Bell first correction and robust error functions
DefinitionAnalyticCombinatorics_Ch8_SaddlePoint_BellSecondOrderanalytic-combinatoricslean4saddle-point
The third and fourth logarithmic saddle functions are the stated Touchard polynomials times exp(r). The scale c2=|b4|/b²+b3²/b³, the actual complex saddle denominator, and first correction define explicit series/number error functions. Local Taylor remainders, scaled cubic/quartic terms, Gaussian polynomial dominators and tail bounds supply a second-order record.
Definition code
import Definitions.Def_AnalyticCombinatorics_Ch8_SaddlePoint_BellHAdmissible
import Definitions.Def_AnalyticCombinatorics_Ch8_SaddlePoint_SecondOrderHAdmissible
import Mathlib.Analysis.Real.Pi.Bounds
import Mathlib
import Mathlib.Analysis.SpecialFunctions.Stirling
import Mathlib.Probability.Distributions.Gaussian.Real
import Mathlib.RingTheory.PowerSeries.Exp
import Mathlib.Analysis.SpecialFunctions.ExpDeriv
import Mathlib.Analysis.SpecialFunctions.Exponential
import Mathlib.Analysis.Analytic.Composition
import Mathlib.RingTheory.PowerSeries.Substitution
import Mathlib.Analysis.Analytic.Basic
import Mathlib.Analysis.Analytic.OfScalars
import Mathlib.Analysis.Analytic.RadiusLiminf
import Mathlib.RingTheory.PowerSeries.Basic
import Mathlib.Algebra.Algebra.Rat
import Mathlib.Data.Complex.Basic
import Mathlib.RingTheory.PowerSeries.Inverse
import Mathlib.RingTheory.PowerSeries.NoZeroDivisors
import Mathlib.RingTheory.PowerSeries.PiTopology
import Mathlib.Combinatorics.Enumerative.Composition
import Mathlib.Data.Nat.Fib.Basic
import Mathlib.Order.Partition.Finpartition
import Mathlib.RingTheory.PowerSeries.Derivative
import Mathlib.Data.Fintype.EquivFin
import Mathlib.RingTheory.PowerSeries.WellKnown
import Mathlib.GroupTheory.Perm.Basic
import Mathlib.Data.Fintype.Perm
import Mathlib.Data.Fintype.Card
import Mathlib.Algebra.BigOperators.NatAntidiagonal
import Mathlib.Data.Fintype.Powerset
set_option autoImplicit true
set_option relaxedAutoImplicit false
set_option maxSynthPendingDepth 3
/-!
# Second-order saddle data for Bell numbers
This file adds the second-order correction layer for
`exp (exp z - 1)`. The logarithmic saddle derivatives are
`b₃(r) = (r + 3r² + r³)e^r` and
`b₄(r) = (r + 7r² + 6r³ + r⁴)e^r`.
-/
open Complex Filter Asymptotics MeasureTheory
open scoped Topology Real NNReal ENNReal Interval PowerSeries
set_option maxHeartbeats 800000
noncomputable section
/-- Third logarithmic saddle derivative for Bell numbers. -/
def bellSaddleB3Real (r : ℝ) : ℝ :=
(r + 3 * r ^ 2 + r ^ 3) * Real.exp r
/-- Fourth logarithmic saddle derivative for Bell numbers. -/
def bellSaddleB4Real (r : ℝ) : ℝ :=
(r + 7 * r ^ 2 + 6 * r ^ 3 + r ^ 4) * Real.exp r
/-- Robust second-order correction scale for Bell numbers. -/
def bellSecondCorrScale (r : ℝ) : ℝ :=
saddleCorrectionScale bellSaddleBReal bellSaddleB3Real bellSaddleB4Real r
/-- Complex saddle scale used by the Bell second-order statement. -/
def bellSecondOrderSaddleScale (n : ℕ) : ℂ :=
saddleScale (fun z : ℂ => Complex.exp (Complex.exp z - 1))
bellSaddleRadius (fun n => bellSaddleBReal (bellSaddleRadius n)) n
/-- The explicit Bell second-order correction at the saddle radius. -/
def bellSecondCorrectionAtSaddle (n : ℕ) : ℝ :=
saddleCorrection bellSaddleBReal bellSaddleB3Real bellSaddleB4Real
(bellSaddleRadius n)
/-- Error term for the formal Bell EGF coefficient. -/
def bellSecondOrderSeriesError (n : ℕ) : ℂ :=
bellSeries.coeff n / bellSecondOrderSaddleScale n -
(1 + (bellSecondCorrectionAtSaddle n : ℂ))
/-- Error term for `B_n / n!`. -/
def bellSecondOrderNumberError (n : ℕ) : ℂ :=
(((AnalyticCombinatorics.Ch1.CombClass.posInt.set.counts n : ℝ) /
n.factorial : ℝ) : ℂ) /
bellSecondOrderSaddleScale n -
(1 + (bellSecondCorrectionAtSaddle n : ℂ))
lemma bellSecondCorrScale_eq_formula {r : ℝ} (hr : 0 < r) :
bellSecondCorrScale r =
Real.exp (-r) *
(2 * r ^ 4 + 13 * r ^ 3 + 24 * r ^ 2 + 14 * r + 2) /
(r * (r + 1) ^ 3) := by
unfold bellSecondCorrScale saddleCorrectionScale bellSaddleBReal bellSaddleB3Real
bellSaddleB4Real
have hb4_nonneg : 0 ≤ (r + 7 * r ^ 2 + 6 * r ^ 3 + r ^ 4) * Real.exp r := by
positivity
rw [abs_of_nonneg hb4_nonneg]
rw [Real.exp_neg]
have her : Real.exp r ≠ 0 := (Real.exp_pos r).ne'
have hrn : r ≠ 0 := hr.ne'
have hr1 : r + 1 ≠ 0 := by positivity
field_simp [hrn, her, hr1]
ring
lemma bellSecondCorrScale_pos_eventually :
∀ᶠ r : ℝ in atTop, 0 < bellSecondCorrScale r := by
filter_upwards [eventually_ge_atTop (1 : ℝ)] with r hr
unfold bellSecondCorrScale saddleCorrectionScale bellSaddleBReal bellSaddleB3Real
bellSaddleB4Real
have hbpos : 0 < r * (r + 1) * Real.exp r := by positivity
have hb4pos : 0 < (r + 7 * r ^ 2 + 6 * r ^ 3 + r ^ 4) * Real.exp r := by
positivity
have hfirst :
0 < |(r + 7 * r ^ 2 + 6 * r ^ 3 + r ^ 4) * Real.exp r| /
(r * (r + 1) * Real.exp r) ^ 2 :=
div_pos (abs_pos.mpr hb4pos.ne') (sq_pos_of_pos hbpos)
have hsecond :
0 ≤ ((r + 3 * r ^ 2 + r ^ 3) * Real.exp r) ^ 2 /
(r * (r + 1) * Real.exp r) ^ 3 := by
positivity
exact add_pos_of_pos_of_nonneg hfirst hsecond
lemma bellSecondCorrScale_nonneg_eventually :
∀ᶠ r : ℝ in atTop, 0 ≤ bellSecondCorrScale r :=
bellSecondCorrScale_pos_eventually.mono fun _ h => h.le
lemma bellSecondCorrScale_lower_exp_eventually :
∀ᶠ r : ℝ in atTop,
(1 / 8 : ℝ) * Real.exp (-r) ≤ bellSecondCorrScale r := by
filter_upwards [eventually_ge_atTop (1 : ℝ)] with r hr
have hrpos : 0 < r := lt_of_lt_of_le zero_lt_one hr
rw [bellSecondCorrScale_eq_formula hrpos]
have hdenpos : 0 < r * (r + 1) ^ 3 := by positivity
have hpoly :
(1 / 8 : ℝ) ≤
(2 * r ^ 4 + 13 * r ^ 3 + 24 * r ^ 2 + 14 * r + 2) /
(r * (r + 1) ^ 3) := by
field_simp [hdenpos.ne']
ring_nf
nlinarith [hr, sq_nonneg r, show 0 ≤ r ^ 3 by positivity,
show 0 ≤ r ^ 4 by positivity]
calc
(1 / 8 : ℝ) * Real.exp (-r)
≤ ((2 * r ^ 4 + 13 * r ^ 3 + 24 * r ^ 2 + 14 * r + 2) /
(r * (r + 1) ^ 3)) * Real.exp (-r) :=
mul_le_mul_of_nonneg_right hpoly (Real.exp_pos _).le
_ = Real.exp (-r) *
(2 * r ^ 4 + 13 * r ^ 3 + 24 * r ^ 2 + 14 * r + 2) /
(r * (r + 1) ^ 3) := by ring
lemma bellSecondCorrScale_le_exp_eventually :
∀ᶠ r : ℝ in atTop, bellSecondCorrScale r ≤ 100 * Real.exp (-r) := by
filter_upwards [eventually_ge_atTop (1 : ℝ)] with r hr
have hrpos : 0 < r := lt_of_lt_of_le zero_lt_one hr
rw [bellSecondCorrScale_eq_formula hrpos]
have hdenpos : 0 < r * (r + 1) ^ 3 := by positivity
have hpoly :
(2 * r ^ 4 + 13 * r ^ 3 + 24 * r ^ 2 + 14 * r + 2) /
(r * (r + 1) ^ 3) ≤ 100 := by
field_simp [hdenpos.ne']
ring_nf
nlinarith [hr, sq_nonneg r, show 0 ≤ r ^ 3 by positivity,
show 0 ≤ r ^ 4 by positivity]
calc
Real.exp (-r) *
(2 * r ^ 4 + 13 * r ^ 3 + 24 * r ^ 2 + 14 * r + 2) /
(r * (r + 1) ^ 3)
= Real.exp (-r) *
((2 * r ^ 4 + 13 * r ^ 3 + 24 * r ^ 2 + 14 * r + 2) /
(r * (r + 1) ^ 3)) := by ring
_ ≤ Real.exp (-r) * 100 :=
mul_le_mul_of_nonneg_left hpoly (Real.exp_pos _).le
_ = 100 * Real.exp (-r) := by ring
lemma bellSecondCorrScale_tendsto_zero :
Tendsto bellSecondCorrScale atTop (𝓝 0) := by
have hneg : Tendsto (fun r : ℝ => (-1 : ℝ) * r) atTop atBot :=
Tendsto.const_mul_atTop_of_neg (by norm_num : (-1 : ℝ) < 0) tendsto_id
have hexp : Tendsto (fun r : ℝ => Real.exp ((-1 : ℝ) * r)) atTop (𝓝 0) :=
Real.tendsto_exp_atBot.comp hneg
have hupper : Tendsto (fun r : ℝ => 100 * Real.exp (-r)) atTop (𝓝 0) := by
have hmul :
Tendsto (fun r : ℝ => 100 * Real.exp ((-1 : ℝ) * r)) atTop (𝓝 0) := by
simpa only [mul_zero] using hexp.const_mul (100 : ℝ)
refine Tendsto.congr' ?_ hmul
exact Eventually.of_forall fun r => by
ring_nf
exact squeeze_zero' bellSecondCorrScale_nonneg_eventually
bellSecondCorrScale_le_exp_eventually hupper
/-- A stronger Bell tail envelope, small enough for the second-order scale. -/
def bellSecondTailE (r : ℝ) : ℝ :=
Real.exp (-((1 / Real.pi ^ 2 : ℝ) * r * Real.exp ((1 / 5 : ℝ) * r)))
lemma bellSecondTailE_nonneg_eventually :
∀ᶠ r : ℝ in atTop, 0 ≤ bellSecondTailE r :=
Eventually.of_forall fun r => by
unfold bellSecondTailE
positivity
lemma bell_second_tail_bound_eventually :
∀ᶠ r : ℝ in atTop, ∀ n : ℕ, ∀ θ : ℝ,
bellSaddleDeltaReal r ≤ |θ| → |θ| ≤ Real.pi →
‖saddleGAt (fun z : ℂ => Complex.exp (Complex.exp z - 1)) r n θ‖ ≤
bellSecondTailE r := by
set_option backward.isDefEq.respectTransparency false in
let c : ℝ := 2 / Real.pi ^ 2
have hcpos : 0 < c := by
dsimp [c]
positivity
have hy_small :
∀ᶠ r : ℝ in atTop,
c * (r * (bellSaddleDeltaReal r) ^ 2) ≤ 1 / 2 := by
simpa [c] using bell_tail_y_tendsto_zero.eventually_le_const
(by norm_num : (0 : ℝ) < 1 / 2)
filter_upwards [eventually_ge_atTop (1 : ℝ), hy_small] with
r hr hy_small_r n θ hδθ hθπ
have hr0 : 0 ≤ r := le_trans zero_le_one hr
have hδpos : 0 ≤ bellSaddleDeltaReal r := by
unfold bellSaddleDeltaReal
positivity
have hθsq_ge : (bellSaddleDeltaReal r) ^ 2 ≤ θ ^ 2 := by
calc
(bellSaddleDeltaReal r) ^ 2 ≤ |θ| ^ 2 :=
pow_le_pow_left₀ hδpos hδθ 2
_ = θ ^ 2 := by rw [sq_abs]
let y : ℝ := c * (r * (bellSaddleDeltaReal r) ^ 2)
have hy0 : 0 ≤ y := by
