Lemma 4.6.2 — Gaussian matrix mgf and cgf identities
ProvedTroppMatrixConcentration.ch4_gaussian_mgf_cgfLet be a fixed Hermitian complex matrix, , and let be a measurable standard real normal random variable on a probability space. For every real ,
The scalar distribution is specified by its actual pushforward Gaussian probability measure with mean zero and variance one. Expectations are Bochner integrals; matrix exponential integrability follows from this law and is part of the proof obligation. No restriction on the sign of or on the Hermitian eigenvalues is imposed.
import Definitions.Def_TroppMatrixConcentration_ch4_scalar_laws import Definitions.Def_TroppMatrixConcentration_dilation open MeasureTheory ProbabilityTheory open scoped Matrix.Norms.L2Operator
namespace TroppMatrixConcentration
theorem ch4_gaussian_mgf_cgf {Ω : Type*} [MeasurableSpace Ω]
(μ : Measure Ω) [IsProbabilityMeasure μ] {d : ℕ} [NeZero d]
(A : Matrix (Fin d) (Fin d) ℂ) (hA : A.IsHermitian)
(g : Ω → ℝ) (hMeas : Measurable g) (hLaw : standardGaussianLaw μ g) (θ : ℝ) :
(∫ ω, matrixExp ((θ * g ω) • A) ∂μ) = matrixExp ((θ ^ 2 / 2) • A ^ 2) ∧
matrixLog (∫ ω, matrixExp ((θ * g ω) • A) ∂μ) = ((θ ^ 2 / 2) • A ^ 2) := 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 , every natural number , every complex matrix indexed by with , every measurable function whose pushforward measure is the Gaussian probability measure with mean and variance (equivalently, with density with respect to Lebesgue measure), and every real number , both equalities and hold. Here is the conjugate transpose, uses matrix multiplication, real scalar multiplication acts entrywise on the complex matrices, is the normed-algebra matrix exponential , and the integral is the matrix-valued Bochner integral, with the convention that a nonintegrable function has integral zero; no separate integrability hypothesis is imposed. The notation means the real continuous functional calculus applied to the total real logarithm: for Hermitian it is , where for and ; the functional-calculus operation returns zero if its self-adjointness or continuity-on-the-real-spectrum condition fails (continuity on the finite spectrum holds for these finite matrices). The dimension-zero case is excluded, the probability-measure hypothesis excludes empty , and there is no nonzero, invertibility, or positivity requirement on and no sign or nonzero requirement on ; in particular, or gives the identity matrix in the first equality and the zero matrix in the second.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.