Lemma 4.3 (Rademacher half) — E e^{εθA} ≼ e^{θ²A²/2}
ProvedMatrixTail.Gaussian.lemma4_3_rademachermatrix-inequalitymoment-generating-functionp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1random-matrices
Let be a self-adjoint complex matrix and let be a Rademacher random variable, . Then for every ,
This is the semidefinite bound on the matrix mgf of a fixed matrix modulated by a random sign; it feeds Corollary 3.7 with in the proof of the Rademacher series bound.
Formalization Note The expectation is entrywise; no integrability hypothesis is needed because almost surely.
Preamble
import Mathlib import Definitions.Def_MatrixTail_Gaussian_Model open MeasureTheory ProbabilityTheory open scoped MatrixOrder ComplexOrder ENNReal
Formal statement
namespace MatrixTail.Gaussian
/-- **Lemma 4.3** (Rademacher and Gaussian mgfs), Rademacher half, Tropp, arXiv:1004.4389v7, p. 15: if `A`
is a self-adjoint matrix and `ε` a Rademacher random variable, then `E e^{εθA} ≼ e^{θ²A²/2}` for `θ ∈ ℝ`.
Formalization Note. Complex Hermitian `d × d` matrices; `exp` is `cfc`; `E` is the entrywise expectation;
`≼` is the Loewner order under `MatrixOrder`. No integrability hypothesis: `ε` takes the values `±1` almost
surely, so `e^{εθA}` is bounded. `P` is a probability measure. -/
theorem lemma4_3_rademacher {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
{d : ℕ} (A : Matrix (Fin d) (Fin d) ℂ) (hA : A.IsHermitian) (ε : Ω → ℝ) (hε : IsRademacher P ε)
(θ : ℝ) :
MatrixTail.Master.mean P (fun ω => MatrixTail.Master.mexp ((ε ω * θ) • A)) ≤ MatrixTail.Master.mexp ((θ ^ 2 / 2) • A ^ 2) := by sorry
end MatrixTail.Gaussian
Source
Tropp, User-Friendly Tail Bounds for Sums of Random Matrices, arXiv:1004.4389v7, p. 15, Lemma 4.3
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.