Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Abstract and Sections 2–7: equivalent quantum capacity definitions

Open
QAlgorithms.Papers.Independent.Gap20.P07.capacityVariations

by Goku · Oct 22, 2026 · Mathlib 0df444a (Lean v4.33.1)

prove2quantum

Let TTT be a CPTP channel between nonzero complex Hilbert spaces, with independent memoryless uses, arbitrary CPTP coding, and no initial shared entanglement. Let Q(T)Q(T)Q(T) be the supremum of positive rates in Definition 2.1, measured in qubits per use, using the full completely bounded trace norm. Then

Q(T)=Qmin⁡(T)=Qe,min⁡(T)=Qav(T)=Qc(T)=Q1→1(T)=Qspor(T)=Qexp⁡(T)=QQAEP(T)=Qexpdim(T)=Qgen(T)=Qiso(T)=Qcf(T).Q(T)=Q_{\min}(T)=Q_{e,\min}(T)=Q_{\rm av}(T)=Q_c(T)=Q_{1\to1}(T)=Q_{\rm spor}(T)=Q_{\exp}(T)=Q_{\rm QAEP}(T)=Q_{\rm expdim}(T)=Q_{\rm gen}(T)=Q_{\rm iso}(T)=Q_{\rm cf}(T).Q(T)=Qmin​(T)=Qe,min​(T)=Qav​(T)=Qc​(T)=Q1→1​(T)=Qspor​(T)=Qexp​(T)=QQAEP​(T)=Qexpdim​(T)=Qgen​(T)=Qiso​(T)=Qcf​(T).

The terms denote, respectively, minimum pure-state fidelity, worst-case entanglement fidelity, Haar average fidelity, maximally mixed source fidelity, ordinary induced trace-norm error, sporadic block lengths, some positive exponential error exponent, QAEP entropy sources, exponentially bounded source dimensions with a source-dependent growth constant, entanglement generation from arbitrary normalized inputs, isometric encoding, and unrestricted finite forward messages per block. Each is an extended nonnegative supremum. Also Q(T⊗S)=Q(T)Q(T\otimes S)=Q(T)Q(T⊗S)=Q(T) for every operational measurement-and-preparation channel SSS; every positive R<Q(T)R<Q(T)R<Q(T) admits all-block codes with error at most Ce−anC e^{-an}Ce−an for some C,a>0C,a>0C,a>0; and

lim⁡ϵ↓0Qϵ(T)=Q(T).\lim_{\epsilon\downarrow0}Q_\epsilon(T)=Q(T).ϵ↓0lim​Qϵ​(T)=Q(T).

Here QϵQ_\epsilonQϵ​ uses eventual full-cb error at most ϵ\epsilonϵ. These are equalities of capacities, without asserting endpoint achievability. Zero-error and unconstrained entropy capacities are exceptions. This result identifies the qualified equivalence asserted in the abstract.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Gap20_P07
Formal statement
noncomputable section

open Filter MeasureTheory

open scoped Topology ENNReal NNReal Matrix ComplexOrder ComplexConjugate

attribute [local instance] Classical.propDecidable

attribute [local instance] Classical.decEq

attribute [local instance] IsStarNormal.instContinuousFunctionalCalculus

namespace QAlgorithms.Papers.Independent.Gap20.P07

open QAlgorithms

open QAlgorithms.RealRegisters (Hilbert)

open QAlgorithms.Papers.Independent.Dependency50.P33
  (State IsTraceClass IsTraceClassChannel traceSize rankOne tensorApply)

open QAlgorithms.Papers.Independent.Frontier30.P14
  (Operation isometryCoeff conjugateKernel)

open QAlgorithms.Papers.Independent.Claimed50.P35
  (stabilizedTraceClassNorm traceClassTensorMap traceClassChannelPower
   transpose transposeBound TraceLinear tracePair MeasurePrepare)

open QAlgorithms.Papers.Independent.Gap48.P29 (fe FiniteCPMap)

