math / data /lean /OAI /Analysis /Fourier /SquareFunction /KernelVariation.lean
introvoyz045's picture
Migrated from GitHub
4a56cd6 verified
Raw History Blame Contribute Delete
4.61 kB
import OAI.Analysis.Fourier.SquareFunction.EuclideanKernel
import OAI.Analysis.Fourier.SquareFunction.EuclideanSharp
open OAI.Analysis.Fourier.Wave
/-! Mean-value and maximal-function bounds for the low-frequency group. -/
namespace OAI.Analysis.Fourier.SquareFunction
open _root_.MeasureTheory Metric
open scoped ENNReal
/-- Polynomial decay changes by a bounded factor within one rescaled unit. -/
theorem polynomial_decay_nearby {E : Type*} [NormedAddCommGroup E]
{q a : ℝ} (hq : 0 ≀ q) (ha : 0 ≀ a) {u v : E}
(huv : a * dist u v ≀ 1) :
(a * β€–uβ€– + 1) ^ (-q) ≀ (2 : ℝ) ^ q * (a * β€–vβ€– + 1) ^ (-q) := by
have htri : β€–vβ€– ≀ β€–uβ€– + dist u v := by
simpa only [dist_eq_norm, norm_sub_rev u v] using norm_le_insert' v u
have hb : (a * β€–vβ€– + 1) / 2 ≀ a * β€–uβ€– + 1 := by
have h := mul_le_mul_of_nonneg_left htri ha
nlinarith [norm_nonneg u, mul_nonneg ha (norm_nonneg u)]
calc
_ ≀ ((a * β€–vβ€– + 1) / 2) ^ (-q) :=
Real.rpow_le_rpow_of_nonpos (by positivity) hb (neg_nonpos.mpr hq)
_ = _ := by
rw [Real.div_rpow (by positivity) (by norm_num) (-q),
Real.rpow_neg (by norm_num : (0 : ℝ) ≀ 2), div_inv_eq_mul]
ring
/-- A decaying derivative controls oscillation between two points lying
within one rescaled unit of an arbitrary comparison point. -/
theorem kernel_variation_of_deriv {n : β„•} {H : Type*}
[NormedAddCommGroup H] [NormedSpace ℝ H]
{a q C : ℝ} (ha : 0 < a) (hq : 0 ≀ q) (hC : 0 ≀ C)
{K : Space n β†’ H} (hKd : Differentiable ℝ K)
(hK : βˆ€ u, β€–fderiv ℝ K uβ€– ≀ C * (a ^ n * a) * (a * β€–uβ€– + 1) ^ (-q))
{x c z : Space n} (hx : a * dist x z ≀ 1) (hc : a * dist c z ≀ 1) :
β€–K x - K cβ€– ≀ (C * 2 ^ q * a * dist x c) *
(a ^ n * (a * β€–zβ€– + 1) ^ (-q)) := by
have hxb : x ∈ closedBall z (1 / a) := (mem_closedBall).2 ((le_div_iffβ‚€ ha).2 (by nlinarith [hx]))
have hcb : c ∈ closedBall z (1 / a) := (mem_closedBall).2 ((le_div_iffβ‚€ ha).2 (by nlinarith [hc]))
have hb (u : Space n) (hu : u ∈ closedBall z (1 / a)) :
β€–fderiv ℝ K uβ€– ≀ C * (a ^ n * a) * (2 ^ q * (a * β€–zβ€– + 1) ^ (-q)) := by
apply (hK u).trans
apply mul_le_mul_of_nonneg_left
Β· apply polynomial_decay_nearby hq ha.le
have hu' := (le_div_iffβ‚€ ha).1 (mem_closedBall.mp hu)
nlinarith
Β· positivity
have h := (convex_closedBall z (1 / a)).norm_image_sub_le_of_norm_fderiv_le
(fun u _ => hKd u) hb hcb hxb
calc
_ ≀ (C * (a ^ n * a) * (2 ^ q * (a * β€–zβ€– + 1) ^ (-q))) * β€–x - cβ€– := h
_ = _ := by rw [dist_eq_norm]; ring
/-- Actual convolution oscillation, bounded by the scalar maximal function
at any nearby point. This supplies the small factor `a * dist x c`. -/
theorem convolution_variation_le_maximal {n : β„•} {H : Type*}
[NormedAddCommGroup H] [NormedSpace β„‚ H] [CompleteSpace H]
{a q C : ℝ} (ha : 0 < a) (hq : 0 < q) (hC : 0 ≀ C)
{K : Space n β†’ H} (hKd : Differentiable ℝ K)
(hK : βˆ€ u, β€–fderiv ℝ K uβ€– ≀ C * (a ^ n * a) * (a * β€–uβ€– + 1) ^ (-q))
{f : Space n β†’ β„‚} (hf : Measurable f) {x c z : Space n}
(hx : a * dist x z ≀ 1) (hc : a * dist c z ≀ 1)
(hix : Integrable (fun y => f y β€’ K (x - y)))
(hic : Integrable (fun y => f y β€’ K (c - y))) :
β€–(∫ y, f y β€’ K (x - y)) - ∫ y, f y β€’ K (c - y)β€–β‚‘ ≀
ENNReal.ofReal (C * 2 ^ q * a * dist x c) *
(polynomialKernelMass n q * ballMaximal volume (fun y => β€–f yβ€–β‚‘) z) := by
let D : Space n β†’ H := fun w => K (w + (x - z)) - K (w + (c - z))
have hD (w : Space n) : β€–D wβ€–β‚‘ ≀ ENNReal.ofReal (C * 2 ^ q * a * dist x c) *
(ENNReal.ofReal (a ^ n) * ENNReal.ofReal ((a * β€–wβ€– + 1) ^ (-q))) := by
have h := kernel_variation_of_deriv ha hq.le hC hKd hK
(x := w + (x - z)) (c := w + (c - z)) (z := w)
(by simpa only [dist_eq_norm, add_sub_cancel_left] using hx)
(by simpa only [dist_eq_norm, add_sub_cancel_left] using hc)
have hd : dist (w + (x - z)) (w + (c - z)) = dist x c := by
simp only [dist_eq_norm]
congr 1
abel
rw [hd] at h
simpa only [D, ofReal_norm, ENNReal.ofReal_mul (by positivity : 0 ≀ C * 2 ^ q * a * dist x c),
ENNReal.ofReal_mul (pow_nonneg ha.le n)] using ENNReal.ofReal_le_ofReal h
have h := enorm_kernel_convolution_le_maximal ha hq hf hD z
rw [← integral_sub hix hic]
convert h using 1
congr 1
apply integral_congr_ae
filter_upwards [] with y
dsimp [D]
rw [smul_sub]
congr 2 <;> abel_nf
end OAI.Analysis.Fourier.SquareFunction