Model tail
ProvedAnalyticCombinatorics.Ch8.Partitions.Erdos.model_tail_leanalytic-combinatoricslean4partitions-remaindersmooth-driftsupporting-lemma
Model tail (§5 brick 5): the m > ⌊n^{2/3}⌋ tail of |modelSummand| is O(1/n). The three
moment tails are exp-small (sigma_geom_tail_le + sharp #119); the surviving exp(−(λ/4)n^{1/6})
beats 1/n (poly_mul_exp_neg_sixthRoot_le_inv).
The full literal source contract is:
theorem model_tail_le :
∃ K : ℝ, 0 < K ∧ ∀ᶠ n : ℕ in Filter.atTop,
(∑' m : ℕ, (if m ≤ ⌊(n:ℝ) ^ (2/3 : ℝ)⌋₊ then (0:ℝ) else |modelSummand n m|))
≤ K / (n:ℝ)
Preamble
import Definitions.Def_AnalyticCombinatorics_Ch8_Partitions_SigmaRecurrence
import Definitions.Def_AnalyticCombinatorics_Ch8_Partitions_UpperBound
import Definitions.Def_AnalyticCombinatorics_Ch8_Partitions_ErdosKernel
import Definitions.Def_AnalyticCombinatorics_Ch8_Partitions_MassRateMoments
import Definitions.Def_AnalyticCombinatorics_Ch8_Partitions_MassRateApprox2
import Mathlib
namespace AnalyticCombinatorics.Ch8.Partitions.Erdos.Close end AnalyticCombinatorics.Ch8.Partitions.Erdos.Close
set_option autoImplicit true
set_option relaxedAutoImplicit false
set_option maxSynthPendingDepth 3
/-!
# Mass-rate campaign: §5 the second-order divisor calc
`kernelMass n = ∑_{m=1}^{n-1} σ(m)/(n−m)·exp(−C(√n−√(n−m)))` versus
`kernelMassApprox2 n = (1/n)M₀(t)+(1/n²)M₁(t)−(C/(8n²√n))M₂(t)` at `t=λ/√n`.
Brick 1 here: `kernelMassApprox2 n = ∑' m, modelSummand n m`, where
`modelSummand n m = σ(m)·e^{−tm}·(1/n + m/n² − Cm²/(8n²√n))` — i.e. the moment package collapses
to a single divisor-weighted Lambert sum. This is the object the per-term estimate
`erdosWeight_coef_second_order` (#97), multiplied by `σ(m)≥0`, is compared against. Opus-authored.
-/
set_option maxHeartbeats 1000000
open Filter
open scoped Topology BigOperators
open AnalyticCombinatorics AnalyticCombinatorics.Ch8 AnalyticCombinatorics.Ch8.Partitions AnalyticCombinatorics.Ch8.Partitions.Erdos
open AnalyticCombinatorics.Ch8.Partitions.Erdos.Close
Formal statement
theorem AnalyticCombinatorics.Ch8.Partitions.Erdos.model_tail_le :
∃ K : ℝ, 0 < K ∧ ∀ᶠ n : ℕ in Filter.atTop,
(∑' m : ℕ, (if m ≤ ⌊(n:ℝ) ^ (2/3 : ℝ)⌋₊ then (0:ℝ) else |modelSummand n m|))
≤ K / (n:ℝ) := by sorrySource