/-- PDF p.1 and Sections 2--7: the qualified abstract claim, excluding the
rejected exact-zero-error and unconstrained entropy variants. The capacities,
not endpoint sets of achievable rates, coincide. -/
theorem capacityVariations {I J : Type} [Nonempty I] [Nonempty J]
    (T : Operation I J) (hT : IsTraceClassChannel T) :
    (∀ c : FidelityCriterion, fidelityCapacity T c = qubitReferenceCapacity T) ∧
    sporadicCapacity T = qubitReferenceCapacity T ∧
    ordinaryCapacity T = qubitReferenceCapacity T ∧
    exponentialCapacity T = qubitReferenceCapacity T ∧
    sourceEntropyCapacity T QuantumSource.QAEP = qubitReferenceCapacity T ∧
    sourceEntropyCapacity T QuantumSource.ExponentialDimension = qubitReferenceCapacity T ∧
    variableSourceGenerationCapacity T = qubitReferenceCapacity T ∧
    isometricCapacity T = qubitReferenceCapacity T ∧
    instrumentForwardCapacity T (fun _ _ => True) = qubitReferenceCapacity T ∧
    (∀ {K L : Type} [Nonempty K] [Nonempty L] (S : Operation K L),
      IsTraceClassChannel S → MeasurePrepare S →
        qubitReferenceCapacity (traceClassTensorMap T S) = qubitReferenceCapacity T) ∧
    (∀ R : ℝ, 0 < R → ENNReal.ofReal R < qubitReferenceCapacity T → ExponentialReliable T R) ∧
    Tendsto (finiteErrorCapacity T) (𝓝[>] (0 : ℝ)) (𝓝 (qubitReferenceCapacity T)) := by sorry

end QAlgorithms.Papers.Independent.Gap20.P07

end
Source
Tema Con Variazioni: Quantum Channel Capacity, arXiv:quant-ph/0311037v1, PDF p. 1, Abstract and Sections 2–7: equivalent quantum capacity definitions
Read-back

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

Let I,JI,JI,J be arbitrary nonempty types and let TTT map complex coefficient matrices on III to those on JJJ. The following definitions specify the hypotheses and capacities in the assertion. Write HA=ℓ2(A;C)H_A=\ell^2(A;\mathbb C)HA​=ℓ2(A;C), with coordinate vectors eae_aea​; a coefficient matrix XXX is trace class when it has an expansion Xij=∑r≥0ur(i)vr(j)‾X_{ij}=\sum_{r\ge0}u_r(i)\overline{v_r(j)}Xij​=∑r≥0​ur​(i)vr​(j)​, with ur,vr∈HAu_r,v_r\in H_Aur​,vr​∈HA​ and ∑r∥ur∥∥vr∥<∞\sum_r\|u_r\|\|v_r\|<\infty∑r​∥ur​∥∥vr​∥<∞, and write ν(X)\nu(X)ν(X) for the infimum of these sums in [0,∞][0,\infty][0,∞], with value +∞+\infty+∞ if there is no expansion. A density coefficient matrix is trace class, has real nonnegative quadratic forms ∑i,j∈Fzi‾Xijzj\sum_{i,j\in F}\overline{z_i}X_{ij}z_j∑i,j∈F​zi​​Xij​zj​ for every finite F⊆AF\subseteq AF⊆A and every z:A→Cz:A\to\mathbb Cz:A→C, and has summable real diagonal entries whose sum is 111. Call a map a channel when it preserves trace class, is additive and complex homogeneous on trace-class inputs, and, for every q∈Nq\in\mathbb Nq∈N, its extension

