Prove2Me
Navigate
DiscoverCollectionsFormalpediaBlogsUsersMomentumMy Missions+
Prove2Me
⌕
Log in
← Formalpedia

OAI.SKGapCutoff.Recipe.buffered_root_direction_weak: intermediate lemma from OpenAI, A spectral gap throughout the high-temperature Sherrington-Kirkpatrick phase (result family 227)

Proved
OAI.SKGapCutoff.Recipe.buffered_root_direction_weak

by dbenbenn · Oct 28, 2026 · Mathlib 0df444a (Lean v4.33.1)

openai-math

An intermediate lemma of the split of OpenAI's proof of OAI.SKGap.sk_main, carved from OpenAI's openai/math development for this platform. The statement and its proof are OpenAI's (OpenAI, A spectral gap throughout the high-temperature Sherrington-Kirkpatrick phase (result family 227), OpenAI Math Release, September 24, 2026, result family 227, github.com/openai/math, Apache License 2.0); in Lean it is the declaration OAI.SKGapCutoff.Recipe.buffered_root_direction_weak of lean/OAI/Probability/SKGap/Localization/LocalExtraction.lean at commit 1581d4c, ported to Lean 4.33.1 and Mathlib 0df444a.

The statement, as published (namespace OAI.SKGapCutoff.Recipe, with open scoped BigOperators Matrix.Norms.Frobenius, open SKGap.Stein Primary Static):

theorem buffered_root_direction_weak {j K B Aroot ε c r₀ R ρ : ℝ}
    (hj : 0≤j) (hK : 0≤K) (hB : 0≤B) (hA : 1≤Aroot) (hc : 0 < c)
    (hr₀ : 0 < r₀) (hρ : 0<ρ) (hε : 0≤ε) (hAR : Real.exp (R/2)≤Aroot)
    (hbuffer : (K+4*j)*(2*ρ) < r₀)
    (hR : (1+2*j)*(1+(1+(K+3*j)/c)*(K+4*j))*(2*ρ)≤R)
    (hsmall : (3*Real.exp (R/2)+2)*((1+2*j)*(1+(1+(K+3*j)/c)*(K+4*j))*(2*ρ))≤ε)
    (himplicit : (1+2*j)*(1+(1+(K+3*j)/c)*(K+4*j))*(2*ρ)≤
      c/(2*(|j| *Real.exp (R/2)*(3*Real.exp (R/2)+16)+1)))
    (habsorb : (|j| *|literalSizeBudget j c R K
        (KernelExpr.dr (KernelExpr.atom 0 0 false))|)*
      ((1+2*j)*(1+(1+(K+3*j)/c)*(K+4*j))*(2*ρ))≤(1:ℝ)/2)
    (k : ℕ) :
    ∃A≥1,∀W C : ℝ,0≤W→0≤C→∀(Ω : Type u) (n : Ω→ℕ)
      (J : ∀b,Interaction (n b)) (h e : ∀b,Fin (n b)→ℝ) (P : ∀b,Observables (n b)),
      (∀b,0<n b)→(∀b,(J b).IsSymm)→(∀b,vectorNorm (e b)≤1)→
      (∀b,StartedMatrixEvent j K (Real.exp (R/2)) A (c/2) B W C (k+3) (2+2*(k+1)) (J b))→
      (∀b,¬SKGap.rootBad j Aroot ε c r₀ (J b) (h b))→
      (∀b,4*(residualDerivativeBudget j K B k+residualDerivativeBudget j K B (k+1))≤ρ*Real.sqrt (n b:ℝ))→
      (∀b x,0≤P b x)→(∀b,∑x,P b x=1)→
      (∀b x i,conditionalMean (P b) x i=mag j (J b) (h b) 1 x i)→
      ∃r : ∀b,Fin (n b)→ℝ,(∀b,SKGap.tapField j (J b) (h b) (r b)=0 ∧
          ∀v,SKGap.tapField j (J b) (h b) v=0→v=r b) ∧
      ∃D≥0,∀b (G : Observables (n b)),
        |∑x,P b x*(residualCutoff ρ j (J b) (h b) k x*G x)*
          (∑i,(Real.tanh (fld j (J b) (h b) (k+1) x i)-Real.tanh (r b i))*e b i)|≤D*starNorm (P b) G

Definitions. The statement is written with OAI.SKGap.Field, OAI.SKGap.rootBad, OAI.SKGap.tapField, OAI.SKGapCutoff.Interaction, OAI.SKGapCutoff.Observables, OAI.SKGapCutoff.Spin, OAI.SKGapCutoff.Primary.fld, OAI.SKGapCutoff.Primary.mag, OAI.SKGapCutoff.Recipe.vectorNorm, OAI.SKGapCutoff.Static.conditionalMean, OAI.SKGap.Stein.KernelExpr and OAI.SKGap.Stein.KernelExpr.dr (definitions bundle OAIQ072A000 of this split); OAI.SKGapCutoff.Recipe.StartedMatrixEvent, OAI.SKGapCutoff.Recipe.literalSizeBudget, OAI.SKGapCutoff.Recipe.residualCutoff, OAI.SKGapCutoff.Recipe.residualDerivativeBudget and OAI.SKGapCutoff.Static.starNorm (definitions bundle OAIQ072A002 of this split).

