Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Commuting rotations give a state-dependent linear recovery bound

Open
QAlgorithms.Papers.Independent.Support50.P44.commuting_rotations_linear_bound

by Goku · Oct 23, 2026 · Mathlib 0df444a (Lean v4.33.1)

prove2quantum

Let ρABC\rho_{ABC}ρABC​ be a density operator on finite-dimensional complex systems A,B,CA,B,CA,B,C, with marginals ρBC\rho_{BC}ρBC​, ρB\rho_BρB​ and ρAB\rho_{AB}ρAB​. There exist a CPTP map R:B→B⊗CR:B\to B\otimes CR:B→B⊗C and unitaries UB,VBCU_B,V_{BC}UB​,VBC​ commuting with ρB,ρBC\rho_B,\rho_{BC}ρB​,ρBC​ respectively, such that, for every XXX supported on supp⁡ρB\operatorname{supp}\rho_BsuppρB​,

R(X)=VBCρBC1/2(ρB−1/2UBXUB†ρB−1/2⊗IC)ρBC1/2VBC†,R(X)=V_{BC}\rho_{BC}^{1/2}(\rho_B^{-1/2}U_B XU_B^\dagger\rho_B^{-1/2}\otimes I_C)\rho_{BC}^{1/2}V_{BC}^\dagger,R(X)=VBC​ρBC1/2​(ρB−1/2​UB​XUB†​ρB−1/2​⊗IC​)ρBC1/2​VBC†​,

and

F(ρABC,(idA⊗R)(ρAB))≥1−ln⁡22I(A:C∣B)ρ.F(\rho_{ABC},(\mathrm{id}_A\otimes R)(\rho_{AB}))\ge1-\frac{\ln2}{2}I(A:C\mid B)_\rho.F(ρABC​,(idA​⊗R)(ρAB​))≥1−2ln2​I(A:C∣B)ρ​.

Inverse square roots are generalized inverses. The full CPTP map completes the supported formula outside that support, and may depend on the entire state. This is the nonuniversal structural recovery result used by the corollary.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Support50_P44
Formal statement
noncomputable section

namespace QAlgorithms.Papers.Independent.Support50.P44

open QAlgorithms.Papers.Independent.Dependency50.P13 (IsTraceClass compression)

open QAlgorithms.Papers.Independent.Dependency50.P21 (marginalAB)

open QAlgorithms.Papers.Independent.Renew46.P37 (classicalKL)

open Filter

open scoped ENNReal ComplexOrder Matrix

attribute [local instance] Classical.propDecidable

attribute [local instance] Classical.decEq

variable {H : Type*} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H]

variable {A B C E G : Type*}

/-- PDF pp.19–20, Lemma 6.1, (116)–(117). Source correction: the formula is
restricted to supported inputs as explicitly stipulated in (115); require a full
CPTP completion. Singular rhoB is allowed. R may depend on the full state. -/
theorem commuting_rotations_linear_bound [Fintype A] [Fintype B] [Fintype C]
    [DecidableEq A] [DecidableEq B] [DecidableEq C]
    (rho : Matrix (A × (B × C)) (A × (B × C)) ℂ) (hrho : IsDensityMatrix rho) :
    let rhoBC := traceLeft rho
    let rhoB : Matrix B B ℂ := fun b b2 => ∑ c, rhoBC (b,c) (b2,c)
    ∃ R : Superop B (B × C), IsChannel R ∧
      ∃ U : Matrix.unitaryGroup B ℂ, ∃ V : Matrix.unitaryGroup (B × C) ℂ,
        (U : Matrix B B ℂ) * rhoB = rhoB * (U : Matrix B B ℂ) ∧
        (V : Matrix (B × C) (B × C) ℂ) * rhoBC =
          rhoBC * (V : Matrix (B × C) (B × C) ℂ) ∧
        (∀ X, finiteSupport rhoB * X * finiteSupport rhoB = X →
          R X = rotatedTranspose rhoBC U V X) ∧
        1 - Real.log 2 / 2 * finiteCMI rho ≤
          rootFidelity rho (idTensorSuperop R (marginalAB rho)) := by sorry

end QAlgorithms.Papers.Independent.Support50.P44

end
Source
Universal recovery map for approximate Markov chains, arXiv:1504.07251v3, PDF p. 19, Lemma 6.1
Read-back