(Φ⊗idq)(X)(j,a),(j′,b)=Φ ⁣((X(i,a),(i′,b))i,i′)j,j′(\Phi\otimes\mathrm{id}_q)(X)_{(j,a),(j',b)} =\Phi\!\left((X_{(i,a),(i',b)})_{i,i'}\right)_{j,j'}(Φ⊗idq​)(X)(j,a),(j′,b)​=Φ((X(i,a),(i′,b)​)i,i′​)j,j′​

takes every density coefficient matrix on I×Fin⁡qI\times\operatorname{Fin}qI×Finq to a density coefficient matrix; assume TTT is such a channel. Define \|\Phi\|_*=\sup_{q\ge0}\sup_{\text{\(X trace class},,\nu(X)\le1}\nu((\Phi\otimes\mathrm{id}_q)X)). Tensor powers act independently on coordinates by successive application to coefficient blocks; the zeroth power is the identity on the one-dimensional register indexed by the empty function. For positive integers n,mn,mn,m, let eT(n,m)e_T(n,m)eT​(n,m) be the infimum of ∥D∘T⊗n∘E−id2⊗m∥∗\|D\circ T^{\otimes n}\circ E-\mathrm{id}_{2}^{\otimes m}\|_*∥D∘T⊗n∘E−id2⊗m​∥∗​ over channels E,DE,DE,D of the required input and output types. Let Q(T)Q(T)Q(T) be the supremum of positive real RRR such that, for every positive-integer sequence na,man_a,m_ana​,ma​ with na→∞n_a\to\inftyna​→∞ and lim sup⁡a(ma/na)≤R\limsup_a(m_a/n_a)\le Rlimsupa​(ma​/na​)≤R, one has eT(na,ma)→0e_T(n_a,m_a)\to0eT​(na​,ma​)→0; all capacity suprema here are in [0,∞][0,\infty][0,∞], with empty supremum 000. A finite block code of length bbb and dimension ddd consists of channels E:Fin⁡d→IbE:\operatorname{Fin}d\to I^bE:Find→Ib and D:Jb→Fin⁡dD:J^b\to\operatorname{Fin}dD:Jb→Find, understood as maps on coefficient matrices, with effective map A=D∘T⊗b∘EA=D\circ T^{\otimes b}\circ EA=D∘T⊗b∘E and error ∥A−id∥∗\|A-\mathrm{id}\|_*∥A−id∥∗​. For finite matrices put ∥X∥1=Re⁡tr⁡X†X\|X\|_1=\operatorname{Re}\operatorname{tr}\sqrt{X^\dagger X}∥X∥1​=RetrX†X​, pψ=ψψ†p_\psi=\psi\psi^\daggerpψ​=ψψ†, and

fe(ρ,A)=Re⁡∑i,j,k,lρjiρlk‾ [A(Ejl)]ik,f_e(\rho,A)=\operatorname{Re}\sum_{i,j,k,l} \rho_{ji}\overline{\rho_{lk}}\,[A(E_{jl})]_{ik},fe​(ρ,A)=Rei,j,k,l∑​ρji​ρlk​​[A(Ejl​)]ik​,

