Documentation

LeanPool.HighContrastHomogenization.Provider.PolynomialHomogenization.RootInterfaceArithmetic

High-contrast homogenization: Provider.PolynomialHomogenization.RootInterfaceArithmetic #

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

Arithmetic reconciliation for the homogenization root #

The root statement carries one polynomial exponent and one error exponent that several clauses share. Every occurrence of the constant is weakened by enlarging it and every occurrence of the error exponent is weakened by shrinking it, so providers producing different values can be reconciled by a maximum and a minimum. This module records the four comparisons that make that reconciliation legal.

The block-to-matrix identification is no longer an arithmetic gap: the root retains the annealed sandwich and contrast decay until the canonical block is identified. The quenched side likewise retains the source burn bound. The finite enlargement of the normalized reference family is part of the GoodScale certificate, so it neither changes the real random scale nor adds an arithmetic weakening to the frozen conclusion.

theorem HCPolySupport.HighContrast.RootInterface.le_rpow_of_le_rpow_of_exponent_le {base c c' y : ℝ} (hbase : 1 ≤ base) (hc : c ≤ c') (hy : y ≤ base ^ c) :
y ≤ base ^ c'

A polynomial length bound at a base of at least one is weakened by enlarging the exponent.

theorem HCPolySupport.HighContrast.RootInterface.rpow_le_rpow_of_exponent_ge_of_le_one {x kap kap' : ℝ} (hx : 0 < x) (hx1 : x ≤ 1) (hkap : kap' ≤ kap) :
x ^ kap ≤ x ^ kap'

The Dirichlet error factor is read at a base of at most one, so it is weakened by shrinking the exponent.

theorem HCPolySupport.HighContrast.RootInterface.rpow_neg_le_rpow_neg_of_exponent_ge {x kap kap' : ℝ} (hx : 1 ≤ x) (hkap : kap' ≤ kap) :
x ^ (-kap) ≤ x ^ (-kap')

The corrector error factor is read at a base of at least one and a negative exponent, so it too is weakened by shrinking the exponent.