OAI/Dynamics/SmoothObstruction/SmoothScales.lean
32 declarations, 0 contain sorry, 0 declared as axioms
- lemma
OAI.SmoothObstruction.SmoothScales.gap_scale_lowerline 15Source documentation
The source floor forces the gap×short-time scale to infinity. This is
g L ≥ b^(3/4)nforg = sqrt b,L = ceil(b^(1/4)n).lemma gap_scale_lower {b n : ℝ} (hn : 1≤n) (hb : n^(-1/10:ℝ)≤b) : n^(37/40:ℝ)≤b^(3/4:ℝ)*n := by … - theorem
OAI.SmoothObstruction.SmoothScales.gap_cost_absorptionline 29Source documentation
All fixed exponential gauge losses are absorbed by the genuine spectral gap×short-time exponent. Neither the gap nor the rates need be bounded away from zero.
theorem gap_cost_absorption {α : Type*} {l : Filter α} {b n : α → ℝ} (hbpos : ∀ᶠ k in l,0<b k) (hb : Tendsto b l (𝓝 0)) (hn : Tendsto n l atTop) (hfloor : ∀ᶠ k in l,n k^(-1/10:ℝ)≤b k) (C : ℝ) {c : ℝ} (hc : 0<c) : Tendsto (fun k => Real.exp (C*b k*n k-c*b k^(3/4:ℝ)*n k)) l (𝓝 0) := by … - theorem
OAI.SmoothObstruction.SmoothScales.weighted_gap_errorline 70Source documentation
Uniform exponentially small common-projector errors absorb every fixed gauge cost. This is the form needed by the finite leading-minor lower bound.
theorem weighted_gap_error {α : Type*} {l : Filter α} {b n δ : α → ℝ} (hbpos : ∀ᶠ k in l,0<b k) (hb : Tendsto b l (𝓝 0)) (hn : Tendsto n l atTop) (hfloor : ∀ᶠ k in l,n k^(-1/10:ℝ)≤b k) {c A : ℝ} (hc : 0<c) (hA : 0≤A) (C : ℝ) (hδ : ∀ᶠ k in l,0≤δ k ∧ δ k≤Real.exp (-c*b k^(3/4:ℝ)*n k)) : Tendsto (fun k => A*Real.exp (C*b k*n k)*δ k) l (𝓝 0) := by … - theorem
OAI.SmoothObstruction.SmoothScales.common_frame_budgetline 97Source documentation
The literal dimension/gauge cost in the common-fast leading-minor proof is eventually small. Its parameters are fixed before bins and frames are selected, and the good-frame error has the source gap×short-time scale.
theorem common_frame_budget {α : Type*} {l : Filter α} {b n δ : α → ℝ} (hbpos : ∀ᶠ k in l,0<b k) (hb : Tendsto b l (𝓝 0)) (hn : Tendsto n l atTop) (hfloor : ∀ᶠ k in l,n k^(-1/10:ℝ)≤b k) {c F C N : ℝ} (hc : 0<c) (hF : 0≤F) (hN : 0≤N) (K₀ : ℝ) (p : ℕ) (hδ : ∀ᶠ k in l,0≤δ k ∧ δ k≤Real.exp (-c*b k^(3/4:ℝ)*n k)) : ∀ᶠ k in l, (F*δ k)*(1+((C*(Real.exp (K₀*b k*n k))^p)^2*(C*2^p))* (C*(C^2*(Real.exp (K₀*b k*n k))^(2*p)*N)))≤1/2 := by … - theorem
OAI.SmoothObstruction.SmoothScales.weighted_full_gapline 138Source documentation
The full-time gap dominates the short-time gap scale and every fixed gauge cost. No uniform positive lower bound on the spectral gap is assumed.
theorem weighted_full_gap {α : Type*} {l : Filter α} {b n : α → ℝ} (hbpos : ∀ᶠ k in l,0<b k) (hb : Tendsto b l (𝓝 0)) (hn : Tendsto n l atTop) (hfloor : ∀ᶠ k in l,n k^(-1/10:ℝ)≤b k) {c A : ℝ} (hc : 0<c) (hA : 0≤A) (C : ℝ) : Tendsto (fun k => A*Real.exp (C*b k*n k)* Real.exp (-c*b k^(1/2:ℝ)*n k)) l (𝓝 0) := by … - theorem
OAI.SmoothObstruction.SmoothScales.common_graph_budgetline 157Source documentation
The literal finite-gauge Cramer bound tends to zero, uniformly over all common bins satisfying the same scalar data. Fixed constants account for the dimension, endpoint projections and unit wedge norms.
