Documentation

LeanPool.TrilinearHilbertTransform.Auto.Sec155OneDistributionCalculationFinalSums.ThreePathTotal

Trilinear Hilbert transform: three_path_total #

theorem TrilinearHT.Auto.ActiveSetup.BasicFamilies.FamParams.SideInput.three_path_total {κ' : LocalKernelClass} {c₀ : ℕ} {A : ActiveSetup κ' c₀} {bf : A.BasicFamilies} {fp : bf.FamParams} (D : fp.SideInput) {Cj CA CT CK : ℕ} {δs W : ℝ} (hV : D.S.Valid Cj CA CT CK δs W) (hCA : 1 ≤ CA) (hCT : 4 ≤ CT) (hd2 : 2 ≤ D.S.dmax) (hD : 2 ≤ D.S.Dstar) (hB : 2 ≤ D.S.Bstar) (hQ0 : 1 ≤ D.S.Q0) (hSP : D.SP = D.S.sideParams δs) (hdD : ↑D.S.dmax ≤ D.S.Dstar) (qG : bf.Grp → ℕ) (hqG : ∀ (G : bf.Grp), 1 ≤ qG G ∧ qG G ≤ D.S.Q0) (J : ℕ) (hJ : 1 ≤ J) (Aw Iw : ℕ) (hIw : 2 ^ Iw ≤ 2 * D.U0) (l : Fin 4) (T : Finset (ℤ × ℤ)) (gs : ℤ × ℤ → Finset bf.Grp) (z : bf.Grp → ChartTag) (hGc : ∀ I ∈ T, ∀ G ∈ gs I, (∀ (j : Fin 4), 8 * (D.Ms j G).Λ ≤ D.S.Ξ) ∧ (∀ (j : Fin 4), (D.Ms j G).zc = z G) ∧ (z G).dim ≤ D.S.dmax) (H : Finset ℕ) (hH : ∀ h ∈ H, h ≤ D.SP.Jst) (rep : ℕ → (G : bf.Grp) → (j : Fin 4) → ℕ → ℕ → ℤ × (ℤ × ℤ) × (D.Ms j G).Md → (D.Ms j G).Md) (hrep : ∀ h ∈ H, ∀ I ∈ T, ∀ G ∈ gs I, ∀ (j : Fin 4), (D.withQ0 (qG G)).RepOK G j h T (rep h G j)) (anc : (I : ℤ × ℤ) → (G : bf.Grp) → D.AncNe3 G I l → ℕ → (j : Fin 4) → j ≠ l → Finset (D.Ms j G).Md → ↥((D.Ms j G).alive (ancAt A.G (D.preAt I) (A.blkStart D.S.K I.1 - 4 * ↑D.S.K)))) (hanc : ∀ (h : ℕ) (hh : h ∈ H) (I : ℤ × ℤ) (hI : I ∈ T) (G : bf.Grp) (hG : G ∈ gs I) (hA : D.AncNe3 G I l) (j : Fin 4) (hj : j ≠ l) (F : Finset ((D.withQ0 (qG G)).Ms j G).Md) (hF : F ∈ (D.withQ0 (qG G)).goodCls G I j h), anc I G hA h j hj F = (D.withQ0 (qG G)).repAnchor G I j h (rep h G j) ⋯ hF) {U ε : ℝ} (hε : 0 < ε) (hεU : ε ≤ U) (hU : ∀ I ∈ T, ∀ G ∈ gs I, ∀ h ∈ H, ∀ (j : Fin 4), j ≠ l → ∀ F ∈ (D.withQ0 (qG G)).goodCls G I j h, (D.withQ0 (qG G)).clsSize G I j F ≤ U) (hUs : ∀ I ∈ T, ∀ G ∈ gs I, √((dyadicGridLength A.G I.1)⁻¹ * ∫ (y : ℝ) in dyadicGridLeft A.G I.1 I.2..dyadicGridLeft A.G I.1 I.2 + dyadicGridLength A.G I.1, ‖augPhys ((D.withQ0 (qG G)).toLeaf.sv l G I) y‖ ^ 2) ≤ U) {K K₁ : ℝ} (hK : 0 ≤ K) (hK₁ : 0 ≤ K₁) (y : ℝ) (hcap : ∀ h ∈ H, 10 ≤ h → ∀ (c : ℝ × ℝ × ℕ × ℝ × (Fin 4 → ℝ)), 0 < c.1 → c.1 ≤ 1 → 1 ≤ c.2.1 → (∀ (i : Fin 4), 0 < threeAmp l c i) → (∀ (j : Fin 4), ε ≤ threeAmp l c j ∧ threeAmp l c j ≤ U ∧ ∃ (k : ℤ), threeAmp l c j = 2 ^ k) → ∀ (κ : Fin 4 → ℕ × ℕ), (∀ (i : Fin 4) (hi : i ≠ l), ∑ G ∈ pathGroups T gs y, ↑((D.withQ0 (qG G)).threeUsedReps G h y l (pathCell A.G T y) (fun (s : ℤ) => anc (pathCell A.G T y s) G) Iw c κ (pathDepths T gs y G) i hi).card ≤ K * threeAmp l c i ^ (-2)) ∧ ∑ G ∈ pathGroups T gs y, ∑ s ∈ pathDepths T gs y G, (D.withQ0 (qG G)).sideWeight G l (pathCell A.G T y) c.2.2.2.1 s ≤ K * threeAmp l c l ^ (-2)) :
∑ I ∈ T with y ∈ dyadicGridCell A.G I.1 I.2, ∑ x ∈ gs I ×ˢ H ×ˢ D.threeTypes l, (K₁ * ((↑H.card)⁻¹ * ∑ c ∈ (D.withQ0 (qG x.1)).threeRetBinsA x.1 I y l (anc I x.1) Iw x.2.1 ε, threeBinVal κ' J Aw l c * (D.withQ0 (qG x.1)).threeCntA x.1 I y l (anc I x.1) Iw x.2.1 c x.2.2)) ^ (1 - 1 / 300 / 4) ≤ ↑H.card * ↑(D.threeTypes l).card * ((K₁ * (8 * entryCU κ' J Aw * 2 ^ (Aw + 4) * 12 ^ 3)) ^ (1 - 1 / 300 / 4) * (D.fourCc * K ^ (2 - 1 / 300 / 2)) * ((↑Iw + 1) * ((Real.logb 2 (U / ε) + 1) ^ 4 * bandC (1 - 1 / 300 / 4) J)))

