Download data/lean/OAI/Analysis/Fourier/WavePackets/LinearKernel.lean from introvoyz045/math: direct link, hf CLI and curl.
- Browser
- Download file 4.06 kB
-
https://huggingface.co/datasets/introvoyz045/math/resolve/main/data/lean/OAI/Analysis/Fourier/WavePackets/LinearKernel.lean
- Command line
-
hf download hf://datasets/introvoyz045/math/data/lean/OAI/Analysis/Fourier/WavePackets/LinearKernel.lean
-
curl -L -o LinearKernel.lean https://huggingface.co/datasets/introvoyz045/math/resolve/main/data/lean/OAI/Analysis/Fourier/WavePackets/LinearKernel.lean
4.06 kB
| import OAI.Analysis.Fourier.WavePackets.Poisson | |
| namespace OAI.Analysis.Fourier.WavePackets | |
| open _root_.MeasureTheory | |
| open scoped ENNReal FourierTransform RealInnerProductSpace | |
| theorem det_adjoint_real {n : ℕ} (A : EuclideanSpace ℝ (Fin n) →ₗ[ℝ] | |
| EuclideanSpace ℝ (Fin n)) : LinearMap.det A.adjoint = LinearMap.det A := by | |
| let b := EuclideanSpace.basisFun (Fin n) ℝ | |
| rw [← LinearMap.det_toMatrix b.toBasis, LinearMap.toMatrix_adjoint b b, | |
| Matrix.det_conjTranspose, LinearMap.det_toMatrix] | |
| simp | |
| /-- Exact change of variables for the Fourier kernel under an arbitrary | |
| invertible linear frequency transformation, including anisotropic ones. -/ | |
| theorem fourierInv_comp_linearEquiv {n : ℕ} | |
| (A : EuclideanSpace ℝ (Fin n) ≃L[ℝ] EuclideanSpace ℝ (Fin n)) | |
| (φ : EuclideanSpace ℝ (Fin n) → ℂ) (x : EuclideanSpace ℝ (Fin n)) : | |
| 𝓕⁻ (φ ∘ A) x = |(LinearMap.det A.toLinearEquiv.toLinearMap)⁻¹| • | |
| 𝓕⁻ φ (A.symm.toLinearEquiv.toLinearMap.adjoint x) := by | |
| rw [Real.fourierInv_eq, Real.fourierInv_eq] | |
| let g := fun ξ : EuclideanSpace ℝ (Fin n) => | |
| Real.fourierChar ⟪A.symm ξ, x⟫ • φ ξ | |
| have hchange := A.toHomeomorph.measurableEmbedding.integral_map (μ := volume) g | |
| change (∫ ξ, g ξ ∂Measure.map A volume) = ∫ ξ, g (A ξ) at hchange | |
| have hd : LinearMap.det A.toLinearEquiv.toLinearMap ≠ 0 := | |
| A.toLinearEquiv.isUnit_det'.ne_zero | |
| have hmap : Measure.map A volume = | |
| ENNReal.ofReal |(LinearMap.det A.toLinearEquiv.toLinearMap)⁻¹| • volume := | |
| Measure.map_linearMap_addHaar_eq_smul_addHaar volume hd | |
| rw [hmap, integral_smul_measure, ENNReal.toReal_ofReal (abs_nonneg _)] at hchange | |
| calc | |
| _ = ∫ ξ, g (A ξ) := by simp [g] | |
| _ = |(LinearMap.det A.toLinearEquiv.toLinearMap)⁻¹| • ∫ ξ, g ξ := hchange.symm | |
| _ = _ := by | |
| congr 1 | |
| apply integral_congr_ae | |
| filter_upwards [] with ξ | |
| dsimp [g] | |
| rw [LinearMap.adjoint_inner_right] | |
| rfl | |
| /-- Jacobian-normalized spatial changes of variables preserve kernel L1 | |
| norms. The transpose determinant is what cancels the frequency Jacobian. -/ | |
| theorem eLpNorm_normalized_linear_kernel {n : ℕ} | |
| (A : EuclideanSpace ℝ (Fin n) ≃L[ℝ] EuclideanSpace ℝ (Fin n)) | |
| {K : EuclideanSpace ℝ (Fin n) → ℂ} (hK : Measurable K) : | |
| eLpNorm (fun x => |(LinearMap.det A.toLinearEquiv.toLinearMap)⁻¹| • | |
| K (A.symm.toLinearEquiv.toLinearMap.adjoint x)) 1 volume = eLpNorm K 1 volume := by | |
| let B := A.symm.toLinearEquiv.toLinearMap.adjoint | |
| let d := LinearMap.det A.toLinearEquiv.toLinearMap | |
| have hd : d ≠ 0 := A.toLinearEquiv.isUnit_det'.ne_zero | |
| have hB : LinearMap.det B = d⁻¹ := by | |
| rw [det_adjoint_real] | |
| exact A.toLinearEquiv.det_coe_symm | |
| have hBm : Measurable B := B.continuous_of_finiteDimensional.measurable | |
| have hmap : Measure.map B volume = ENNReal.ofReal |d| • volume := by | |
| rw [Measure.map_linearMap_addHaar_eq_smul_addHaar volume (by rw [hB]; exact inv_ne_zero hd), hB, inv_inv] | |
| have hKB : Measurable (fun x => K (B x)) := hK.comp hBm | |
| change eLpNorm (|d⁻¹| • (fun x => K (B x))) 1 volume = _ | |
| rw [eLpNorm_const_smul, Real.enorm_eq_ofReal (abs_nonneg _), | |
| eLpNorm_one_eq_lintegral_enorm hKB.aestronglyMeasurable, | |
| eLpNorm_one_eq_lintegral_enorm hK.aestronglyMeasurable] | |
| rw [← lintegral_map hK.enorm hBm, hmap, lintegral_smul_measure, smul_eq_mul, ← mul_assoc, | |
| ← ENNReal.ofReal_mul (abs_nonneg _)] | |
| simp [hd] | |
| /-- In particular, rotating and anisotropically rescaling a multiplier | |
| does not change the L1 norm of its inverse-Fourier kernel. -/ | |
| theorem eLpNorm_fourierInv_comp_linearEquiv {n : ℕ} | |
| (A : EuclideanSpace ℝ (Fin n) ≃L[ℝ] EuclideanSpace ℝ (Fin n)) | |
| (φ : EuclideanSpace ℝ (Fin n) → ℂ) (hφ : Measurable (𝓕⁻ φ)) : | |
| eLpNorm (𝓕⁻ (φ ∘ A)) 1 volume = eLpNorm (𝓕⁻ φ) 1 volume := by | |
| have he := funext (fourierInv_comp_linearEquiv A φ) | |
| rw [he] | |
| exact eLpNorm_normalized_linear_kernel A hφ | |
| end OAI.Analysis.Fourier.WavePackets | |