v21 Section 3.2: analytic normalized map and uniform jets

Lean contributors (project-level attribution): Zichen Wang.

Source header · Attribution source

public sectionnoncomputable sectionopen Filteropen scoped Topologynamespace DFP.Polar.Normalized

The ordered family of scalar denominators in the two normalized stages and rescaling.

def denominators (L κ t : ℝ) (x : ℝ × ℝ × ℝ) : Fin 9 → ℝ := let ζOdd := feedback L κ t x let x₁ := cycleStage L κ 0 t x x ![unitRoot t, unitRoot t ^ 2 + unitRoot t + 1, sinFactor 0 t, sinFactor 1 t, x.1, x.2.1, stageSigma 0 t x 1 ζOdd, x₁.2.1, stageSigma 1 t x₁ ζOdd 1]

The explicit ordered formula defining all denominators used by Filter.map.{u_1, u_2} {α : Type u_1} {β : Type u_2} (m : α → β) (f : Filter α) : Filter βmap.

theorem denominators_apply (L κ t : ℝ) (x : ℝ × ℝ × ℝ) : denominators L κ t x = let ζOdd := feedback L κ t x let x₁ := cycleStage L κ 0 t x x ![unitRoot t, unitRoot t ^ 2 + unitRoot t + 1, sinFactor 0 t, sinFactor 1 t, x.1, x.2.1, stageSigma 0 t x 1 ζOdd, x₁.2.1, stageSigma 1 t x₁ ζOdd 1] := by -- The denominator family is defined by exactly this displayed vector. rfl

A two-entry vector is analytic when both scalar coordinates are analytic.

private lemma analyticAtFin2Vec {E : Type*} [NormedAddCommGroup E] [NormedSpace ℝ E] {f : E → Fin 2 → ℝ} {z : E} (h0 : AnalyticAt ℝ (fun y : E ↦ f y 0) z) (h1 : AnalyticAt ℝ (fun y : E ↦ f y 1) z) : AnalyticAt ℝ f z := by -- Reduce the finite vector to its two coordinate germs. rw [analyticAt_pi_iff] intro j fin_cases j · simpa using h0 · simpa using h1

The coordinate update as a function of its normalized gradient change and rotation matrix.

private def stageFromFactors (x : ℝ × ℝ × ℝ) (d : (Fin 2 → ℝ) × (Fin 2 → Fin 2 → ℝ)) : ℝ × ℝ × ℝ := let X := RealSymmetric2.matrix x.1 x.2.1 x.2.2 let v := stepV x let w := d.1 let σ := v ⬝ᵥ w let P := Matrix.of d.2 let Y := P * ((1 - σ⁻¹ • Matrix.vecMulVec w v) * X * (1 - σ⁻¹ • Matrix.vecMulVec v w) + σ⁻¹ • Matrix.vecMulVec w w) * P.transpose (Y 0 0, Y 0 1, Y 1 1)

The normalized stage equals the update formed from its gradient change and rotation matrix.

private lemma stage_eq_fromFactors (i : Fin 2) (t : ℝ) (x : ℝ × ℝ × ℝ) (ζ η : ℝ) : stage i t x ζ η = stageFromFactors x (stepW i t x ζ η, fun r c ↦ rotation i t r c) := by -- The interface uses exactly the source's factored bracket and rotation congruence. simp only [stage_apply, stageMatrix_def, stageSigma_def, stageFromFactors] rfl

The factorized coordinate update preserves analytic state and factor paths with a nonzero secant scalar.

private lemma stageFromFactorsAlong_analyticAt {E : Type*} [NormedAddCommGroup E] [NormedSpace ℝ E] {X : E → ℝ × ℝ × ℝ} {d : E → (Fin 2 → ℝ) × (Fin 2 → Fin 2 → ℝ)} {z : E} (hX : AnalyticAt ℝ X z) (hd : AnalyticAt ℝ d z) (hSlope : AnalyticAt ℝ (fun y ↦ slope (X y)) z) (hσ0 : stepV (X z) ⬝ᵥ (d z).1 ≠ 0) : AnalyticAt ℝ (fun y ↦ stageFromFactors (X y) (d y)) z := by -- Expose each state coordinate through analytic projections. have hA : AnalyticAt ℝ (fun y ↦ (X y).1) z := AnalyticAt.comp (f := X) analyticAt_fst hX have hCV : AnalyticAt ℝ (fun y ↦ (X y).2) z := AnalyticAt.comp (f := X) analyticAt_snd hX have hC : AnalyticAt ℝ (fun y ↦ (X y).2.1) z := AnalyticAt.comp (f := fun y ↦ (X y).2) analyticAt_fst hCV have hV : AnalyticAt ℝ (fun y ↦ (X y).2.2) z := AnalyticAt.comp (f := fun y ↦ (X y).2) analyticAt_snd hCV -- The finite factor coordinates supply the vector and rotation entries. have hW : AnalyticAt ℝ (fun y ↦ (d y).1) z := AnalyticAt.comp (f := d) analyticAt_fst hd have hP : AnalyticAt ℝ (fun y ↦ (d y).2) z := AnalyticAt.comp (f := d) analyticAt_snd hd have hW0 := analyticAt_pi_iff.mp hW 0 have hW1 := analyticAt_pi_iff.mp hW 1 have hP0 := analyticAt_pi_iff.mp hP 0 have hP1 := analyticAt_pi_iff.mp hP 1 have hP00 := analyticAt_pi_iff.mp hP0 0 have hP01 := analyticAt_pi_iff.mp hP0 1 have hP10 := analyticAt_pi_iff.mp hP1 0 have hP11 := analyticAt_pi_iff.mp hP1 1 have hV0 : AnalyticAt ℝ (fun y ↦ stepV (X y) 0) z := by simp only [stepV_def, Fin.isValue, Matrix.cons_val_zero] exact hSlope.neg have hV1 : AnalyticAt ℝ (fun y ↦ stepV (X y) 1) z := by simp only [stepV_def, Fin.isValue, Matrix.cons_val_one, Matrix.cons_val_fin_one] exact analyticAt_const have hσ : AnalyticAt ℝ (fun y ↦ stepV (X y) ⬝ᵥ (d y).1) z := by simp only [dotProduct, Fin.sum_univ_two] exact (hV0.mul hW0).add (hV1.mul hW1) have hσinv := hσ.inv hσ0 -- The inverse secant is the only nonpolynomial operation in the factored bracket. simp only [stageFromFactors, Matrix.mul_apply, Fin.sum_univ_two, Matrix.transpose_apply, Matrix.smul_apply, Matrix.sub_apply, Matrix.add_apply, Matrix.vecMulVec_apply, Matrix.one_apply, Matrix.of_apply, smul_eq_mul, RealSymmetric2.matrix_eq, Fin.isValue, Matrix.cons_val_zero, Matrix.cons_val_one, Matrix.cons_val_fin_one] fun_prop

The factorized coordinate update is analytic in its vector and rotation at a nonzero secant scalar.