dsimp [y, c]
positivity
have hyhalf : y ≤ 1 / 2 := by
simpa [y, c] using hy_small_r
have hone_sub : y / 2 ≤ 1 - Real.exp (-y) :=
bell_one_sub_exp_neg_ge_half hy0 hyhalf
have hcos := Real.cos_le_one_sub_mul_cos_sq hθπ
have hrcos_le : r * Real.cos θ ≤ r - c * (r * (bellSaddleDeltaReal r) ^ 2) := by
have hcos' : Real.cos θ ≤ 1 - c * θ ^ 2 := by
dsimp [c] at hcos ⊢
linarith
have hmul := mul_le_mul_of_nonneg_left hcos' hr0
have hsquare : c * (r * (bellSaddleDeltaReal r) ^ 2) ≤ c * (r * θ ^ 2) := by
exact mul_le_mul_of_nonneg_left
(mul_le_mul_of_nonneg_left hθsq_ge hr0) hcpos.le
calc
r * Real.cos θ ≤ r * (1 - c * θ ^ 2) := hmul
_ = r - c * (r * θ ^ 2) := by ring_nf
_ ≤ r - c * (r * (bellSaddleDeltaReal r) ^ 2) := by linarith
have hexp_rcos_le : Real.exp (r * Real.cos θ) ≤ Real.exp r * Real.exp (-y) := by
have harg : r * Real.cos θ ≤ r - y := by
simpa [y] using hrcos_le
calc
Real.exp (r * Real.cos θ) ≤ Real.exp (r - y) := Real.exp_le_exp.mpr harg
_ = Real.exp r * Real.exp (-y) := by
rw [show r - y = r + -y by ring_nf, Real.exp_add]
have hdiff :
Real.exp (r * Real.cos θ) - Real.exp r ≤
-Real.exp r * (1 - Real.exp (-y)) := by
calc
Real.exp (r * Real.cos θ) - Real.exp r
≤ Real.exp r * Real.exp (-y) - Real.exp r :=
sub_le_sub_right hexp_rcos_le _
_ = -Real.exp r * (1 - Real.exp (-y)) := by ring_nf
have hstrong :
Real.exp (r * Real.cos θ) - Real.exp r ≤
-((1 / Real.pi ^ 2) * r * Real.exp ((1 / 5 : ℝ) * r)) := by
have hmul : Real.exp r * (y / 2) ≤ Real.exp r * (1 - Real.exp (-y)) :=
mul_le_mul_of_nonneg_left hone_sub (Real.exp_pos r).le
have hneg :
-Real.exp r * (1 - Real.exp (-y)) ≤ -Real.exp r * (y / 2) := by
nlinarith
calc
Real.exp (r * Real.cos θ) - Real.exp r
≤ -Real.exp r * (1 - Real.exp (-y)) := hdiff
_ ≤ -Real.exp r * (y / 2) := hneg
_ = -((1 / Real.pi ^ 2) * r * Real.exp ((1 / 5 : ℝ) * r)) := by
dsimp [y, c, bellSaddleDeltaReal]
rw [show (Real.exp (-(2 / 5 : ℝ) * r)) ^ 2 =
Real.exp (2 * (-(2 / 5 : ℝ) * r)) by
rw [← Real.exp_nat_mul]
norm_num]
calc
-Real.exp r * (2 / Real.pi ^ 2 * (r * Real.exp (2 * (-(2 / 5 : ℝ) * r))) / 2)
= -(1 / Real.pi ^ 2 * r *
(Real.exp r * Real.exp (2 * (-(2 / 5 : ℝ) * r)))) := by ring_nf
_ = -(1 / Real.pi ^ 2 * r *
Real.exp (r + 2 * (-(2 / 5 : ℝ) * r))) := by
rw [← Real.exp_add]
_ = -((1 / Real.pi ^ 2) * r * Real.exp ((1 / 5 : ℝ) * r)) := by
ring_nf
calc
‖saddleGAt (fun z : ℂ => Complex.exp (Complex.exp z - 1)) r n θ‖
≤ Real.exp (Real.exp (r * Real.cos θ) - Real.exp r) :=
bell_saddleGAt_norm_le_exp r n θ
_ ≤ Real.exp (-((1 / Real.pi ^ 2) * r * Real.exp ((1 / 5 : ℝ) * r))) :=
Real.exp_le_exp.mpr hstrong
_ = bellSecondTailE r := by rfl
lemma bellSecondTailE_decay :
Tendsto
(fun r : ℝ =>
Real.sqrt (2 * Real.pi * bellSaddleBReal r) * bellSecondTailE r /
bellSecondCorrScale r)
atTop (𝓝 0) := by
have hcoef_top :
Tendsto (fun r : ℝ => (1 / Real.pi ^ 2 : ℝ) *
Real.exp ((1 / 5 : ℝ) * r)) atTop atTop := by
exact Tendsto.const_mul_atTop (by positivity : (0 : ℝ) < 1 / Real.pi ^ 2)
bell_exp_one_fifth_tendsto_atTop
have hcoef_big :
∀ᶠ r : ℝ in atTop,
3 ≤ (1 / Real.pi ^ 2 : ℝ) * Real.exp ((1 / 5 : ℝ) * r) :=
hcoef_top.eventually_ge_atTop 3
have hbase :
Tendsto (fun r : ℝ => r ^ (1 : ℝ) * Real.exp (-(3 / 2 : ℝ) * r))
atTop (𝓝 0) :=
tendsto_rpow_mul_exp_neg_mul_atTop_nhds_zero (1 : ℝ) (3 / 2 : ℝ)
(by norm_num : (0 : ℝ) < 3 / 2)
have hupper :
Tendsto (fun r : ℝ =>
(8 * Real.sqrt (4 * Real.pi)) *
(r * Real.exp (-(3 / 2 : ℝ) * r))) atTop (𝓝 0) := by
have hcongr :
(fun r : ℝ => r * Real.exp (-(3 / 2 : ℝ) * r)) =ᶠ[atTop]
fun r : ℝ => r ^ (1 : ℝ) * Real.exp (-(3 / 2 : ℝ) * r) := by
filter_upwards [eventually_gt_atTop (0 : ℝ)] with r hr
rw [Real.rpow_one]
simpa using
((Tendsto.congr' hcongr.symm hbase).const_mul
(8 * Real.sqrt (4 * Real.pi)))
refine squeeze_zero' ?_ ?_ hupper
· filter_upwards [bellSecondCorrScale_pos_eventually] with r hcorr
exact div_nonneg
(mul_nonneg (Real.sqrt_nonneg _) (by unfold bellSecondTailE; positivity)) hcorr.le
· filter_upwards [eventually_ge_atTop (1 : ℝ), bellSecondCorrScale_lower_exp_eventually,
bellSecondCorrScale_pos_eventually, hcoef_big] with r hr hcorrLower hcorrPos hcoef
have hr0 : 0 ≤ r := le_trans zero_le_one hr
have hαpos : 0 < (1 / 8 : ℝ) * Real.exp (-r) := by positivity
have hE_nonneg : 0 ≤ bellSecondTailE r := by
unfold bellSecondTailE
positivity
have hE_le : bellSecondTailE r ≤ Real.exp (-(3 * r)) := by
unfold bellSecondTailE
have hmul :
3 * r ≤ (1 / Real.pi ^ 2 : ℝ) * r * Real.exp ((1 / 5 : ℝ) * r) := by
calc
3 * r ≤ ((1 / Real.pi ^ 2 : ℝ) * Real.exp ((1 / 5 : ℝ) * r)) * r :=
mul_le_mul_of_nonneg_right hcoef hr0
_ = (1 / Real.pi ^ 2 : ℝ) * r * Real.exp ((1 / 5 : ℝ) * r) := by
ring
exact Real.exp_le_exp.mpr (neg_le_neg hmul)
have hB_le : 2 * Real.pi * bellSaddleBReal r ≤
4 * Real.pi * (r ^ 2 * Real.exp r) := by
have hrp1 : r + 1 ≤ 2 * r := by linarith
have hmul : r * (r + 1) ≤ r * (2 * r) :=
mul_le_mul_of_nonneg_left hrp1 hr0
have hB : bellSaddleBReal r ≤ 2 * r ^ 2 * Real.exp r := by
unfold bellSaddleBReal
calc
r * (r + 1) * Real.exp r ≤ r * (2 * r) * Real.exp r :=
mul_le_mul_of_nonneg_right hmul (Real.exp_pos r).le
_ = 2 * r ^ 2 * Real.exp r := by ring_nf
calc
2 * Real.pi * bellSaddleBReal r
≤ 2 * Real.pi * (2 * r ^ 2 * Real.exp r) :=
mul_le_mul_of_nonneg_left hB (by positivity)
_ = 4 * Real.pi * (r ^ 2 * Real.exp r) := by ring_nf
have hsqrt :
Real.sqrt (2 * Real.pi * bellSaddleBReal r) ≤
Real.sqrt (4 * Real.pi) * (r * Real.exp (r / 2)) := by
calc
Real.sqrt (2 * Real.pi * bellSaddleBReal r)
≤ Real.sqrt (4 * Real.pi * (r ^ 2 * Real.exp r)) :=
Real.sqrt_le_sqrt hB_le
_ = Real.sqrt (4 * Real.pi) * (r * Real.exp (r / 2)) := by
have h4pi : 0 ≤ 4 * Real.pi := by positivity
rw [Real.sqrt_mul h4pi (r ^ 2 * Real.exp r)]
rw [Real.sqrt_mul (sq_nonneg r) (Real.exp r)]
rw [Real.sqrt_sq hr0]
rw [← Real.exp_half]
have hnum :
Real.sqrt (2 * Real.pi * bellSaddleBReal r) * bellSecondTailE r ≤
Real.sqrt (4 * Real.pi) * (r * Real.exp (r / 2)) * Real.exp (-(3 * r)) := by
calc
Real.sqrt (2 * Real.pi * bellSaddleBReal r) * bellSecondTailE r
≤ (Real.sqrt (4 * Real.pi) * (r * Real.exp (r / 2))) *
bellSecondTailE r :=
mul_le_mul_of_nonneg_right hsqrt hE_nonneg
_ ≤ (Real.sqrt (4 * Real.pi) * (r * Real.exp (r / 2))) *
Real.exp (-(3 * r)) :=
mul_le_mul_of_nonneg_left hE_le (by positivity)
calc
Real.sqrt (2 * Real.pi * bellSaddleBReal r) * bellSecondTailE r /
bellSecondCorrScale r
≤ Real.sqrt (2 * Real.pi * bellSaddleBReal r) * bellSecondTailE r /
((1 / 8 : ℝ) * Real.exp (-r)) :=
div_le_div_of_nonneg_left
(mul_nonneg (Real.sqrt_nonneg _) hE_nonneg) hαpos hcorrLower
_ ≤ (Real.sqrt (4 * Real.pi) * (r * Real.exp (r / 2)) *
Real.exp (-(3 * r))) / ((1 / 8 : ℝ) * Real.exp (-r)) :=
div_le_div_of_nonneg_right hnum hαpos.le
_ = 8 * Real.sqrt (4 * Real.pi) *
(r * Real.exp (-(3 / 2 : ℝ) * r)) := by
have hexp_ne : Real.exp (-r) ≠ 0 := (Real.exp_pos (-r)).ne'
have hexp_calc :
Real.exp (r / 2) * Real.exp (-(3 * r)) / Real.exp (-r) =
Real.exp (-(3 / 2 : ℝ) * r) := by
rw [← Real.exp_add]
rw [← Real.exp_sub]
congr 1
ring
calc
(Real.sqrt (4 * Real.pi) * (r * Real.exp (r / 2)) *
Real.exp (-(3 * r))) / ((1 / 8 : ℝ) * Real.exp (-r))
= 8 * Real.sqrt (4 * Real.pi) *
(r * (Real.exp (r / 2) * Real.exp (-(3 * r)) /
Real.exp (-r))) := by
field_simp [hexp_ne]
_ = 8 * Real.sqrt (4 * Real.pi) *
(r * Real.exp (-(3 / 2 : ℝ) * r)) := by
rw [hexp_calc]
lemma bell_tail_second_uniform :
∃ E : ℝ → ℝ,
(∀ᶠ r : ℝ in atTop, 0 ≤ E r) ∧
Tendsto
(fun r : ℝ => Real.sqrt (2 * Real.pi * bellSaddleBReal r) * E r /
bellSecondCorrScale r)
atTop (𝓝 0) ∧
(∀ᶠ r : ℝ in atTop, ∀ n : ℕ, ∀ θ : ℝ,
bellSaddleDeltaReal r ≤ |θ| → |θ| ≤ Real.pi →
‖saddleGAt (fun z : ℂ => Complex.exp (Complex.exp z - 1)) r n θ‖ ≤ E r) := by
exact ⟨bellSecondTailE, bellSecondTailE_nonneg_eventually, bellSecondTailE_decay,
bell_second_tail_bound_eventually⟩
private def expLocalRemainderFifth (θ : ℝ) : ℂ :=
ExpStirling.expLocalRemainder θ -
((((θ : ℂ) * Complex.I) ^ 3) / (((3 : ℕ).factorial : ℕ) : ℂ) +
(((θ : ℂ) * Complex.I) ^ 4) / (((4 : ℕ).factorial : ℕ) : ℂ))
private lemma expLocalRemainderFifth_norm_le {θ : ℝ} (hθ : |θ| ≤ 2) :
‖expLocalRemainderFifth θ‖ ≤ Real.exp 2 * |θ| ^ 5 := by
let u : ℂ := (θ : ℂ) * Complex.I
have htail := Complex.norm_exp_sub_sum_le_norm_mul_exp u 5
have hnorm : ‖u‖ = |θ| := by simp [u]
have hleft :
ExpStirling.expLocalRemainder θ -
(u ^ 3 / (((3 : ℕ).factorial : ℕ) : ℂ) +
u ^ 4 / (((4 : ℕ).factorial : ℕ) : ℂ)) =
Complex.exp u - ∑ m ∈ Finset.range 5, u ^ m / (m.factorial : ℂ) := by
simp [ExpStirling.expLocalRemainder, u, Finset.sum_range_succ]
ring
change
‖ExpStirling.expLocalRemainder θ -
(u ^ 3 / (((3 : ℕ).factorial : ℕ) : ℂ) +
u ^ 4 / (((4 : ℕ).factorial : ℕ) : ℂ))‖ ≤
Real.exp 2 * |θ| ^ 5
rw [hleft]
calc
‖Complex.exp u - ∑ m ∈ Finset.range 5, u ^ m / (m.factorial : ℂ)‖
≤ ‖u‖ ^ 5 * Real.exp ‖u‖ := htail
_ = |θ| ^ 5 * Real.exp |θ| := by rw [hnorm]
_ ≤ |θ| ^ 5 * Real.exp 2 := by
exact mul_le_mul_of_nonneg_left (Real.exp_le_exp.mpr hθ)
(pow_nonneg (abs_nonneg θ) 5)
_ = Real.exp 2 * |θ| ^ 5 := by ring