Used by. The proof of OAI.SKGapCutoff.Recipe.buffered_posterior_covariance of this split imports it.

Preamble
import Mathlib
import Definitions.Def_OAIQ072A002
Formal statement
section

set_option linter.all false
set_option linter.unusedVariables false
set_option linter.unusedTactic false
set_option linter.unreachableTactic false

namespace OAI

section

noncomputable section
open scoped BigOperators Matrix.Norms.Frobenius
namespace SKGapCutoff.Recipe
open SKGap.Stein Primary Static
universe u

theorem buffered_root_direction_weak {j K B Aroot ε c r₀ R ρ : ℝ}
    (hj : 0≤j) (hK : 0≤K) (hB : 0≤B) (hA : 1≤Aroot) (hc : 0 < c)
    (hr₀ : 0 < r₀) (hρ : 0<ρ) (hε : 0≤ε) (hAR : Real.exp (R/2)≤Aroot)
    (hbuffer : (K+4*j)*(2*ρ) < r₀)
    (hR : (1+2*j)*(1+(1+(K+3*j)/c)*(K+4*j))*(2*ρ)≤R)
    (hsmall : (3*Real.exp (R/2)+2)*((1+2*j)*(1+(1+(K+3*j)/c)*(K+4*j))*(2*ρ))≤ε)
    (himplicit : (1+2*j)*(1+(1+(K+3*j)/c)*(K+4*j))*(2*ρ)≤
      c/(2*(|j| *Real.exp (R/2)*(3*Real.exp (R/2)+16)+1)))
    (habsorb : (|j| *|literalSizeBudget j c R K
        (KernelExpr.dr (KernelExpr.atom 0 0 false))|)*
      ((1+2*j)*(1+(1+(K+3*j)/c)*(K+4*j))*(2*ρ))≤(1:ℝ)/2)
    (k : ℕ) :
    ∃A≥1,∀W C : ℝ,0≤W→0≤C→∀(Ω : Type u) (n : Ω→ℕ)
      (J : ∀b,Interaction (n b)) (h e : ∀b,Fin (n b)→ℝ) (P : ∀b,Observables (n b)),
      (∀b,0<n b)→(∀b,(J b).IsSymm)→(∀b,vectorNorm (e b)≤1)→
      (∀b,StartedMatrixEvent j K (Real.exp (R/2)) A (c/2) B W C (k+3) (2+2*(k+1)) (J b))→
      (∀b,¬SKGap.rootBad j Aroot ε c r₀ (J b) (h b))→
      (∀b,4*(residualDerivativeBudget j K B k+residualDerivativeBudget j K B (k+1))≤ρ*Real.sqrt (n b:ℝ))→
      (∀b x,0≤P b x)→(∀b,∑x,P b x=1)→
      (∀b x i,conditionalMean (P b) x i=mag j (J b) (h b) 1 x i)→
      ∃r : ∀b,Fin (n b)→ℝ,(∀b,SKGap.tapField j (J b) (h b) (r b)=0 ∧
          ∀v,SKGap.tapField j (J b) (h b) v=0→v=r b) ∧
      ∃D≥0,∀b (G : Observables (n b)),
        |∑x,P b x*(residualCutoff ρ j (J b) (h b) k x*G x)*
          (∑i,(Real.tanh (fld j (J b) (h b) (k+1) x i)-Real.tanh (r b i))*e b i)|≤D*starNorm (P b) G := by
  sorry

end SKGapCutoff.Recipe

end
end

end OAI
end
Source
OpenAI, A spectral gap throughout the high-temperature Sherrington-Kirkpatrick phase (result family 227), OpenAI Math Release, September 24, 2026, https://github.com/openai/math (result family 227; paper: https://github.com/openai/math/blob/1581d4cbdd69a0faf34b2089614fe2a814316a6a/preprints/A-spectral-gap-throughout-the-high-temperature-Sherrington-Kirkpatrick-phase-September-24-2026/main.pdf; Apache License 2.0); Lean: https://github.com/openai/math/blob/1581d4cbdd69a0faf34b2089614fe2a814316a6a/lean/OAI/Probability/SKGap/Localization/LocalExtraction.lean (declaration OAI.SKGapCutoff.Recipe.buffered_root_direction_weak)

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 paper
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me