theorem common_graph_budget {α : Type*} {l : Filter α} {b n δ : α → ℝ} (hbpos : ∀ᶠ k in l,0<b k) (hb : Tendsto b l (𝓝 0)) (hn : Tendsto n l atTop) (hfloor : ∀ᶠ k in l,n k^(-1/10:ℝ)≤b k) {c C N V W : ℝ} (hc : 0<c) (hN : 0≤N) (hV : 0≤V) (hW : 0≤W) (K₀ : ℝ) (p : ℕ) (hδ : ∀ᶠ k in l,0≤δ k ∧ δ k≤Real.exp (-c*b k^(3/4:ℝ)*n k)) : Tendsto (fun k => 2*((C*(Real.exp (K₀*b k*n k))^p)^2*(C*2^p))* ((C*(C*(Real.exp (K₀*b k*n k))^p)^2*N*V)* Real.exp (-2*b k^(1/2:ℝ)*n k)+ (C*(C^2*(Real.exp (K₀*b k*n k))^(2*p)*N))*W*δ k)) l (𝓝 0) := by … - theorem
OAI.SmoothObstruction.SmoothScales.perturbed_graph_decayline 186Source documentation
Relative nonlinear errors preserve the vanishing Cramer bound.
theorem perturbed_graph_decay {α : Type*} {l : Filter α} {θ ε : α → ℝ} (hθ : Tendsto θ l (𝓝 0)) (hε : Tendsto ε l (𝓝 0)) : Tendsto (fun k => (θ k+ε k)/(1-ε k)) l (𝓝 0) := by … - lemma
OAI.SmoothObstruction.SmoothScales.ceil_gap_lowerline 200Source documentation
The manuscript's exact natural short length, including its ceiling.
lemma ceil_gap_lower {b n : ℝ} (hb : 0<b) : b^(3/4:ℝ)*n≤Real.sqrt b*(Nat.ceil (b^(1/4:ℝ)*n):ℝ) := by … - lemma
OAI.SmoothObstruction.SmoothScales.ceil_gap_errorline 208 - theorem
OAI.SmoothObstruction.SmoothScales.common_frame_budget_shortline 218Source documentation
Discharges the literal finite common-frame budget directly from the source's
g = sqrt bandL = ceil(b^(1/4)n), without dropping the rounding or imposing a positive gap floor.theorem common_frame_budget_short {α : Type*} {l : Filter α} {b n δ : α → ℝ} (hbpos : ∀ᶠ k in l,0<b k) (hb : Tendsto b l (𝓝 0)) (hn : Tendsto n l atTop) (hfloor : ∀ᶠ k in l,n k^(-1/10:ℝ)≤b k) {c F C N : ℝ} (hc : 0<c) (hF : 0≤F) (hN : 0≤N) (K₀ : ℝ) (p : ℕ) (hδ : ∀ᶠ k in l,0≤δ k ∧ δ k≤ Real.exp (-c*Real.sqrt (b k)*(Nat.ceil (b k^(1/4:ℝ)*n k):ℝ))) : ∀ᶠ k in l, (F*δ k)*(1+((C*(Real.exp (K₀*b k*n k))^p)^2*(C*2^p))* (C*(C^2*(Real.exp (K₀*b k*n k))^(2*p)*N)))≤1/2 := by … - theorem
OAI.SmoothObstruction.SmoothScales.uniform_common_frame_budgetline 238Source documentation
Uniformity over every possible common-bin rank, including empty groups. Constants may depend on the rank but are fixed before bins, points and times are chosen. Thus no hidden point-dependent large-scale threshold remains.