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

Let A,B,CA,B,CA,B,C be finite types equipped with decidable equality, and let ρ\rhoρ be a positive semidefinite complex matrix of trace one indexed by A×(B×C)A\times(B\times C)A×(B×C). Define ρBC\rho_{BC}ρBC​ by tracing out AAA, ρAB\rho_{AB}ρAB​ by tracing out CCC, and ρB\rho_BρB​ by tracing out CCC from ρBC\rho_{BC}ρBC​; explicitly, their entries are respectively ∑aρ(a,(b,c)),(a,(b′,c′))\sum_a\rho_{(a,(b,c)),(a,(b',c'))}∑a​ρ(a,(b,c)),(a,(b′,c′))​, ∑cρ(a,(b,c)),(a′,(b′,c))\sum_c\rho_{(a,(b,c)),(a',(b',c))}∑c​ρ(a,(b,c)),(a′,(b′,c))​, and ∑c(ρBC)(b,c),(b′,c)\sum_c(\rho_{BC})_{(b,c),(b',c)}∑c​(ρBC​)(b,c),(b′,c)​. Put S2(T)=−Re⁡Tr⁡(f(T))S_2(T)=-\operatorname{Re}\operatorname{Tr}(f(T))S2​(T)=−ReTr(f(T)) for f(x)=xlog⁡2xf(x)=x\log_2xf(x)=xlog2​x evaluated by real continuous functional calculus, and put I=S2(ρAB)+S2(ρBC)−S2(ρB)−S2(ρ)I=S_2(\rho_{AB})+S_2(\rho_{BC})-S_2(\rho_B)-S_2(\rho)I=S2​(ρAB​)+S2​(ρBC​)−S2​(ρB​)−S2​(ρ). Then there exist a map RRR from BBB-matrices to (B×C)(B\times C)(B×C)-matrices and unitary matrices UUU on BBB and VVV on B×CB\times CB×C with the following properties: R(X)=∑j<mKjXKj∗R(X)=\sum_{j<m}K_jXK_j^*R(X)=∑j<m​Kj​XKj∗​ for some natural number mmm and matrices KjK_jKj​ satisfying ∑jKj∗Kj=IB\sum_jK_j^*K_j=I_B∑j​Kj∗​Kj​=IB​; UρB=ρBUU\rho_B=\rho_BUUρB​=ρB​U and VρBC=ρBCVV\rho_{BC}=\rho_{BC}VVρBC​=ρBC​V; and, writing Q=ρBg(ρB)Q=\rho_Bg(\rho_B)Q=ρB​g(ρB​) with g(x)=x−1g(x)=x^{-1}g(x)=x−1, D=h(ρB)D=h(\rho_B)D=h(ρB​) with h(x)=(x)−1h(x)=(\sqrt x)^{-1}h(x)=(x​)−1, and S=ρBCS=\sqrt{\rho_{BC}}S=ρBC​​, every matrix XXX satisfying QXQ=XQXQ=XQXQ=X obeys

R(X)=VS((DUXU∗D)⊗IC)SV∗.R(X)=VS\bigl((DUXU^*D)\otimes I_C\bigr)SV^*.R(X)=VS((DUXU∗D)⊗IC​)SV∗.

Furthermore, if τ=(id⁡A⊗R)(ρAB)\tau=(\operatorname{id}_A\otimes R)(\rho_{AB})τ=(idA​⊗R)(ρAB​), meaning τ(a,u),(a′,v)=R((ρAB)(a,⋅),(a′,⋅))u,v\tau_{(a,u),(a',v)}=R((\rho_{AB})_{(a,\cdot),(a',\cdot)})_{u,v}τ(a,u),(a′,v)​=R((ρAB​)(a,⋅),(a′,⋅)​)u,v​, then

1−ln⁡22I≤Re⁡Tr⁡ρ τρ.1-\frac{\ln2}{2}I \le \operatorname{Re}\operatorname{Tr}\sqrt{\sqrt\rho\,\tau\sqrt\rho}.1−2ln2​I≤ReTrρ​τρ​​.

All matrix functions above use real continuous functional calculus, with reciprocal zero defined as zero. No nonemptiness assumptions are stated on the index types, but the trace-one hypothesis cannot hold if any of A,B,CA,B,CA,B,C is empty.

Human review
  • Endorsed by Shuze Chen · Oct 24, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Goku · Oct 24, 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