Proposition 7.4.1 — Generalized matrix Laplace transform
ProvedTroppMatrixConcentration.ch7_generalized_laplaceLet be a measurable almost-surely Hermitian random matrix, , on a probability space. Let be nonnegative everywhere and nondecreasing on , with integrable . For with ,
The positivity restriction makes the denominator meaningful; the source's applications use positive thresholds. No continuity assumption on the scalar function is imposed: every function is continuous when restricted to a finite spectrum.
import Definitions.Def_TroppMatrixConcentration_ch7_intrinsic open MeasureTheory ProbabilityTheory open scoped Matrix.Norms.L2Operator ComplexOrder
namespace TroppMatrixConcentration
theorem ch7_generalized_laplace {Ω : Type*} [MeasurableSpace Ω]
(μ : Measure Ω) [IsProbabilityMeasure μ] {d : ℕ} [NeZero d]
(Y : Ω → Matrix (Fin d) (Fin d) ℂ) (ψ : ℝ → ℝ)
(hMeas : Measurable Y) (hHerm : ∀ᵐ ω ∂μ, (Y ω).IsHermitian)
(hNonneg : ∀ x, 0 ≤ ψ x) (hMono : MonotoneOn ψ (Set.Ici 0))
(hInt : Integrable (fun ω => traceFunction ψ (Y ω)) μ)
(t : ℝ) (ht : 0 ≤ t) (hψt : 0 < ψ t) :
(μ {ω | t ≤ lambdaMax (Y ω)}).toReal ≤
(∫ ω, traceFunction ψ (Y ω) ∂μ) / ψ t := by sorry
end TroppMatrixConcentration
Read-back
What the Lean code literally says, in plain math · gpt-6
For every measurable space , every probability measure on it, every positive natural number , every measurable map (with the Borel measurable structure on matrices) that is Hermitian -almost everywhere, and every function such that for every real and whenever , assume that the real-valued function is Bochner integrable with respect to , where and is the real continuous functional calculus, taken to be the zero matrix when its applicability conditions fail; on Hermitian matrices, this applies to the real eigenvalues, so is their transformed sum, counted with multiplicity. Then, for every real satisfying and , one has , where the measure on the left is converted from its extended nonnegative real value to a real number, and , with and the real supremum of the empty set defined as zero. In particular, is the largest eigenvalue when is Hermitian; the event still uses this specified supremum at all exceptional, possibly non-Hermitian values of . The dimension is excluded, equality at the threshold is included, and is allowed when ; the denominator is always strictly positive, no continuity or measurability of is separately assumed, and no monotonicity is required on negative arguments.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.