private lemma bellSaddleBReal_pos_of_ge_one {r : ℝ} (hr : 1 ≤ r) :
0 < bellSaddleBReal r := by
unfold bellSaddleBReal
positivity
private lemma bell_sqrtB_ge_r_exp_half {r : ℝ} (hr : 1 ≤ r) :
r * Real.exp (r / 2) ≤ Real.sqrt (bellSaddleBReal r) := by
have hr0 : 0 ≤ r := le_trans zero_le_one hr
have hsq :
(r * Real.exp (r / 2)) ^ 2 ≤ bellSaddleBReal r := by
unfold bellSaddleBReal
rw [Real.exp_half]
have hpoly : r ^ 2 ≤ r * (r + 1) := by nlinarith [hr0]
calc
(r * Real.sqrt (Real.exp r)) ^ 2 = r ^ 2 * Real.exp r := by
rw [mul_pow, Real.sq_sqrt (Real.exp_pos r).le]
_ ≤ (r * (r + 1)) * Real.exp r :=
mul_le_mul_of_nonneg_right hpoly (Real.exp_pos r).le
exact Real.le_sqrt_of_sq_le hsq
private lemma bell_sqrtB_sq {r : ℝ} (hb : 0 < bellSaddleBReal r) :
((Real.sqrt (bellSaddleBReal r) : ℂ) ^ 2) =
(bellSaddleBReal r : ℂ) := by
exact_mod_cast (Real.sq_sqrt hb.le)
private def bellScaledCubic (r x : ℝ) : ℂ :=
-Complex.I *
((bellSaddleB3Real r : ℂ) /
(6 * (Real.sqrt (bellSaddleBReal r) : ℂ) ^ 3)) * (x : ℂ) ^ 3
private def bellScaledQuartic (r x : ℝ) : ℂ :=
((bellSaddleB4Real r : ℂ) / (24 * (bellSaddleBReal r : ℂ) ^ 2)) *
(x : ℂ) ^ 4
private def bellScaledRemainder (r x : ℝ) : ℂ :=
bellLocalExponent r (x / Real.sqrt (bellSaddleBReal r)) -
bellScaledCubic r x - bellScaledQuartic r x
private lemma bellLocalExponent_scaled_split (r x : ℝ) :
bellLocalExponent r (x / Real.sqrt (bellSaddleBReal r)) =
bellScaledCubic r x + bellScaledQuartic r x + bellScaledRemainder r x := by
unfold bellScaledRemainder
ring
private lemma bell_saddleSecondPoly_eq_scaled_terms {r x : ℝ}
(hb : 0 < bellSaddleBReal r) :
saddleSecondPoly bellSaddleBReal bellSaddleB3Real bellSaddleB4Real r x =
1 + bellScaledCubic r x + bellScaledQuartic r x +
(bellScaledCubic r x) ^ 2 / 2 := by
have hsqrt_sq := bell_sqrtB_sq hb
have hsqrt_ne : (Real.sqrt (bellSaddleBReal r) : ℂ) ≠ 0 := by
exact Complex.ofReal_ne_zero.mpr (Real.sqrt_pos.mpr hb).ne'
have hb_ne : (bellSaddleBReal r : ℂ) ≠ 0 :=
Complex.ofReal_ne_zero.mpr hb.ne'
have hsqrt_pow6 :
(Real.sqrt (bellSaddleBReal r) : ℂ) ^ 6 =
(bellSaddleBReal r : ℂ) ^ 3 := by
calc
(Real.sqrt (bellSaddleBReal r) : ℂ) ^ 6 =
((Real.sqrt (bellSaddleBReal r) : ℂ) ^ 2) ^ 3 := by ring
_ = (bellSaddleBReal r : ℂ) ^ 3 := by rw [hsqrt_sq]
unfold saddleSecondPoly bellScaledCubic bellScaledQuartic
field_simp [hsqrt_ne, hb_ne]
ring_nf
rw [hsqrt_pow6, Complex.I_sq]
norm_num [Complex.ofReal_pow]
ring
private lemma bellScaledCubic_norm_le {r x : ℝ} (hr : 1 ≤ r) :
‖bellScaledCubic r x‖ ≤ 10 * Real.exp (-(r / 2)) * |x| ^ 3 := by
have hrpos : 0 < r := lt_of_lt_of_le zero_lt_one hr
have hbpos : 0 < bellSaddleBReal r := bellSaddleBReal_pos_of_ge_one hr
have hsqrt_pos : 0 < Real.sqrt (bellSaddleBReal r) := Real.sqrt_pos.mpr hbpos
have hsqrt_ge : r * Real.exp (r / 2) ≤ Real.sqrt (bellSaddleBReal r) :=
bell_sqrtB_ge_r_exp_half hr
have hsqrt3_ge :
(r * Real.exp (r / 2)) ^ 3 ≤ (Real.sqrt (bellSaddleBReal r)) ^ 3 :=
pow_le_pow_left₀ (by positivity : 0 ≤ r * Real.exp (r / 2)) hsqrt_ge 3
have hb3_nonneg : 0 ≤ bellSaddleB3Real r := by
unfold bellSaddleB3Real
positivity
have hb3_le : bellSaddleB3Real r ≤ 5 * r ^ 3 * Real.exp r := by
unfold bellSaddleB3Real
have hpoly : r + 3 * r ^ 2 + r ^ 3 ≤ 5 * r ^ 3 := by
nlinarith [hr, sq_nonneg r, show 0 ≤ r ^ 3 by positivity]
exact mul_le_mul_of_nonneg_right hpoly (Real.exp_pos r).le
have hcoef :
‖(bellSaddleB3Real r : ℂ) /
(6 * (Real.sqrt (bellSaddleBReal r) : ℂ) ^ 3)‖ ≤
10 * Real.exp (-(r / 2)) := by
calc
‖(bellSaddleB3Real r : ℂ) /
(6 * (Real.sqrt (bellSaddleBReal r) : ℂ) ^ 3)‖
= bellSaddleB3Real r /
(6 * (Real.sqrt (bellSaddleBReal r)) ^ 3) := by
rw [norm_div, Complex.norm_real, Real.norm_eq_abs,
abs_of_nonneg hb3_nonneg]
simp only [norm_mul, norm_ofNat, norm_pow, Complex.norm_real, Real.norm_eq_abs]
rw [abs_of_pos hsqrt_pos]
_ ≤ (5 * r ^ 3 * Real.exp r) /
(6 * (r * Real.exp (r / 2)) ^ 3) := by
exact div_le_div₀ (by positivity) hb3_le (by positivity)
(by nlinarith [hsqrt3_ge])
_ = (5 / 6 : ℝ) * Real.exp (-(r / 2)) := by
have hr_ne : r ≠ 0 := hrpos.ne'
field_simp [hr_ne, (Real.exp_pos (r / 2)).ne']
have hexp_pow3 :
(Real.exp (r / 2)) ^ 3 = Real.exp (3 * (r / 2)) := by
rw [← Real.exp_nat_mul]
norm_num
rw [hexp_pow3, ← Real.exp_add]
congr 1
ring
_ ≤ 10 * Real.exp (-(r / 2)) := by
exact mul_le_mul_of_nonneg_right (by norm_num : (5 / 6 : ℝ) ≤ 10)
(Real.exp_pos _).le
calc
‖bellScaledCubic r x‖
= ‖(bellSaddleB3Real r : ℂ) /
(6 * (Real.sqrt (bellSaddleBReal r) : ℂ) ^ 3)‖ * |x| ^ 3 := by
unfold bellScaledCubic
rw [norm_mul, norm_mul, norm_neg, norm_I, one_mul, norm_pow,
Complex.norm_real, Real.norm_eq_abs]
_ ≤ (10 * Real.exp (-(r / 2))) * |x| ^ 3 :=
mul_le_mul_of_nonneg_right hcoef (pow_nonneg (abs_nonneg x) 3)
_ = 10 * Real.exp (-(r / 2)) * |x| ^ 3 := by ring
private lemma bellScaledQuartic_norm_le {r x : ℝ} (hr : 1 ≤ r) :
‖bellScaledQuartic r x‖ ≤ 10 * Real.exp (-r) * |x| ^ 4 := by
have hrpos : 0 < r := lt_of_lt_of_le zero_lt_one hr
have hbpos : 0 < bellSaddleBReal r := bellSaddleBReal_pos_of_ge_one hr
have hb_ge : r ^ 2 * Real.exp r ≤ bellSaddleBReal r := by
unfold bellSaddleBReal
have hpoly : r ^ 2 ≤ r * (r + 1) := by nlinarith [hr, sq_nonneg r]
exact mul_le_mul_of_nonneg_right hpoly (Real.exp_pos r).le
have hb_sq_ge : (r ^ 2 * Real.exp r) ^ 2 ≤ (bellSaddleBReal r) ^ 2 :=
pow_le_pow_left₀ (by positivity : 0 ≤ r ^ 2 * Real.exp r) hb_ge 2
have hb4_nonneg : 0 ≤ bellSaddleB4Real r := by
unfold bellSaddleB4Real
positivity
have hb4_le : bellSaddleB4Real r ≤ 15 * r ^ 4 * Real.exp r := by
unfold bellSaddleB4Real
have hpoly : r + 7 * r ^ 2 + 6 * r ^ 3 + r ^ 4 ≤ 15 * r ^ 4 := by
nlinarith [hr, sq_nonneg r, show 0 ≤ r ^ 3 by positivity,
show 0 ≤ r ^ 4 by positivity]
exact mul_le_mul_of_nonneg_right hpoly (Real.exp_pos r).le
have hcoef :
‖(bellSaddleB4Real r : ℂ) / (24 * (bellSaddleBReal r : ℂ) ^ 2)‖ ≤
10 * Real.exp (-r) := by
calc
‖(bellSaddleB4Real r : ℂ) / (24 * (bellSaddleBReal r : ℂ) ^ 2)‖
= bellSaddleB4Real r / (24 * (bellSaddleBReal r) ^ 2) := by
rw [norm_div, Complex.norm_real, Real.norm_eq_abs,
abs_of_nonneg hb4_nonneg]
simp only [norm_mul, norm_ofNat, norm_pow, Complex.norm_real, Real.norm_eq_abs]
rw [abs_of_pos hbpos]
_ ≤ (15 * r ^ 4 * Real.exp r) /
(24 * (r ^ 2 * Real.exp r) ^ 2) := by
exact div_le_div₀ (by positivity) hb4_le (by positivity)
(by nlinarith [hb_sq_ge])
_ = (15 / 24 : ℝ) * Real.exp (-r) := by
have hr_ne : r ≠ 0 := hrpos.ne'
have hexp_ne : Real.exp r ≠ 0 := (Real.exp_pos r).ne'
field_simp [hr_ne, hexp_ne]
ring_nf
rw [Real.exp_neg]
rw [mul_inv_cancel₀ hexp_ne]
_ ≤ 10 * Real.exp (-r) := by
exact mul_le_mul_of_nonneg_right (by norm_num : (15 / 24 : ℝ) ≤ 10)
(Real.exp_pos _).le
calc
‖bellScaledQuartic r x‖
= ‖(bellSaddleB4Real r : ℂ) /
(24 * (bellSaddleBReal r : ℂ) ^ 2)‖ * |x| ^ 4 := by
unfold bellScaledQuartic
rw [norm_mul, norm_pow, Complex.norm_real, Real.norm_eq_abs]
_ ≤ (10 * Real.exp (-r)) * |x| ^ 4 :=
mul_le_mul_of_nonneg_right hcoef (pow_nonneg (abs_nonneg x) 4)
_ = 10 * Real.exp (-r) * |x| ^ 4 := by ring
private def complexExpRemainderFifth (z : ℂ) : ℂ :=
Complex.exp z - ∑ m ∈ Finset.range 5, z ^ m / (m.factorial : ℂ)
private lemma complexExpRemainderFifth_norm_le {z : ℂ} (hz : ‖z‖ ≤ 1) :
‖complexExpRemainderFifth z‖ ≤ Real.exp 1 * ‖z‖ ^ 5 := by
unfold complexExpRemainderFifth
calc
‖Complex.exp z - ∑ m ∈ Finset.range 5, z ^ m / (m.factorial : ℂ)‖
≤ ‖z‖ ^ 5 * Real.exp ‖z‖ :=
Complex.norm_exp_sub_sum_le_norm_mul_exp z 5
_ ≤ ‖z‖ ^ 5 * Real.exp 1 :=
mul_le_mul_of_nonneg_left (Real.exp_le_exp.mpr hz)
(pow_nonneg (norm_nonneg z) 5)
_ = Real.exp 1 * ‖z‖ ^ 5 := by ring
private def bellLocalTaylorConst : ℝ :=
10000 * Real.exp 2
private lemma bellLocalTaylorConst_pos : 0 < bellLocalTaylorConst := by
unfold bellLocalTaylorConst
positivity
private lemma bellLocalExponent_fifth_remainder_norm_le {r θ : ℝ}
(hr : 1 ≤ r) (hθ : |θ| ≤ 1) (hsmall : 2 * (r * |θ|) ≤ 1) :
‖bellLocalExponent r θ -
(-Complex.I * ((bellSaddleB3Real r : ℂ) / 6) * (θ : ℂ) ^ 3 +
((bellSaddleB4Real r : ℂ) / 24) * (θ : ℂ) ^ 4)‖
≤ bellLocalTaylorConst * Real.exp r * r ^ 5 * |θ| ^ 5 := by
let y : ℝ := |θ|
let w : ℂ := (θ : ℂ) * Complex.I
let u : ℂ := (r : ℂ) * (Complex.exp w - 1)
let m : ℂ := (r : ℂ) * w
let s2 : ℂ := (r : ℂ) * w ^ 2 / 2
let s3 : ℂ := (r : ℂ) * w ^ 3 / 6
let outer : ℂ := Complex.exp u - ∑ k ∈ Finset.range 3, u ^ k / (k.factorial : ℂ)
let inner : ℂ := (r : ℂ) * ExpStirling.expLocalRemainder θ
let cross : ℂ := (u ^ 2 - m ^ 2) / 2
let outerPoly : ℂ := m ^ 3 / 6 + m ^ 2 * s2 / 2 + m ^ 4 / 24
let innerPoly : ℂ := (r : ℂ) * (w ^ 3 / 6 + w ^ 4 / 24)
let crossPoly : ℂ := m * s2 + m * s3 + s2 ^ 2 / 2
let P : ℂ :=
-Complex.I * ((bellSaddleB3Real r : ℂ) / 6) * (θ : ℂ) ^ 3 +
((bellSaddleB4Real r : ℂ) / 24) * (θ : ℂ) ^ 4
have hrpos : 0 < r := lt_of_lt_of_le zero_lt_one hr
have hr0 : 0 ≤ r := hrpos.le
have hy0 : 0 ≤ y := by
dsimp [y]
exact abs_nonneg θ
have hy1 : y ≤ 1 := by
simpa [y] using hθ
have hE2_ge_one : 1 ≤ Real.exp 2 := by
calc
(1 : ℝ) = Real.exp 0 := by rw [Real.exp_zero]
