Equation 8.1.2 — Variational trace exponential
ProvedTroppMatrixConcentration.ch8_variational_trace_expFor every Hermitian complex matrix and positive-definite complex matrix of the same positive dimension,
The maximum is over positive-definite matrices of the same size and is asserted to be attained. This gives the exact variational objective used in the source’s Lieb concavity argument.
import Definitions.Def_TroppMatrixConcentration_ch8_entropy open scoped Matrix.Norms.L2Operator ComplexOrder
namespace TroppMatrixConcentration
theorem ch8_variational_trace_exp {d : ℕ} [NeZero d]
(H A : Matrix (Fin d) (Fin d) ℂ) (hH : H.IsHermitian) (hA : A.PosDef) :
IsGreatest {r : ℝ | ∃ T : Matrix (Fin d) (Fin d) ℂ,
T.PosDef ∧ r = (Matrix.trace (T * H)).re + (Matrix.trace A).re -
ch8_relativeEntropy T A}
(traceExp (H + matrixLog A)) := by sorry
end TroppMatrixConcentration
Read-back
What the Lean code literally says, in plain math · gpt-6
For every nonzero natural number , every complex matrix that is Hermitian, and every complex positive definite matrix , define to be the continuous functional calculus of the real logarithm applied to a matrix , and let denote the matrix exponential. Then the real number is the greatest element of the set
In particular, every positive definite gives a displayed value at most , and at least one such gives equality; uniqueness is not asserted. Positive definiteness here includes being Hermitian and having strictly positive quadratic form on each nonzero complex vector. The dimension-zero case is excluded, and the matrices over which the greatest value is attained are required to be positive definite, so singular positive semidefinite matrices are excluded.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.