all x,y∈Mx,y\in M, the canonical Dirac map satisfies the distance identity (6). ∥δx−δy∥=d(x,y).\|\delta_x-\delta_y\|=d(x,y).(6)✦

In the paper, the result reads: for a pointed metric space (M,d,o)(M,d,o), Lip⁡0(M)\operatorname{Lip}_0(M) consists of real Lipschitz functions vanishing at oo, with the Lipschitz norm, F(M)F(M) is the closed linear span of evaluations δx\delta_x in Lip⁡0(M)*\operatorname{Lip}_0(M)^*, δo=0\delta_o=0 and ∥δx−δy∥=d(x,y)\|\delta_x-\delta_y\|=d(x,y), AP means uniform approximation of the identity on every norm-compact set by bounded finite-rank linear operators, and, for λ≥1\lambda\geq1, λ\lambda-BAP additionally bounds their norms by λ\lambda, while the metric approximation property (MAP) is 11-BAP.

What Is Checked

The definitions and distance identity are stated over arbitrary pointed metric spaces, and the approximation conditions over real normed spaces, with a real bound λ\lambda for λ\lambda-BAP.

The basepoint-vanishing Lipschitz functions form a real vector subspace of ℝM\mathbb R^M, and each evaluation has norm at most d(x,o)d(x,o).