Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Middle-interval radial gain from bad-direction exclusion

Open
FalconerPath.one_scale_middle_radial_estimates

by dangnguyen · Nov 3, 2026 · Mathlib 0df444a (Lean v4.33.1)

bad-directionsfalconerfiber-estimates

Fix k≥2k\ge2k≥2, k−1≤q<k−12k-1\le q<k-\tfrac12k−1≤q<k−21​, k−12<t<kk-\tfrac12<t<kk−21​<t<k, and ζ,L,a,C>0\zeta,L,a,C>0ζ,L,a,C>0, with sss any real parameter. Take a sequence of normalized regular-class data for these parameters, with Nj→∞N_j\to\inftyNj​→∞, hj→0h_j\to0hj​→0, and profiles converging uniformly on [0,1][0,1][0,1] to mmm. Suppose 0<A<B<10<A<B<10<A<B<1, B<2AB<2AB<2A, and m(w)−m(A)≥t(w−A)m(w)-m(A)\ge t(w-A)m(w)−m(A)≥t(w−A) for every w∈[A,B]w\in[A,B]w∈[A,B]. Let W>0W>0W>0.

There is c1>0c_1>0c1​>0 such that for all sufficiently large jjj there is a Borel Hj⊂GjH_j\subset G_jHj​⊂Gj​ deleting product mass at most Rj−c1(B−A)/2R_j^{-c_1(B-A)/2}Rj−c1​(B−A)/2​. At the four depths a0=0a_0=0a0​=0, a1=⌊ANj⌋a_1=\lfloor AN_j\rfloora1​=⌊ANj​⌋, a2=⌊BNj⌋a_2=\lfloor BN_j\rfloora2​=⌊BNj​⌋, a3=Nja_3=N_ja3​=Nj​, for every i<3i<3i<3, every retained witness (x,z)(x,z)(x,z), and every depth-aia_iai​ cell QQQ containing zzz,

ρj{z′∈Q:∣πx(z′)−πx(z)∣≤W2−ai+1}≤fiρj(Q),\rho_j\{z^\prime\in Q:|\pi_x(z^\prime)-\pi_x(z)|\le W2^{-a_{i+1}}\}\le f_i\rho_j(Q),ρj​{z′∈Q:∣πx​(z′)−πx​(z)∣≤W2−ai+1​}≤fi​ρj​(Q),

where f1=Rj−(q+c1/2)(B−A)f_1=R_j^{-(q+c_1/2)(B-A)}f1​=Rj−(q+c1​/2)(B−A)​ and f0=f2=1f_0=f_2=1f0​=f2​=1.

Here a normalized regular-class datum consists of an integer NjN_jNj​, probabilities νj,ρj\nu_j,\rho_jνj​,ρj​ on Rk+1\mathbb R^{k+1}Rk+1 with supports in B‾(0,L)\overline B(0,L)B(0,L) and cross distances at least aaa, a translated half-open unit root carrying ρj\rho_jρj​, a nonnegative error hjh_jhj​, and a profile mjm_jmj​ with mj(0)=0m_j(0)=0mj​(0)=0, nondecreasing and (k+1)(k+1)(k+1)-Lipschitz on [0,1][0,1][0,1], satisfying mj(u)≥su−hjm_j(u)\ge su-h_jmj​(u)≥su−hj​. In its translated dyadic grid, every occupied depth-bbb cell has ρj\rho_jρj​-mass between Rj−mj(b/Nj)−hjR_j^{-m_j(b/N_j)-h_j}Rj−mj​(b/Nj​)−hj​​ and Rj−mj(b/Nj)+hjR_j^{-m_j(b/N_j)+h_j}Rj−mj​(b/Nj​)+hj​​, for b≤Njb\le N_jb≤Nj​, where Rj=2NjR_j=2^{N_j}Rj​=2Nj​. For occupied nested cells P⊂QP\subset QP⊂Q at depths c≥bc\ge bc≥b, c≤Njc\le N_jc≤Nj​, the ratio ρj(P)/ρj(Q)\rho_j(P)/\rho_j(Q)ρj​(P)/ρj​(Q) lies between Rj−[mj(c/Nj)−mj(b/Nj)]−hjR_j^{-[m_j(c/N_j)-m_j(b/N_j)]-h_j}Rj−[mj​(c/Nj​)−mj​(b/Nj​)]−hj​​ and Rj−[mj(c/Nj)−mj(b/Nj)]+hjR_j^{-[m_j(c/N_j)-m_j(b/N_j)]+h_j}Rj−[mj​(c/Nj​)−mj​(b/Nj​)]+hj​​. The datum includes a Borel Gj⊂supp⁡νj×supp⁡ρjG_j\subset\operatorname{supp}\nu_j\times\operatorname{supp}\rho_jGj​⊂suppνj​×suppρj​ such that, for ρj\rho_jρj​-almost every zzz, the unnormalized reverse law (πz)#(νj∣Gjz)(\pi_z)_\#(\nu_j|_{G_j^z})(πz​)#​(νj​∣Gjz​​) obeys cap bounds RjhjuqR_j^{h_j}u^qRjhj​​uq and directional plate bounds max⁡(Cuζ,0)\max(Cu^\zeta,0)max(Cuζ,0) for Rj−1≤u≤1R_j^{-1}\le u\le1Rj−1​≤u≤1. The geometry constants L,aL,aL,a are positive. The Lean statement isolates the proof's Borel middle-interval deletion and bounds the full conditional source law at each retained witness.

