OctoInformBeta

OpenAI preprint, September 2026 · §2. Affine labels with a prescribed tolerance

Lemma 2.1: Positive tolerances on the exponent cube admit labels with small offsets and control of active limits

Finite affine labels attain prescribed slope tolerances.

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

Affine functions on positive coordinate space can be compared at states whose coordinates are powers of a small parameter. The question is how closely the slope of an active branch can follow a limiting exponent.

Let mm be a nonnegative integer and let a<ba<b be real numbers. The exponent cube [a,b]m[a,b]^m is the set of vectors whose coordinates satisfy (1): [a,b]m={p∈ℝm:a≤pi≤b for every i}.[a,b]^m=\{p\in\mathbb R^m:a\leq p_i\leq b\text{ for every }i\}.(1)✦ Write ∥⋅∥∞\|\cdot\|_\infty for the supremum norm on exponent vectors. For a real base hh and an exponent vector p∈ℝmp\in\mathbb R^m, the power point Xh(p)X_h(p) is the vector of coordinate powers in (2): Xh(p)=(hpi)i=1m.X_h(p)=(h^{p_i})_{i=1}^m.(2)✦

A parameterized affine label L=(rL,cL)L=(r_L,c_L) in dimension mm consists of a fixed slope rL∈ℝmr_L\in\mathbb R^m and an offset function cL:ℝ→ℝc_L:\mathbb R\to\mathbb R✦; its value at a real parameter hh and vector x∈ℝmx\in\mathbb R^m is (3): Lh(x)=rL⋅x+cL(h)=∑i=1m(rL)ixi+cL(h).L_h(x)=r_L\cdot x+c_L(h) =\sum_{i=1}^m(r_L)_i x_i+c_L(h).(3)✦ For a finite nonempty family Λ\Lambda of affine labels, a label L∈ΛL\in\Lambda is active at (h,x)(h,x) if it attains the minimum in (4), including every tied minimizer: Lh(x)=minJ∈ΛJh(x).L_h(x)=\min_{J\in\Lambda}J_h(x).(4)

Affine Approximation Certificates

Let E:ℝm→ℝE:\mathbb R^m\to\mathbb R be a tolerance function positive on [a,b]m[a,b]^m, with no regularity assumption. A finite family Λ\Lambda of affine labels in dimension mm has sequential slope approximation for a,b,Ea,b,E if the following condition holds: for a fixed label L∈ΛL\in\Lambda, a positive sequence hn→0h_n\to0, and a sequence pn∈ℝmp^n\in\mathbb R^m converging to p∈[a,b]mp\in[a,b]^m, activity of LL at Xhn(pn)X_{h_n}(p^n) for every nn entails (5): ∥p−rL∥∞<E(rL).\|p-r_L\|_\infty<E(r_L).(5)✦

An affine approximation certificate Λ\Lambda for m,a,b,Em,a,b,E is a finite nonempty family of affine labels with sequential slope approximation whose slopes and offsets satisfy (6): rL∈[a,b]m,cL(h)=o(ha)(h→0+),L∈Λ.r_L\in[a,b]^m,\qquad c_L(h)=o(h^a)\quad(h\to0^+),\qquad L\in\Lambda.(6)✦ Let 𝒜(m,a,b,E)\mathcal A(m,a,b,E) denote the set of these certificates.

The Claim

Theorem 1. Let mm be a nonnegative integer, let a<ba<b be real numbers, and let E:ℝm→ℝE:\mathbb R^m\to\mathbb R be positive on [a,b]m[a,b]^m. The existence assertion is (7). 𝒜(m,a,b,E)≠⌀.\mathcal A(m,a,b,E)\ne\varnothing.(7)✦ For every nonnegative integer mm, real numbers a<ba<b, and tolerance function E:ℝm→ℝE:\mathbb R^m\to\mathbb R positive on [a,b]m[a,b]^m, the full arbitrary-tolerance existence assertion is (8). 𝒜(m,a,b,E)≠⌀.\mathcal A(m,a,b,E)\ne\varnothing.(8)✦

In the paper, the result reads: for an integer m≥1m\geq1, real numbers a<ba<b, and any function E:[a,b]m→(0,∞)E:[a,b]^m\to(0,\infty) with no regularity assumption, there is a finite nonempty family of affine labels with slopes rL∈[a,b]mr_L\in[a,b]^m and offsets cL(h)=o(ha)c_L(h)=o(h^a) as h→0+h\to0^+ such that, for every positive sequence hn→0h_n\to0, every sequence pn∈ℝmp^n\in\mathbb R^m converging to p∈[a,b]mp\in[a,b]^m, and every fixed label LL active at Xhn(pn)X_{h_n}(p^n) for all nn, one has ∥p−rL∥∞<E(rL)\|p-r_L\|_\infty<E(r_L).

What Is Checked

The claim is stated over affine approximation certificates satisfying (6), with activity defined by (4) and exponent limits taken through the power coordinates (2).