Blueprint lines 2783-2817 at one point y for three cores with side role l: the powered retained principal quantities of the cells of the path of y (over their groups, the heights and the role types) sum to at most |H| |types| (K₁ C_J 2 · 12⁴)^p C_c K^{2-c/2} (log₂(U/ε) + 1)⁴ C_band.

theorem TrilinearHT.Auto.ActiveSetup.BasicFamilies.FamParams.SideInput.three_core_final_side {κ' : LocalKernelClass} {c₀ : ℕ} {A : ActiveSetup κ' c₀} {bf : A.BasicFamilies} {fp : bf.FamParams} (D : fp.SideInput) {Cj CA CT CK : ℕ} {δs W : ℝ} (hV : D.S.Valid Cj CA CT CK δs W) (hCA : 1 ≤ CA) (hCT : 4 ≤ CT) (hd2 : 2 ≤ D.S.dmax) (hD : 2 ≤ D.S.Dstar) (hB : 2 ≤ D.S.Bstar) (hQ0 : 1 ≤ D.S.Q0) (hSP : D.SP = D.S.sideParams δs) (hdD : ↑D.S.dmax ≤ D.S.Dstar) (qG : bf.Grp → ℕ) (hqG : ∀ (G : bf.Grp), 1 ≤ qG G ∧ qG G ≤ D.S.Q0) (J : ℕ) (hJ : 1 ≤ J) (Aw Iw : ℕ) (hIw : 2 ^ Iw ≤ 2 * D.U0) (l : Fin 4) (R : ℤ × ℤ) (T : Finset (ℤ × ℤ)) (hT : ∀ I ∈ T, dyadicGridCell A.G I.1 I.2 ⊆ dyadicGridCell A.G R.1 R.2) (gs : ℤ × ℤ → Finset bf.Grp) (z : bf.Grp → ChartTag) (hGc : ∀ I ∈ T, ∀ G ∈ gs I, (∀ (j : Fin 4), 8 * (D.Ms j G).Λ ≤ D.S.Ξ) ∧ (∀ (j : Fin 4), (D.Ms j G).zc = z G) ∧ (z G).dim ≤ D.S.dmax) (H : Finset ℕ) (hH : ∀ h ∈ H, h ≤ D.SP.Jst) (rep : ℕ → (G : bf.Grp) → (j : Fin 4) → ℕ → ℕ → ℤ × (ℤ × ℤ) × (D.Ms j G).Md → (D.Ms j G).Md) (hrep : ∀ h ∈ H, ∀ I ∈ T, ∀ G ∈ gs I, ∀ (j : Fin 4), (D.withQ0 (qG G)).RepOK G j h T (rep h G j)) (anc : (I : ℤ × ℤ) → (G : bf.Grp) → D.AncNe3 G I l → ℕ → (j : Fin 4) → j ≠ l → Finset (D.Ms j G).Md → ↥((D.Ms j G).alive (ancAt A.G (D.preAt I) (A.blkStart D.S.K I.1 - 4 * ↑D.S.K)))) (hanc : ∀ (h : ℕ) (hh : h ∈ H) (I : ℤ × ℤ) (hI : I ∈ T) (G : bf.Grp) (hG : G ∈ gs I) (hA : D.AncNe3 G I l) (j : Fin 4) (hj : j ≠ l) (F : Finset ((D.withQ0 (qG G)).Ms j G).Md) (hF : F ∈ (D.withQ0 (qG G)).goodCls G I j h), anc I G hA h j hj F = (D.withQ0 (qG G)).repAnchor G I j h (rep h G j) ⋯ hF) (hcard : ∀ I ∈ T, ∀ G ∈ gs I, ∀ (j : Fin 4), ↑((D.Ms j G).alive (ancAt A.G (D.preAt I) (A.blkStart D.S.K I.1 - 4 * ↑D.S.K))).card ≤ D.S.Dstar) {U ε : ℝ} (hε : 0 < ε) (hεU : ε ≤ U) (hU : ∀ I ∈ T, ∀ G ∈ gs I, ∀ h ∈ H, ∀ (j : Fin 4), j ≠ l → ∀ F ∈ (D.withQ0 (qG G)).goodCls G I j h, (D.withQ0 (qG G)).clsSize G I j F ≤ U) (hUs : ∀ I ∈ T, ∀ G ∈ gs I, √((dyadicGridLength A.G I.1)⁻¹ * ∫ (y : ℝ) in dyadicGridLeft A.G I.1 I.2..dyadicGridLeft A.G I.1 I.2 + dyadicGridLength A.G I.1, ‖augPhys ((D.withQ0 (qG G)).toLeaf.sv l G I) y‖ ^ 2) ≤ U) {K K₁ : ℝ} (hK : 0 ≤ K) (hK₁ : 0 ≤ K₁) (hcap : ∀ (y : ℝ), ∀ h ∈ H, 10 ≤ h → ∀ (c : ℝ × ℝ × ℕ × ℝ × (Fin 4 → ℝ)), 0 < c.1 → c.1 ≤ 1 → 1 ≤ c.2.1 → (∀ (i : Fin 4), 0 < threeAmp l c i) → (∀ (j : Fin 4), ε ≤ threeAmp l c j ∧ threeAmp l c j ≤ U ∧ ∃ (k : ℤ), threeAmp l c j = 2 ^ k) → ∀ (κ : Fin 4 → ℕ × ℕ), (∀ (i : Fin 4) (hi : i ≠ l), ∑ G ∈ pathGroups T gs y, ↑((D.withQ0 (qG G)).threeUsedReps G h y l (pathCell A.G T y) (fun (s : ℤ) => anc (pathCell A.G T y s) G) Iw c κ (pathDepths T gs y G) i hi).card ≤ K * threeAmp l c i ^ (-2)) ∧ ∑ G ∈ pathGroups T gs y, ∑ s ∈ pathDepths T gs y G, (D.withQ0 (qG G)).sideWeight G l (pathCell A.G T y) c.2.2.2.1 s ≤ K * threeAmp l c l ^ (-2)) (Λ3 Err : ℤ × ℤ → bf.Grp → ℝ) (hErr : ∀ (I : ℤ × ℤ) (G : bf.Grp), 0 ≤ Err I G) (hP : ∀ I ∈ T, ∀ G ∈ gs I, ∀ y ∈ dyadicGridCell A.G I.1 I.2, Λ3 I G ≤ K₁ * ∑ h ∈ H, (↑H.card)⁻¹ * (D.withQ0 (qG G)).threeBinSumA J Aw Iw G I y l (anc I G) h + Err I G) (CI : ℤ × ℤ → ℝ) (hCI : ∀ I ∈ T, CI I ≤ ∑ G ∈ gs I, Λ3 I G) {η : ℝ} (hη : 0 ≤ η) :
∑ I ∈ T, dyadicGridLength A.G I.1 * min η (CI I) ≤ η ^ (1 - (1 - 1 / 300 / 4)) * (dyadicGridLength A.G R.1 * (↑H.card * ↑(D.threeTypes l).card * ((K₁ * (8 * entryCU κ' J Aw * 2 ^ (Aw + 4) * 12 ^ 3)) ^ (1 - 1 / 300 / 4) * (D.fourCc * K ^ (2 - 1 / 300 / 2)) * ((↑Iw + 1) * ((Real.logb 2 (U / ε) + 1) ^ 4 * bandC (1 - 1 / 300 / 4) J))))) + ∑ I ∈ T, dyadicGridLength A.G I.1 * ∑ G ∈ gs I, (Err I G + K₁ * (8 * entryCU κ' J Aw * 2 ^ (Aw + 4) * 12 ^ 3 * ε * U ^ 3 * (D.S.Dstar ^ 3 * (↑Iw + 1))))

