Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Optimum recovery is an attained semidefinite program

Proved
QAlgorithms.Papers.Independent.Dependency50.P40.optimumRecoverySDP

by Goku · Oct 19, 2026 · Mathlib 0df444a (Lean v4.33.1)

prove2quantum

Let H=ChH=\mathbb C^hH=Ch and K=CkK=\mathbb C^kK=Ck, with h,k>0h,k>0h,k>0, and fix a density operator ρ\rhoρ on HHH and a CPTP noise channel E:L(H)→L(K)\mathcal E:L(H)\to L(K)E:L(H)→L(K) with finite Kraus operators EjE_jEj​. For any normalized purification ∣ψ⟩|\psi\rangle∣ψ⟩ of ρ\rhoρ, define Fe(ρ,B)=⟨ψ∣(I⊗B)(∣ψ⟩⟨ψ∣)∣ψ⟩F_e(\rho,\mathcal B)=\langle\psi|(I\otimes\mathcal B)(|\psi\rangle\langle\psi|)|\psi\rangleFe​(ρ,B)=⟨ψ∣(I⊗B)(∣ψ⟩⟨ψ∣)∣ψ⟩. With output-first, unnormalized operator kets, put Cρ,E=∑j∣ρEj†⟩⟩⟨⟨ρEj†∣C_{\rho,\mathcal E}=\sum_j|\rho E_j^\dagger\rangle\rangle\langle\langle\rho E_j^\dagger|Cρ,E​=∑j​∣ρEj†​⟩⟩⟨⟨ρEj†​∣. Then

max⁡R  CPTPFe(ρ,R∘E)=max⁡X⪰0,  tr⁡HX=IKtr⁡(XCρ,E).\max_{\mathcal R\;\mathrm{CPTP}}F_e(\rho,\mathcal R\circ\mathcal E) =\max_{X\succeq0,\;\operatorname{tr}_H X=I_K}\operatorname{tr}(XC_{\rho,\mathcal E}).RCPTPmax​Fe​(ρ,R∘E)=X⪰0,trH​X=IK​max​tr(XCρ,E​).

Both maxima are attained. Every recovery gives a feasible X=XRX=X_{\mathcal R}X=XR​ with this objective, and every feasible XXX is the Choi matrix of a CPTP recovery. This identifies the exact recovery problem with an SDP; it supplies no bit-complexity claim.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Dependency50_P40
Formal statement
noncomputable section

namespace QAlgorithms.Papers.Independent.Dependency50.P40

open scoped ComplexConjugate ComplexOrder Matrix

open QAlgorithms.Papers.Independent.Information100.P34 (dyad)

/-- PDF pp.1–4, abstract and (7), (22): the exact recovery SDP has an attained maximum.
Every feasible Choi matrix is included, and every CPTP recovery is compared. -/
theorem optimumRecoverySDP {h k a m : ℕ} (hh : 0 < h) (hk : 0 < k)
    (rho : Matrix (Fin h) (Fin h) ℂ) (hrho : IsDensityMatrix rho)
    (psi : EuclideanSpace ℂ (Fin a × Fin h)) (hpsi : IsPurification rho psi)
    (E : Fin m → Matrix (Fin k) (Fin h) ℂ) (hE : IsKrausFamily E) :
    (∀ R : Superop (Fin k) (Fin h), IsChannel R →
      ChoiFeasible (outputFirstChoi R) ∧ purificationFidelity psi (R ∘ krausApply E) =
        (outputFirstChoi R * recoveryCoefficient rho E).trace.re) ∧
    (∀ X : Matrix (Fin h × Fin k) (Fin h × Fin k) ℂ, ChoiFeasible X →
      ∃ R : Superop (Fin k) (Fin h), IsChannel R ∧ outputFirstChoi R = X) ∧
    ∃ (R : Superop (Fin k) (Fin h))
      (X : Matrix (Fin h × Fin k) (Fin h × Fin k) ℂ),
      IsChannel R ∧ outputFirstChoi R = X ∧ ChoiFeasible X ∧
      purificationFidelity psi (R ∘ krausApply E) =
        (X * recoveryCoefficient rho E).trace.re ∧
      (∀ Y, ChoiFeasible Y → (Y * recoveryCoefficient rho E).trace.re ≤
        (X * recoveryCoefficient rho E).trace.re) ∧
      (∀ S : Superop (Fin k) (Fin h), IsChannel S →
        purificationFidelity psi (S ∘ krausApply E) ≤
          purificationFidelity psi (R ∘ krausApply E)) := by sorry

end QAlgorithms.Papers.Independent.Dependency50.P40

end
Source
Optimum Quantum Error Recovery using Semidefinite Programming, arXiv:quant-ph/0606035v1, PDF p. 4, Abstract; IV, Eq. (22): optimum entanglement-fidelity recovery is an attained SDP
Read-back