private lemma stageFromFactors_analyticAt (x : ℝ × ℝ × ℝ) (d : (Fin 2 → ℝ) × (Fin 2 → Fin 2 → ℝ)) (hσ0 : stepV x ⬝ᵥ d.1 ≠ 0) : AnalyticAt ℝ (stageFromFactors x) d := by -- Fixed-state quotients are scalar constants, including when C is zero. exact stageFromFactorsAlong_analyticAt analyticAt_const analyticAt_id analyticAt_const hσ0

The canceled normalized stage is analytic along analytic scale, state, and multiplier paths.

private lemma stageAlong_analyticAt {E : Type*} [NormedAddCommGroup E] [NormedSpace ℝ E] (i : Fin 2) (τ : E → ℝ) (X : E → ℝ × ℝ × ℝ) (ζ ζNext : E → ℝ) (z : E) (hτ : AnalyticAt ℝ τ z) (hX : AnalyticAt ℝ X z) (hζ : AnalyticAt ℝ ζ z) (hζNext : AnalyticAt ℝ ζNext z) (hSlope : AnalyticAt ℝ (fun y ↦ slope (X y)) z) (hτ0 : τ z = 0) (hσ0 : stageSigma i (τ z) (X z) (ζ z) (ζNext z) ≠ 0) : AnalyticAt ℝ (fun y ↦ stage i (τ y) (X y) (ζ y) (ζNext y)) z := by -- Compose only the regularized trigonometric germs with the scale path. have hAngle : AnalyticAt ℝ (fun y ↦ stageAngle i (τ y)) z := (stageAngle_analyticAt i).comp_of_eq hτ hτ0 have hSin : AnalyticAt ℝ (fun y ↦ sinFactor i (τ y)) z := (sinFactor_analyticAt i).comp_of_eq hτ hτ0 have hCot : AnalyticAt ℝ (fun y ↦ cotFactor i (τ y)) z := (cotFactor_analyticAt i).comp_of_eq hτ hτ0 have hW : AnalyticAt ℝ (fun y ↦ stepW i (τ y) (X y) (ζ y) (ζNext y)) z := by apply analyticAtFin2Vec · simp only [stepW_def, Fin.isValue, Matrix.cons_val_zero] fun_prop · simpa only [stepW_def, Fin.isValue, Matrix.cons_val_one, Matrix.cons_val_fin_one] using hζNext have hP : AnalyticAt ℝ (fun y ↦ fun r c : Fin 2 ↦ rotation i (τ y) r c) z := by refine analyticAt_pi_iff.mpr ?_ intro r refine analyticAt_pi_iff.mpr ?_ intro c fin_cases r all_goals fin_cases c all_goals simp only [rotation_def, Matrix.of_apply, Fin.zero_eta, Fin.mk_one, Fin.isValue, Matrix.cons_val_zero, Matrix.cons_val_one, Matrix.cons_val_fin_one] all_goals fun_prop have hsecant : stepV (X z) ⬝ᵥ stepW i (τ z) (X z) (ζ z) (ζNext z) ≠ 0 := by simpa only [stageSigma_def] using hσ0 -- Assemble the exact source bracket through its factor interface. simp only [stage_eq_fromFactors] exact stageFromFactorsAlong_analyticAt hX (hW.prod hP) hSlope hsecant

The normalized secant scalar is the dot product of the normalized step and gradient change.

private lemma stageSigma_scalar (i : Fin 2) (t : ℝ) (x : ℝ × ℝ × ℝ) (ζ η : ℝ) : stageSigma i t x ζ η = -slope x * ((η - ζ) * cotFactor i t - ζ * t ^ 4 * slope x) + η := by -- Evaluate the two entries of the normalized dot product. simp [stageSigma_def, stepV_def, stepW_def]

Slope preserves analyticity along a state path with a nonzero off-diagonal coordinate.

private lemma slopeAlong_analyticAt {E : Type*} [NormedAddCommGroup E] [NormedSpace ℝ E] {X : E → ℝ × ℝ × ℝ} {z : E} (hX : AnalyticAt ℝ X z) (hC : (X z).2.1 ≠ 0) : AnalyticAt ℝ (fun y ↦ slope (X y)) z := by -- Compose the two coordinate projections before dividing by C. have hCV : AnalyticAt ℝ (fun y ↦ (X y).2) z := AnalyticAt.comp (f := X) analyticAt_snd hX have hC' : AnalyticAt ℝ (fun y ↦ (X y).2.1) z := AnalyticAt.comp (f := fun y ↦ (X y).2) analyticAt_fst hCV have hV : AnalyticAt ℝ (fun y ↦ (X y).2.2) z := AnalyticAt.comp (f := fun y ↦ (X y).2) analyticAt_snd hCV simp only [slope_def] exact hV.div hC' hC

Feedback is analytic along scale and boundary-state paths with nonzero radial coordinate.

private lemma feedbackAlong_analyticAt {E : Type*} [NormedAddCommGroup E] [NormedSpace ℝ E] (L κ : ℝ) {τ : E → ℝ} {X : E → ℝ × ℝ × ℝ} {z : E} (hτ : AnalyticAt ℝ τ z) (hX : AnalyticAt ℝ X z) (hA : (X z).1 ≠ 0) : AnalyticAt ℝ (fun y ↦ feedback L κ (τ y) (X y)) z := by -- The only state quotient is κ / A; the remaining operations are entire. have hA' : AnalyticAt ℝ (fun y ↦ (X y).1) z := AnalyticAt.comp (f := X) analyticAt_fst hX simp only [feedback_def] fun_prop

The normalized secant scalar is analytic along regular analytic input paths.

private lemma stageSigmaAlong_analyticAt {E : Type*} [NormedAddCommGroup E] [NormedSpace ℝ E] (i : Fin 2) {τ : E → ℝ} {X : E → ℝ × ℝ × ℝ} {ζ η : E → ℝ} {z : E} (hτ : AnalyticAt ℝ τ z) (hζ : AnalyticAt ℝ ζ z) (hη : AnalyticAt ℝ η z) (hSlope : AnalyticAt ℝ (fun y ↦ slope (X y)) z) (hτ0 : τ z = 0) : AnalyticAt ℝ (fun y ↦ stageSigma i (τ y) (X y) (ζ y) (η y)) z := by -- Use the canceled cotangent rather than its singular quotient representation. have hCot : AnalyticAt ℝ (fun y ↦ cotFactor i (τ y)) z := (cotFactor_analyticAt i).comp_of_eq hτ hτ0 simp only [stageSigma_scalar] fun_prop

A cycle stage is analytic along paths with analytic feedback and slope at zero scale.