Blueprint eq:core-final (lines 2808-2817) for the three-core patterns with side role l on the side states: let T be the retained cells under the root R with their groups g(I), the anchors the hereditary representatives, and assume the principal bounds of eq:principal-core (‖Λ_{I,G}‖ ≤ K₁ × the height-averaged bin sum + err_{I,G}), the retained amplitudes σ ≤ U, at most D_* alive modes, and the path caps of eq:top-count along every retained path. Removing the negligible amplitudes (< ε), eq:min-power with subadditivity of x^p, the path estimates (three_path_total) and eq:path give, with p = 1 - 1/1200, Σ_I |I| min(η, C_I) ≤ η^{1/1200} |R| M_p + Σ_I |I| Σ_G (err_{I,G} + K₁ · negligible), where M_p = |H| |types| (K₁ C₃)^p C_c K^{2-c/2} (I_w + 1) (log₂(U/ε) + 1)⁴ C_band is polynomial in the parameters (P^{C₄}).

theorem TrilinearHT.Auto.min_sum_le_sum_min {ι : Type u_1} (s : Finset ι) (x : ι → ℝ) (hx : ∀ i ∈ s, 0 ≤ x i) {η : ℝ} (hη : 0 ≤ η) :
min η (∑ i ∈ s, x i) ≤ ∑ i ∈ s, min η (x i)

min(η, ·) is subadditive on nonnegative sums (η ≥ 0).

theorem TrilinearHT.Auto.core_final_patterns (G : RealDyadicGrid) (T : Finset (ℤ × ℤ)) {τ : Type u_1} (P : Finset τ) (x : τ → ℤ × ℤ → ℝ) (hx : ∀ t ∈ P, ∀ I ∈ T, 0 ≤ x t I) (CI : ℤ × ℤ → ℝ) (hCI : ∀ I ∈ T, CI I ≤ ∑ t ∈ P, x t I) {η : ℝ} (hη : 0 ≤ η) :
∑ I ∈ T, dyadicGridLength G I.1 * min η (CI I) ≤ ∑ t ∈ P, ∑ I ∈ T, dyadicGridLength G I.1 * min η (x t I)

Blueprint eq:core-final over the patterns with at least three cores: if the core quantity of each cell is at most the sum of the pattern quantities x_τ(I) ≥ 0, then Σ_I |I| min(η, C_I) ≤ Σ_τ Σ_I |I| min(η, x_τ(I)), so the bounds for the four-core pattern and the three-core patterns (four_core_final_side, three_core_final_side) add up.

The exported core estimate (declaration core-estimate, blueprint lines 2820-2822) #