Lemma-14
ProvedQAlgorithms.Papers.Independent.Gap20.P04.lemma14Let be a generalized-dephasing channel with isometry , where each environment vector has norm one and the vectors need not be orthogonal. Let be any physical finite ensemble of pure reference-input states. Form another such ensemble with the same probabilities and conditional input marginals , choosing fresh pure purifications on any finite reference. Both ensembles pass through the same .
For every ,
where and in bits. This permits restricting CE optimization to diagonal input marginals, including both weight endpoints.
import Definitions.Def_QAlgorithms_Papers_Independent_Gap20_P04
noncomputable section
namespace QAlgorithms.Papers.Independent.Gap20.P04
open QAlgorithms
open QAlgorithms.Papers.Independent.Claimed50.P24
(Dilation PhysicalDilation ChannelEB Hadamard EqualMax AdditiveMax
dephasingDilation dephasingGamma cloningDilation)
open QAlgorithms.Papers.Independent.Information100.P67
(receiverChannel complementaryChannel dilationPower)
open QAlgorithms.Papers.Independent.Information100.P32 (diagonal)
open QAlgorithms.Papers.Independent.Dependency50.P24 (ensembleHolevoBits)
open QAlgorithms.Papers.Independent.Gap48.P44 (cqTimeSharing ceTimeSharing)
open scoped BigOperators Matrix ComplexConjugate
attribute [local instance] Classical.propDecidable Classical.decEq
variable {I B E J D F : Type}
variable [Fintype I] [Fintype B] [Fintype E]
variable [Fintype J] [Fintype D] [Fintype F]
variable [DecidableEq I] [DecidableEq B] [DecidableEq E]
variable [DecidableEq J] [DecidableEq D] [DecidableEq F]
variable [Nonempty I] [Nonempty B] [Nonempty E]
variable [Nonempty J] [Nonempty D] [Nonempty F]
/-- PDF p.23, Lemma 14: dephase marginals and repurify; both CE weight endpoints included. -/
theorem lemma14 (v : I → EuclideanSpace ℂ E) (hv : ∀ i, ‖v i‖ = 1)
(e t : PurifiedInputFamily I) (he : e.Valid) (ht : t.Valid)
(hreplace : DephasedReplacement e t) (lam : ℝ) (h0 : 0 ≤ lam) (h1 : lam ≤ 1) :
ceScore (gdDilation v) lam e ≤ ceScore (gdDilation v) lam t := by sorry
end QAlgorithms.Papers.Independent.Gap20.P04
end
Read-back
What the Lean code literally says, in plain math · gpt-6.1-sol
Let be finite nonempty index types with decidable equality, and let be unit vectors. Define , with receiver and environment maps
Let each have positive finite label and reference dimensions, nonnegative probabilities summing to , and unit purifications on reference . Write for their reduced input matrices. Assume a bijection of label sets satisfies and , where retains the diagonal entries and sets every off-diagonal entry to zero. Define by continuous functional calculus and
For every real , . Both endpoints for are included; the reference dimensions need not agree, and the replacement condition applies even at zero-probability labels.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.