private lemma cycleStageAlong_analyticAt {E : Type*} [NormedAddCommGroup E] [NormedSpace ℝ E] (L κ : ℝ) (i : Fin 2) {τ : E → ℝ} {X₀ X : E → ℝ × ℝ × ℝ} {z : E} (hτ : AnalyticAt ℝ τ z) (hX : AnalyticAt ℝ X z) (hfb : AnalyticAt ℝ (fun y ↦ feedback L κ (τ y) (X₀ y)) z) (hSlope : AnalyticAt ℝ (fun y ↦ slope (X y)) z) (hτ0 : τ z = 0) : AnalyticAt ℝ (fun y ↦ cycleStage L κ i (τ y) (X₀ y) (X y)) z := by -- Each phase selects one constant multiplier and the same boundary feedback. let ζ : E → ℝ := fun y ↦ (![1, feedback L κ (τ y) (X₀ y)] : Fin 2 → ℝ) i let η : E → ℝ := fun y ↦ (![feedback L κ (τ y) (X₀ y), 1] : Fin 2 → ℝ) i have hζ : AnalyticAt ℝ ζ z := by fin_cases i · simpa only [ζ, Fin.zero_eta, Fin.isValue, Matrix.cons_val_zero] using (analyticAt_const : AnalyticAt ℝ (fun _ : E ↦ (1 : ℝ)) z) · simpa [ζ] using hfb have hη : AnalyticAt ℝ η z := by fin_cases i · simpa [η] using hfb · simpa [η] using (analyticAt_const : AnalyticAt ℝ (fun _ : E ↦ (1 : ℝ)) z) have hfb0 : feedback L κ (τ z) (X₀ z) = 1 := by rw [hτ0] simp [feedback_def] have hζ0 : ζ z = 1 := by dsimp only [ζ] rw [hfb0] fin_cases i all_goals rfl have hη0 : η z = 1 := by dsimp only [η] rw [hfb0] fin_cases i all_goals rfl have hσ0 : stageSigma i (τ z) (X z) (ζ z) (η z) ≠ 0 := by rw [hτ0, hζ0, hη0] simp [stageSigma_scalar] simpa only [cycleStage_apply, ζ, η] using stageAlong_analyticAt i τ X ζ η z hτ hX hζ hη hSlope hτ0 hσ0

Rescaling preserves analyticity of state and scale paths based at zero scale.

private lemma rescaleAlong_analyticAt {E : Type*} [NormedAddCommGroup E] [NormedSpace ℝ E] {τ : E → ℝ} {X : E → ℝ × ℝ × ℝ} {z : E} (hτ : AnalyticAt ℝ τ z) (hX : AnalyticAt ℝ X z) (hτ0 : τ z = 0) : AnalyticAt ℝ (fun y ↦ rescale (τ y) (X y)) z := by -- Coordinate multiplication by the regularized rescaling factor is analytic. have hA : AnalyticAt ℝ (fun y ↦ (X y).1) z := AnalyticAt.comp (f := X) analyticAt_fst hX have hCV : AnalyticAt ℝ (fun y ↦ (X y).2) z := AnalyticAt.comp (f := X) analyticAt_snd hX have hC : AnalyticAt ℝ (fun y ↦ (X y).2.1) z := AnalyticAt.comp (f := fun y ↦ (X y).2) analyticAt_fst hCV have hV : AnalyticAt ℝ (fun y ↦ (X y).2.2) z := AnalyticAt.comp (f := fun y ↦ (X y).2) analyticAt_snd hCV have hρ : AnalyticAt ℝ (fun y ↦ rescaleFactor (τ y)) z := rescaleFactor_analyticAt.comp_of_eq hτ hτ0 simp only [rescale_apply] exact ((hρ.pow 2).mul hA).prod ((hρ.mul hC).prod hV)

The root branch is analytic at its unit base value.

private lemma unitRootAnalytic : AnalyticAt ℝ unitRoot 0 := by -- The logarithm argument equals one at zero. have hpoly : AnalyticAt ℝ (fun t : ℝ ↦ 1 + t ^ 3) 0 := by fun_prop have hnonzero : 0 < (1 : ℝ) + 0 ^ 3 := by norm_num rw [funext unitRoot_def] exact ((hpoly.log hnonzero).div_const).rexp'

The regularized sine factor has the prescribed limiting angular coefficient.

private lemma sinFactor_zero (i : Fin 2) : sinFactor i 0 = limitAngle i := by -- Evaluate the canceled sinc formula, whose value at zero is one. fin_cases i all_goals norm_num [sinFactor_def, stageAngle_def, stageAngleFactor_def, angleFactor_def, unitRoot_def, limitAngle_zero, limitAngle_one, stageWeight_zero, stageWeight_one] all_goals ring

At zero scale, the denominators reduce to limiting angular coefficients and state coordinates.

private lemma denominators_zero (L κ : ℝ) (x : ℝ × ℝ × ℝ) : denominators L κ 0 x = ![1, 3, limitAngle 0, limitAngle 1, x.1, x.2.1, 1, (leadingStage (limitAngle 0) x).2.1, 1] := by -- Unit feedback makes both secant scalars exactly one at zero. simp [denominators_apply, unitRoot_def, sinFactor_zero, feedback_def, cycleStage_zero, stageSigma_scalar] norm_num

The finite denominator family is jointly analytic at zero scale and a positive state.

private lemma denominators_analyticAt_zero (L κ : ℝ) (x : ℝ × ℝ × ℝ) (hA : 0 < x.1) (hC : 0 < x.2.1) (hC₁ : 0 < (leadingStage (limitAngle 0) x).2.1) : AnalyticAt ℝ (fun y : ℝ × (ℝ × ℝ × ℝ) ↦ denominators L κ y.1 y.2) (0, x) := by -- The first-stage state and both secant scalars use the same analytic feedback. have hτ : AnalyticAt ℝ (fun y : ℝ × (ℝ × ℝ × ℝ) ↦ y.1) (0, x) := analyticAt_fst have hX : AnalyticAt ℝ (fun y : ℝ × (ℝ × ℝ × ℝ) ↦ y.2) (0, x) := analyticAt_snd have hfb := feedbackAlong_analyticAt L κ hτ hX hA.ne' have hSlope := slopeAlong_analyticAt hX hC.ne' have hX₁ := cycleStageAlong_analyticAt L κ 0 hτ hX hfb hSlope rfl have hC₁₀ : (cycleStage L κ 0 0 x x).2.1 ≠ 0 := by rw [cycleStage_zero] exact hC₁.ne' have hSlope₁ := slopeAlong_analyticAt hX₁ hC₁₀ have hσ₀ := stageSigmaAlong_analyticAt 0 (ζ := fun _ ↦ 1) hτ analyticAt_const hfb hSlope rfl have hσ₁ := stageSigmaAlong_analyticAt 1 (η := fun _ ↦ 1) hτ hfb analyticAt_const hSlope₁ rfl have hroot : AnalyticAt ℝ (fun y : ℝ × (ℝ × ℝ × ℝ) ↦ unitRoot y.1) (0, x) := unitRootAnalytic.comp_of_eq hτ rfl have hsin0 : AnalyticAt ℝ (fun y : ℝ × (ℝ × ℝ × ℝ) ↦ sinFactor 0 y.1) (0, x) := (sinFactor_analyticAt 0).comp_of_eq hτ rfl have hsin1 : AnalyticAt ℝ (fun y : ℝ × (ℝ × ℝ × ℝ) ↦ sinFactor 1 y.1) (0, x) := (sinFactor_analyticAt 1).comp_of_eq hτ rfl have hCV : AnalyticAt ℝ (fun y : ℝ × (ℝ × ℝ × ℝ) ↦ y.2.2) (0, x) := AnalyticAt.comp (f := fun y : ℝ × (ℝ × ℝ × ℝ) ↦ y.2) analyticAt_snd hX have hA' : AnalyticAt ℝ (fun y : ℝ × (ℝ × ℝ × ℝ) ↦ y.2.1) (0, x) := AnalyticAt.comp (f := fun y : ℝ × (ℝ × ℝ × ℝ) ↦ y.2) analyticAt_fst hX have hC' : AnalyticAt ℝ (fun y : ℝ × (ℝ × ℝ × ℝ) ↦ y.2.2.1) (0, x) := AnalyticAt.comp (f := fun y : ℝ × (ℝ × ℝ × ℝ) ↦ y.2.2) analyticAt_fst hCV have hX₁CV : AnalyticAt ℝ (fun y : ℝ × (ℝ × ℝ × ℝ) ↦ (cycleStage L κ 0 y.1 y.2 y.2).2) (0, x) := AnalyticAt.comp (f := fun y : ℝ × (ℝ × ℝ × ℝ) ↦ cycleStage L κ 0 y.1 y.2 y.2) analyticAt_snd hX₁ have hX₁C : AnalyticAt ℝ (fun y : ℝ × (ℝ × ℝ × ℝ) ↦ (cycleStage L κ 0 y.1 y.2 y.2).2.1) (0, x) := AnalyticAt.comp (f := fun y : ℝ × (ℝ × ℝ × ℝ) ↦ (cycleStage L κ 0 y.1 y.2 y.2).2) analyticAt_fst hX₁CV -- Evaluate each finite index after proving all meaningful component germs. refine analyticAt_pi_iff.mpr ?_ intro i fin_cases i all_goals simp only [denominators_apply, Fin.zero_eta, Fin.mk_one, Fin.isValue, Fin.reduceFinMk, Matrix.cons_val] all_goals fun_prop