_ ≤ Real.exp 2 := Real.exp_le_exp.mpr (by norm_num : (0 : ℝ) ≤ 2)
have hE1_le_E2 : Real.exp 1 ≤ Real.exp 2 :=
Real.exp_le_exp.mpr (by norm_num : (1 : ℝ) ≤ 2)
have hw_norm : ‖w‖ = y := by
dsimp [w, y]
simp
have hm_norm : ‖m‖ = r * y := by
dsimp [m]
rw [norm_mul, Complex.norm_real, Real.norm_eq_abs, abs_of_nonneg hr0, hw_norm]
have hs2_norm_le : ‖s2‖ ≤ r * y ^ 2 := by
dsimp [s2]
calc
‖(r : ℂ) * w ^ 2 / 2‖ = r * y ^ 2 / 2 := by
rw [norm_div, norm_mul, Complex.norm_real, Real.norm_eq_abs,
abs_of_nonneg hr0, norm_pow, hw_norm]
norm_num
_ ≤ r * y ^ 2 := by
nlinarith [hr0, pow_nonneg hy0 2]
have hs3_norm_le : ‖s3‖ ≤ r * y ^ 3 := by
dsimp [s3]
calc
‖(r : ℂ) * w ^ 3 / 6‖ = r * y ^ 3 / 6 := by
rw [norm_div, norm_mul, Complex.norm_real, Real.norm_eq_abs,
abs_of_nonneg hr0, norm_pow, hw_norm]
norm_num
_ ≤ r * y ^ 3 := by
nlinarith [hr0, pow_nonneg hy0 3]
have hu_norm : ‖u‖ ≤ 2 * (r * y) := by
have hbase := bell_exp_i_sub_one_norm_le_two_abs hθ
calc
‖u‖ = r * ‖Complex.exp w - 1‖ := by
dsimp [u]
rw [norm_mul, Complex.norm_real, Real.norm_eq_abs, abs_of_nonneg hr0]
_ ≤ r * (2 * y) := by
exact mul_le_mul_of_nonneg_left (by simpa [w, y] using hbase) hr0
_ = 2 * (r * y) := by ring
have hu_le_one : ‖u‖ ≤ 1 := hu_norm.trans hsmall
have hs_norm_le : ‖u - m‖ ≤ r * y ^ 2 := by
have hbase := bell_exp_i_sub_one_sub_id_norm_le_sq hθ
have hdiff : u - m = (r : ℂ) *
(Complex.exp w - 1 - (θ : ℂ) * Complex.I) := by
dsimp [u, m, w]
ring
rw [hdiff]
calc
‖(r : ℂ) * (Complex.exp w - 1 - (θ : ℂ) * Complex.I)‖
= r * ‖Complex.exp w - 1 - (θ : ℂ) * Complex.I‖ := by
rw [norm_mul, Complex.norm_real, Real.norm_eq_abs, abs_of_nonneg hr0]
_ ≤ r * y ^ 2 := by
exact mul_le_mul_of_nonneg_left (by simpa [w, y] using hbase) hr0
have htRest_norm_le :
‖u - m - s2‖ ≤ Real.exp 2 * r * y ^ 3 := by
have hrest : u - m - s2 = (r : ℂ) * ExpStirling.expLocalRemainder θ := by
dsimp [u, m, s2, w]
rw [ExpStirling.expLocalRemainder_eq θ]
ring_nf
norm_num [Complex.ofReal_mul]
ring
rw [hrest]
have hrem := ExpStirling.expLocalRemainder_norm_le_exp_one θ hθ
calc
‖(r : ℂ) * ExpStirling.expLocalRemainder θ‖
= r * ‖ExpStirling.expLocalRemainder θ‖ := by
rw [norm_mul, Complex.norm_real, Real.norm_eq_abs, abs_of_nonneg hr0]
_ ≤ r * (Real.exp 1 * y ^ 3) := by
exact mul_le_mul_of_nonneg_left (by simpa [y] using hrem) hr0
_ ≤ Real.exp 2 * r * y ^ 3 := by
calc
r * (Real.exp 1 * y ^ 3) = Real.exp 1 * (r * y ^ 3) := by ring
_ ≤ Real.exp 2 * (r * y ^ 3) :=
mul_le_mul_of_nonneg_right hE1_le_E2
(mul_nonneg hr0 (pow_nonneg hy0 3))
_ = Real.exp 2 * r * y ^ 3 := by ring
have htFourth_norm_le :
‖u - m - s2 - s3‖ ≤ 2 * Real.exp 2 * r * y ^ 4 := by
let t : ℂ := (r : ℂ) * expLocalRemainderFifth θ
let s4 : ℂ := (r : ℂ) * w ^ 4 / 24
have hrest : u - m - s2 - s3 = s4 + t := by
dsimp [u, m, s2, s3, s4, t, w, expLocalRemainderFifth]
rw [ExpStirling.expLocalRemainder_eq θ]
ring_nf
norm_num [Complex.ofReal_mul]
ring
have hs4 : ‖s4‖ ≤ r * y ^ 4 := by
dsimp [s4]
calc
‖(r : ℂ) * w ^ 4 / 24‖ = r * y ^ 4 / 24 := by
rw [norm_div, norm_mul, Complex.norm_real, Real.norm_eq_abs,
abs_of_nonneg hr0, norm_pow, hw_norm]
norm_num
_ ≤ r * y ^ 4 := by
nlinarith [hr0, pow_nonneg hy0 4]
have ht : ‖t‖ ≤ Real.exp 2 * r * y ^ 4 := by
have hrem := expLocalRemainderFifth_norm_le
(by linarith [hy1] : |θ| ≤ 2)
dsimp [t]
calc
‖(r : ℂ) * expLocalRemainderFifth θ‖
= r * ‖expLocalRemainderFifth θ‖ := by
rw [norm_mul, Complex.norm_real, Real.norm_eq_abs, abs_of_nonneg hr0]
_ ≤ r * (Real.exp 2 * y ^ 5) := by
exact mul_le_mul_of_nonneg_left (by simpa [y] using hrem) hr0
_ ≤ Real.exp 2 * r * y ^ 4 := by
have hy5_le_y4 : y ^ 5 ≤ y ^ 4 := by
calc
y ^ 5 = y ^ 4 * y := by ring
_ ≤ y ^ 4 * 1 :=
mul_le_mul_of_nonneg_left hy1 (pow_nonneg hy0 4)
_ = y ^ 4 := by ring
calc
r * (Real.exp 2 * y ^ 5) = (Real.exp 2 * r) * y ^ 5 := by ring
_ ≤ (Real.exp 2 * r) * y ^ 4 :=
mul_le_mul_of_nonneg_left hy5_le_y4
(mul_nonneg (Real.exp_pos 2).le hr0)
_ = Real.exp 2 * r * y ^ 4 := by ring
rw [hrest]
calc
‖s4 + t‖ ≤ ‖s4‖ + ‖t‖ := norm_add_le _ _
_ ≤ r * y ^ 4 + Real.exp 2 * r * y ^ 4 := add_le_add hs4 ht
_ ≤ 2 * Real.exp 2 * r * y ^ 4 := by
have hleft : r * y ^ 4 ≤ Real.exp 2 * r * y ^ 4 := by
calc
r * y ^ 4 = 1 * (r * y ^ 4) := by ring
_ ≤ Real.exp 2 * (r * y ^ 4) :=
mul_le_mul_of_nonneg_right hE2_ge_one
(mul_nonneg hr0 (pow_nonneg hy0 4))
_ = Real.exp 2 * r * y ^ 4 := by ring
linarith
have hW_decomp :
bellLocalExponent r θ = Complex.exp (r : ℂ) * (outer + inner + cross) := by
simpa [outer, inner, cross, u, m, w] using bellLocalExponent_decomp r θ
have hP_decomp :
P = Complex.exp (r : ℂ) * (outerPoly + innerPoly + crossPoly) := by
dsimp [P, outerPoly, innerPoly, crossPoly, m, s2, s3, w,
bellSaddleB3Real, bellSaddleB4Real]
rw [← Complex.ofReal_exp r]
ring_nf
norm_num [Complex.ofReal_add, Complex.ofReal_mul, Complex.ofReal_pow]
ring
have hdiff :
bellLocalExponent r θ - P =
Complex.exp (r : ℂ) *
((outer - outerPoly) + (inner - innerPoly) + (cross - crossPoly)) := by
rw [hW_decomp, hP_decomp]
ring
have houter_eq :
outer - outerPoly =
complexExpRemainderFifth u +
(u ^ 3 - m ^ 3 - 3 * m ^ 2 * s2) / 6 +
(u ^ 4 - m ^ 4) / 24 := by
dsimp [outer, outerPoly, complexExpRemainderFifth]
simp [Finset.sum_range_succ]
ring
have hcube_eq :
u ^ 3 - m ^ 3 - 3 * m ^ 2 * s2 =
3 * m ^ 2 * (u - m - s2) + 3 * m * (u - m) ^ 2 + (u - m) ^ 3 := by
ring
have hcube_norm :
‖u ^ 3 - m ^ 3 - 3 * m ^ 2 * s2‖ ≤
20 * Real.exp 2 * r ^ 5 * y ^ 5 := by
rw [hcube_eq]
have hraw :
‖3 * m ^ 2 * (u - m - s2) + 3 * m * (u - m) ^ 2 + (u - m) ^ 3‖
≤ 3 * ‖m‖ ^ 2 * ‖u - m - s2‖ +
3 * ‖m‖ * ‖u - m‖ ^ 2 + ‖u - m‖ ^ 3 := by
calc
‖3 * m ^ 2 * (u - m - s2) + 3 * m * (u - m) ^ 2 + (u - m) ^ 3‖
≤ ‖3 * m ^ 2 * (u - m - s2)‖ +
‖3 * m * (u - m) ^ 2‖ + ‖(u - m) ^ 3‖ := by
calc
‖3 * m ^ 2 * (u - m - s2) + 3 * m * (u - m) ^ 2 +
(u - m) ^ 3‖
≤ ‖3 * m ^ 2 * (u - m - s2) + 3 * m * (u - m) ^ 2‖ +
‖(u - m) ^ 3‖ := norm_add_le _ _
_ ≤ (‖3 * m ^ 2 * (u - m - s2)‖ +
‖3 * m * (u - m) ^ 2‖) + ‖(u - m) ^ 3‖ :=
add_le_add (norm_add_le _ _) (le_refl _)
_ = ‖3 * m ^ 2 * (u - m - s2)‖ +
‖3 * m * (u - m) ^ 2‖ + ‖(u - m) ^ 3‖ := by ring
_ = 3 * ‖m‖ ^ 2 * ‖u - m - s2‖ +
3 * ‖m‖ * ‖u - m‖ ^ 2 + ‖u - m‖ ^ 3 := by
simp [norm_pow]
have hbound :
3 * ‖m‖ ^ 2 * ‖u - m - s2‖ +
3 * ‖m‖ * ‖u - m‖ ^ 2 + ‖u - m‖ ^ 3
≤ 20 * Real.exp 2 * r ^ 5 * y ^ 5 := by
have hm_le : ‖m‖ ≤ r * y := by rw [hm_norm]
have hr3_le_r5 : r ^ 3 ≤ r ^ 5 := by
have h1_le_r2 : (1 : ℝ) ≤ r ^ 2 := by nlinarith [hr, sq_nonneg r]
calc
r ^ 3 = r ^ 3 * 1 := by ring
_ ≤ r ^ 3 * r ^ 2 :=
mul_le_mul_of_nonneg_left h1_le_r2 (by positivity)
_ = r ^ 5 := by ring
have hy6_le_y5 : y ^ 6 ≤ y ^ 5 := by
calc
y ^ 6 = y ^ 5 * y := by ring
_ ≤ y ^ 5 * 1 := mul_le_mul_of_nonneg_left hy1 (pow_nonneg hy0 5)
_ = y ^ 5 := by ring
calc
3 * ‖m‖ ^ 2 * ‖u - m - s2‖ +
3 * ‖m‖ * ‖u - m‖ ^ 2 + ‖u - m‖ ^ 3
≤ 3 * (r * y) ^ 2 * (Real.exp 2 * r * y ^ 3) +
3 * (r * y) * (r * y ^ 2) ^ 2 +
(r * y ^ 2) ^ 3 := by
gcongr
_ ≤ 20 * Real.exp 2 * r ^ 5 * y ^ 5 := by
have hnon_r3 : 0 ≤ r ^ 3 := by positivity
have hnon_y5 : 0 ≤ y ^ 5 := pow_nonneg hy0 5
have h3plain :
3 * r ^ 3 * y ^ 5 ≤ 3 * Real.exp 2 * r ^ 3 * y ^ 5 := by
calc
3 * r ^ 3 * y ^ 5 = 3 * (1 * (r ^ 3 * y ^ 5)) := by ring
_ ≤ 3 * (Real.exp 2 * (r ^ 3 * y ^ 5)) := by
gcongr
_ = 3 * Real.exp 2 * r ^ 3 * y ^ 5 := by ring
have hy6term :
r ^ 3 * y ^ 6 ≤ Real.exp 2 * r ^ 3 * y ^ 5 := by
calc
r ^ 3 * y ^ 6 ≤ r ^ 3 * y ^ 5 :=
mul_le_mul_of_nonneg_left hy6_le_y5 hnon_r3
_ = 1 * (r ^ 3 * y ^ 5) := by ring
_ ≤ Real.exp 2 * (r ^ 3 * y ^ 5) := by
gcongr
_ = Real.exp 2 * r ^ 3 * y ^ 5 := by ring
calc
3 * (r * y) ^ 2 * (Real.exp 2 * r * y ^ 3) +
3 * (r * y) * (r * y ^ 2) ^ 2 + (r * y ^ 2) ^ 3
= 3 * Real.exp 2 * r ^ 3 * y ^ 5 +
3 * r ^ 3 * y ^ 5 + r ^ 3 * y ^ 6 := by ring
_ ≤ 3 * Real.exp 2 * r ^ 3 * y ^ 5 +
3 * Real.exp 2 * r ^ 3 * y ^ 5 +
Real.exp 2 * r ^ 3 * y ^ 5 :=
add_le_add (add_le_add (le_refl _) h3plain) hy6term
_ = 7 * Real.exp 2 * r ^ 3 * y ^ 5 := by ring
_ ≤ 20 * Real.exp 2 * r ^ 5 * y ^ 5 := by
gcongr
norm_num
exact hraw.trans hbound
have hquartic_norm :
‖u ^ 4 - m ^ 4‖ ≤ 20 * r ^ 5 * y ^ 5 := by
have hquartic_eq :
u ^ 4 - m ^ 4 = (u - m) * (u ^ 3 + u ^ 2 * m + u * m ^ 2 + m ^ 3) := by
ring
rw [hquartic_eq]
have hsum :
‖u ^ 3 + u ^ 2 * m + u * m ^ 2 + m ^ 3‖ ≤
20 * r ^ 3 * y ^ 3 := by
have hm_le : ‖m‖ ≤ r * y := by rw [hm_norm]
have hu_le : ‖u‖ ≤ 2 * r * y := by
simpa [mul_assoc] using hu_norm
calc
‖u ^ 3 + u ^ 2 * m + u * m ^ 2 + m ^ 3‖
≤ ‖u ^ 3‖ + ‖u ^ 2 * m‖ + ‖u * m ^ 2‖ + ‖m ^ 3‖ := by
calc
‖u ^ 3 + u ^ 2 * m + u * m ^ 2 + m ^ 3‖
≤ ‖u ^ 3 + u ^ 2 * m + u * m ^ 2‖ + ‖m ^ 3‖ := norm_add_le _ _
_ ≤ (‖u ^ 3 + u ^ 2 * m‖ + ‖u * m ^ 2‖) + ‖m ^ 3‖ :=
add_le_add (norm_add_le _ _) (le_refl _)
_ ≤ ((‖u ^ 3‖ + ‖u ^ 2 * m‖) + ‖u * m ^ 2‖) + ‖m ^ 3‖ :=
add_le_add (add_le_add (norm_add_le _ _) (le_refl _)) (le_refl _)
_ = ‖u ^ 3‖ + ‖u ^ 2 * m‖ + ‖u * m ^ 2‖ + ‖m ^ 3‖ := by ring
_ = ‖u‖ ^ 3 + ‖u‖ ^ 2 * ‖m‖ + ‖u‖ * ‖m‖ ^ 2 + ‖m‖ ^ 3 := by
simp [norm_pow]
_ ≤ 20 * r ^ 3 * y ^ 3 := by
