Jensen inequality on a nonclosed convex domain
ProvedTroppMatrixConcentration.ch3_lieb_jensenmatrix-concentrationrandom-matrices
Let be a real Banach space, , and continuous and concave on (so is convex). On a probability space, suppose and are Bochner integrable, almost surely, and . Then
The domain need not be closed. The explicit mean-membership hypothesis allows Jensen’s inequality to be used on the positive-definite cone.
Preamble
import Mathlib.Analysis.Convex.Integral open MeasureTheory set_option autoImplicit false
Formal statement
namespace TroppMatrixConcentration
/-- Jensen's inequality on a possibly nonclosed convex domain, provided the mean belongs to it. -/
theorem ch3_lieb_jensen {Ω E : Type*} [MeasurableSpace Ω]
[NormedAddCommGroup E] [NormedSpace ℝ E] [CompleteSpace E]
(μ : Measure Ω) [IsProbabilityMeasure μ] (s : Set E) (g : E → ℝ) (f : Ω → E)
(hg : ConcaveOn ℝ s g) (hgc : ContinuousOn g s)
(hfs : ∀ᵐ ω ∂μ, f ω ∈ s) (hf : Integrable f μ)
(hgf : Integrable (fun ω => g (f ω)) μ) (hm : (∫ ω, f ω ∂μ) ∈ s) :
(∫ ω, g (f ω) ∂μ) ≤ g (∫ ω, f 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; Jensen inequality (2.2.2), printed p. 25, and its use in Corollary 3.4.2, printed p. 35. Auxiliary Banach-space formulation with explicit integrability, continuity, and mean-membership hypotheses.