Both limiting angular coefficients are strictly positive.

private lemma limitAngle_pos (i : Fin 2) : 0 < limitAngle i := by -- Both phase weights multiply the positive prescribed base angle. have hbase : 0 < DFP.Polar.baseAngle := by rw [DFP.Polar.baseAngle_def] positivity fin_cases i · simpa only [Fin.zero_eta, Fin.isValue, limitAngle_zero] using mul_pos (two_pos : (0 : ℝ) < 2) hbase · simpa only [Fin.mk_one, Fin.isValue, limitAngle_one] using hbase

All zero-scale denominators are positive on a positive state neighborhood.

private lemma denominators_pos_zero (L κ : ℝ) (x : ℝ × ℝ × ℝ) (hA : 0 < x.1) (hC : 0 < x.2.1) (hC₁ : 0 < (leadingStage (limitAngle 0) x).2.1) (i : Fin 9) : 0 < denominators L κ 0 x i := by -- Read positivity from the exact boundary denominator vector. rw [denominators_zero] fin_cases i all_goals simp only [Nat.succ_eq_add_one, Nat.reduceAdd, Fin.isValue, Fin.reduceFinMk, Matrix.cons_val] all_goals first | exact hA | exact hC | exact hC₁ | exact limitAngle_pos _ | norm_num

On a compact positive-state set, all normalized denominators have one positive lower bound for uniformly small scale.

theorem denominators_uniformOn (L κ : ℝ) (𝒦 : Set (ℝ × ℝ × ℝ)) (h𝒦 : IsCompact 𝒦) (hA : ∀ x ∈ 𝒦, 0 < x.1) (hC : ∀ x ∈ 𝒦, 0 < x.2.1) (hC₁ : ∀ x ∈ 𝒦, 0 < (leadingStage (limitAngle 0) x).2.1) : ∃ ε > 0, ∃ d > 0, ∀ t ∈ Set.Icc (-ε) ε, ∀ x ∈ 𝒦, ∀ i : Fin 9, d ≤ |denominators L κ t x i| := (deterministic) timeout at `whnf`, maximum number of heartbeats (10000000) has been reached Note: Use `set_option maxHeartbeats <num>` to set the limit. Hint: Additional diagnostic information may be available using the `set_option diagnostics true` command.by -- Take one positive minimum on the fixed compact state/index fiber. have hcompact : IsCompact (𝒦 ×ˢ (Set.univ : Set (Fin 9))) := h𝒦.prod isCompact_univ have hcontinuous : ContinuousOn (fun xi : (ℝ × ℝ × ℝ) × Fin 9 ↦ |denominators L κ 0 xi.1 xi.2|) (𝒦 ×ˢ Set.univ) := by intro xi hxi have hD := (denominators_analyticAt_zero L κ xi.1 (hA xi.1 hxi.1) (hC xi.1 hxi.1) (hC₁ xi.1 hxi.1)).continuousAt have hbase : ContinuousAt (fun x : ℝ × ℝ × ℝ ↦ denominators L κ 0 x) xi.1 := hD.comp (continuousAt_const.prodMk continuousAt_id) have heval : ContinuousAt (fun z : (ℝ × ℝ × ℝ) × Fin 9 ↦ denominators L κ 0 z.1 z.2) xi := by apply continuousAt_prod_of_discrete_right.mpr exact (continuousAt_apply xi.2 _).comp hbase exact heval.abs.continuousWithinAt have hpositive : ∀ xi ∈ (𝒦 ×ˢ (Set.univ : Set (Fin 9))), 0 < |denominators L κ 0 xi.1 xi.2| := by intro xi hxi exact abs_pos.mpr (denominators_pos_zero L κ xi.1 (hA xi.1 hxi.1) (hC xi.1 hxi.1) (hC₁ xi.1 hxi.1) xi.2).ne' obtain ⟨b, hb, hbound⟩ := hcompact.exists_forall_le' hcontinuous hpositive -- Joint continuity preserves half this minimum on one common scale radius. have hnear : ∀ xi ∈ (𝒦 ×ˢ (Set.univ : Set (Fin 9))), ∀ᶠ z : ℝ × ((ℝ × ℝ × ℝ) × Fin 9) in 𝓝 (0, xi), b / 2 ≤ |denominators L κ z.1 z.2.1 z.2.2| := by intro xi hxi have hD := (denominators_analyticAt_zero L κ xi.1 (hA xi.1 hxi.1) (hC xi.1 hxi.1) (hC₁ xi.1 hxi.1)).continuousAt have hinput : ContinuousAt (fun z : ℝ × ((ℝ × ℝ × ℝ) × Fin 9) ↦ (z.1, z.2.1)) (0, xi) := continuousAt_fst.prodMk (continuousAt_snd.fst) have hindex : ContinuousAt (fun z : ℝ × ((ℝ × ℝ × ℝ) × Fin 9) ↦ z.2.2) (0, xi) := continuousAt_snd.snd have hfull : ContinuousAt (fun z : (ℝ × (ℝ × ℝ × ℝ)) × Fin 9 ↦ |denominators L κ z.1.1 z.1.2 z.2|) ((0, xi.1), xi.2) := by apply continuousAt_prod_of_discrete_right.mpr exact ((continuousAt_apply xi.2 _).comp hD).abs have heval : ContinuousAt (fun z : ℝ × ((ℝ × ℝ × ℝ) × Fin 9) ↦ |denominators L κ z.1 z.2.1 z.2.2|) (0, xi) := by simpa only [Function.comp_def] using (ContinuousAt.comp (g := fun z : (ℝ × (ℝ × ℝ × ℝ)) × Fin 9 ↦ |denominators L κ z.1.1 z.1.2 z.2|) (f := fun z : ℝ × ((ℝ × ℝ × ℝ) × Fin 9) ↦ ((z.1, z.2.1), z.2.2)) (deterministic) timeout at `isDefEq`, maximum number of heartbeats (10000000) has been reached Note: Use `set_option maxHeartbeats <num>` to set the limit. Hint: Additional diagnostic information may be available using the `set_option diagnostics true` command.hfull (hinput.prodMk hindex)) have hgap : b / 2 < |denominators L κ 0 xi.1 xi.2| := (half_lt_self hb).trans_le (hbound xi hxi) (deterministic) timeout at `«tactic execution»`, maximum number of heartbeats (10000000) has been reached Note: Use `set_option maxHeartbeats <num>` to set the limit. Hint: Additional diagnostic information may be available using the `set_option diagnostics true` command.exact (heval.eventually (Ioi_mem_nhds hgap)).mono fun _ hz ↦ hz.le obtain ⟨δ, hδ, hδbound⟩ := FiniteTaylorJet.compactFiberNormRadius (X := ℝ) (P := fun t xi ↦ b / 2 ≤ |denominators L κ t xi.1 xi.2|) hcompact hnear refine ⟨δ / 2, half_pos hδ, b / 2, half_pos hb, ?_⟩ intro t ht x hx i have htNorm : ‖t‖ ≤ δ / 2 := by rw [Real.norm_eq_abs, abs_le] exact ht exact hδbound (x, i) ⟨hx, Set.mem_univ i⟩ t (htNorm.trans_lt (half_lt_self hδ))