calc
‖u‖ ^ 3 + ‖u‖ ^ 2 * ‖m‖ + ‖u‖ * ‖m‖ ^ 2 + ‖m‖ ^ 3
≤ (2 * r * y) ^ 3 + (2 * r * y) ^ 2 * (r * y) +
(2 * r * y) * (r * y) ^ 2 + (r * y) ^ 3 := by
gcongr
_ ≤ 20 * r ^ 3 * y ^ 3 := by
calc
(2 * r * y) ^ 3 + (2 * r * y) ^ 2 * (r * y) +
(2 * r * y) * (r * y) ^ 2 + (r * y) ^ 3
= 15 * r ^ 3 * y ^ 3 := by ring
_ ≤ 20 * r ^ 3 * y ^ 3 := by
gcongr
norm_num
calc
‖(u - m) * (u ^ 3 + u ^ 2 * m + u * m ^ 2 + m ^ 3)‖
= ‖u - m‖ * ‖u ^ 3 + u ^ 2 * m + u * m ^ 2 + m ^ 3‖ := by rw [norm_mul]
_ ≤ (r * y ^ 2) * (20 * r ^ 3 * y ^ 3) :=
mul_le_mul hs_norm_le hsum (norm_nonneg _) (by positivity)
_ ≤ 20 * r ^ 5 * y ^ 5 := by
have hr4_le_r5 : r ^ 4 ≤ r ^ 5 := by
calc
r ^ 4 = r ^ 4 * 1 := by ring
_ ≤ r ^ 4 * r :=
mul_le_mul_of_nonneg_left hr (by positivity)
_ = r ^ 5 := by ring
nlinarith [hr4_le_r5, hy0, pow_nonneg hy0 5]
have houter :
‖outer - outerPoly‖ ≤ 100 * Real.exp 2 * r ^ 5 * y ^ 5 := by
rw [houter_eq]
have htail :
‖complexExpRemainderFifth u‖ ≤ 32 * Real.exp 2 * r ^ 5 * y ^ 5 := by
have htail0 := complexExpRemainderFifth_norm_le hu_le_one
calc
‖complexExpRemainderFifth u‖ ≤ Real.exp 1 * ‖u‖ ^ 5 := htail0
_ ≤ Real.exp 1 * (2 * (r * y)) ^ 5 :=
mul_le_mul_of_nonneg_left
(pow_le_pow_left₀ (norm_nonneg u) hu_norm 5) (Real.exp_pos 1).le
_ ≤ 32 * Real.exp 2 * r ^ 5 * y ^ 5 := by
calc
Real.exp 1 * (2 * (r * y)) ^ 5 =
32 * Real.exp 1 * r ^ 5 * y ^ 5 := by ring
_ ≤ 32 * Real.exp 2 * r ^ 5 * y ^ 5 := by
gcongr
calc
‖complexExpRemainderFifth u +
(u ^ 3 - m ^ 3 - 3 * m ^ 2 * s2) / 6 +
(u ^ 4 - m ^ 4) / 24‖
≤ ‖complexExpRemainderFifth u‖ +
‖(u ^ 3 - m ^ 3 - 3 * m ^ 2 * s2) / 6‖ +
‖(u ^ 4 - m ^ 4) / 24‖ := by
calc
‖complexExpRemainderFifth u +
(u ^ 3 - m ^ 3 - 3 * m ^ 2 * s2) / 6 +
(u ^ 4 - m ^ 4) / 24‖
≤ ‖complexExpRemainderFifth u +
(u ^ 3 - m ^ 3 - 3 * m ^ 2 * s2) / 6‖ +
‖(u ^ 4 - m ^ 4) / 24‖ := norm_add_le _ _
_ ≤ (‖complexExpRemainderFifth u‖ +
‖(u ^ 3 - m ^ 3 - 3 * m ^ 2 * s2) / 6‖) +
‖(u ^ 4 - m ^ 4) / 24‖ :=
add_le_add (norm_add_le _ _) (le_refl _)
_ = ‖complexExpRemainderFifth u‖ +
‖(u ^ 3 - m ^ 3 - 3 * m ^ 2 * s2) / 6‖ +
‖(u ^ 4 - m ^ 4) / 24‖ := by ring
_ ≤ 32 * Real.exp 2 * r ^ 5 * y ^ 5 +
20 * Real.exp 2 * r ^ 5 * y ^ 5 +
20 * r ^ 5 * y ^ 5 := by
have hcube_div :
‖(u ^ 3 - m ^ 3 - 3 * m ^ 2 * s2) / 6‖ ≤
20 * Real.exp 2 * r ^ 5 * y ^ 5 := by
calc
‖(u ^ 3 - m ^ 3 - 3 * m ^ 2 * s2) / 6‖
≤ ‖u ^ 3 - m ^ 3 - 3 * m ^ 2 * s2‖ := by
rw [norm_div, Complex.norm_ofNat]
nlinarith [norm_nonneg (u ^ 3 - m ^ 3 - 3 * m ^ 2 * s2)]
_ ≤ 20 * Real.exp 2 * r ^ 5 * y ^ 5 := hcube_norm
have hquartic_div :
‖(u ^ 4 - m ^ 4) / 24‖ ≤ 20 * r ^ 5 * y ^ 5 := by
calc
‖(u ^ 4 - m ^ 4) / 24‖ ≤ ‖u ^ 4 - m ^ 4‖ := by
rw [norm_div, Complex.norm_ofNat]
nlinarith [norm_nonneg (u ^ 4 - m ^ 4)]
_ ≤ 20 * r ^ 5 * y ^ 5 := hquartic_norm
exact add_le_add (add_le_add htail hcube_div) hquartic_div
_ ≤ 100 * Real.exp 2 * r ^ 5 * y ^ 5 := by
have hlast :
20 * r ^ 5 * y ^ 5 ≤ 20 * Real.exp 2 * r ^ 5 * y ^ 5 := by
calc
20 * r ^ 5 * y ^ 5 = 20 * (1 * (r ^ 5 * y ^ 5)) := by ring
_ ≤ 20 * (Real.exp 2 * (r ^ 5 * y ^ 5)) := by
gcongr
_ = 20 * Real.exp 2 * r ^ 5 * y ^ 5 := by ring
calc
32 * Real.exp 2 * r ^ 5 * y ^ 5 +
20 * Real.exp 2 * r ^ 5 * y ^ 5 + 20 * r ^ 5 * y ^ 5
≤ 32 * Real.exp 2 * r ^ 5 * y ^ 5 +
20 * Real.exp 2 * r ^ 5 * y ^ 5 +
20 * Real.exp 2 * r ^ 5 * y ^ 5 := by
exact add_le_add (le_refl _) hlast
_ = 72 * Real.exp 2 * r ^ 5 * y ^ 5 := by ring
_ ≤ 100 * Real.exp 2 * r ^ 5 * y ^ 5 := by
gcongr
norm_num
have hinner :
‖inner - innerPoly‖ ≤ Real.exp 2 * r ^ 5 * y ^ 5 := by
have hinner_eq :
inner - innerPoly = (r : ℂ) * expLocalRemainderFifth θ := by
dsimp [inner, innerPoly, expLocalRemainderFifth]
ring
rw [hinner_eq]
have hrem := expLocalRemainderFifth_norm_le
(by linarith [hy1] : |θ| ≤ 2)
calc
‖(r : ℂ) * expLocalRemainderFifth θ‖
= r * ‖expLocalRemainderFifth θ‖ := by
rw [norm_mul, Complex.norm_real, Real.norm_eq_abs, abs_of_nonneg hr0]
_ ≤ r * (Real.exp 2 * y ^ 5) := by
exact mul_le_mul_of_nonneg_left (by simpa [y] using hrem) hr0
_ ≤ Real.exp 2 * r ^ 5 * y ^ 5 := by
have hr_le_r5 : r ≤ r ^ 5 := by
calc
r = r * 1 := by ring
_ ≤ r * r ^ 4 :=
mul_le_mul_of_nonneg_left
(by
have h1_le_r2 : (1 : ℝ) ≤ r ^ 2 := by nlinarith [hr, sq_nonneg r]
calc
(1 : ℝ) = 1 * 1 := by ring
_ ≤ r ^ 2 * r ^ 2 :=
mul_le_mul h1_le_r2 h1_le_r2 zero_le_one (by positivity)
_ = r ^ 4 := by ring)
hr0
_ = r ^ 5 := by ring
calc
r * (Real.exp 2 * y ^ 5) = (Real.exp 2 * y ^ 5) * r := by ring
_ ≤ (Real.exp 2 * y ^ 5) * r ^ 5 :=
mul_le_mul_of_nonneg_left hr_le_r5
(mul_nonneg (Real.exp_pos 2).le (pow_nonneg hy0 5))
_ = Real.exp 2 * r ^ 5 * y ^ 5 := by ring
have hcross :
‖cross - crossPoly‖ ≤ 10 * Real.exp 2 * r ^ 5 * y ^ 5 := by
have hcross_eq :
cross - crossPoly =
m * (u - m - s2 - s3) + ((u - m) ^ 2 - s2 ^ 2) / 2 := by
dsimp [cross, crossPoly]
ring
rw [hcross_eq]
have hs_plus_s2 :
‖u - m + s2‖ ≤ 2 * r * y ^ 2 := by
calc
‖u - m + s2‖ ≤ ‖u - m‖ + ‖s2‖ := norm_add_le _ _
_ ≤ r * y ^ 2 + r * y ^ 2 := add_le_add hs_norm_le hs2_norm_le
_ = 2 * r * y ^ 2 := by ring
have hsquare_diff :
‖((u - m) ^ 2 - s2 ^ 2) / 2‖ ≤
2 * Real.exp 2 * r ^ 2 * y ^ 5 := by
have hsq_eq :
(u - m) ^ 2 - s2 ^ 2 = (u - m - s2) * (u - m + s2) := by
ring
calc
‖((u - m) ^ 2 - s2 ^ 2) / 2‖
≤ ‖(u - m) ^ 2 - s2 ^ 2‖ := by
rw [norm_div, Complex.norm_ofNat]
norm_num
_ = ‖(u - m - s2) * (u - m + s2)‖ := by rw [hsq_eq]
_ = ‖u - m - s2‖ * ‖u - m + s2‖ := by rw [norm_mul]
_ ≤ (Real.exp 2 * r * y ^ 3) * (2 * r * y ^ 2) :=
mul_le_mul htRest_norm_le hs_plus_s2 (norm_nonneg _) (by positivity)
_ = 2 * Real.exp 2 * r ^ 2 * y ^ 5 := by ring
calc
‖m * (u - m - s2 - s3) + ((u - m) ^ 2 - s2 ^ 2) / 2‖
≤ ‖m * (u - m - s2 - s3)‖ +
‖((u - m) ^ 2 - s2 ^ 2) / 2‖ := norm_add_le _ _
_ ≤ (r * y) * (2 * Real.exp 2 * r * y ^ 4) +
2 * Real.exp 2 * r ^ 2 * y ^ 5 := by
exact add_le_add
(by
rw [norm_mul, hm_norm]
exact mul_le_mul_of_nonneg_left htFourth_norm_le (by positivity))
hsquare_diff
_ ≤ 10 * Real.exp 2 * r ^ 5 * y ^ 5 := by
have hr2_le_r5 : r ^ 2 ≤ r ^ 5 := by
calc
r ^ 2 = r ^ 2 * 1 := by ring
_ ≤ r ^ 2 * r ^ 3 :=
mul_le_mul_of_nonneg_left
(by
calc
(1 : ℝ) ≤ r := hr
_ ≤ r ^ 3 := by
nlinarith [hr, sq_nonneg r, show 0 ≤ r ^ 3 by positivity])
(by positivity)
_ = r ^ 5 := by ring
calc
(r * y) * (2 * Real.exp 2 * r * y ^ 4) +
2 * Real.exp 2 * r ^ 2 * y ^ 5
= 4 * Real.exp 2 * r ^ 2 * y ^ 5 := by ring
_ ≤ 10 * Real.exp 2 * r ^ 5 * y ^ 5 := by
have hnon : 0 ≤ Real.exp 2 * r ^ 2 * y ^ 5 := by positivity
calc
4 * Real.exp 2 * r ^ 2 * y ^ 5
≤ 10 * Real.exp 2 * r ^ 2 * y ^ 5 := by
nlinarith
_ ≤ 10 * Real.exp 2 * r ^ 5 * y ^ 5 := by
gcongr
rw [show bellLocalExponent r θ -
(-Complex.I * ((bellSaddleB3Real r : ℂ) / 6) * (θ : ℂ) ^ 3 +
((bellSaddleB4Real r : ℂ) / 24) * (θ : ℂ) ^ 4) =
bellLocalExponent r θ - P by rfl]
rw [hdiff]
rw [norm_mul]
have hexp_norm : ‖Complex.exp (r : ℂ)‖ = Real.exp r := by simp
rw [hexp_norm]
have hinside :
‖(outer - outerPoly) + (inner - innerPoly) + (cross - crossPoly)‖
≤ bellLocalTaylorConst * r ^ 5 * y ^ 5 := by
calc
‖(outer - outerPoly) + (inner - innerPoly) + (cross - crossPoly)‖
≤ ‖outer - outerPoly‖ + ‖inner - innerPoly‖ + ‖cross - crossPoly‖ := by
calc
‖(outer - outerPoly) + (inner - innerPoly) + (cross - crossPoly)‖
≤ ‖(outer - outerPoly) + (inner - innerPoly)‖ +
‖cross - crossPoly‖ := norm_add_le _ _
_ ≤ (‖outer - outerPoly‖ + ‖inner - innerPoly‖) +
‖cross - crossPoly‖ :=
add_le_add (norm_add_le _ _) (le_refl _)
_ = ‖outer - outerPoly‖ + ‖inner - innerPoly‖ +
‖cross - crossPoly‖ := by ring
_ ≤ 100 * Real.exp 2 * r ^ 5 * y ^ 5 +
Real.exp 2 * r ^ 5 * y ^ 5 +
10 * Real.exp 2 * r ^ 5 * y ^ 5 :=
add_le_add (add_le_add houter hinner) hcross
_ ≤ bellLocalTaylorConst * r ^ 5 * y ^ 5 := by
unfold bellLocalTaylorConst
calc
100 * Real.exp 2 * r ^ 5 * y ^ 5 +
Real.exp 2 * r ^ 5 * y ^ 5 +
10 * Real.exp 2 * r ^ 5 * y ^ 5
= 111 * (Real.exp 2 * r ^ 5 * y ^ 5) := by ring
_ ≤ 10000 * (Real.exp 2 * r ^ 5 * y ^ 5) := by
gcongr
norm_num
_ = 10000 * Real.exp 2 * r ^ 5 * y ^ 5 := by ring
calc
Real.exp r *
‖(outer - outerPoly) + (inner - innerPoly) + (cross - crossPoly)‖
≤ Real.exp r * (bellLocalTaylorConst * r ^ 5 * y ^ 5) :=
mul_le_mul_of_nonneg_left hinside (Real.exp_pos r).le
_ = bellLocalTaylorConst * Real.exp r * r ^ 5 * |θ| ^ 5 := by
dsimp [y]
ring
private lemma bellScaledRemainder_eq_unscaled {r x : ℝ}
(hb : 0 < bellSaddleBReal r) :
bellScaledRemainder r x =
bellLocalExponent r (x / Real.sqrt (bellSaddleBReal r)) -
(-Complex.I * ((bellSaddleB3Real r : ℂ) / 6) *
((x / Real.sqrt (bellSaddleBReal r) : ℝ) : ℂ) ^ 3 +
((bellSaddleB4Real r : ℂ) / 24) *
((x / Real.sqrt (bellSaddleBReal r) : ℝ) : ℂ) ^ 4) := by
have hsqrt_ne : (Real.sqrt (bellSaddleBReal r) : ℂ) ≠ 0 := by
exact Complex.ofReal_ne_zero.mpr (Real.sqrt_pos.mpr hb).ne'
have hb_ne : (bellSaddleBReal r : ℂ) ≠ 0 :=
Complex.ofReal_ne_zero.mpr hb.ne'
have hsqrt_sq := bell_sqrtB_sq hb
have hsqrt_pow4 :
(Real.sqrt (bellSaddleBReal r) : ℂ) ^ 4 =
(bellSaddleBReal r : ℂ) ^ 2 := by
calc
(Real.sqrt (bellSaddleBReal r) : ℂ) ^ 4 =
