Proposition 8.3.5 — Generalized Klein inequality
ProvedTroppMatrixConcentration.ch8_generalized_kleinLet be an interval of the real line, and let be a finite family of scalar functions such that for every . If are Hermitian complex matrices of the same positive dimension with all eigenvalues in , then
No scalar continuity assumption is required, because the matrix functional calculus evaluates each function on a finite spectrum. This transfers scalar inequalities involving two variables to trace inequalities.
import Definitions.Def_TroppMatrixConcentration_ch8_entropy open scoped Matrix.Norms.L2Operator ComplexOrder
namespace TroppMatrixConcentration
theorem ch8_generalized_klein {d N : ℕ} [NeZero d]
(I : Set ℝ) (hI : Convex ℝ I) (f g : Fin N → ℝ → ℝ)
(hfg : ∀ a ∈ I, ∀ h ∈ I, 0 ≤ ∑ i, f i a * g i h)
(A H : Matrix (Fin d) (Fin d) ℂ) (hA : A.IsHermitian) (hH : H.IsHermitian)
(hAI : spectrum ℝ A ⊆ I) (hHI : spectrum ℝ H ⊆ I) :
0 ≤ ∑ i, (Matrix.trace (ch8_matrixFunction (f i) A *
ch8_matrixFunction (g i) H)).re := by sorry
end TroppMatrixConcentration
Read-back
What the Lean code literally says, in plain math · gpt-6
For every positive natural number , every natural number (including ), every convex subset , and every two families of functions indexed by , suppose that for every and every . For any two Hermitian complex matrices and whose real spectra are contained in , the following inequality holds: . Here each matrix function is defined using the real continuous functional calculus: for a Hermitian matrix with spectral decomposition , the notation means . No continuity, differentiability, or convexity of the scalar functions is assumed. The matrices need not commute. When , both scalar and matrix sums are empty and equal to zero. Although convex sets may in general be empty, the spectral containment assumptions for positive-size Hermitian matrices exclude an empty .
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.