The normalized two-stage map is jointly analytic at zero scale and a positive state.

theorem map_analyticAt_zero (L κ : ℝ) (x : ℝ × ℝ × ℝ) (hA : 0 < x.1) (hC : 0 < x.2.1) (hC₁ : 0 < (leadingStage (limitAngle 0) x).2.1) : AnalyticAt ℝ (Function.uncurry (map L κ)) (0, x) := by -- First stage, evaluated at its exact limiting state. have hτ : AnalyticAt ℝ (fun y : ℝ × (ℝ × ℝ × ℝ) ↦ y.1) (0, x) := analyticAt_fst have hX : AnalyticAt ℝ (fun y : ℝ × (ℝ × ℝ × ℝ) ↦ y.2) (0, x) := analyticAt_snd have hfb := feedbackAlong_analyticAt L κ hτ hX hA.ne' have hSlope := slopeAlong_analyticAt hX hC.ne' have hX₁ := cycleStageAlong_analyticAt L κ 0 hτ hX hfb hSlope rfl have hC₁₀ : (cycleStage L κ 0 0 x x).2.1 ≠ 0 := by rw [cycleStage_zero] exact hC₁.ne' have hSlope₁ := slopeAlong_analyticAt hX₁ hC₁₀ have hX₂ := cycleStageAlong_analyticAt L κ 1 hτ hX₁ hfb hSlope₁ rfl have hrescale := rescaleAlong_analyticAt hτ hX₂ rfl -- Compose the two stages before the source's final change of scale. change AnalyticAt ℝ (fun y : ℝ × (ℝ × ℝ × ℝ) ↦ map L κ y.1 y.2) (0, x) simp only [map_def, twoStep_def] exact hrescale

Every finite Taylor order of the normalized map is uniform on a fixed compact positive-state set.

theorem map_uniformJets (L κ : ℝ) (m : ℕ) (𝒦 : Set (ℝ × ℝ × ℝ)) (h𝒦 : IsCompact 𝒦) (hA : ∀ x ∈ 𝒦, 0 < x.1) (hC : ∀ x ∈ 𝒦, 0 < x.2.1) (hC₁ : ∀ x ∈ 𝒦, 0 < (leadingStage (limitAngle 0) x).2.1) : FiniteTaylorJet.IsUniformOn (fun x : ℝ × ℝ × ℝ ↦ fun t : ℝ ↦ map L κ t x) (fun x ↦ FiniteTaylorJet.ofFunction ℝ m (fun t ↦ map L κ t x) 0) 0 𝒦 := by -- Compact-fiber Taylor uniformity applies after swapping parameter and scale coordinates. refine FiniteTaylorJet.isUniformOn_of_contDiffAt m (fun x : ℝ × ℝ × ℝ ↦ fun t : ℝ ↦ map L κ t x) 0 𝒦 h𝒦 ?_ intro x hx have hm := map_analyticAt_zero L κ x (hA x hx) (hC x hx) (hC₁ x hx) have hswap : AnalyticAt ℝ (fun y : (ℝ × ℝ × ℝ) × ℝ ↦ (y.2, y.1)) (x, 0) := by exact (analyticAt_snd : AnalyticAt ℝ (fun y : (ℝ × ℝ × ℝ) × ℝ ↦ y.2) (x, 0)).prod (analyticAt_fst : AnalyticAt ℝ (fun y : (ℝ × ℝ × ℝ) × ℝ ↦ y.1) (x, 0)) have hswap0 : (fun y : (ℝ × ℝ × ℝ) × ℝ ↦ (y.2, y.1)) (x, 0) = (0, x) := rfl have hcomp := hm.comp_of_eq hswap hswap0 have htarget : AnalyticAt ℝ (Function.uncurry (fun x : ℝ × ℝ × ℝ ↦ fun t : ℝ ↦ map L κ t x)) (x, 0) := by change AnalyticAt ℝ (fun y : (ℝ × ℝ × ℝ) × ℝ ↦ map L κ y.2 y.1) (x, 0) exact hcomp exact htarget.contDiffAt

Each feedback-controlled normalized stage is analytic in scale at zero.

theorem cycleStage_analyticAt_zero (L κ : ℝ) (i : Fin 2) (x₀ x : ℝ × ℝ × ℝ) : AnalyticAt ℝ (fun t ↦ cycleStage L κ i t x₀ x) 0 := by -- Fixed-state quotients are constants; cubic feedback is an analytic scale path. have hfb : AnalyticAt ℝ (fun t : ℝ ↦ feedback L κ t x₀) 0 := by simp only [feedback_def] fun_prop exact cycleStageAlong_analyticAt L κ i analyticAt_id analyticAt_const hfb analyticAt_const rfl

The cubic root differs from one by order three in scale.

private lemma unitRoot_sub_one_isBigO : (fun t : ℝ ↦ unitRoot t - 1) =O[𝓝 0] (fun t ↦ t ^ 3) := by -- Compose the regular exp-log branch with the cubic scale variable. let f : ℝ → ℝ := fun u ↦ Real.exp (Real.log (1 + u) / 3) have ha : AnalyticAt ℝ f 0 := by have hpos : 0 < (1 : ℝ) + 0 := by norm_num dsimp only [f] fun_prop have hcubic : Tendsto (fun t : ℝ ↦ t ^ 3) (𝓝 0) (𝓝 0) := by have hcont : ContinuousAt (fun t : ℝ ↦ t ^ 3) 0 := by fun_prop simpa [ContinuousAt] using hcont have hb := ha.differentiableAt.isBigO_sub.comp_tendsto hcubic simpa [f, unitRoot_def, Function.comp_def] using hb

