A conditional truncated-kernel heat transfer bound
ProvedAnalyticCombinatorics.Ch8.Partitions.Erdos.truncated_block_heat_transferanalytic-combinatoricsfinite-kernel-formslean4partitions-remainder
On a finite type with decidable equality, let be nonnegative row-stochastic kernels and a nonnegative unit-mass distribution. Assume the rowwise bound , the stated truncated heat lower bound , and . Then
Every kernel, horizon and loss factor is retained; no producer for the truncated estimate is asserted.
This support estimate carries the following complete literal source contract:
theorem truncated_block_heat_transfer
(K Ktr : α → α → ℝ) (μ0 : α → ℝ) (D : α → ℝ)
(W c0 ε : ℝ) (M : ℕ)
(hKnn : ∀ x z, 0 ≤ K x z) (hKsum : ∀ x, ∑ z, K x z = 1)
(hKtrnn : ∀ x z, 0 ≤ Ktr x z) (hKtrsum : ∀ x, ∑ z, Ktr x z = 1)
(hμ0nn : ∀ x, 0 ≤ μ0 x) (hμ0sum : ∑ x, μ0 x = 1)
(hclose : ∀ x, (∑ z, |K x z - Ktr x z|) ≤ ε)
(htrunc :
c0 / Real.sqrt ((M : ℝ) + 1)
≤ ∑ t ∈ Finset.range M,
∑ x, distPow Ktr μ0 t x * (if |D x| ≤ W then (1 : ℝ) else 0))
(herr :
(∑ t ∈ Finset.range M, (t : ℝ)) * ε
≤ c0 / (2 * Real.sqrt ((M : ℝ) + 1))) :
c0 / (2 * Real.sqrt ((M : ℝ) + 1))
≤ ∑ t ∈ Finset.range M,
∑ x, distPow K μ0 t x * (if |D x| ≤ W then (1 : ℝ) else 0)
Preamble
import Definitions.Def_AnalyticCombinatorics_Ch8_Partitions_TanakaTelescope
import Mathlib
set_option autoImplicit true
set_option relaxedAutoImplicit false
set_option maxSynthPendingDepth 3
/-!
# Truncation transfer (renewal route B, Layer-2 steps A & C)
The Erdős rank-difference kernel is soft/nonlocal, but its rank-drop tail is *exponential*
(`RankDropKer.Pker_rankDrop_tail_majorant`, `γ = C/60`). The Layer-2 block-heat lower bound is
therefore proved for a *truncated* kernel (jumps capped at `L_m ≍ log m`, mass renormalized) and
transferred back to the real kernel via the banked `OccupationTransfer.occupation_transfer`
(`L¹`-distance `t·ε` over a horizon). This file banks the two kernel-agnostic, finite-sum bridges:
* `normalize_truncate_L1_eq_two_tail` — the per-row `L¹` distance between a probability row and its
normalized truncation `P' = P·1_Good/q` is exactly `2·τ`, where `τ` is the truncated tail mass.
* `truncated_block_heat_transfer` — a truncated block-heat lower bound `c0/√(M+1)` transfers to the
real kernel as `c0/(2√(M+1))`, provided the cumulative `L¹` error `(∑_{t<M} t)·ε ≤ c0/(2√(M+1))`.
(Route + constants from ChatGPT ac2 R9, verified against the repo's `occupation_transfer` /
`Pker_rankDrop_tail_majorant`.)
-/
open Finset
open AnalyticCombinatorics AnalyticCombinatorics.Ch8 AnalyticCombinatorics.Ch8.Partitions AnalyticCombinatorics.Ch8.Partitions.Erdos
variable {α : Type*} [Fintype α] [DecidableEq α]
Formal statement
theorem AnalyticCombinatorics.Ch8.Partitions.Erdos.truncated_block_heat_transfer
(K Ktr : α → α → ℝ) (μ0 : α → ℝ) (D : α → ℝ)
(W c0 ε : ℝ) (M : ℕ)
(hKnn : ∀ x z, 0 ≤ K x z) (hKsum : ∀ x, ∑ z, K x z = 1)
(hKtrnn : ∀ x z, 0 ≤ Ktr x z) (hKtrsum : ∀ x, ∑ z, Ktr x z = 1)
(hμ0nn : ∀ x, 0 ≤ μ0 x) (hμ0sum : ∑ x, μ0 x = 1)
(hclose : ∀ x, (∑ z, |K x z - Ktr x z|) ≤ ε)
(htrunc :
c0 / Real.sqrt ((M : ℝ) + 1)
≤ ∑ t ∈ Finset.range M,
∑ x, distPow Ktr μ0 t x * (if |D x| ≤ W then (1 : ℝ) else 0))
(herr :
(∑ t ∈ Finset.range M, (t : ℝ)) * ε
≤ c0 / (2 * Real.sqrt ((M : ℝ) + 1))) :
c0 / (2 * Real.sqrt ((M : ℝ) + 1))
≤ ∑ t ∈ Finset.range M,
∑ x, distPow K μ0 t x * (if |D x| ≤ W then (1 : ℝ) else 0) := by sorrySource
Human review
Confirmed by the mission captain (proposal self-audit).