Lemma 4, (25)–(27): finite-block spectral converse
OpenQAlgorithms.Papers.Independent.Intakenext100.P20.visible_spectral_converseLet be an arbitrary discrete label set and let be a nonzero finite-dimensional complex Hilbert space. The source consists of normalized pure states and probabilities with ; write . Each memory space is finite and nonzero. Blind encoders and decoders are CPTP; a visible encoder maps labels to normalized density operators. A blind encoder is fixed before its input label.
For any visible code let . For every real , put and , the orthogonal spectral projectors including equality only in . Then
Moreover, for every , with ,
This converse applies at any finite block by substituting the product ensemble, without a full-rank assumption.
import Definitions.Def_QAlgorithms_Papers_Independent_Intakenext100_P20
noncomputable section
open scoped Matrix ENNReal ComplexOrder Topology
namespace QAlgorithms.Papers.Independent.Intakenext100.P20
open QAlgorithms.Papers.Independent.Information100.P08 (IsPure)
open QAlgorithms.Papers.Independent.Information100.P47 (IsProbability)
open QAlgorithms.Papers.Independent.Information100.P53 (SpectralData)
open QAlgorithms.Papers.Independent.Gap48.P19 (shannonNats)
open QAlgorithms.Papers.Independent.Dependency50.P28 (classicalD)
attribute [local instance] Classical.propDecidable
/-- Lemma 4, PDF pp.5-6, (25)-(27): finite-block visible spectral converse,
including the equality threshold in the upper projector and every real lambda. -/
theorem visible_spectral_converse {L : Type*} {d : ℕ} (S : DiscretePureSource L d)
(C : VisibleCode L (Fin d)) (lambda : ℝ) :
let epsilon := averageLoss S.weight S.state (fun i => C.decode (C.encode i)) false
epsilon + Real.exp lambda * C.memory ≥
(S.average * thresholdProjection S.average lambda false).trace.re ∧
1 - epsilon ≤ Real.exp lambda * C.memory +
(S.average * thresholdProjection S.average lambda true).trace.re ∧
∀ s : ℝ, 1 ≤ s → 1 - epsilon ≤ Real.exp lambda * C.memory +
Real.exp ((1 - s) * lambda + logPowerTrace S.average s) := by sorry
end QAlgorithms.Papers.Independent.Intakenext100.P20
end
Read-back
What the Lean code literally says, in plain math · gpt-6.1-sol
Let be any type, a natural number, and specify , nonnegative summable weights with , and density matrices for unit vectors , where a density matrix means a positive semidefinite matrix of trace . Let a visible code have positive integer memory dimension , assigned memory density matrices , and a decoder for a finite family with . For any real , put
Define and by applying respectively the scalar functions and to through functional calculus. Then
and, for every real ,
where is defined by functional calculus using real scalar powers and is the natural logarithm. Eigenvalues equal to occur in , and not in ; need not be finite or countable.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.