sum_norm_sq_rank_one_outer_product_loewner_le_scalar_mul_sum
ProvedPSD-monotone Loewner domination step. For a finite family of rank-one outer products y_c (x) y_c = vecMulVec y_c y_c and a scalar M dominating every squared norm ||y_c||^2 = y_c . y_c, the weighted Gram sum is dominated in the Loewner order: sum_c ||y_c||^2 (y_c (x) y_c) ⪯ M . sum_c (y_c (x) y_c), stated as the difference M . sum (y_c(x)y_c) - sum ||y_c||^2 (y_c(x)y_c) being positive semidefinite. This is the PSD-monotone domination immediately after the Gram-square collapse sum_c (y_c(x)y_c)^2 = sum_c ||y_c||^2 (y_c(x)y_c) in Rudelson's noncommutative-Khintchine selection lemma. Proof: the difference equals sum_c (M - ||y_c||^2) (y_c (x) y_c); each y_c (x) y_c is PSD, each scalar M - ||y_c||^2 >= 0, a nonnegative scalar multiple of a PSD matrix is PSD, and a finite sum of PSD matrices is PSD.
import Mathlib.LinearAlgebra.Matrix.PosDef import Mathlib.Data.Matrix.Mul import Mathlib.Data.Real.Basic import Mathlib.Data.Real.StarOrdered open scoped BigOperators Matrix
theorem sum_norm_sq_rank_one_outer_product_loewner_le_scalar_mul_sum {d : Nat} {ι : Type*} (s : Finset ι) (y : ι → Fin d → Real) (M : Real) (hM : ∀ c ∈ s, (y c ⬝ᵥ y c) ≤ M) : (M • (∑ c ∈ s, Matrix.vecMulVec (y c) (y c)) - ∑ c ∈ s, (y c ⬝ᵥ y c) • Matrix.vecMulVec (y c) (y c)).PosSemidef := by sorry