Finite Holevo optimization and attaining unitary encoding
ProvedQAlgorithms.Papers.Independent.Support50.P39.finiteHolevoOptimizationLet be a normalized finite dimensional density operator, with both local dimensions positive and not necessarily equal. Fix resource copies and an output dimension . All entropies are in bits. Define to be the supremum of the Holevo score
over every finite probability distribution and CPTP maps . Then, with ,
An optimal channel exists, together with unitaries such that for every traceless matrix . With probabilities , the encoders attain . Moreover, the same unitary family works for every CPTP preprocessing : its encoding ensemble is admissible and, writing , its score is exactly
The last identity uses the signed entropy difference, even if negative. Only the optimized expression can use clipped coherent information. The statement includes , when the optimized value is zero and the unitary family has one member.
This is the finite optimization identity proved on pp.4–5. It supplies an explicit attaining ensemble without equating the objective to the unrestricted fixed- operational capacity.
import Definitions.Def_QAlgorithms_Papers_Independent_Support50_P39
namespace QAlgorithms.Papers.Independent.Support50.P39
noncomputable section
open QAlgorithms
open Filter Topology
open scoped Matrix ComplexConjugate ComplexOrder
attribute [local instance] Classical.propDecidable Classical.decEq
/-- arXiv:quant-ph/0106080v3, PDF pp.4--6, Eqs. (9)--(13): the finite
Holevo objective H_d equals log2 d plus optimized clipped coherent information,
and an optimal CPTP map and d^2-element unitary encoding attain it.
The SAME twirling family gives, for EVERY preprocessing map, the signed
per-map identity chi=log2 d+S(B^n)-S(sigma). Clipping is used only at the
optimized value. This includes d=1 (both sides zero) and does not identify
H_d with the unrestricted fixed-(d,n) operational capacity. -/
theorem finiteHolevoOptimization {A B : Type*} [Fintype A] [Fintype B]
[DecidableEq A] [DecidableEq B] [Nonempty A] [Nonempty B]
(rho : Matrix (A × B) (A × B) ℂ) (hrho : IsDensityMatrix rho)
(n d : ℕ) (hn : 0 < n) (hd : 0 < d) :
denseCodingHolevoValue rho n d = Real.logb 2 d + finiteCoherentOpt rho n d ∧
∃ Phi : Superop (Fin n → A) (Fin d), IsChannel Phi ∧
clippedCoherentInformation (aliceAction Phi (noisyResourceCopies rho n)) =
finiteCoherentOpt rho n d ∧
∃ U : Fin (d*d) → Matrix.unitaryGroup (Fin d) ℂ,
TracelessTwirling U ∧
(∀ Psi : Superop (Fin n → A) (Fin d), IsChannel Psi →
IsEncoding (uniformWeight d) (unitaryEncoders Psi U) ∧
encodingHolevo rho n (uniformWeight d) (unitaryEncoders Psi U) =
Real.logb 2 d + vnEntropy (traceLeft (noisyResourceCopies rho n)) -
vnEntropy (aliceAction Psi (noisyResourceCopies rho n))) ∧
encodingHolevo rho n (uniformWeight d) (unitaryEncoders Phi U) =
denseCodingHolevoValue rho n d := by sorry
end
end QAlgorithms.Papers.Independent.Support50.P39
Read-back
What the Lean code literally says, in plain math · gpt-6.1-sol
Let be finite, nonempty types with decidable equality, let be a positive semidefinite complex matrix on with trace , and let satisfy , including the possibility . Write , with applied by continuous functional calculus. Let , and, for a map on Alice's matrices, let , meaning that is applied to each Alice matrix block with Bob's row and column fixed. Write for the partial trace over Alice and set . A channel is a map with a finite Kraus representation satisfying . For a finite ensemble of channels , indexed by , with , real , and , define
Let be the real supremum of all these , and let be the real supremum of over all channels . The declaration asserts , and asserts that there exist such a channel and a family of unitary matrices satisfying and
For this same family , every channel gives an ensemble of channels , with positive ensemble size , nonnegative weights summing to , and
For the particular channel chosen above, this ensemble additionally satisfies .
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.