The regular angle factor is analytic at zero.

private lemma stageAngleFactorAnalytic (i : Fin 2) : AnalyticAt ℝ (stageAngleFactor i) 0 := by -- Its only denominator has the value three at zero. have hd := (unitRootAnalytic.pow 2).add unitRootAnalytic |>.add (analyticAt_const : AnalyticAt ℝ (fun _ : ℝ ↦ (1 : ℝ)) 0) have hd0 : unitRoot 0 ^ 2 + unitRoot 0 + 1 ≠ 0 := by norm_num [unitRoot_def] rw [funext (stageAngleFactor_def i)] simp only [angleFactor_def] exact analyticAt_const.mul (analyticAt_const.div hd hd0)

The canceled angle factor has its limiting coefficient at zero.

private lemma stageAngleFactor_zero (i : Fin 2) : stageAngleFactor i 0 = limitAngle i := by -- At zero the sinc multiplier is one. have h := sinFactor_zero i simpa [sinFactor_def, stageAngle_def] using h

The canceled angle factor differs from its limiting value only at cubic order.

private lemma stageAngleFactor_sub_limit_isBigO (i : Fin 2) : (fun t : ℝ ↦ stageAngleFactor i t - limitAngle i) =O[𝓝 0] (fun t ↦ t ^ 3) := by -- Exact cubic cancellation leaves an analytic rational function of the unit root. let f : ℝ → ℝ := fun u ↦ stageWeight i * (9 * DFP.Polar.baseAngle / (u ^ 2 + u + 1)) have hd0 : (1 : ℝ) ^ 2 + 1 + 1 ≠ 0 := by norm_num have ha : AnalyticAt ℝ f 1 := by dsimp only [f] fun_prop have hroot : Tendsto unitRoot (𝓝 0) (𝓝 1) := by simpa [ContinuousAt, unitRoot_def] using unitRootAnalytic.continuousAt have hb := (ha.differentiableAt.isBigO_sub.comp_tendsto hroot).trans unitRoot_sub_one_isBigO have hf0 : f 1 = limitAngle i := by have h := stageAngleFactor_zero i simpa [f, stageAngleFactor_def, angleFactor_def, unitRoot_def] using h simp only [hf0, Function.comp_def] at hb simpa only [f, stageAngleFactor_def, angleFactor_def] using hb

The stage angle is of quadratic scale order.

private lemma stageAngle_isBigO (i : Fin 2) : (stageAngle i) =O[𝓝 0] (fun t : ℝ ↦ t ^ 2) := by -- The analytic canceled coefficient is bounded at zero. have hb := (stageAngleFactorAnalytic i).continuousAt.isBigO_one ℝ have hm := (Asymptotics.isBigO_refl (fun t : ℝ ↦ t ^ 2) (𝓝 0)).mul hb rw [funext (stageAngle_def i)] simpa only [mul_one] using hm

The sinc defect factors through the square of its argument with an analytic cofactor.

private lemma sinc_sub_one_factor : ∃ q : ℝ → ℝ, AnalyticAt ℝ q 0 ∧ ∀ u : ℝ, Real.sinc u - 1 = u ^ 2 * q u := by -- The sine Taylor formula has zero constant and quadratic coefficients. obtain ⟨q, hq, hformula⟩ := (Real.analyticAt_sin : AnalyticAt ℝ Real.sin 0).exists_eq_sum_add_pow_mul 3 have hsin : ∀ u : ℝ, Real.sin u = u + u ^ 3 * q u := by intro u simpa [Finset.sum_range_succ, iteratedDeriv_succ', iteratedDeriv_zero, Real.deriv_sin, Real.deriv_cos, smul_eq_mul] using hformula u refine ⟨q, hq, ?_⟩ intro u by_cases hu : u = 0 · subst u simp · rw [Real.sinc_of_ne_zero hu] rw [sub_eq_iff_eq_add] apply (div_eq_iff hu).mpr rw [hsin] ring

The cosine defect has an analytic quadratic factor.

private lemma cos_sub_one_factor : ∃ q : ℝ → ℝ, AnalyticAt ℝ q 0 ∧ ∀ u : ℝ, Real.cos u - 1 = u ^ 2 * q u := by -- Cosine has value one and zero linear coefficient at the origin. obtain ⟨q, hq, hformula⟩ := (Real.analyticAt_cos : AnalyticAt ℝ Real.cos 0).exists_eq_sum_add_pow_mul 2 refine ⟨q, hq, ?_⟩ intro u have h := hformula u have hcos : Real.cos u = 1 + u ^ 2 * q u := by simpa [Finset.sum_range_succ, iteratedDeriv_succ', iteratedDeriv_zero, Real.deriv_cos, smul_eq_mul] using h linarith

Quadratic analytic trigonometric defects become fourth-order scale errors.

private lemma quadraticFactorAlong_isBigO (i : Fin 2) (q : ℝ → ℝ) (hq : AnalyticAt ℝ q 0) : (fun t : ℝ ↦ stageAngle i t ^ 2 * q (stageAngle i t)) =O[𝓝 0] (fun t ↦ t ^ 4) := by -- The angle is O(t²), and the composed analytic cofactor stays bounded. have hangle0 : stageAngle i 0 = 0 := by simp [stageAngle_def] have hcomp := hq.comp_of_eq (stageAngle_analyticAt i) hangle0 have hb := ((stageAngle_isBigO i).pow 2).mul (hcomp.continuousAt.isBigO_one ℝ) simpa only [mul_one, ← pow_mul, Nat.reduceMul, Function.comp_def] using hb

The regularized sine coefficient has no scale terms of order one or two.

private lemma sinFactor_sub_limitAngle_isBigO (i : Fin 2) : (fun t : ℝ ↦ sinFactor i t - limitAngle i) =O[𝓝 0] (fun t ↦ t ^ 3) := by -- Separate cubic variation of the canceled angle factor from the fourth-order sinc defect. obtain ⟨q, hq, hfactor⟩ := sinc_sub_one_factor have hsinc4 : (fun t : ℝ ↦ Real.sinc (stageAngle i t) - 1) =O[𝓝 0] (fun t ↦ t ^ 4) := by simpa only [hfactor] using quadraticFactorAlong_isBigO i q hq have hfour : (fun t : ℝ ↦ t ^ 4) =O[𝓝 0] (fun t : ℝ ↦ t ^ 3) := (Asymptotics.isLittleO_pow_pow (Nat.lt_succ_self 3)).isBigO have hbounded : (fun t : ℝ ↦ Real.sinc (stageAngle i t)) =O[𝓝 0] (fun _ : ℝ ↦ (1 : ℝ)) := (Real.continuous_sinc.continuousAt.comp (stageAngle_analyticAt i).continuousAt).isBigO_one ℝ have hfirst := (stageAngleFactor_sub_limit_isBigO i).mul hbounded have hsecond := (hsinc4.trans hfour).const_mul_left (limitAngle i) have heq (t : ℝ) : sinFactor i t - limitAngle i = (stageAngleFactor i t - limitAngle i) * Real.sinc (stageAngle i t) + limitAngle i * (Real.sinc (stageAngle i t) - 1) := by rw [sinFactor_def] ring have hfirst' : (fun t : ℝ ↦ (stageAngleFactor i t - limitAngle i) * Real.sinc (stageAngle i t)) =O[𝓝 0] (fun t ↦ t ^ 3) := by simpa only [mul_one] using hfirst simpa only [← heq] using hfirst'.add hsecond