((Real.sqrt (bellSaddleBReal r) : ℂ) ^ 2) ^ 2 := by ring
_ = (bellSaddleBReal r : ℂ) ^ 2 := by rw [hsqrt_sq]
have hθ3 :
((x / Real.sqrt (bellSaddleBReal r) : ℝ) : ℂ) ^ 3 =
(x : ℂ) ^ 3 / (Real.sqrt (bellSaddleBReal r) : ℂ) ^ 3 := by
norm_num [Complex.ofReal_div]
ring
have hθ4 :
((x / Real.sqrt (bellSaddleBReal r) : ℝ) : ℂ) ^ 4 =
(x : ℂ) ^ 4 / (Real.sqrt (bellSaddleBReal r) : ℂ) ^ 4 := by
norm_num [Complex.ofReal_div]
ring
unfold bellScaledRemainder bellScaledCubic bellScaledQuartic
rw [hθ3, hθ4, hsqrt_pow4]
field_simp [hsqrt_ne, hb_ne]
ring
private lemma bellScaledRemainder_norm_le {r x : ℝ}
(hr : 1 ≤ r)
(hθδ : |x / Real.sqrt (bellSaddleBReal r)| ≤ bellSaddleDeltaReal r)
(hδle : bellSaddleDeltaReal r ≤ 1)
(hsmall : 2 * (r * bellSaddleDeltaReal r) ≤ 1) :
‖bellScaledRemainder r x‖ ≤
bellLocalTaylorConst * Real.exp (-(3 / 2 : ℝ) * r) * |x| ^ 5 := by
have hrpos : 0 < r := lt_of_lt_of_le zero_lt_one hr
have hbpos : 0 < bellSaddleBReal r := bellSaddleBReal_pos_of_ge_one hr
have hsqrt_pos : 0 < Real.sqrt (bellSaddleBReal r) := Real.sqrt_pos.mpr hbpos
have hsqrt_ge : r * Real.exp (r / 2) ≤ Real.sqrt (bellSaddleBReal r) :=
bell_sqrtB_ge_r_exp_half hr
have hθone : |x / Real.sqrt (bellSaddleBReal r)| ≤ 1 :=
hθδ.trans hδle
have hθsmall : 2 * (r * |x / Real.sqrt (bellSaddleBReal r)|) ≤ 1 := by
calc
2 * (r * |x / Real.sqrt (bellSaddleBReal r)|)
≤ 2 * (r * bellSaddleDeltaReal r) := by
gcongr
_ ≤ 1 := hsmall
have hrem := bellLocalExponent_fifth_remainder_norm_le
(r := r) (θ := x / Real.sqrt (bellSaddleBReal r)) hr hθone hθsmall
rw [bellScaledRemainder_eq_unscaled hbpos]
have hθ_abs_le :
|x / Real.sqrt (bellSaddleBReal r)| ≤
|x| / (r * Real.exp (r / 2)) := by
rw [abs_div, abs_of_pos hsqrt_pos]
exact div_le_div_of_nonneg_left (abs_nonneg x)
(by positivity : 0 < r * Real.exp (r / 2)) hsqrt_ge
have hθ5_le :
|x / Real.sqrt (bellSaddleBReal r)| ^ 5 ≤
(|x| / (r * Real.exp (r / 2))) ^ 5 :=
pow_le_pow_left₀ (abs_nonneg _) hθ_abs_le 5
calc
‖bellLocalExponent r (x / Real.sqrt (bellSaddleBReal r)) -
(-Complex.I * ((bellSaddleB3Real r : ℂ) / 6) *
((x / Real.sqrt (bellSaddleBReal r) : ℝ) : ℂ) ^ 3 +
((bellSaddleB4Real r : ℂ) / 24) *
((x / Real.sqrt (bellSaddleBReal r) : ℝ) : ℂ) ^ 4)‖
≤ bellLocalTaylorConst * Real.exp r * r ^ 5 *
|x / Real.sqrt (bellSaddleBReal r)| ^ 5 := hrem
_ ≤ bellLocalTaylorConst * Real.exp r * r ^ 5 *
(|x| / (r * Real.exp (r / 2))) ^ 5 := by
exact mul_le_mul_of_nonneg_left hθ5_le
(mul_nonneg
(mul_nonneg bellLocalTaylorConst_pos.le (Real.exp_pos r).le)
(pow_nonneg hrpos.le 5))
_ = bellLocalTaylorConst * Real.exp (-(3 / 2 : ℝ) * r) * |x| ^ 5 := by
have hr_ne : r ≠ 0 := hrpos.ne'
have hexp_ne : Real.exp (r / 2) ≠ 0 := (Real.exp_pos (r / 2)).ne'
field_simp [hr_ne, hexp_ne]
rw [show (Real.exp (r / 2)) ^ 5 = Real.exp (5 * (r / 2)) by
rw [← Real.exp_nat_mul]
norm_num]
have hexp_calc :
Real.exp (5 * (r / 2)) * Real.exp (-(r * 3 / 2)) = Real.exp r := by
rw [← Real.exp_add]
congr 1
ring
calc
bellLocalTaylorConst * Real.exp r * |x| ^ 5 =
bellLocalTaylorConst * |x| ^ 5 *
(Real.exp (5 * (r / 2)) * Real.exp (-(r * 3 / 2))) := by
rw [hexp_calc]
ring
_ = bellLocalTaylorConst * |x| ^ 5 * Real.exp (5 * (r / 2)) *
Real.exp (-(r * 3 / 2)) := by ring
private lemma complex_exp_second_error_bound (C Q R W : ℂ)
(hW : W = C + Q + R) (hWnorm : ‖W‖ ≤ 1) :
‖Complex.exp W - (1 + C + Q + C ^ 2 / 2)‖ ≤
Real.exp 1 * ‖W‖ ^ 3 + ‖R‖ +
(‖Q‖ + ‖R‖) * (2 * ‖C‖ + ‖Q‖ + ‖R‖) := by
have hsum :
(∑ m ∈ Finset.range 3, W ^ m / (m.factorial : ℂ)) =
1 + W + W ^ 2 / 2 := by
simp [Finset.sum_range_succ]
have htail :
‖Complex.exp W - (1 + W + W ^ 2 / 2)‖ ≤ Real.exp 1 * ‖W‖ ^ 3 := by
calc
‖Complex.exp W - (1 + W + W ^ 2 / 2)‖ =
‖Complex.exp W - ∑ m ∈ Finset.range 3, W ^ m / (m.factorial : ℂ)‖ := by
rw [hsum]
_ ≤ ‖W‖ ^ 3 * Real.exp ‖W‖ :=
Complex.norm_exp_sub_sum_le_norm_mul_exp W 3
_ ≤ ‖W‖ ^ 3 * Real.exp 1 :=
mul_le_mul_of_nonneg_left (Real.exp_le_exp.mpr hWnorm) (pow_nonneg (norm_nonneg W) 3)
_ = Real.exp 1 * ‖W‖ ^ 3 := by ring
have hdecomp :
Complex.exp W - (1 + C + Q + C ^ 2 / 2) =
(Complex.exp W - (1 + W + W ^ 2 / 2)) +
(R + (W ^ 2 - C ^ 2) / 2) := by
rw [hW]
ring
rw [hdecomp]
calc
‖(Complex.exp W - (1 + W + W ^ 2 / 2)) + (R + (W ^ 2 - C ^ 2) / 2)‖
≤ ‖Complex.exp W - (1 + W + W ^ 2 / 2)‖ +
‖R + (W ^ 2 - C ^ 2) / 2‖ := norm_add_le _ _
_ ≤ Real.exp 1 * ‖W‖ ^ 3 + (‖R‖ + ‖(W ^ 2 - C ^ 2) / 2‖) := by
exact add_le_add htail (norm_add_le _ _)
_ ≤ Real.exp 1 * ‖W‖ ^ 3 + (‖R‖ + ‖W ^ 2 - C ^ 2‖) := by
have hhalf : ‖(W ^ 2 - C ^ 2) / 2‖ ≤ ‖W ^ 2 - C ^ 2‖ := by
rw [norm_div, Complex.norm_ofNat]
nlinarith [norm_nonneg (W ^ 2 - C ^ 2)]
exact add_le_add (le_refl _) (add_le_add_right hhalf _)
_ ≤ Real.exp 1 * ‖W‖ ^ 3 + ‖R‖ +
(‖Q‖ + ‖R‖) * (2 * ‖C‖ + ‖Q‖ + ‖R‖) := by
have hsq :
W ^ 2 - C ^ 2 = (Q + R) * (2 * C + Q + R) := by
rw [hW]
ring
have hprod :
‖W ^ 2 - C ^ 2‖ ≤ (‖Q‖ + ‖R‖) * (2 * ‖C‖ + ‖Q‖ + ‖R‖) := by
rw [hsq]
calc
‖(Q + R) * (2 * C + Q + R)‖ =
‖Q + R‖ * ‖2 * C + Q + R‖ := by rw [norm_mul]
_ ≤ (‖Q‖ + ‖R‖) * (2 * ‖C‖ + ‖Q‖ + ‖R‖) := by
have hQR : ‖Q + R‖ ≤ ‖Q‖ + ‖R‖ := norm_add_le _ _
have h2C : ‖(2 : ℂ) * C‖ = 2 * ‖C‖ := by
rw [norm_mul, Complex.norm_ofNat]
have hCQR : ‖2 * C + Q + R‖ ≤ 2 * ‖C‖ + ‖Q‖ + ‖R‖ := by
calc
‖2 * C + Q + R‖ ≤ ‖2 * C + Q‖ + ‖R‖ := norm_add_le _ _
_ ≤ (‖2 * C‖ + ‖Q‖) + ‖R‖ :=
add_le_add (norm_add_le _ _) (le_refl _)
_ = 2 * ‖C‖ + ‖Q‖ + ‖R‖ := by rw [h2C]
exact mul_le_mul hQR hCQR (norm_nonneg _) (by positivity)
nlinarith [hprod]
private def bellGaussianDom (x : ℝ) : ℝ :=
Real.exp (-(x ^ 2) / 2) *
(|x| ^ 5 + |x| ^ 6 + |x| ^ 7 + |x| ^ 8 + |x| ^ 9 + |x| ^ 10 +
|x| ^ 11 + |x| ^ 12 + |x| ^ 13 + |x| ^ 14 + |x| ^ 15)
private lemma bellGaussianDom_nonneg (x : ℝ) : 0 ≤ bellGaussianDom x := by
unfold bellGaussianDom
positivity
private lemma gaussian_abs_monomial_integrable (k : ℕ) :
Integrable (fun x : ℝ => Real.exp (-(x ^ 2) / 2) * |x| ^ k) := by
have hk_nonneg : (0 : ℝ) ≤ (k : ℝ) := by exact_mod_cast (Nat.zero_le k)
have hk : (-1 : ℝ) < (k : ℝ) := by linarith
have hbase : Integrable (fun x : ℝ => x ^ k * Real.exp (-(1 / 2 : ℝ) * x ^ 2)) := by
simpa [Real.rpow_natCast] using
(integrable_rpow_mul_exp_neg_mul_sq (by norm_num : (0 : ℝ) < 1 / 2) hk)
have hbase2 : Integrable (fun x : ℝ => Real.exp (-(x ^ 2) / 2) * x ^ k) := by
convert hbase using 1
ext x
ring_nf
have hnorm := hbase2.norm
convert hnorm using 1
ext x
rw [Real.norm_eq_abs, abs_mul, abs_of_pos (Real.exp_pos _), abs_pow]
private lemma bellGaussianDom_integrable : Integrable bellGaussianDom := by
let g : ℕ → ℝ → ℝ := fun k x => Real.exp (-(x ^ 2) / 2) * |x| ^ k
have hsum :
Integrable (fun x : ℝ =>
g 5 x + g 6 x + g 7 x + g 8 x + g 9 x + g 10 x +
g 11 x + g 12 x + g 13 x + g 14 x + g 15 x) := by
simpa [g, add_assoc] using
((((((((((gaussian_abs_monomial_integrable 5).add
(gaussian_abs_monomial_integrable 6)).add
(gaussian_abs_monomial_integrable 7)).add
(gaussian_abs_monomial_integrable 8)).add
(gaussian_abs_monomial_integrable 9)).add
(gaussian_abs_monomial_integrable 10)).add
(gaussian_abs_monomial_integrable 11)).add
(gaussian_abs_monomial_integrable 12)).add
(gaussian_abs_monomial_integrable 13)).add
(gaussian_abs_monomial_integrable 14)).add
(gaussian_abs_monomial_integrable 15)
convert hsum using 1
ext x
unfold bellGaussianDom
dsimp [g]
ring_nf
private lemma bellGaussianDom_integral_nonneg :
0 ≤ ∫ x : ℝ, bellGaussianDom x := by
exact integral_nonneg (fun x => bellGaussianDom_nonneg x)
private def bellLocalL1Const : ℝ :=
27 * Real.exp 1 * (20 + bellLocalTaylorConst) ^ 3 + bellLocalTaylorConst +
6 * (10 + bellLocalTaylorConst) * (30 + bellLocalTaylorConst)
private lemma bellLocalL1Const_pos : 0 < bellLocalL1Const := by
unfold bellLocalL1Const bellLocalTaylorConst
positivity
private lemma bell_local_integrand_bound {r x : ℝ}
(hr : 1 ≤ r)
(hθδ : |x / Real.sqrt (bellSaddleBReal r)| ≤ bellSaddleDeltaReal r)
(hδle : bellSaddleDeltaReal r ≤ 1)
(hsmall : 2 * (r * bellSaddleDeltaReal r) ≤ 1)
(hWnorm : ‖bellLocalExponent r (x / Real.sqrt (bellSaddleBReal r))‖ ≤ 1) :
‖Complex.exp (-(x ^ 2) / 2) *
(saddleLocalRatio (fun z : ℂ => Complex.exp (Complex.exp z - 1))
bellSaddleAReal bellSaddleBReal r
(x / Real.sqrt (bellSaddleBReal r)) -
saddleSecondPoly bellSaddleBReal bellSaddleB3Real bellSaddleB4Real r x)‖
≤ bellLocalL1Const * Real.exp (-(3 / 2 : ℝ) * r) * bellGaussianDom x := by
have hrpos : 0 < r := lt_of_lt_of_le zero_lt_one hr
have hr0 : 0 ≤ r := hrpos.le
have hbpos : 0 < bellSaddleBReal r := bellSaddleBReal_pos_of_ge_one hr
let C : ℂ := bellScaledCubic r x
let Q : ℂ := bellScaledQuartic r x
let R : ℂ := bellScaledRemainder r x
let W : ℂ := bellLocalExponent r (x / Real.sqrt (bellSaddleBReal r))
let y : ℝ := |x|
let A : ℝ := Real.exp (-(r / 2))
let B : ℝ := Real.exp (-r)
let D : ℝ := Real.exp (-(3 / 2 : ℝ) * r)
let K : ℝ := bellLocalTaylorConst
let S : ℝ := 10 * y ^ 3 + 10 * y ^ 4 + K * y ^ 5
let T : ℝ := (10 * y ^ 4 + K * y ^ 5) * (20 * y ^ 3 + 10 * y ^ 4 + K * y ^ 5)
let G : ℝ := y ^ 5 + y ^ 6 + y ^ 7 + y ^ 8 + y ^ 9 + y ^ 10 +
y ^ 11 + y ^ 12 + y ^ 13 + y ^ 14 + y ^ 15
