Uniform leading norm multiplicativity and entropy additivity for weak noise
OpenQAlgorithms.Papers.Independent.Support50.P31.weakNoiseLeadingAdditivityFix , positive dimensions , and a norm exponent . For channels on -dimensional complex spaces define and . Let be maximum output Schatten purity, minimum output natural entropy, and . Then, uniformly over the channel tuple as ,
Precisely, for each there is a cutoff , chosen before the channels, such that for every channel tuple and , the absolute residuals in the norm expansions are at most , and the entropy residual is at most . The cutoff may depend on . This is leading-order additivity, without asserting exact equality for positive noise.
import Definitions.Def_QAlgorithms_Papers_Independent_Support50_P31
noncomputable section
open Filter Asymptotics
open scoped ComplexConjugate ENNReal
namespace QAlgorithms.Papers.Independent.Support50.P31
attribute [local instance] Classical.propDecidable Classical.decEq
open QAlgorithms.Papers.Independent.Information100.P56 (entropy)
open QAlgorithms.Papers.Independent.Information100.P63 (schatten)
open QAlgorithms.Papers.Independent.Gap48.P03 (holevoQuantity)
open QAlgorithms.Papers.Independent.Claimed50.P38 (tensorSuperop)
open QAlgorithms.Papers.Independent.Dependency50.P44 (tensorChannels)
/-- PDF pp.7–9, Theorem and (17): leading norm multiplicativity and entropy
additivity uniformly over channel tuples. The cutoff precedes the channels and
may depend on fixed dimensions, n and p. Distinct norm and entropy scales. -/
theorem weakNoiseLeadingAdditivity {n : ℕ} (hn : 0 < n) (d : Fin n → ℕ) (hd : ∀ i, 0 < d i)
(p : ℝ≥0∞) (hp : 1 ≤ p) :
∀ eta : ℝ, 0 < eta → ∃ delta : ℝ, 0 < delta ∧
∀ Phi : (i : Fin n) → Superop (Fin (d i)) (Fin (d i)),
(∀ i, IsChannel (Phi i)) → ∀ eps : ℝ, 0 < eps → eps < delta → eps < 1 →
(|nuP p (tensorChannels (fun i => weakNoise eps (Phi i))) -
∏ i, nuP p (weakNoise eps (Phi i))| ≤ eta * eps) ∧
(|nuH (tensorChannels (fun i => weakNoise eps (Phi i))) -
∑ i, nuH (weakNoise eps (Phi i))| ≤ eta * |eps * Real.log eps|) ∧
(|nuP ⊤ (tensorChannels (fun i => weakNoise eps (Phi i))) -
(1 - eps * ∑ i, (1 - nuFlat (Phi i)))| ≤ eta * eps) ∧
(∀ i, |nuP ⊤ (weakNoise eps (Phi i)) -
(1 - eps + eps * nuFlat (Phi i))| ≤ eta * eps) := by sorry
end QAlgorithms.Papers.Independent.Support50.P31
end
Read-back
What the Lean code literally says, in plain math · gpt-6.1-sol
For every positive natural number , every family of positive natural dimensions indexed by , and every extended nonnegative real , the following uniform assertion holds. For a map on a finite complex matrix space, write for the supremum over positive semidefinite trace-one inputs of when , and of when ; write for the infimum over these inputs of , with natural logarithms and ; and write . For every real there exists a real such that, for every family in which each map has a finite Kraus representation with , and every real with , put and let be the tensor map whose entries are . Then
The chosen precedes the quantification over the maps, so it applies to every such family with the fixed dimensions and exponent; dimensions equal to one and are included.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.