Cutoff Hessian Entry Difference-Quotient Bound
ProvedFunctionalInequalities.Analysis.exists_hessian_cutoff_sq_difference_bound_modularfunctional-inequalitieskls-conjecturekls-modularletwin2026log-concave
This formalizes a supporting component of Theorem 2.5 of Letwin (arXiv:2607.24164v1).
Let . Assume are convex, is Lipschitz, is Lipschitz, and for some the almost-everywhere Hessian data satisfy , , and . Assume also that for all , and that . For every and , there exists such that, whenever and ,
This is a global difference-quotient estimate for cutoff Hessian entries.
Superseded upload. This snapshot was replaced by FunctionalInequalities.Analysis.exists_hessian_cutoff_sq_difference_bound_readable: open the readable replacement. The replacement preserves the frozen source and mathematical statement, with a compact declaration context. This historical node remains available for existing imports.
Preamble
import Definitions.Def_KLSModular_FunctionalInequalities_Analysis_Convex_HessianIncrement noncomputable section open MeasureTheory Set Filter open scoped Topology RealInnerProductSpace open FunctionalInequalities.Probability.Stein noncomputable section open MeasureTheory Set Filter open scoped Topology ContDiff RealInnerProductSpace open FunctionalInequalities.Probability.Stein section namespace FunctionalInequalities end FunctionalInequalities open FunctionalInequalities section namespace FunctionalInequalities.Analysis end FunctionalInequalities.Analysis open FunctionalInequalities FunctionalInequalities.Analysis set_option maxHeartbeats 800000
Formal statement
theorem FunctionalInequalities.Analysis.exists_hessian_cutoff_sq_difference_bound_modular {n : ℕ}
{u V χ : EuclideanSpace ℝ (Fin n) → ℝ}
(hu : ContDiff ℝ 1 u) (hconv : ConvexOn ℝ univ u) {G : NNReal}
(hG : LipschitzWith G (gradient u)) (hV : ContDiff ℝ 1 V) (hVc : ConvexOn ℝ univ V)
{LF : NNReal} (hF : LipschitzWith LF (fun x => -u x + V (gradient u x)))
{K : ℝ} (hK : 0 < K)
(hdata : ∀ᵐ x ∂(volume : Measure (EuclideanSpace ℝ (Fin n))),
(hessianMatrix u x).PosDef ∧
Real.log (hessianMatrix u x).det = -u x + V (gradient u x) ∧
(K • (1 : Matrix (Fin n) (Fin n) ℝ) - hessianMatrix u x).PosSemidef)
{D : ℝ} (hw : ∀ x h, u (x + h) + u (x - h) - 2 * u x ≤ D * ‖h‖ ^ 2)
(hχ : ContDiff ℝ 1 χ) (hχc : HasCompactSupport χ)
(i j : Fin n) (v : EuclideanSpace ℝ (Fin n)) :
∃ C : ℝ, 0 ≤ C ∧ ∀ t : ℝ, t ≠ 0 → |t| ≤ 1 →
(∫ x, (χ (x + t • v) * hessianMatrix u (x + t • v) i j -
χ x * hessianMatrix u x i j) ^ 2) ≤ C * t ^ 2 := by sorry
Source
Source: Letwin, The KLS constant is O(log^(1/4) n) (arXiv:2607.24164v1), Theorem 2.5. Lean source: https://github.com/human-ai-mathematics/kls-lean/blob/408915f/formalization/FunctionalInequalities/Analysis/Convex/HessianIncrement.lean#L263-L457 (repository human-ai-mathematics/kls-lean, commit 408915f, module `FunctionalInequalities.Analysis.Convex.HessianIncrement`, declaration `FunctionalInequalities.Analysis.exists_hessian_cutoff_sq_difference_bound`; the repository is private at upload time).