The normalized rotation has only cubic and higher scale perturbations of its limiting shear.

private lemma rotation_sub_zero_isBigO (i : Fin 2) : (fun t : ℝ ↦ fun r c : Fin 2 ↦ rotation i t r c - rotation i 0 r c) =O[𝓝 0] (fun t ↦ t ^ 3) := by -- The diagonal cosine and upper-right entry are O(t⁴); the lower-left error is O(t³). obtain ⟨q, hq, hfactor⟩ := cos_sub_one_factor have hcos4 : (fun t : ℝ ↦ Real.cos (stageAngle i t) - 1) =O[𝓝 0] (fun t ↦ t ^ 4) := by simpa only [hfactor] using quadraticFactorAlong_isBigO i q hq have hfour : (fun t : ℝ ↦ t ^ 4) =O[𝓝 0] (fun t : ℝ ↦ t ^ 3) := (Asymptotics.isLittleO_pow_pow (Nat.lt_succ_self 3)).isBigO have hupper := (Asymptotics.isBigO_refl (fun t : ℝ ↦ t ^ 4) (𝓝 0)).mul ((sinFactor_analyticAt i).continuousAt.isBigO_one ℝ) have hsin := sinFactor_sub_limitAngle_isBigO i have hupper' : (fun t : ℝ ↦ t ^ 4 * sinFactor i t) =O[𝓝 0] (fun t ↦ t ^ 4) := by simpa only [mul_one] using hupper have hangle0 : stageAngle i 0 = 0 := by simp [stageAngle_def] refine Asymptotics.isBigO_pi.mpr ?_ intro r refine Asymptotics.isBigO_pi.mpr ?_ intro c fin_cases r all_goals fin_cases c · simpa [rotation_def, hangle0, sinFactor_zero] using hcos4.trans hfour · simpa [rotation_def, hangle0, sinFactor_zero] using hupper'.trans hfour · simpa [rotation_def, hangle0, sinFactor_zero, neg_sub, sub_eq_add_neg, add_comm] using hsin.neg_left · simpa [rotation_def, hangle0, sinFactor_zero] using hcos4.trans hfour

Fixed-boundary feedback is a cubic perturbation of one.

private lemma feedback_sub_one_isBigO (L κ : ℝ) (x₀ : ℝ × ℝ × ℝ) : (fun t : ℝ ↦ feedback L κ t x₀ - 1) =O[𝓝 0] (fun t ↦ t ^ 3) := by -- The cubic exponent is the sole scale input to an analytic exponential. let f : ℝ → ℝ := fun u ↦ Real.exp (-L * (κ / x₀.1) ^ 2 * u) have ha : AnalyticAt ℝ f 0 := by dsimp only [f] fun_prop have hcubic : Tendsto (fun t : ℝ ↦ t ^ 3) (𝓝 0) (𝓝 0) := by have hc : ContinuousAt (fun t : ℝ ↦ t ^ 3) 0 := by fun_prop simpa [ContinuousAt] using hc have hb := ha.differentiableAt.isBigO_sub.comp_tendsto hcubic simpa [f, feedback_def, Function.comp_def] using hb

Cubic multiplier perturbations produce a cubic error in the canceled gradient change.

private lemma stepW_sub_zero_isBigO (i : Fin 2) (x : ℝ × ℝ × ℝ) (ζ η : ℝ → ℝ) (hζ : ContinuousAt ζ 0) (hζ0 : ζ 0 = 1) (hη0 : η 0 = 1) (hζ3 : (fun t ↦ ζ t - 1) =O[𝓝 0] (fun t : ℝ ↦ t ^ 3)) (hη3 : (fun t ↦ η t - 1) =O[𝓝 0] (fun t : ℝ ↦ t ^ 3)) : (fun t : ℝ ↦ fun j : Fin 2 ↦ stepW i t x (ζ t) (η t) j - stepW i 0 x (ζ 0) (η 0) j) =O[𝓝 0] (fun t ↦ t ^ 3) := by -- The multiplier difference is cubic and cotFactor is bounded after regularization. have hdiff : (fun t ↦ η t - ζ t) =O[𝓝 0] (fun t : ℝ ↦ t ^ 3) := by simpa only [sub_sub_sub_cancel_right] using hη3.sub hζ3 have hcot := (cotFactor_analyticAt i).continuousAt.isBigO_one ℝ have hfirst : (fun t ↦ (η t - ζ t) * cotFactor i t) =O[𝓝 0] (fun t : ℝ ↦ t ^ 3) := by simpa only [mul_one] using hdiff.mul hcot have hfour : (fun t : ℝ ↦ t ^ 4) =O[𝓝 0] (fun t : ℝ ↦ t ^ 3) := (Asymptotics.isLittleO_pow_pow (Nat.lt_succ_self 3)).isBigO have hζbounded := hζ.isBigO_one ℝ have hsecond : (fun t ↦ ζ t * t ^ 4 * slope x) =O[𝓝 0] (fun t : ℝ ↦ t ^ 3) := by have hb : (fun t : ℝ ↦ ζ t * t ^ 4) =O[𝓝 0] (fun t ↦ t ^ 4) := by simpa only [one_mul] using hζbounded.mul (Asymptotics.isBigO_refl (fun t : ℝ ↦ t ^ 4) (𝓝 0)) simpa only [mul_comm (slope x)] using (hb.const_mul_left (slope x)).trans hfour refine Asymptotics.isBigO_pi.mpr ?_ intro j fin_cases j · simpa [stepW_def, hζ0, hη0] using hfirst.sub hsecond · simpa [stepW_def, hζ0, hη0] using hη3

The feedback-controlled stage differs from its limiting stage by a cubic scale error.

