Choi and root-fidelity primal semidefinite programs
OpenQAlgorithms.Papers.Independent.Intakenext100.P75.recovery_primal_sdpLet and be density operators on finite-dimensional nonzero complex systems, and let purify . Set , reindexed onto , and . Then both following maxima are attained:
Here is squared Uhlmann fidelity. The Choi variable is positive but unnormalized; the second program optimizes root fidelity. These are the primal optimization formulations used in the duality argument, including for singular purifications.
import Definitions.Def_QAlgorithms_Papers_Independent_Intakenext100_P75
namespace QAlgorithms.Papers.Independent.Intakenext100.P75
open scoped Matrix ComplexOrder
open QAlgorithms.Papers.Independent.Information100.P63 (Mat)
open QAlgorithms.Papers.Independent.Information100.P53 (fidelity)
open QAlgorithms.Papers.Independent.Information100.P08 (IsPure)
open QAlgorithms.Papers.Independent.Dependency50.P18 (bipartiteTensor)
variable {A B C D Ap Bp Cp R I J K L : Type}
variable [Fintype A] [DecidableEq A] [Nonempty A]
variable [Fintype B] [DecidableEq B] [Nonempty B]
variable [Fintype C] [DecidableEq C] [Nonempty C]
variable [Fintype D] [DecidableEq D] [Nonempty D]
variable [Fintype Ap] [DecidableEq Ap] [Nonempty Ap]
variable [Fintype Bp] [DecidableEq Bp] [Nonempty Bp]
variable [Fintype Cp] [DecidableEq Cp] [Nonempty Cp]
variable [Fintype R] [DecidableEq R] [Nonempty R]
/-- PDF p.3, (15),(16): both primal maxima are attained; (16) optimizes
ROOT fidelity, whereas (15) optimizes squared fidelity. -/
theorem recovery_primal_sdp
[Nonempty A] [Nonempty B] [Nonempty C] [Nonempty D] (rho : Mat (A × B)) (sigma : Mat (A × C))
(purification : Mat ((A × C) × D))
(hrho : IsDensityMatrix rho) (hsigma : IsDensityMatrix sigma)
(hpure : IsPure purification) (hmarginal : traceRight purification = sigma) :
IsGreatest {v : ℝ | ∃ t : Mat ((A × B) × D), t.PosSemidef ∧
(Matrix.of fun x y : A × D => ∑ b, t ((x.1, b), x.2) ((y.1, b), y.2)) = 1 ∧
v = fidelity rho (choiOutput (marginalAD purification) t)} (generalizedRecoveryFidelity rho sigma) ∧
IsGreatest {v : ℝ | ∃ t : Mat ((A × B) × D), ∃ z : Mat (A × B),
t.PosSemidef ∧
(Matrix.of fun x y : A × D => ∑ b, t ((x.1, b), x.2) ((y.1, b), y.2)) = 1 ∧
(Matrix.fromBlocks rho z z.conjTranspose
(choiOutput (marginalAD purification) t)).PosSemidef ∧
v = (z + z.conjTranspose).trace.re / 2} (Real.sqrt (generalizedRecoveryFidelity rho sigma)) := by sorry
end QAlgorithms.Papers.Independent.Intakenext100.P75
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 positive semidefinite trace-one matrices on , and let on , where and . Set , and, for a matrix on , define , where . With , let be the supremum of over finite Kraus channels , whose Kraus operators obey . Both of the following maxima are attained and have the stated values:
\max_{\substack{t\ge0,\ z\text{ any matrix on }A\times B\\ \sum_b t_{((a,b),d),((a',b),d')}=\delta_{a,a'}\delta_{d,d'}\\ \begin{psmallmatrix}\rho&z\\z^\dagger&O_S(t)\end{psmallmatrix}\ge0}} \frac{\operatorname{Re}\operatorname{tr}(z+z^\dagger)}2=\sqrt G.Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.