theorem uniform_common_frame_budget {α : Type*} {l : Filter α} {b n δ : α → ℝ} (hbpos : ∀ᶠ k in l,0<b k) (hb : Tendsto b l (𝓝 0)) (hn : Tendsto n l atTop) (hfloor : ∀ᶠ k in l,n k^(-1/10:ℝ)≤b k) (d : ℕ) {c : ℝ} (hc : 0<c) (F C N K₀ : Fin (d+1) → ℝ) (hF : ∀ p,0≤F p) (hN : ∀ p,0≤N p) (hδ : ∀ᶠ k in l,0≤δ k ∧ δ k≤ Real.exp (-c*Real.sqrt (b k)*(Nat.ceil (b k^(1/4:ℝ)*n k):ℝ))) : ∀ᶠ k in l,∀ p : Fin (d+1), (F p*δ k)*(1+((C p*(Real.exp (K₀ p*b k*n k))^p.val)^2*(C p*2^p.val))* (C p*((C p)^2*(Real.exp (K₀ p*b k*n k))^(2*p.val)*N p)))≤1/2 := by … - theorem
OAI.SmoothObstruction.SmoothScales.uniform_common_graph_budgetline 254Source documentation
The actual Cramer error is small for all ranks simultaneously. Fixed rank-dependent costs do not change the source's quantifier order.
theorem uniform_common_graph_budget {α : Type*} {l : Filter α} {b n δ : α → ℝ} (hbpos : ∀ᶠ k in l,0<b k) (hb : Tendsto b l (𝓝 0)) (hn : Tendsto n l atTop) (hfloor : ∀ᶠ k in l,n k^(-1/10:ℝ)≤b k) (d : ℕ) {c : ℝ} (hc : 0<c) (C N V W K₀ : Fin (d+1) → ℝ) (hN : ∀ p,0≤N p) (hV : ∀ p,0≤V p) (hW : ∀ p,0≤W p) (hδ : ∀ᶠ k in l,0≤δ k ∧ δ k≤ Real.exp (-c*Real.sqrt (b k)*(Nat.ceil (b k^(1/4:ℝ)*n k):ℝ))) {ε : ℝ} (hε : 0<ε) : ∀ᶠ k in l,∀ p : Fin (d+1), 2*((C p*(Real.exp (K₀ p*b k*n k))^p.val)^2*(C p*2^p.val))* ((C p*(C p*(Real.exp (K₀ p*b k*n k))^p.val)^2*N p*V p)* … - theorem
OAI.SmoothObstruction.ExteriorStability.exterior_segmentline 284Source documentation
Exterior powers commute with the source's time-ordered product.
theorem exterior_segment (p : ℕ) (A : ℕ → E →L[ℝ] E) (s n : ℕ) : exterior p (segment A s n) = segment (fun t => exterior p (A t)) s n := by … - theorem
OAI.SmoothObstruction.ExteriorStability.relative_exterior_coreline 295Source documentation
Finite-product assertion of manuscript Lemma
lem:relative-exterior, in the canonical exterior Euclidean operator norm. The one-step constant is uniform over all exterior degreesp ≤ d, all products and all lengths. The proof actually does not needa ≥ 0,m ≥ 1, orK ≥ 1; the source assumptions are retained in the final source-specialized statement.theorem relative_exterior_core (d : ℕ) (K : ℝ) : ∃ C : ℝ, 0 ≤ C ∧ ∀ (p : ℕ), p ≤ d → ∀ (A B : ℕ → E →L[ℝ] E) (m : ℕ) (σ a ε : ℝ), 0 ≤ ε → ε ≤ 1 → (∀ t < m, ‖A t‖ ≤ K) → (∀ t < m, ‖B t - A t‖ ≤ ε) → (∀ s k : ℕ, s + k ≤ m → ‖exterior p (segment A s k)‖ ≤ Real.exp (σ * k + a)) → ‖exterior p (segment B 0 m) - exterior p (segment A 0 m)‖ ≤ Real.exp (σ * m + a) * ((1 + C * ε * Real.exp (a + |σ|)) ^ m - 1) := by … - theorem
OAI.SmoothObstruction.ExteriorStability.relative_exteriorline 319Source documentation
Exact finite estimate in
lem:relative-exteriorfor reald-dimensional Euclidean operators, hence real matrices in the Euclidean operator norm.theorem relative_exterior (d : ℕ) (_hd : 1 ≤ d) (K : ℝ) (_hK : 1 ≤ K) : ∃ C : ℝ, 0 ≤ C ∧ ∀ (p : ℕ), p ≤ d → ∀ (m : ℕ), 1 ≤ m → ∀ (A B : ℕ → EuclideanSpace ℝ (Fin d) →L[ℝ] EuclideanSpace ℝ (Fin d)) (σ a ε : ℝ), 0 ≤ a → 0 ≤ ε → ε ≤ 1 → (∀ t < m, ‖A t‖ ≤ K) → (∀ t < m, ‖B t - A t‖ ≤ ε) → (∀ s k : ℕ, s + k ≤ m → ‖exterior p (segment A s k)‖ ≤ Real.exp (σ * k + a)) → ‖exterior p (segment B 0 m) - exterior p (segment A 0 m)‖ ≤ Real.exp (σ * m + a) * ((1 + C * ε * Real.exp (a + |σ|)) ^ m - 1) := by … - theorem
OAI.SmoothObstruction.ExteriorStability.exterior_zeroline 340 - theorem
OAI.SmoothObstruction.ExteriorStability.exterior_zero_differenceline 346 - def
OAI.SmoothObstruction.SmoothScales.Bline 352Source documentation
The spatial exponent in the manuscript, with its literal positive root.
