Extension of Normalized Smooth Taylor Bounds to Affine-Orthogonal Functions
ProvedFunctionalInequalities.KLS.LogConcave.Suspension.norm_coordTensor_taylorTensor_sq_le_of_smooth_readableThis auxiliary extension statement formalizes a component of Proposition 7.1, p. 28, of Bizeul–Klartag–Lehec (arXiv:2610.05474v1), rather than the full proposition.
Let , , and . Suppose RegularPotential : , is smooth with derivatives of polynomial growth, and . Let , the normalized law, and assume it is isotropic (KLS.Isotropic). The Euclidean space is EuclideanSpace ℝ (Fin n).
Write , using taylorTensor and coordTensor in the standard orthonormal basis. Assume that every real-valued with a globally bounded second Fréchet derivative, , , and for all , satisfies .
Then every with and for all satisfies
Here the extended norm is eLpNorm, valued in . This is a conditional extension from the stated normalized smooth estimate; no cumulant bound or universal KLS bound is asserted. All natural degrees , including , are allowed.
Original Lean declaration header
Within namespace FunctionalInequalities.KLS.LogConcave.Suspension. Shown in its original module context. Proof omitted.
theorem norm_coordTensor_taylorTensor_sq_le_of_smooth {d : ℕ} {b : ℝ} (hb : 0 ≤ b) {n : ℕ}
(V : EuclideanSpace ℝ (Fin n) → ℝ) {a b' : ℝ} (hV : KLS.RegularPotential V a b')
(hiso : KLS.Isotropic hV.law)
(h113 : ∀ g : EuclideanSpace ℝ (Fin n) → ℝ, ContDiff ℝ 2 g →
(∃ M : ℝ, ∀ x, ‖fderiv ℝ (fderiv ℝ g) x‖ ≤ M) → MemLp g 2 (hV.law : Measure _) →
∫ x, g x ∂(hV.law : Measure _) = 0 →
∫ x, g x ^ 2 ∂(hV.law : Measure _) = 1 →
(∀ u, ∫ x, ⟪u, x⟫ * g x ∂(hV.law : Measure _) = 0) →
‖coordTensor (EuclideanSpace.basisFun (Fin n) ℝ) (taylorTensor (hV.law : Measure _) g d)‖ ^ 2 ≤
b / (d.factorial : ℝ) ^ 2)
{f : EuclideanSpace ℝ (Fin n) → ℝ} (hf : MemLp f 2 (hV.law : Measure _))
(hf0 : ∫ x, f x ∂(hV.law : Measure _) = 0)
(hfx : ∀ u, ∫ x, ⟪u, x⟫ * f x ∂(hV.law : Measure _) = 0) :
‖coordTensor (EuclideanSpace.basisFun (Fin n) ℝ) (taylorTensor (hV.law : Measure _) f d)‖ ^ 2 ≤
b / (d.factorial : ℝ) ^ 2 * (eLpNorm f 2 (hV.law : Measure _)).toReal ^ 2
Module navigation
Module: KLS/BKL/TaylorBound.
Definition modules: Analysis/Tensor/Coefficients, Analysis/Tensor/Coordinate, KLS/Isotropic, Analysis/BakryEmery/Tilt/Basic, KLS/LogConcave/Regular/Basic.
Selected supporting theorems: exists_norm_coordTensor_taylorTensor_le, taylorTensor_add, taylorTensor_const_mul, integrable_exp_mul_norm, contDiff_affineOrthPart.
Next: norm_coordTensor_taylorTensor_sq_le_two_mul.
Source excerpt, lines 387–401 · Frozen source 408915f.
import Definitions.Def_KLSModular_FunctionalInequalities_KLS_BKL_TaylorBound open FunctionalInequalities.Analysis.Tensor set_option autoImplicit false open MeasureTheory FunctionalInequalities.L2 open scoped ContDiff RealInnerProductSpace open FunctionalInequalities FunctionalInequalities.Analysis.BakryEmery
theorem FunctionalInequalities.KLS.LogConcave.Suspension.norm_coordTensor_taylorTensor_sq_le_of_smooth_readable {d : ℕ} {b : ℝ} (hb : 0 ≤ b) {n : ℕ}
(V : EuclideanSpace ℝ (Fin n) → ℝ) {a b' : ℝ} (hV : KLS.RegularPotential V a b')
(hiso : KLS.Isotropic hV.law)
(h113 : ∀ g : EuclideanSpace ℝ (Fin n) → ℝ, ContDiff ℝ 2 g →
(∃ M : ℝ, ∀ x, ‖fderiv ℝ (fderiv ℝ g) x‖ ≤ M) → MemLp g 2 (hV.law : Measure _) →
∫ x, g x ∂(hV.law : Measure _) = 0 →
∫ x, g x ^ 2 ∂(hV.law : Measure _) = 1 →
(∀ u, ∫ x, ⟪u, x⟫ * g x ∂(hV.law : Measure _) = 0) →
‖coordTensor (EuclideanSpace.basisFun (Fin n) ℝ) (taylorTensor (hV.law : Measure _) g d)‖ ^ 2 ≤
b / (d.factorial : ℝ) ^ 2)
{f : EuclideanSpace ℝ (Fin n) → ℝ} (hf : MemLp f 2 (hV.law : Measure _))
(hf0 : ∫ x, f x ∂(hV.law : Measure _) = 0)
(hfx : ∀ u, ∫ x, ⟪u, x⟫ * f x ∂(hV.law : Measure _) = 0) :
‖coordTensor (EuclideanSpace.basisFun (Fin n) ℝ) (taylorTensor (hV.law : Measure _) f d)‖ ^ 2 ≤
b / (d.factorial : ℝ) ^ 2 * (eLpNorm f 2 (hV.law : Measure _)).toReal ^ 2 := by sorry