What the Lean code literally says, in plain math · gpt-6.1-sol

For all natural numbers h,k,a,mh,k,a,mh,k,a,m with h>0h>0h>0 and k>0k>0k>0, let ρ\rhoρ be a complex h×hh\times hh×h positive semidefinite matrix with tr⁡ρ=1\operatorname{tr}\rho=1trρ=1, let ψ∈Ca×h\psi\in\mathbb C^{a\times h}ψ∈Ca×h have Euclidean norm 111 and satisfy ∑rψr,uψr,v‾=ρu,v\sum_r\psi_{r,u}\overline{\psi_{r,v}}=\rho_{u,v}∑r​ψr,u​ψr,v​​=ρu,v​, and let E0,…,Em−1E_0,\ldots,E_{m-1}E0​,…,Em−1​ be complex k×hk\times hk×h matrices satisfying ∑jEj†Ej=Ih\sum_jE_j^\dagger E_j=I_h∑j​Ej†​Ej​=Ih​. Write E(M)=∑jEjMEj†\mathcal E(M)=\sum_jE_jME_j^\daggerE(M)=∑j​Ej​MEj†​. A channel from dimension ppp to dimension qqq means a function on matrices having a finite representation Λ(M)=∑sAsMAs†\Lambda(M)=\sum_sA_sMA_s^\daggerΛ(M)=∑s​As​MAs†​ for every MMM, with ∑sAs†As=Ip\sum_sA_s^\dagger A_s=I_p∑s​As†​As​=Ip​. Define its output-first Choi matrix by JΛ(b,u),(c,v)=Λ(euv)b,cJ_\Lambda{}_{(b,u),(c,v)}=\Lambda(e_{uv})_{b,c}JΛ​(b,u),(c,v)​=Λ(euv​)b,c​, where euve_{uv}euv​ has its only nonzero entry, 111, at (u,v)(u,v)(u,v); call a matrix XXX indexed by h×kh\times kh×k feasible when XXX is positive semidefinite and ∑bX(b,u),(b,v)=δuv\sum_bX_{(b,u),(b,v)}=\delta_{uv}∑b​X(b,u),(b,v)​=δuv​. For a matrix AAA, write ∣A⟩|A\rangle∣A⟩ for the vector with coordinates ∣A⟩(b,u)=Ab,u|A\rangle_{(b,u)}=A_{b,u}∣A⟩(b,u)​=Ab,u​, and set

K=∑j=0m−1∣ρEj†⟩⟨ρEj†∣.K=\sum_{j=0}^{m-1}|\rho E_j^\dagger\rangle\langle\rho E_j^\dagger|.K=j=0∑m−1​∣ρEj†​⟩⟨ρEj†​∣.

Here an outer product ∣v⟩⟨w∣|v\rangle\langle w|∣v⟩⟨w∣ has entries vpwq‾v_p\overline{w_q}vp​wq​​. For an endomorphism Λ\LambdaΛ on h×hh\times hh×h matrices, let

Fψ(Λ)=Re⁡ ⁣(ψ†(id⁡a⊗Λ)(ψψ†)ψ),\mathcal F_\psi(\Lambda) =\operatorname{Re}\!\left(\psi^\dagger(\operatorname{id}_a\otimes\Lambda)(\psi\psi^\dagger)\psi\right),Fψ​(Λ)=Re(ψ†(ida​⊗Λ)(ψψ†)ψ),

where the extension applies Λ\LambdaΛ separately to each h×hh\times hh×h block indexed by a pair of reference indices. Then every channel RRR from dimension kkk to dimension hhh has feasible JRJ_RJR​ and satisfies Fψ(R∘E)=Re⁡tr⁡(JRK)\mathcal F_\psi(R\circ\mathcal E)=\operatorname{Re}\operatorname{tr}(J_RK)Fψ​(R∘E)=Retr(JR​K); every feasible XXX is JRJ_RJR​ for some such channel RRR; and there exist such a channel RRR and feasible X=JRX=J_RX=JR​ for which

Fψ(R∘E)=Re⁡tr⁡(XK),Re⁡tr⁡(YK)≤Re⁡tr⁡(XK) for every feasible Y,Fψ(S∘E)≤Fψ(R∘E) for every channel S.\mathcal F_\psi(R\circ\mathcal E)=\operatorname{Re}\operatorname{tr}(XK), \quad \operatorname{Re}\operatorname{tr}(YK)\le\operatorname{Re}\operatorname{tr}(XK) \ \text{for every feasible }Y, \quad \mathcal F_\psi(S\circ\mathcal E)\le\mathcal F_\psi(R\circ\mathcal E) \ \text{for every channel }S.Fψ​(R∘E)=Retr(XK),Retr(YK)≤Retr(XK) for every feasible Y,Fψ​(S∘E)≤Fψ​(R∘E) for every channel S.

The parameters a,ma,ma,m have no explicit positivity hypotheses, but a=0a=0a=0 is incompatible with ∥ψ∥=1\|\psi\|=1∥ψ∥=1, and m=0m=0m=0 is incompatible with the Kraus identity when h>0h>0h>0.

Human review
  • Endorsed by Shuze Chen · Oct 20, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Goku · Oct 20, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me