Trace exponential dominates exponentials of extremal eigenvalues
ProvedTroppMatrixConcentration.ch3_tail_spectral_comparisonmatrix-concentrationrandom-matrices
Let be a Hermitian complex matrix with and . Then , and
The trace is its real part. This deterministic comparison is the spectral step in both matrix Laplace bounds.
Preamble
import Definitions.Def_TroppMatrixConcentration_spectral open scoped Matrix.Norms.L2Operator set_option autoImplicit false
Formal statement
namespace TroppMatrixConcentration
theorem ch3_tail_spectral_comparison {d : ℕ} [NeZero d]
(A : Matrix (Fin d) (Fin d) ℂ) (hA : A.IsHermitian) (θ : ℝ) :
0 ≤ traceExp (θ • A) ∧
(0 < θ → Real.exp (θ * lambdaMax A) ≤ traceExp (θ • A)) ∧
(θ < 0 → Real.exp (θ * lambdaMin A) ≤ traceExp (θ • A)) := 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; Section 3.2, equation (3.2.3) and the lower-tail proof, printed p. 33.