have hK_nonneg : 0 ≤ K := by
dsimp [K]
exact bellLocalTaylorConst_pos.le
have hy0 : 0 ≤ y := by
dsimp [y]
exact abs_nonneg x
have hA_nonneg : 0 ≤ A := by dsimp [A]; positivity
have hB_nonneg : 0 ≤ B := by dsimp [B]; positivity
have hD_nonneg : 0 ≤ D := by dsimp [D]; positivity
have hB_le_A : B ≤ A := by
dsimp [A, B]
exact Real.exp_le_exp.mpr (by nlinarith [hr0])
have hD_le_A : D ≤ A := by
dsimp [A, D]
exact Real.exp_le_exp.mpr (by nlinarith [hr0])
have hD_le_B : D ≤ B := by
dsimp [B, D]
exact Real.exp_le_exp.mpr (by nlinarith [hr0])
have hA_mul_B : A * B = D := by
dsimp [A, B, D]
rw [← Real.exp_add]
congr 1
ring
have hA_cube : A ^ 3 = D := by
dsimp [A, D]
rw [show (Real.exp (-(r / 2))) ^ 3 = Real.exp (3 * (-(r / 2))) by
rw [← Real.exp_nat_mul]
norm_num]
congr 1
ring
have hWsplit : W = C + Q + R := by
dsimp [W, C, Q, R]
exact bellLocalExponent_scaled_split r x
have hP :
saddleSecondPoly bellSaddleBReal bellSaddleB3Real bellSaddleB4Real r x =
1 + C + Q + C ^ 2 / 2 := by
dsimp [C, Q]
exact bell_saddleSecondPoly_eq_scaled_terms hbpos
have herr := complex_exp_second_error_bound C Q R W hWsplit (by simpa [W] using hWnorm)
have hC : ‖C‖ ≤ 10 * A * y ^ 3 := by
dsimp [C, A, y]
simpa [mul_assoc] using bellScaledCubic_norm_le (r := r) (x := x) hr
have hQ_B : ‖Q‖ ≤ 10 * B * y ^ 4 := by
dsimp [Q, B, y]
simpa [mul_assoc] using bellScaledQuartic_norm_le (r := r) (x := x) hr
have hQ_A : ‖Q‖ ≤ 10 * A * y ^ 4 := by
exact hQ_B.trans (by gcongr)
have hR_D : ‖R‖ ≤ K * D * y ^ 5 := by
dsimp [R, K, D, y]
simpa [mul_assoc] using
bellScaledRemainder_norm_le (r := r) (x := x) hr hθδ hδle hsmall
have hR_A : ‖R‖ ≤ K * A * y ^ 5 := by
exact hR_D.trans (by gcongr)
have hR_B : ‖R‖ ≤ K * B * y ^ 5 := by
exact hR_D.trans (by gcongr)
have hWpoly : ‖W‖ ≤ A * S := by
calc
‖W‖ = ‖C + Q + R‖ := by rw [hWsplit]
_ ≤ ‖C‖ + ‖Q‖ + ‖R‖ := by
calc
‖C + Q + R‖ ≤ ‖C + Q‖ + ‖R‖ := norm_add_le _ _
_ ≤ (‖C‖ + ‖Q‖) + ‖R‖ := add_le_add (norm_add_le _ _) (le_refl _)
_ = ‖C‖ + ‖Q‖ + ‖R‖ := by ring
_ ≤ 10 * A * y ^ 3 + 10 * A * y ^ 4 + K * A * y ^ 5 :=
add_le_add (add_le_add hC hQ_A) hR_A
_ = A * S := by
dsimp [S]
ring
have hWcube :
Real.exp 1 * ‖W‖ ^ 3 ≤ Real.exp 1 * D * S ^ 3 := by
calc
Real.exp 1 * ‖W‖ ^ 3 ≤ Real.exp 1 * (A * S) ^ 3 := by
exact mul_le_mul_of_nonneg_left
(pow_le_pow_left₀ (norm_nonneg W) hWpoly 3) (Real.exp_pos 1).le
_ = Real.exp 1 * D * S ^ 3 := by
rw [mul_pow, hA_cube]
ring
have hQR : ‖Q‖ + ‖R‖ ≤ B * (10 * y ^ 4 + K * y ^ 5) := by
calc
‖Q‖ + ‖R‖ ≤ 10 * B * y ^ 4 + K * B * y ^ 5 := add_le_add hQ_B hR_B
_ = B * (10 * y ^ 4 + K * y ^ 5) := by ring
have hCQR : 2 * ‖C‖ + ‖Q‖ + ‖R‖ ≤ A * (20 * y ^ 3 + 10 * y ^ 4 + K * y ^ 5) := by
calc
2 * ‖C‖ + ‖Q‖ + ‖R‖
≤ 2 * (10 * A * y ^ 3) + 10 * A * y ^ 4 + K * A * y ^ 5 := by
exact add_le_add (add_le_add
(mul_le_mul_of_nonneg_left hC (by norm_num : (0 : ℝ) ≤ 2)) hQ_A) hR_A
_ = A * (20 * y ^ 3 + 10 * y ^ 4 + K * y ^ 5) := by ring
have hprod :
(‖Q‖ + ‖R‖) * (2 * ‖C‖ + ‖Q‖ + ‖R‖) ≤ D * T := by
calc
(‖Q‖ + ‖R‖) * (2 * ‖C‖ + ‖Q‖ + ‖R‖)
≤ (B * (10 * y ^ 4 + K * y ^ 5)) *
(A * (20 * y ^ 3 + 10 * y ^ 4 + K * y ^ 5)) := by
have hupper_nonneg : 0 ≤ B * (10 * y ^ 4 + K * y ^ 5) := by
positivity
exact mul_le_mul hQR hCQR (by positivity) hupper_nonneg
_ = D * T := by
dsimp [T]
rw [show B * (10 * y ^ 4 + K * y ^ 5) *
(A * (20 * y ^ 3 + 10 * y ^ 4 + K * y ^ 5)) =
(A * B) * ((10 * y ^ 4 + K * y ^ 5) *
(20 * y ^ 3 + 10 * y ^ 4 + K * y ^ 5)) by ring]
rw [hA_mul_B]
have hpoly :
Real.exp 1 * S ^ 3 + K * y ^ 5 + T ≤ bellLocalL1Const * G := by
have hK20_nonneg : 0 ≤ 20 + K := by positivity
have hK10_nonneg : 0 ≤ 10 + K := by positivity
have hK30_nonneg : 0 ≤ 30 + K := by positivity
have hY345_nonneg : 0 ≤ y ^ 3 + y ^ 4 + y ^ 5 := by positivity
have hY45_nonneg : 0 ≤ y ^ 4 + y ^ 5 := by positivity
have hG_nonneg : 0 ≤ G := by
dsimp [G]
positivity
have hy5_le_G : y ^ 5 ≤ G := by
dsimp [G]
nlinarith [pow_nonneg hy0 6, pow_nonneg hy0 7, pow_nonneg hy0 8,
pow_nonneg hy0 9, pow_nonneg hy0 10, pow_nonneg hy0 11,
pow_nonneg hy0 12, pow_nonneg hy0 13, pow_nonneg hy0 14,
pow_nonneg hy0 15]
have hcube_support :
(y ^ 3 + y ^ 4 + y ^ 5) ^ 3 ≤ 27 * G := by
dsimp [G]
ring_nf
nlinarith [pow_nonneg hy0 5, pow_nonneg hy0 6, pow_nonneg hy0 7,
pow_nonneg hy0 8, pow_nonneg hy0 9, pow_nonneg hy0 10,
pow_nonneg hy0 11, pow_nonneg hy0 12, pow_nonneg hy0 13,
pow_nonneg hy0 14, pow_nonneg hy0 15]
have hprod_support :
(y ^ 4 + y ^ 5) * (y ^ 3 + y ^ 4 + y ^ 5) ≤ 6 * G := by
dsimp [G]
ring_nf
nlinarith [pow_nonneg hy0 5, pow_nonneg hy0 6, pow_nonneg hy0 7,
pow_nonneg hy0 8, pow_nonneg hy0 9, pow_nonneg hy0 10]
have hS_le : S ≤ (20 + K) * (y ^ 3 + y ^ 4 + y ^ 5) := by
dsimp [S]
calc
10 * y ^ 3 + 10 * y ^ 4 + K * y ^ 5
≤ (20 + K) * y ^ 3 + (20 + K) * y ^ 4 + (20 + K) * y ^ 5 := by
gcongr
· linarith
· linarith
· linarith
_ = (20 + K) * (y ^ 3 + y ^ 4 + y ^ 5) := by ring
have hS_cube :
S ^ 3 ≤ (20 + K) ^ 3 * (27 * G) := by
calc
S ^ 3 ≤ ((20 + K) * (y ^ 3 + y ^ 4 + y ^ 5)) ^ 3 :=
pow_le_pow_left₀ (by positivity) hS_le 3
_ = (20 + K) ^ 3 * (y ^ 3 + y ^ 4 + y ^ 5) ^ 3 := by ring
_ ≤ (20 + K) ^ 3 * (27 * G) :=
mul_le_mul_of_nonneg_left hcube_support (pow_nonneg hK20_nonneg 3)
have hT_le :
T ≤ (10 + K) * (30 + K) * (6 * G) := by
have hleft : 10 * y ^ 4 + K * y ^ 5 ≤ (10 + K) * (y ^ 4 + y ^ 5) := by
calc
10 * y ^ 4 + K * y ^ 5 ≤ (10 + K) * y ^ 4 + (10 + K) * y ^ 5 := by
gcongr
· linarith
· linarith
_ = (10 + K) * (y ^ 4 + y ^ 5) := by ring
have hright :
20 * y ^ 3 + 10 * y ^ 4 + K * y ^ 5 ≤
(30 + K) * (y ^ 3 + y ^ 4 + y ^ 5) := by
calc
20 * y ^ 3 + 10 * y ^ 4 + K * y ^ 5
≤ (30 + K) * y ^ 3 + (30 + K) * y ^ 4 + (30 + K) * y ^ 5 := by
gcongr
· linarith
· linarith
· linarith
_ = (30 + K) * (y ^ 3 + y ^ 4 + y ^ 5) := by ring
calc
T ≤ ((10 + K) * (y ^ 4 + y ^ 5)) *
((30 + K) * (y ^ 3 + y ^ 4 + y ^ 5)) := by
dsimp [T]
exact mul_le_mul hleft hright (by positivity) (by positivity)
_ = (10 + K) * (30 + K) *
((y ^ 4 + y ^ 5) * (y ^ 3 + y ^ 4 + y ^ 5)) := by ring
_ ≤ (10 + K) * (30 + K) * (6 * G) := by
gcongr
calc
Real.exp 1 * S ^ 3 + K * y ^ 5 + T
≤ Real.exp 1 * ((20 + K) ^ 3 * (27 * G)) + K * G +
(10 + K) * (30 + K) * (6 * G) := by
exact add_le_add (add_le_add
(mul_le_mul_of_nonneg_left hS_cube (Real.exp_pos 1).le)
(mul_le_mul_of_nonneg_left hy5_le_G hK_nonneg)) hT_le
_ = (27 * Real.exp 1 * (20 + K) ^ 3 + K +
6 * (10 + K) * (30 + K)) * G := by ring
_ = bellLocalL1Const * G := by
dsimp [K]
unfold bellLocalL1Const
ring
have hdiff :
‖Complex.exp W - (1 + C + Q + C ^ 2 / 2)‖ ≤
bellLocalL1Const * D * G := by
calc
‖Complex.exp W - (1 + C + Q + C ^ 2 / 2)‖
≤ Real.exp 1 * ‖W‖ ^ 3 + ‖R‖ +
(‖Q‖ + ‖R‖) * (2 * ‖C‖ + ‖Q‖ + ‖R‖) := herr
_ ≤ Real.exp 1 * D * S ^ 3 + K * D * y ^ 5 + D * T := by
exact add_le_add (add_le_add hWcube hR_D) hprod
_ = D * (Real.exp 1 * S ^ 3 + K * y ^ 5 + T) := by ring
_ ≤ D * (bellLocalL1Const * G) :=
mul_le_mul_of_nonneg_left hpoly hD_nonneg
_ = bellLocalL1Const * D * G := by ring
rw [bell_saddleLocalRatio_eq]
rw [hP]
change
‖Complex.exp (-(x ^ 2) / 2) *
(Complex.exp W - (1 + C + Q + C ^ 2 / 2))‖ ≤
bellLocalL1Const * D * bellGaussianDom x
rw [norm_mul]
have hgauss :
‖Complex.exp (-(x ^ 2) / 2)‖ = Real.exp (-(x ^ 2) / 2) := by
rw [Complex.norm_exp]
congr 1
simp [pow_two]
rw [hgauss]
unfold bellGaussianDom
dsimp [D, G, y]
calc
Real.exp (-(x ^ 2) / 2) *
‖Complex.exp W - (1 + C + Q + C ^ 2 / 2)‖
≤ Real.exp (-(x ^ 2) / 2) * (bellLocalL1Const * D * G) :=
mul_le_mul_of_nonneg_left hdiff (Real.exp_pos _).le
_ = bellLocalL1Const * D *
(Real.exp (-(x ^ 2) / 2) *
(|x| ^ 5 + |x| ^ 6 + |x| ^ 7 + |x| ^ 8 + |x| ^ 9 + |x| ^ 10 +
|x| ^ 11 + |x| ^ 12 + |x| ^ 13 + |x| ^ 14 + |x| ^ 15)) := by
ring
private lemma bell_local_integrand_continuous (r : ℝ) :
Continuous fun x : ℝ =>
‖Complex.exp (-(x ^ 2) / 2) *
(saddleLocalRatio (fun z : ℂ => Complex.exp (Complex.exp z - 1))
bellSaddleAReal bellSaddleBReal r
(x / Real.sqrt (bellSaddleBReal r)) -
saddleSecondPoly bellSaddleBReal bellSaddleB3Real bellSaddleB4Real r x)‖ := by
have h :
Continuous fun x : ℝ =>
‖Complex.exp (-(x ^ 2) / 2) *
(Complex.exp (bellLocalExponent r (x / Real.sqrt (bellSaddleBReal r))) -
saddleSecondPoly bellSaddleBReal bellSaddleB3Real bellSaddleB4Real r x)‖ := by
unfold bellLocalExponent saddleSecondPoly
fun_prop
simpa [bell_saddleLocalRatio_eq] using h
private lemma bellGaussianDom_continuous : Continuous bellGaussianDom := by
unfold bellGaussianDom
fun_prop
private lemma bellGaussianDom_window_le_integral {L : ℝ} (hL : 0 ≤ L) :
(∫ x in -L..L, bellGaussianDom x) ≤ ∫ x : ℝ, bellGaussianDom x := by
have hle : -L ≤ L := by linarith
rw [intervalIntegral.integral_of_le hle]
exact setIntegral_le_integral bellGaussianDom_integrable
(Eventually.of_forall bellGaussianDom_nonneg)
/--
Bell-specific hard local estimate: the fifth-order expansion of
`exp (exp (r e^{iθ}) - exp r - i a(r)θ + b(r)θ²/2)` after the scaling
`θ = x / sqrt (b(r))`, with the cubic, quartic, and sextic Gaussian correction
subtracted, is `L¹-o(bellSecondCorrScale r)`.
