Display (2.6) — Golden–Thompson inequality: tr e^{A+H} ≤ tr(e^A e^H)
ProvedMatrixTail.Azuma.display2_6golden-thompsonmatrix-analysisp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1trace-inequality
Let and be self-adjoint (Hermitian) complex matrices. The Golden–Thompson inequality states that
The matrix exponential does not turn sums into products, and this trace inequality is the limited substitute for that rule. In the paper it is cited from Bhatia, Matrix Analysis, Sec. IX.3, and it is the first step in the symmetrization lemma (Lemma 7.6) behind the matrix Azuma inequality. The analogous inequality for three matrices is false.
Formalization Note The exponentials are defined by the continuous functional calculus. The trace is real for Hermitian , and the statement compares its real part.
Preamble
import Mathlib import Definitions.Def_MatrixTail_Azuma_Model open MeasureTheory ProbabilityTheory open scoped MatrixOrder ComplexOrder ENNReal
Formal statement
namespace MatrixTail.Azuma
/-- **Display (2.6), the Golden–Thompson inequality.** Tropp, *User-Friendly Tail Bounds for Sums of Random
Matrices*, arXiv:1004.4389v7, §2.4, p. 8: "The Golden–Thompson inequality [Bha97, Sec. IX.3] states that
`tr e^{A+H} ≤ tr(e^A e^H)` for all s.a. `A, H`." The paper cites it without proof; it is the first step of the
proof of Lemma 7.6 (p. 29).
**Formalization Note.** Complex Hermitian `d × d` matrices; `e^A` is `cfc Real.exp A` (`mexp`) and
`tr e^{A+H}` is the real part of the trace (`trExp`). The trace `tr(e^A e^H)` is real for Hermitian `A, H`
(it equals `tr(e^{A/2} e^H e^{A/2})`), so taking its real part loses nothing. -/
theorem display2_6 {d : ℕ} (A H : Matrix (Fin d) (Fin d) ℂ) (hA : A.IsHermitian) (hH : H.IsHermitian) :
trExp (A + H) ≤ (Matrix.trace (mexp A * mexp H)).re := by sorry
end MatrixTail.Azuma
Source
Tropp, User-Friendly Tail Bounds for Sums of Random Matrices, arXiv:1004.4389v7, p. 8, §2.4, display (2.6) (cited from [Bha97, Sec. IX.3])
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.