Data processing of Petz overlaps and their log moments
OpenQAlgorithms.Papers.Independent.Intakethird100.P04.petz_data_processingLet be arbitrary density operators on a complex Hilbert space , and let be a trace-preserving completely positive linear map on trace-class operators, with arbitrary output Hilbert space. For define
Powers of exponent zero are the identity. The fractional-power product is trace class; its trace is nonnegative, and the logarithm is extended with . For every ,
At either endpoint the overlaps equal one. The paper credits Uhlmann and Petz for earlier proofs and presents these inequalities as a consequence of its operational approach; their statements do not require its conversion formula as a hypothesis.
import Definitions.Def_QAlgorithms_Papers_Independent_Intakethird100_P04
noncomputable section
open Filter
open scoped ENNReal NNReal BigOperators ComplexOrder
namespace QAlgorithms.Papers.Independent.Intakethird100.P04
attribute [local instance] Classical.propDecidable Classical.decEq
attribute [local instance] IsStarNormal.instContinuousFunctionalCalculus
open QAlgorithms.Papers.Independent.Dependency50.P13
(State operatorTrace relativeEntropy)
open QAlgorithms.Papers.Independent.Support50.P20
(Effect traceAcceptance IsHilbertCPTP BlockHilbert blockBasis)
open QAlgorithms.Papers.Independent.Frontier18.P18 (boundedOperator)
variable {H K : Type*}
variable [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H]
variable [NormedAddCommGroup K] [InnerProductSpace ℂ K] [CompleteSpace K]
/-- Remark 2, PDF p.7, (31)--(32): both overlap and extended log-moment
inequalities on [0,1]. Ordinary A^0=I makes endpoints equalities, also for singular
states. Source credits Uhlmann [14] and Petz [15], claiming a new proof. -/
theorem petz_data_processing {I J : Type*}
(b : HilbertBasis I ℂ H) (d : HilbertBasis J ℂ K)
(rho sigma : State b)
(Phi : (H →L[ℂ] H) → (K →L[ℂ] K)) (hPhi : IsHilbertCPTP b d Phi)
(s : ℝ) (hs : 0 ≤ s) (hs1 : s ≤ 1) :
ENNReal.ofReal (traceAcceptance b
(CFC.rpow rho.val.val (1 - s)) (CFC.rpow sigma.val.val s)) ≤
ENNReal.ofReal (traceAcceptance d
(CFC.rpow (Phi rho.val.val) (1 - s)) (CFC.rpow (Phi sigma.val.val) s)) ∧
logMoment b rho.val.val sigma.val.val s ≤
logMoment d (Phi rho.val.val) (Phi sigma.val.val) s := by sorry
end QAlgorithms.Papers.Independent.Intakethird100.P04
end
Read-back
What the Lean code literally says, in plain math · gpt-6.1-sol
Let be complete complex inner-product spaces with Hilbert bases and , where are arbitrary types, and define , and similarly for . Let be trace-class operators on , meaning that each admits a representation with ; assume each is self-adjoint, satisfies for every , and has complex trace . Let be a function on bounded operators from to bounded operators on , additive and complex homogeneous on trace-class operators, mapping them to trace-class operators with . Assume also that for every and every trace-class operator block , reality and nonnegativity of for every tuple imply the corresponding reality and nonnegativity after applying to every block entry. For every real with , define using the continuous-functional-calculus real power appearing in the statement, and put
The theorem asserts both
The quantities are regarded as extended nonnegative reals, and the logarithms are extended natural logarithms, taking value at zero and at infinity. Thus the first inequality explicitly uses the nonnegative truncation of the real trace, and the second is the expanded meaning of the two logarithmic moments. Both endpoints are included, with no invertibility assumption on either state. No countability, separability, or nonzero-dimension assumption is imposed; the trace sums use total summation, returning zero for nonsummable families, and the hypotheses have no instances when normalized states do not exist.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.