Lemma 8: convexity of operational ss-private capacity
OpenQAlgorithms.Papers.Independent.Support50.P07.ssPrivateCapacity_convexLet have compatible input and receiver spaces, and let . Let be any Stinespring realization of the receiver map . Then
Each operational capacity uses its channel's full complementary output.
All quantum registers are finite and nonzero, states have trace one, pure vectors have norm one, and entropy and logarithms are in bits. Operational private capacities use memoryless powers, positive blocklengths and message counts, uniform messages, arbitrary mixed block inputs, and normalized decoding POVMs. A code is good when its joint decoded message/environment state has full trace norm less than the error from a perfectly correlated uniform pair tensor its actual common environment marginal. Rates use every positive error and rate slack at all sufficiently large blocklengths. Symmetric assistance allows one arbitrary finite symmetric channel per block without charging its dimension.
This result supplies a capacity identity or bound for the channel family and resource model stated above.
import Definitions.Def_QAlgorithms_Papers_Independent_Support50_P07
noncomputable section
open scoped BigOperators Matrix ComplexConjugate ComplexOrder
open Filter
namespace QAlgorithms.Papers.Independent.Support50.P07
open QAlgorithms.Papers.Independent.Information100.P67
(IsIsometry receiverChannel complementaryChannel IsDegradable IsAntidegradable
dilationPower idealJoint maximumCoherentInformation)
open QAlgorithms.Papers.Independent.Claimed50.P24 (productDilation)
open QAlgorithms.Papers.Independent.Dependency50.P19 (mutualInformation)
variable {A B E C D F : Type}
variable [Fintype A] [DecidableEq A] [Fintype B] [DecidableEq B]
variable [Fintype E] [DecidableEq E] [Fintype C] [DecidableEq C]
variable [Fintype D] [DecidableEq D] [Fintype F] [DecidableEq F]
/-- Lemma 8, PDF p.5: convexity compares RECEIVER maps. Each channel's full
complementaryChannel comes from its own isometry; no mixture of isometries is taken. -/
theorem ssPrivateCapacity_convex [Nonempty A] [Nonempty B] [Nonempty E]
{G H : Type} [Fintype G] [DecidableEq G]
[Nonempty G] [Fintype H] [DecidableEq H] [Nonempty H]
(U : Matrix (B × E) A ℂ) (V : Matrix (B × G) A ℂ)
(W : Matrix (B × H) A ℂ) (hU : IsIsometry U) (hV : IsIsometry V)
(hW : IsIsometry W) (p : ℝ) (hp : p ∈ Set.Icc 0 1)
(hmix : ∀ rho, receiverChannel W rho = ((1 - p : ℝ) : ℂ) • receiverChannel U rho +
(p : ℂ) • receiverChannel V rho) :
ssPrivateCapacity W ≤ (1 - p) * ssPrivateCapacity U + p * ssPrivateCapacity V := by sorry
end QAlgorithms.Papers.Independent.Support50.P07
end
Read-back
What the Lean code literally says, in plain math · gpt-6.1-sol
Let be finite nonempty types with decidable equality. Let , , and satisfy , and let . Assume that, for every complex matrix on , including matrices that are not density matrices,
For any dilation , define as the real supremum of nonnegative rates with the following property: for every , all sufficiently large positive integers admit integers , density matrices on , and a receiver POVM , such that and , where swaps the auxiliary factors. Density matrices are positive semidefinite with trace one, and POVM elements are positive semidefinite and sum to . Let apply independently times and route one auxiliary factor to each output, put , and put ; require , where . The auxiliary dimension may depend on , with no stated bound. Then
Both endpoints and are included.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.