Documentation

LeanPool.HighContrastHomogenization.Provider.Quenched.SmallContrastSymmetricSkew

High-contrast homogenization: Provider.Quenched.SmallContrastSymmetricSkew #

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

The symmetric Schur skew at small diagonal contrast #

The large-contrast version of the symmetric-skew estimate fails, while its perturbative form is sufficient for the small-contrast centering estimate. This file proves that form directly: if a Schur block with a SYMMETRIC skew coordinate dominates its own sharp, then the σ*-normalized skew is bounded by half of the normalized diagonal gap — first order in θ − 1, with an absolute constant.

The proof tests the sharp ordering at the single doubled vector (x, t·k x) with t = (θ+1)/(θ−1): at this shift the cross terms cancel and the quadratic gain appears with the exact constant 4/(θ−1).

theorem HCPolySupport.HighContrast.Quenched.symmetricSchurSkew_quad_le_of_sharp_le {d : ℕ} {S SStar k : Mat d} (hS : Matrix.PosDef S) (hStar : Matrix.PosDef SStar) (hk : Matrix.conjTranspose k = k) (hsharp : fullBlockSharp (schurBlock S SStar k) ≤ schurBlock S SStar k) {theta : ℝ} (htheta : 1 < theta) (hS_le : S ≤ theta • SStar) (x : Vec d) :
Matrix.mulVec k x ⬝ᵥ SStar⁻¹.mulVec (Matrix.mulVec k x) ≤ (theta - 1) ^ 2 / 4 * x ⬝ᵥ Matrix.mulVec SStar x

The symmetric Schur skew is first order in the diagonal gap. If the Schur block of (S, S₊, k) with symmetric k dominates its sharp and S ≤ θ·S₊, then the S₊⁻¹-quadratic of k is bounded by ((θ−1)/2)² times the S₊-quadratic. This is the perturbative half of the printed display HC (2.81), proved without the refuted large-contrast half.

theorem HCPolySupport.HighContrast.Quenched.symmetricSchurSkew_conj_le_of_sharp_le {d : ℕ} {S SStar k : Mat d} (hS : Matrix.PosDef S) (hStar : Matrix.PosDef SStar) (hk : Matrix.conjTranspose k = k) (hsharp : fullBlockSharp (schurBlock S SStar k) ≤ schurBlock S SStar k) {theta : ℝ} (htheta : 1 < theta) (hS_le : S ≤ theta • SStar) :
Matrix.conjTranspose k * SStar⁻¹ * k ≤ ((theta - 1) ^ 2 / 4) • SStar

Matrix-order form of the symmetric-skew bound.