Documentation

LeanPool.HighContrastHomogenization.Provider.Quenched.SmallContrastCorrectedGenerationCap

High-contrast homogenization: Provider.Quenched.SmallContrastCorrectedGenerationCap #

Imported from the Apache-2.0 HighContrastHomogenization development at commit 7a13dbcd8d6609264a713373f5c69ceeac870472.

The corrected generation cap #

This file packages the three finite generation delays used by the endpoint. The burn-in, deep-cap, and reverse-transfer delays are combined at the printed base, with the coefficient and exponent table used by the corrected cadence.

theorem HCPolySupport.HighContrast.Quenched.three_pow_burnSplitDepth_le_corrected_base (d : ℕ) (hd : 2 ≤ d) {Cd g Pi base : ℝ} (hCd : 1 ≤ Cd) (hzeta : 1 ≤ zetaG g) (hPi : 1 ≤ Pi) (hPiBase : Pi ≤ base) :
3 ^ burnSplitDepth d Cd g Pi ≤ 3 * (100 / 99 * Cd * zetaG g * √↑d) * base

The burn-split depth is polynomially bounded at the endpoint base.

theorem HCPolySupport.HighContrast.Quenched.max_reverseAdapterFactor_le_corrected_generation_scale {d : ℕ} {g K base cW Cd Cdz C5 CB CF cF : ℝ} {E : BlockMat d} {mAl : Mat d} {G D : ℕ} (hg : g ∈ Set.Ico 0 1) (hK : 1 ≤ K) (hbase : 3 ≤ base) (hcW : 0 ≤ cW) (hCd : 1 ≤ Cd) (hzeta : 1 ≤ zetaG g) (hecc : witnessEccentricity mAl ≤ base) (hkappa1 : 1 ≤ kappaRef E) (hkappa : kappaRef E ≤ 9 / 8) (hgrowth : growthBar K ≤ 2 * base.rpow cW) (hGD : G + 1 = D) (hdepth : 3 ^ D ≤ C5 * base) (hCB : 0 ≤ CB) (hCdz : Cdz = Cd * zetaG g) (hCF : CF = max 1 (CB * (45 / 2) * Cdz * C5)) (hcF : cF = 4 + 3 * cW) :
max 1 (CB * reverseAdapterFactor Cd g K E mAl G) ≤ CF * base.rpow cF

The reverse-adapter factor is absorbed by the transfer coefficient and the exponent 4 + 3*cW appearing in the corrected generation table.

theorem HCPolySupport.HighContrast.Quenched.corrected_generation_power_cap (d : ℕ) (hd : 2 ≤ d) {H D G ns nB nE nF : ℕ} {g K Pi base cW rate Cd Cdz C5 CE CB CF cF Cns cns : ℝ} {E : BlockMat d} {mAl : Mat d} (hg : g ∈ Set.Ico 0 1) (hK : 1 ≤ K) (hPi : 1 ≤ Pi) (hPiBase : Pi ≤ base) (hbase : 3 ≤ base) (hcW : 0 ≤ cW) (hrate : 0 < rate) (hCd : 1 ≤ Cd) (hzeta : 1 ≤ zetaG g) (hecc : witnessEccentricity mAl ≤ base) (hkappa1 : 1 ≤ kappaRef E) (hkappa : kappaRef E ≤ 9 / 8) (hgrowth : growthBar K ≤ 2 * base.rpow cW) (hD : D = burnSplitDepth d Cd g Pi) (hGD : G + 1 = D) (hCB : 0 ≤ CB) (hCdz : Cdz = Cd * zetaG g) (hC5 : C5 = 3 * (100 / 99 * Cd * zetaG g * √↑d)) (hCE : CE = 8100 * 3 ^ (3 * H) * C5 ^ 4) (hCF : CF = max 1 (CB * (45 / 2) * Cdz * C5)) (hcF : cF = 4 + 3 * cW) (hCns : Cns = 3 ^ (H + 1) * 624 * CE * (3 * CF.rpow rate⁻¹) * 3) (hcns : cns = 4 * cW + 4 + cF / rate) (hnsSum : ns ≤ H + 1 + nB + nE + nF) (hB : 3 ^ nB ≤ 3 * (13 * growthBar K ^ 4)) (hE : 3 ^ nE ≤ 8100 * 3 ^ (3 * H) * 3 ^ (4 * D)) (hF : 3 ^ nF ≤ 3 * (max 1 (CB * reverseAdapterFactor Cd g K E mAl G)).rpow rate⁻¹) :
3 ^ ns ≤ Cns * base.rpow cns

The maximum of the three finite delay components obeys the generation coefficient and exponent table used by the initial cadence level schedule.