noncomputable def B (b : ℝ) : ℝ := b ^ (1 / 8 : ℝ)
- lemma
OAI.SmoothObstruction.SmoothScales.positiveline 354 - lemma
OAI.SmoothObstruction.SmoothScales.ratioline 356 - lemma
OAI.SmoothObstruction.SmoothScales.root_sixteenline 362 - lemma
OAI.SmoothObstruction.SmoothScales.lower_boundline 366 - lemma
OAI.SmoothObstruction.SmoothScales.square_dominatesline 379 - lemma
OAI.SmoothObstruction.SmoothScales.ratio_tendsto_zeroline 389 - theorem
OAI.SmoothObstruction.SmoothScales.decisive_decayline 401Source documentation
Quantitative exponential decay used in the relative-minor transfer. It holds for every fixed real coefficient
C, including all constants coming from the dimension, charts, and inverse basis bounds.theorem decisive_decay {α : Type*} {l : Filter α} {b n : α → ℝ} (hbpos : ∀ᶠ k in l, 0 < b k) (hb : Tendsto b l (𝓝 0)) (hn : Tendsto n l atTop) (hfloor : ∀ᶠ k in l, n k ^ (-1 / 10 : ℝ) ≤ b k) (C : ℝ) : Tendsto (fun k => n k * Real.exp (-(3 / 2 : ℝ) * B (b k) * n k + C * b k * n k)) l (𝓝 0) := by …Mentions: OAI.SmoothObstruction.SmoothScales.B
- lemma
OAI.SmoothObstruction.RelativeAsymptotics.exp_sub_one_leline 432 - lemma
OAI.SmoothObstruction.RelativeAsymptotics.pow_one_add_sub_one_leline 436 - lemma
OAI.SmoothObstruction.RelativeAsymptotics.normalized_estimateline 444 - theorem
OAI.SmoothObstruction.RelativeAsymptotics.uniform_relative_smallline 473Source documentation
The little-oh conclusion of the exterior-product lemma. The hypotheses on
xare the proved finite estimate, not an asymptotic conclusion. This bound is uniform over all products and all exterior degrees covered by that estimate.theorem uniform_relative_small {α : Type*} {l : Filter α} {b n x σ a ε : α → ℝ} {m : α → ℕ} (hbpos : ∀ᶠ k in l, 0 < b k) (hb : Tendsto b l (𝓝 0)) (hn : Tendsto n l atTop) (hfloor : ∀ᶠ k in l, n k ^ (-1 / 10 : ℝ) ≤ b k) (C C₀ C₁ C₂ C₃ : ℝ) (hC : 0 ≤ C) (hC₀ : 0 ≤ C₀) (hC₃ : 0 ≤ C₃) (hdata : ∀ᶠ k in l, 0 ≤ x k ∧ 0 ≤ a k ∧ 0 ≤ ε k ∧ (m k : ℝ) ≤ 2 * n k ∧ a k ≤ C₁ * b k * n k ∧ |σ k| ≤ C₂ ∧ ε k ≤ C₃ * Real.exp (-(3 / 2 : ℝ) * SmoothScales.B (b k) * n k) ∧ x k ≤ Real.exp (σ k * m k + a k) * …Mentions: OAI.SmoothObstruction.SmoothScales.B
- def
OAI.SmoothObstruction.RelativeAsymptotics.envelopeline 545Source documentation
A single, choice-independent numerical envelope for all the relative exterior errors at a given dyadic scale.
noncomputable def envelope (C C₀ C₁ C₂ C₃ b n : ℝ) : ℝ := …
- lemma
OAI.SmoothObstruction.RelativeAsymptotics.uniform_relative_boundline 550 - lemma
OAI.SmoothObstruction.RelativeAsymptotics.envelope_tendstoline 584