Sharp relative-entropy bound for unnormalised equal-trace matrices
OpenQAlgorithms.Papers.Independent.Support50.P17.proposition1For , define
Let and be matrices on the same nonzero finite-dimensional complex Hilbert space, with equal trace and . Put and . Then
For each fixed , the function is nonincreasing on . For every satisfying , the explicit matrices
have the specified traces, minima and trace distance, and attain equality in the bound. Moreover when .
This includes the positive-definite proposition and its explicitly justified semidefinite boundary, without imposing trace one or requiring attainment in each fixed dimension.
import Definitions.Def_QAlgorithms_Papers_Independent_Support50_P17
noncomputable section
open MeasureTheory
open scoped Matrix ComplexOrder Matrix.Norms.L2Operator
namespace QAlgorithms.Papers.Independent.Support50.P17
attribute [local instance] Classical.propDecidable Classical.decEq
-- Reuse the exact frozen support predicate and signed, unnormalised entropy.
open QAlgorithms.Papers.Independent.Dependency50.P21 (IsSupported relativeEntropy)
/-- PDF pp.6–8, Proposition 1, (13): the sharp unnormalised equal-trace bound,
its explicit alpha=0 extension, nonincreasing correction, and the exact 2x2 attaining pair.
The printed “quality” on p.6 means equality, as shown by the p.8 construction. -/
theorem proposition1 :
(∀ {I : Type*} [Fintype I] [DecidableEq I] [Nonempty I]
(A B : Matrix I I ℂ), A.PosSemidef → B.PosDef →
A.trace = B.trace → minEigenvalue B = 1 →
let alpha := minEigenvalue A
let T := traceDistance A B
|alpha - 1| ≤ T ∧
relativeEntropy A B ≤
(((1 + T) * Real.log (1 + T) + zeroExtendedLogCorrection alpha T : ℝ) : EReal)) ∧
(∀ T : ℝ, 0 ≤ T →
AntitoneOn (fun alpha => zeroExtendedLogCorrection alpha T) (Set.Ici 0)) ∧
(∀ alpha T : ℝ, 0 ≤ alpha → 0 ≤ T → |alpha - 1| ≤ T →
let A := propositionA alpha T
let B := propositionB alpha T
A.PosSemidef ∧ B.PosDef ∧ (0 < alpha → A.PosDef) ∧
A.trace = B.trace ∧ minEigenvalue A = alpha ∧ minEigenvalue B = 1 ∧
traceDistance A B = T ∧
relativeEntropy A B =
(((1 + T) * Real.log (1 + T) + zeroExtendedLogCorrection alpha T : ℝ) : EReal)) := by sorry
end QAlgorithms.Papers.Independent.Support50.P17
end
Read-back
What the Lean code literally says, in plain math · gpt-6.1-sol
Define if and otherwise, and define . The theorem asserts three statements jointly: for every finite, nonempty index type with decidable equality and complex matrices indexed by , if is positive semidefinite, is positive definite, their traces are equal, and , then, setting and , one has and ; for every real , the function is nonincreasing on , meaning implies ; and for every real satisfying , the matrices and satisfy , , whenever , equal traces, , , , and . Here , viewed in the extended reals, and matrix functions use continuous functional calculus with . The case is included through the explicitly defined value of .
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.