Preamble
import Definitions.Def_FalconerPath_defs
open Set MeasureTheory Filter
open scoped ENNReal NNReal Topology
Formal statement
namespace FalconerPath
theorem one_scale_middle_radial_estimates {k : ℕ} (hk : 2 ≤ k)
    (s q ζ L a C t : ℝ) (hq : (k:ℝ)-1 ≤ q) (hqS : q < (k:ℝ)-1/2)
    (ht : (k:ℝ)-1/2 < t) (htk : t < k) (hζ : 0 < ζ)
    (hL : 0 < L) (ha : 0 < a) (hC : 0 < C)
    (X : ℕ → RadialClassData k s q ζ L a C)
    (hN : Tendsto (fun j => (X j).N) atTop atTop)
    (herr : Tendsto (fun j => (X j).err) atTop (𝓝 0))
    (m : ℝ → ℝ)
    (hconv : ∀ ε : ℝ, 0 < ε → ∃ J : ℕ, ∀ j ≥ J, ∀ u ∈ Icc (0:ℝ) 1,
      |(X j).m u-m u| < ε)
    (A B : ℝ) (hA : 0 < A) (hAB : A < B) (hB : B < 1) (hBA : B < 2*A)
    (hmiddle : ∀ w ∈ Icc A B, t*(w-A) ≤ m w-m A)
    (W : ℝ) (hW : 0 < W) :
    ∃ c₁ : ℝ, 0 < c₁ ∧ ∀ᶠ j in atTop,
      ∃ H : PairSet (k+1), MeasurableSet H ∧ H ⊆ (X j).G ∧
        ((X j).ν.prod (X j).ρ) ((X j).G \ H) ≤
          ENNReal.ofReal ((scale (X j).N)^(-c₁*(B-A)/2)) ∧
        LocalFiberBounds false (X j).ρ H (fun _ => univ) (X j).v
          (contactDepths (X j).N A B) 3
          (fun i => if i = 1 then (scale (X j).N)^(-(q+c₁/2)*(B-A)) else 1) W := by sorry
end FalconerPath
Source
OpenAI, The Falconer distance conjecture in all dimensions, OpenAI Math Release preprint, September 23, 2026, https://github.com/openai/math/blob/adc7f1241b42e322a6451854ab7e4b4c146bf78a/preprints/The-Falconer-distance-conjecture-in-all-dimensions-September-23-2026/paper.pdf. Section 5.1, Lemma 5.3 (One-scale improvement), label prep:one-scale, proof from “For the middle interval put D=” through “Every remaining witness satisfies”, including the angular envelope deletion, equation prep:witness-bad, witness-to-center comparison, equation prep:linearization-margin, Taylor expansion, and heavy-cell cover.

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me