Integrability of extreme eigenvalues
ProvedTroppMatrixConcentration.ch3_expect_extrema_integrablematrix-concentrationrandom-matrices
Let be a measurable, almost surely Hermitian complex random matrix, , on any measure space. If is Bochner integrable in the Euclidean operator norm, then
No finite-measure or probability assumption is needed. This supporting regularity lemma makes the expectations in the matrix Laplace method well-defined.
Preamble
import Definitions.Def_TroppMatrixConcentration_probability open MeasureTheory open scoped Matrix.Norms.L2Operator
Formal statement
namespace TroppMatrixConcentration
/-- The extreme eigenvalues of an integrable Hermitian random matrix are integrable. -/
theorem ch3_expect_extrema_integrable {Ω : Type*} [MeasurableSpace Ω]
(μ : Measure Ω) {d : ℕ} [NeZero d]
(Y : Ω → Matrix (Fin d) (Fin d) ℂ)
(hMeas : Measurable Y) (hHerm : ∀ᵐ ω ∂μ, (Y ω).IsHermitian)
(hInt : Integrable Y μ) :
Integrable (fun ω => lambdaMax (Y ω)) μ ∧
Integrable (fun ω => lambdaMin (Y ω)) μ := by sorry
end TroppMatrixConcentration
Source
Joel A. Tropp, An Introduction to Matrix Concentration Inequalities, arXiv:1501.01571v1 (7 January 2015), https://arxiv.org/abs/1501.01571v1; Auxiliary regularity lemma for Proposition 3.2.2, printed p. 33; uses spectral norm relation (2.1.22), printed p. 24, and the standing regularity convention of Section 2.2.1, printed p. 25.