Theorem 1: four pure-source coding rates equal entropy
OpenQAlgorithms.Papers.Independent.Intakenext100.P20.pure_source_coding_ratesLet be a discrete probability law on an arbitrary label set and let be normalized pure states on a nonzero finite-dimensional complex space. Set . At each positive block length use the independent product probabilities and tensor product states. A blind encoder and decoder are CPTP; a visible encoder maps labels to memory density matrices and has a CPTP decoder. Each memory has a positive finite dimension. Define the ordinary blind and visible rates as the infima of over code sequences whose average overlap errors tend to zero. Define their converse rates by replacing error convergence with . Then
All logarithms are natural. Infima and memory-rate limits are in the extended reals; no full-rank or finite-label hypothesis is added.
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
/-- Theorem 1, PDF p.2, (1): all four pure-source rates equal the entropy
of the computed average. This is a statement of Hayashi's cited known result;
no equivalence with a fixed-error library theorem is assumed as a hypothesis. -/
theorem pure_source_coding_rates {L : Type*} {d : ℕ} (S : DiscretePureSource L d) :
let H := (QAlgorithms.Papers.Independent.Information100.P56.entropy S.average : EReal)
blindMinimumRate S false = H ∧
visibleMinimumRate S false = H ∧
blindMinimumRate S true = H ∧
visibleMinimumRate S true = H := by sorry
end QAlgorithms.Papers.Independent.Intakenext100.P20
end
Read-back
What the Lean code literally says, in plain math · gpt-6.1-sol
For every type , natural number , and source having , nonnegative summable weights totaling , and positive semidefinite trace-one matrices each equal to for some unit vector , form
where is defined by continuous functional calculus, is the natural logarithm with its total-function value , and the real number is embedded into the extended reals. The assertion is that all four minimum rates—visible or blind, and with either loss criterion below—equal . More explicitly, at each positive block length , a code has a positive memory dimension and a decoding channel ; a visible encoder assigns a memory density matrix to each , whereas a blind encoder is a channel acting on the input matrix , where . Density matrices are positive semidefinite with trace , and every channel must admit a finite Kraus representation satisfying . With , the loss is
where for visible codes and for blind codes. In each case, the minimum rate means the extended-real infimum of over all code sequences satisfying the chosen criterion: either , or extended-real . Each of these four infima equals .
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.