Lemma 1: label-versus-mean Shannon distinguishability
ProvedQAlgorithms.Papers.Independent.Support50.P27.label_mean_shannonLet , and let be an ensemble of density operators on a finite dimensional nonzero Hilbert space. Thus , , and every is positive semidefinite with trace one, including at zero-weight labels. Write . Let be the supremum, over finite POVMs, of the mutual information in bits between the ensemble label and the measurement outcome. Let denote accessible information of the equiprobable binary ensemble consisting of and . Then
The estimate relates a single label's distinguishability from the ensemble mean to the information available from the whole ensemble. No strictly positive weight assumption is imposed.
import Definitions.Def_QAlgorithms_Papers_Independent_Support50_P27
noncomputable section
namespace QAlgorithms.Papers.Independent.Support50.P27
open scoped BigOperators Matrix ComplexConjugate ComplexOrder
open QAlgorithms.Papers.Independent.Dependency50.P07
(CQData ensembleMean ensembleInformation accessibleInformation)
open QAlgorithms.Papers.Independent.Dependency50.P24 (Key)
open QAlgorithms.Papers.Independent.Dependency50.P47 (cqMatrix)
attribute [local instance] Classical.decEq Classical.propDecidable
/-- PDF p.12, Lemma 1: any label's weighted Shannon distinguishability from
the ensemble mean is at most accessible information. Zero label weights and
m=0 are allowed; all conditional states are normalized density matrices. -/
theorem label_mean_shannon {m : ℕ} {E : Type} [Fintype E] [Nonempty E]
(A : CQData (Fin (2 ^ m)) E) (hA : A.Valid) :
∀ x, A.prob x * shannonDistinguishability (A.state x) (ensembleMean A) ≤
accessibleInformation A := by sorry
end QAlgorithms.Papers.Independent.Support50.P27
end
Read-back
What the Lean code literally says, in plain math · gpt-6.1-sol
For every natural number , finite nonempty type , and ensemble consisting of numbers summing to over and positive semidefinite, trace-one matrices on , let . For any finite ensemble on , define as the real supremum over all finite families with at least one outcome and , of , where , , and with zero summands interpreted as . The statement is that every label satisfies
Labels with are included; , giving a single label, is also included.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.