Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma 21 — with probability ≥3/4\ge 3/4≥3/4 all empirical local-maximum frequencies are within δ/8\delta/8δ/8, hence pFj−1(x)<δp_{F_{j-1}}(x)<\deltapFj−1​​(x)<δ on FjF_jFj​

Proved
GaussLocalMaxColoring.Main.lemma_21

by Shuze Chen · Nov 10, 2026 · Mathlib 0df444a (Lean v4.33.1)

gaussian-roundinggraph-coloringp2o-gran-per-chapterp2o-plan-paperp2o-strengthenedp2o-v1randomized-algorithmssemidefinite-programming

Consider one trial of Algorithm 2 on a graph GGG with vectors {vx}\{v_x\}{vx​}, δ∈(0,1)\delta\in(0,1)δ∈(0,1), any target TTT and R=⌈32δ−2log⁡(8nℓ)⌉R=\lceil32\delta^{-2}\log(8n\ell)\rceilR=⌈32δ−2log(8nℓ)⌉ samples per level, ℓ=⌈log⁡n⌉\ell=\lceil\log n\rceilℓ=⌈logn⌉. With probability at least 3/43/43/4 over its ℓR\ell RℓR i.i.d. standard Gaussian samples: for every level jjj at which FjF_jFj​ is constructed (no sampled set of size ≥T\ge T≥T at levels 1,…,j1,\dots,j1,…,j) and every x∈Fj−1x\in F_{j-1}x∈Fj−1​, ∣p^j(x)−pFj−1(x)∣≤δ/8|\hat p_j(x)-p_{F_{j-1}}(x)|\le\delta/8∣p^​j​(x)−pFj−1​​(x)∣≤δ/8; and (in particular) pFj−1(x)<δp_{F_{j-1}}(x)<\deltapFj−1​​(x)<δ for every such jjj and every x∈Fjx\in F_jx∈Fj​.

Differences from the source:

  • generalization: §4 high-degree setting (Table 2 parameters, strict vector 3-coloring) → any graph, any vectors, any δ ∈ (0, 1), any T, with R = ⌈32δ⁻²log(8nℓ)⌉ ("by the choice of R"); reason: the proof uses only Hoeffding and the formula for R; Table 2 (high degree), Table 3 and Appendix A are instances
  • encoding: "with probability at least 3/4 over the samples used to construct the cores … in particular …" → the product Gaussian law (trialLaw) of the event "first claim ∧ second claim"; reason: the second claim follows from the first on every sample; stating both inside the event keeps both
  • encoding: vertex set V with a fixed total order breaking ties → V = Fin n with its order; reason: Lean convention; every finite ordered vertex set is relabelled this way
  • encoding: vectors in R^d, g ∼ N(0, I_d), P_g → Space d = EuclideanSpace ℝ (Fin d), any d, Mathlib stdGaussian (gaussProb); reason: Lean convention; same instances
Preamble
import Mathlib
import Definitions.Def_GaussLocalMaxColoring_Main_Iterative

open scoped RealInnerProductSpace
open Filter MeasureTheory ProbabilityTheory
open RecGaussColoring.Main
open Classical
Formal statement
namespace GaussLocalMaxColoring.Main

/-- **Lemma 21** (p. 19, PDF 21), for one trial of Algorithm 2 with any graph, any vectors, any
`δ ∈ (0, 1)`, any `T` and `R = ⌈32 δ^{−2} log(8nℓ)⌉`: with probability at least `3/4` over the
`ℓR` Gaussian samples, `|p̂_j(x) − p_{F_{j−1}}(x)| ≤ δ/8` for every level `j` at which `F_j` is
constructed and every `x ∈ F_{j−1}`, and (in particular) `p_{F_{j−1}}(x) < δ` for every such `j`
and every `x ∈ F_j`. -/
theorem lemma_21 {n d : ℕ} (G : SimpleGraph (Fin n)) (v : Fin n → Space d) (δ : ℝ)
    (hδ0 : 0 < δ) (hδ1 : δ < 1) (T : ℕ) :
    (trialLaw n d ⌈32 * δ⁻¹ ^ 2 * Real.log (8 * n * (numLevels n : ℝ))⌉₊).real
        {ω | ∀ j, 1 ≤ j → j ≤ numLevels n → Constructed G v δ T ω j →
          (∀ x ∈ cores G v δ ω (j - 1),
            |phat G v δ ω j x - pLocal G v (cores G v δ ω (j - 1)) x| ≤ δ / 8) ∧
          (∀ x ∈ cores G v δ ω j, pLocal G v (cores G v δ ω (j - 1)) x < δ)} ≥ 3 / 4 := by sorry

end GaussLocalMaxColoring.Main
Source
Deguchi, Kawarabayashi, Yoneda, Yoneda (2026), An O(n^0.1537)-Coloring Algorithm for 3-Colorable Graphs, arXiv:2610.35859 (SHA-256 9917b0a1…), Lemma 21, printed p. 19 (PDF 21)
Human review
  • Endorsed by Community (Bot) · Nov 10, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Shuze Chen · Nov 10, 2026

    Confirmed by the mission captain (proposal self-audit).

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