This is the only intentionally isolated analytic gap in this file. It is the
same local Taylor/Gaussian-domination estimate as the involution proof, but with
the nested exponential Bell saddle algebra.
-/
theorem bell_local_second_L1 :
Tendsto
(fun r : ℝ =>
(∫ x in -(bellSaddleDeltaReal r * Real.sqrt (bellSaddleBReal r))..
(bellSaddleDeltaReal r * Real.sqrt (bellSaddleBReal r)),
‖Complex.exp (-(x ^ 2) / 2) *
(saddleLocalRatio (fun z : ℂ => Complex.exp (Complex.exp z - 1))
bellSaddleAReal bellSaddleBReal r
(x / Real.sqrt (bellSaddleBReal r)) -
saddleSecondPoly bellSaddleBReal bellSaddleB3Real bellSaddleB4Real r x)‖) /
bellSecondCorrScale r)
atTop (𝓝 0) := by
let M : ℝ := ∫ x : ℝ, bellGaussianDom x
have hM_nonneg : 0 ≤ M := by
dsimp [M]
exact bellGaussianDom_integral_nonneg
have hsmall_u :
∀ᶠ r : ℝ in atTop, 2 * (r * bellSaddleDeltaReal r) ≤ 1 := by
have ht : Tendsto (fun r : ℝ => 2 * (r * bellSaddleDeltaReal r)) atTop (𝓝 0) := by
simpa using (bell_r_delta_tendsto_zero.const_mul (2 : ℝ))
exact ht.eventually_le_const zero_lt_one
have hupper_tendsto :
Tendsto
(fun r : ℝ =>
8 * bellLocalL1Const * M * Real.exp (-(1 / 2 : ℝ) * r))
atTop (𝓝 0) := by
have hlin : Tendsto (fun r : ℝ => (-(1 / 2 : ℝ)) * r) atTop atBot :=
Tendsto.const_mul_atTop_of_neg (by norm_num : (-(1 / 2 : ℝ)) < 0) tendsto_id
have hexp : Tendsto (fun r : ℝ => Real.exp ((-(1 / 2 : ℝ)) * r)) atTop (𝓝 0) :=
Real.tendsto_exp_atBot.comp hlin
simpa [mul_assoc] using hexp.const_mul (8 * bellLocalL1Const * M)
refine squeeze_zero' ?_ ?_ hupper_tendsto
· filter_upwards [eventually_ge_atTop (1 : ℝ), bellSecondCorrScale_pos_eventually] with
r hr hcorr
let L : ℝ := bellSaddleDeltaReal r * Real.sqrt (bellSaddleBReal r)
have hLnonneg : 0 ≤ L := by
dsimp [L, bellSaddleDeltaReal]
positivity
have hle : -L ≤ L := by linarith
have hint_nonneg :
0 ≤ ∫ x in -L..L,
‖Complex.exp (-(x ^ 2) / 2) *
(saddleLocalRatio (fun z : ℂ => Complex.exp (Complex.exp z - 1))
bellSaddleAReal bellSaddleBReal r
(x / Real.sqrt (bellSaddleBReal r)) -
saddleSecondPoly bellSaddleBReal bellSaddleB3Real bellSaddleB4Real r x)‖ :=
intervalIntegral.integral_nonneg hle (fun _ _ => norm_nonneg _)
exact div_nonneg hint_nonneg hcorr.le
· filter_upwards [eventually_ge_atTop (1 : ℝ), bell_delta_le_one_eventually,
bell_local_exponent_norm_le_eventually,
bellLocalBound_tendsto_zero.eventually_le_const zero_lt_one,
hsmall_u, bellSecondCorrScale_lower_exp_eventually,
bellSecondCorrScale_pos_eventually] with
r hr hδle hloc hlocSmall hsmall hcorrLower hcorrPos
let L : ℝ := bellSaddleDeltaReal r * Real.sqrt (bellSaddleBReal r)
let F : ℝ → ℝ := fun x =>
‖Complex.exp (-(x ^ 2) / 2) *
(saddleLocalRatio (fun z : ℂ => Complex.exp (Complex.exp z - 1))
bellSaddleAReal bellSaddleBReal r
(x / Real.sqrt (bellSaddleBReal r)) -
saddleSecondPoly bellSaddleBReal bellSaddleB3Real bellSaddleB4Real r x)‖
let H : ℝ → ℝ := fun x =>
(bellLocalL1Const * Real.exp (-(3 / 2 : ℝ) * r)) * bellGaussianDom x
have hrpos : 0 < r := lt_of_lt_of_le zero_lt_one hr
have hbpos : 0 < bellSaddleBReal r := bellSaddleBReal_pos_of_ge_one hr
have hsqrt_pos : 0 < Real.sqrt (bellSaddleBReal r) := Real.sqrt_pos.mpr hbpos
have hδnonneg : 0 ≤ bellSaddleDeltaReal r := by
unfold bellSaddleDeltaReal
positivity
have hLnonneg : 0 ≤ L := by
dsimp [L]
positivity
have hle : -L ≤ L := by linarith
have hpoint : ∀ x ∈ Set.Icc (-L) L, F x ≤ H x := by
intro x hx
have hxabs : |x| ≤ L := by
exact abs_le.mpr ⟨by linarith [hx.1], hx.2⟩
have hθδ : |x / Real.sqrt (bellSaddleBReal r)| ≤ bellSaddleDeltaReal r := by
rw [abs_div, abs_of_pos hsqrt_pos]
calc
|x| / Real.sqrt (bellSaddleBReal r) ≤
L / Real.sqrt (bellSaddleBReal r) :=
div_le_div_of_nonneg_right hxabs hsqrt_pos.le
_ = bellSaddleDeltaReal r := by
dsimp [L]
field_simp [hsqrt_pos.ne']
have hWnorm :
‖bellLocalExponent r (x / Real.sqrt (bellSaddleBReal r))‖ ≤ 1 :=
(hloc (x / Real.sqrt (bellSaddleBReal r)) hθδ).trans hlocSmall
have hb := bell_local_integrand_bound hr hθδ hδle hsmall hWnorm
dsimp [F, H]
exact hb
have hFint : IntervalIntegrable F volume (-L) L := by
exact (bell_local_integrand_continuous r).intervalIntegrable _ _
have hHint : IntervalIntegrable H volume (-L) L := by
exact (bellGaussianDom_continuous.const_mul
(bellLocalL1Const * Real.exp (-(3 / 2 : ℝ) * r))).intervalIntegrable _ _
have hIntBound :
(∫ x in -L..L, F x) ≤
(bellLocalL1Const * Real.exp (-(3 / 2 : ℝ) * r)) * M := by
have hconst_nonneg :
0 ≤ bellLocalL1Const * Real.exp (-(3 / 2 : ℝ) * r) :=
mul_nonneg bellLocalL1Const_pos.le (Real.exp_pos _).le
calc
(∫ x in -L..L, F x) ≤ ∫ x in -L..L, H x :=
intervalIntegral.integral_mono_on hle hFint hHint hpoint
_ = (bellLocalL1Const * Real.exp (-(3 / 2 : ℝ) * r)) *
(∫ x in -L..L, bellGaussianDom x) := by
dsimp [H]
rw [intervalIntegral.integral_const_mul]
_ ≤ (bellLocalL1Const * Real.exp (-(3 / 2 : ℝ) * r)) * M := by
exact mul_le_mul_of_nonneg_left
(by dsimp [M]; exact bellGaussianDom_window_le_integral hLnonneg)
hconst_nonneg
have hnum_nonneg :
0 ≤ (bellLocalL1Const * Real.exp (-(3 / 2 : ℝ) * r)) * M :=
mul_nonneg
(mul_nonneg bellLocalL1Const_pos.le (Real.exp_pos _).le) hM_nonneg
have hscale_pos : 0 < (1 / 8 : ℝ) * Real.exp (-r) := by positivity
have hmain :
(∫ x in -L..L, F x) / bellSecondCorrScale r ≤
((bellLocalL1Const * Real.exp (-(3 / 2 : ℝ) * r)) * M) /
((1 / 8 : ℝ) * Real.exp (-r)) := by
calc
(∫ x in -L..L, F x) / bellSecondCorrScale r
≤ ((bellLocalL1Const * Real.exp (-(3 / 2 : ℝ) * r)) * M) /
bellSecondCorrScale r :=
div_le_div_of_nonneg_right hIntBound hcorrPos.le
_ ≤ ((bellLocalL1Const * Real.exp (-(3 / 2 : ℝ) * r)) * M) /
((1 / 8 : ℝ) * Real.exp (-r)) :=
div_le_div_of_nonneg_left hnum_nonneg hscale_pos hcorrLower
calc
((∫ x in -(bellSaddleDeltaReal r * Real.sqrt (bellSaddleBReal r))..
(bellSaddleDeltaReal r * Real.sqrt (bellSaddleBReal r)),
‖Complex.exp (-(x ^ 2) / 2) *
(saddleLocalRatio (fun z : ℂ => Complex.exp (Complex.exp z - 1))
bellSaddleAReal bellSaddleBReal r
(x / Real.sqrt (bellSaddleBReal r)) -
saddleSecondPoly bellSaddleBReal bellSaddleB3Real bellSaddleB4Real r x)‖) /
bellSecondCorrScale r)
= (∫ x in -L..L, F x) / bellSecondCorrScale r := by rfl
_ ≤ ((bellLocalL1Const * Real.exp (-(3 / 2 : ℝ) * r)) * M) /
((1 / 8 : ℝ) * Real.exp (-r)) := hmain
_ = 8 * bellLocalL1Const * M * Real.exp (-(1 / 2 : ℝ) * r) := by
have hexp_ne : Real.exp (-r) ≠ 0 := (Real.exp_pos (-r)).ne'
field_simp [hexp_ne]
have hexp_calc' :
Real.exp (-(3 * r / 2)) =
Real.exp (-(r / 2)) * Real.exp (-r) := by
rw [← Real.exp_add]
congr 1
ring
rw [hexp_calc']
ring
/-- The produced second-order H-admissible Bell instance. -/
def bellSecondOrderHAdmissible : SecondOrderHAdmissible bellHAdmissible where
b3 := bellSaddleB3Real
b4 := bellSaddleB4Real
corrScale := bellSecondCorrScale
corrScale_pos := by
simpa [bellHAdmissible] using bellSecondCorrScale_pos_eventually
corrScale_tendsto_zero := by
simpa [bellHAdmissible] using bellSecondCorrScale_tendsto_zero
corrScale_dominates_correction := by
refine ⟨1, by norm_num, ?_⟩
exact Eventually.of_forall fun r => by
simp [bellHAdmissible, bellSecondCorrScale]
local_second_L1 := by
simpa [bellHAdmissible] using bell_local_second_L1
tail_second_uniform := by
simpa [bellHAdmissible] using bell_tail_second_uniform
Source