Expectation of an almost surely positive-definite matrix
ProvedTroppMatrixConcentration.ch3_lieb_integral_posDefmatrix-concentrationrandom-matrices
Let be a Bochner-integrable random complex matrix on a probability space. If is positive definite almost surely, then
No uniform lower eigenvalue bound is assumed. Dimension zero is permitted. This auxiliary lemma keeps the mean inside the positive-definite domain when applying Jensen’s inequality to Lieb’s concave function.
Preamble
import Definitions.Def_TroppMatrixConcentration_probability open MeasureTheory ProbabilityTheory open scoped Matrix.Norms.L2Operator ComplexOrder set_option autoImplicit false
Formal statement
namespace TroppMatrixConcentration
theorem ch3_lieb_integral_posDef {Ω : Type*} [MeasurableSpace Ω]
(μ : Measure Ω) [IsProbabilityMeasure μ] {d : ℕ}
(Z : Ω → Matrix (Fin d) (Fin d) ℂ)
(hInt : Integrable Z μ) (hPD : ∀ᵐ ω ∂μ, (Z ω).PosDef) :
(∫ ω, Z ω ∂μ).PosDef := 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 positive-definite integration fact used in Corollary 3.4.2, printed p. 35; quadratic-form definition (2.1.10), printed p. 20.