private lemma cycleStage_sub_leading_isBigO (L κ : ℝ) (i : Fin 2) (x₀ x : ℝ × ℝ × ℝ) : (fun t ↦ cycleStage L κ i t x₀ x - leadingStage (limitAngle i) x) =O[𝓝 0] (fun t : ℝ ↦ t ^ 3) := by -- Both phases have unit limits and cubic feedback perturbations. let ζ : ℝ → ℝ := fun t ↦ (![1, feedback L κ t x₀] : Fin 2 → ℝ) i let η : ℝ → ℝ := fun t ↦ (![feedback L κ t x₀, 1] : Fin 2 → ℝ) i have hfb : ContinuousAt (fun t : ℝ ↦ feedback L κ t x₀) 0 := by simp only [feedback_def] fun_prop have hζ : ContinuousAt ζ 0 := by fin_cases i · simpa [ζ] using (continuousAt_const : ContinuousAt (fun _ : ℝ ↦ (1 : ℝ)) 0) · simpa [ζ] using hfb have hζ0 : ζ 0 = 1 := by dsimp only [ζ] have hf0 : feedback L κ 0 x₀ = 1 := by simp [feedback_def] rw [hf0] fin_cases i all_goals rfl have hη0 : η 0 = 1 := by dsimp only [η] have hf0 : feedback L κ 0 x₀ = 1 := by simp [feedback_def] rw [hf0] fin_cases i all_goals rfl have hζ3 : (fun t ↦ ζ t - 1) =O[𝓝 0] (fun t : ℝ ↦ t ^ 3) := by fin_cases i · simpa [ζ] using Asymptotics.isBigO_zero (fun t : ℝ ↦ t ^ 3) (𝓝 0) · simpa [ζ] using feedback_sub_one_isBigO L κ x₀ have hη3 : (fun t ↦ η t - 1) =O[𝓝 0] (fun t : ℝ ↦ t ^ 3) := by fin_cases i · simpa [η] using feedback_sub_one_isBigO L κ x₀ · simpa [η] using Asymptotics.isBigO_zero (fun t : ℝ ↦ t ^ 3) (𝓝 0) have hW3 := stepW_sub_zero_isBigO i x ζ η hζ hζ0 hη0 hζ3 hη3 have hP3 := rotation_sub_zero_isBigO i let d : ℝ → (Fin 2 → ℝ) × (Fin 2 → Fin 2 → ℝ) := fun t ↦ (stepW i t x (ζ t) (η t), fun r c ↦ rotation i t r c) have hd3 : (fun t : ℝ ↦ d t - d 0) =O[𝓝 0] (fun t ↦ t ^ 3) := by have hdataEq : (fun t : ℝ ↦ d t - d 0) = (fun t : ℝ ↦ ((fun j : Fin 2 ↦ stepW i t x (ζ t) (η t) j - stepW i 0 x (ζ 0) (η 0) j), (fun r c : Fin 2 ↦ rotation i t r c - rotation i 0 r c))) := by funext t rfl rw [hdataEq] exact hW3.prod_left hP3 -- The cubic factor perturbation tends to zero; the bracket is regular at that base. have hcubic : Tendsto (fun t : ℝ ↦ t ^ 3) (𝓝 0) (𝓝 0) := by have hc : ContinuousAt (fun t : ℝ ↦ t ^ 3) 0 := by fun_prop simpa [ContinuousAt] using hc have hd : Tendsto d (𝓝 0) (𝓝 (d 0)) := tendsto_sub_nhds_zero_iff.mp (hd3.trans_tendsto hcubic) have hσ0 : stepV x ⬝ᵥ (d 0).1 ≠ 0 := by simp [d, hζ0, hη0, stepV_def, stepW_def] have hregular := (stageFromFactors_analyticAt x (d 0) hσ0).differentiableAt.isBigO_sub have hcore := (hregular.comp_tendsto hd).trans hd3 have heq (t : ℝ) : cycleStage L κ i t x₀ x = stageFromFactors x (d t) := by simp only [cycleStage_apply, d, ζ, η] rw [stage_eq_fromFactors] have hstage : (fun t : ℝ ↦ cycleStage L κ i t x₀ x - cycleStage L κ i 0 x₀ x) =O[𝓝 0] (fun t ↦ t ^ 3) := by simpa only [Function.comp_def, ← heq] using hcore simpa only [cycleStage_zero] using hstage

The quadratic Taylor jet of a normalized stage is its constant limiting jet.

private lemma cycleStage_twoJet_eq_constant (L κ : ℝ) (i : Fin 2) (x₀ x : ℝ × ℝ × ℝ) : FiniteTaylorJet.ofFunction ℝ 2 (fun t ↦ cycleStage L κ i t x₀ x) 0 = FiniteTaylorJet.ofFunction ℝ 2 (fun _ : ℝ ↦ leadingStage (limitAngle i) x) 0 := by -- An O(t³) perturbation is invisible to the order-two Peano jet. have hf : ContDiffAt ℝ 2 (fun t ↦ cycleStage L κ i t x₀ x) 0 := (cycleStage_analyticAt_zero L κ i x₀ x).contDiffAt have hg : ContDiffAt ℝ 2 (fun _ : ℝ ↦ leadingStage (limitAngle i) x) 0 := contDiffAt_const have herr : (fun t : ℝ ↦ cycleStage L κ i (0 + t) x₀ x - (fun _ : ℝ ↦ leadingStage (limitAngle i) x) (0 + t)) =O[𝓝 0] (fun t ↦ t ^ (2 + 1)) := by simpa only [zero_add, Nat.reduceAdd] using cycleStage_sub_leading_isBigO L κ i x₀ x exact FiniteTaylorJet.ofFunction_eq_of_sub_isBigO_succ hf hg herr

Every positive-order scale derivative through order two of a normalized stage vanishes at zero.

private lemma cycleStage_iteratedFDeriv_eq_zero (L κ : ℝ) (i : Fin 2) (x₀ x : ℝ × ℝ × ℝ) (n : Fin 3) (hn : (n : ℕ) ≠ 0) : iteratedFDeriv ℝ (n : ℕ) (fun t ↦ cycleStage L κ i t x₀ x) 0 = 0 := by -- The jet comparison API cancels the nonzero factorial in each coefficient. have hderiv := (FiniteTaylorJet.ofFunction_eq_iff_iteratedDeriv_eq 2 (fun t ↦ cycleStage L κ i t x₀ x) (fun _ : ℝ ↦ leadingStage (limitAngle i) x) 0 0).mp (cycleStage_twoJet_eq_constant L κ i x₀ x) n -- A one-variable multilinear derivative is determined by its value at ones. apply ContinuousMultilinearMap.ext_ring simpa only [iteratedDeriv_eq_iteratedFDeriv, iteratedFDeriv_const_of_ne hn, Pi.zero_apply, zero_apply] using hderiv

A normalized stage has zero first scale derivative at zero.

theorem cycleStage_firstIteratedFDeriv_eq_zero (L κ : ℝ) (i : Fin 2) (x₀ x : ℝ × ℝ × ℝ) : iteratedFDeriv ℝ 1 (fun t ↦ cycleStage L κ i t x₀ x) 0 = 0 := by -- Select the linear derivative of the constant quadratic jet. exact cycleStage_iteratedFDeriv_eq_zero L κ i x₀ x 1 one_ne_zero

A normalized stage has zero second scale derivative at zero.

theorem cycleStage_secondIteratedFDeriv_eq_zero (L κ : ℝ) (i : Fin 2) (x₀ x : ℝ × ℝ × ℝ) : iteratedFDeriv ℝ 2 (fun t ↦ cycleStage L κ i t x₀ x) 0 = 0 := by -- Select the quadratic derivative of the same constant jet. exact cycleStage_iteratedFDeriv_eq_zero L κ i x₀ x 2 two_ne_zero
end DFP.Polar.Normalized