Maximum of the shifted relative entropy
ProvedQAlgorithms.Papers.Independent.Support50.P17.regularisedMaximumLet and be density matrices on the same nonzero finite-dimensional complex Hilbert space. Write for the identity and for the unnormalised Umegaki relative entropy, with natural logarithms, zero spectral weights contributing zero, and value on failure of support containment. Then
Equality holds whenever both densities have rank one and . Such orthogonal pure densities attaining equality exist in every dimension . The inequality also holds in dimension one.
This identifies the global maximum before any normalization constant is applied; the shifted matrices are not divided by their trace.
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 p.2, Section I: the global log 2 maximum and equality for orthogonal pure
states. Such witnesses exist in every dimension d≥2; dimension one retains the bound. -/
theorem regularisedMaximum :
(∀ {I : Type*} [Fintype I] [DecidableEq I] [Nonempty I]
(rho sigma : Matrix I I ℂ), IsDensityMatrix rho → IsDensityMatrix sigma →
relativeEntropy (rho + 1) (sigma + 1) ≤ (Real.log 2 : EReal) ∧
(rho.rank = 1 → sigma.rank = 1 → rho * sigma = 0 →
relativeEntropy (rho + 1) (sigma + 1) = (Real.log 2 : EReal))) ∧
(∀ d : ℕ, 2 ≤ d → ∃ rho sigma : Matrix (Fin d) (Fin d) ℂ,
IsDensityMatrix rho ∧ IsDensityMatrix sigma ∧
rho.rank = 1 ∧ sigma.rank = 1 ∧ rho * sigma = 0 ∧
relativeEntropy (rho + 1) (sigma + 1) = (Real.log 2 : 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
For every finite, nonempty index type with decidable equality and every pair of complex matrices indexed by that are positive semidefinite and have trace , the extended real relative entropy satisfies ; if additionally and both have rank and , equality holds. Moreover, for every natural number , there exist such positive semidefinite, trace-one matrices of rank with and equality. Here is when , and otherwise, with matrix functions defined by continuous functional calculus. The shifts here are positive definite. The first assertion includes one-dimensional spaces, where its additional rank-one orthogonality hypotheses cannot all hold.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.