where EjlE_{jl}Ejl​ is the matrix unit. The four fidelity values are respectively the infimum of Re⁡tr⁡(pψA(pψ))\operatorname{Re}\operatorname{tr}(p_\psi A(p_\psi))Retr(pψ​A(pψ​)) over unit vectors ψ\psiψ; the infimum of fe(ρ,A)f_e(\rho,A)fe​(ρ,A) over positive semidefinite trace-one ρ\rhoρ; the Haar average of Re⁡tr⁡(pψA(pψ))\operatorname{Re}\operatorname{tr}(p_\psi A(p_\psi))Retr(pψ​A(pψ​)), with ψ=Ue0\psi=Ue_0ψ=Ue0​ and normalized Haar-distributed UUU; and Re⁡tr⁡(pΦd(id⊗A)(pΦd))\operatorname{Re}\operatorname{tr}(p_{\Phi_d}(\mathrm{id}\otimes A)(p_{\Phi_d}))Retr(pΦd​​(id⊗A)(pΦd​​)), where Φd=d−1/2∑iei⊗ei\Phi_d=d^{-1/2}\sum_i e_i\otimes e_iΦd​=d−1/2∑i​ei​⊗ei​. For any one of these four criteria ccc, let Fc(T)F_c(T)Fc​(T) be the supremum of positive RRR admitting codes of length a+1a+1a+1, positive dimension dad_ada​, lim sup⁡alog⁡2(da)/(a+1)=R\limsup_a\log_2(d_a)/(a+1)=Rlimsupa​log2​(da​)/(a+1)=R, and fidelity value tending to 111. Let Qsp(T)Q_{\mathrm{sp}}(T)Qsp​(T) instead use the existence of positive-integer sequences na,man_a,m_ana​,ma​, with na→∞n_a\to\inftyna​→∞, lim sup⁡ma/na=R\limsup m_a/n_a=Rlimsupma​/na​=R, and eT(na,ma)→0e_T(n_a,m_a)\to0eT​(na​,ma​)→0. Let Qord(T)Q_{\mathrm{ord}}(T)Qord​(T) use codes of length a+1a+1a+1 and the same dimension-rate condition as FcF_cFc​, with sup⁡∥X∥1≤1∥Aa(X)−X∥1→0\sup_{\|X\|_1\le1}\|A_a(X)-X\|_1\to0sup∥X∥1​≤1​∥Aa​(X)−X∥1​→0. Say RRR is exponentially reliable when there exist a0,C>0a_0,C>0a0​,C>0 and codes at every length b≥1b\ge1b≥1, of dimension 2⌊bR⌋2^{\lfloor bR\rfloor}2⌊bR⌋, whose errors are eventually at most Ce−a0bCe^{-a_0b}Ce−a0​b; Qexp⁡(T)Q_{\exp}(T)Qexp​(T) is the supremum of positive such RRR. A source is a sequence of positive dimensions dad_ada​ and density matrices ρa\rho_aρa​; its entropy rate is h=lim sup⁡aS(ρa)/(a+1)h=\limsup_a S(\rho_a)/(a+1)h=limsupa​S(ρa​)/(a+1), interpreted in [0,∞][0,\infty][0,∞], where S(ρ)=−Re⁡tr⁡(ρlog⁡2ρ)S(\rho)=-\operatorname{Re}\operatorname{tr}(\rho\log_2\rho)S(ρ)=−Retr(ρlog2​ρ) is defined by functional calculus with 0log⁡20=00\log_2 0=00log2​0=0. It is reliable if codes of length a+1a+1a+1 and dimension dad_ada​ have fe(ρa,Aa)→1f_e(\rho_a,A_a)\to1fe​(ρa​,Aa​)→1. For a property P\mathcal PP, let Qsrc,P(T)Q_{\mathrm{src},\mathcal P}(T)Qsrc,P​(T) be the supremum of positive RRR for which a reliable source satisfying P\mathcal PP has R≤hR\le hR≤h. The source property QAEP means there is a finite r≥0r\ge0r≥0 with h=rh=rh=r such that, for every η>0\eta>0η>0, the mass Re⁡tr⁡(PaρaPa)\operatorname{Re}\operatorname{tr}(P_a\rho_aP_a)Retr(Pa​ρa​Pa​) tends to 111, where PaP_aPa​ is the spectral indicator of [2−(a+1)(r+η),2−(a+1)(r−η)][2^{-(a+1)(r+\eta)},2^{-(a+1)(r-\eta)}][2−(a+1)(r+η),2−(a+1)(r−η)]; exponential dimension means ∃τ>0 ∀a, da≤τa+1\exists\tau>0\,\forall a,\ d_a\le\tau^{a+1}∃τ>0∀a, da​≤τa+1. Let Qgen(T)Q_{\mathrm{gen}}(T)Qgen​(T) use the dimension-rate condition of FcF_cFc​, arbitrary unit vectors γa∈Cda⊗Cda\gamma_a\in\mathbb C^{d_a}\otimes\mathbb C^{d_a}γa​∈Cda​⊗Cda​, and Re⁡tr⁡(pΦda(id⊗Aa)(pγa))→1\operatorname{Re}\operatorname{tr}(p_{\Phi_{d_a}}(\mathrm{id}\otimes A_a)(p_{\gamma_a}))\to1Retr(pΦda​​​(id⊗Aa​)(pγa​​))→1. Let Qiso(T)Q_{\mathrm{iso}}(T)Qiso​(T) use the fourth fidelity criterion, additionally requiring every encoder to have an isometry V:HFin⁡da→HIa+1V:H_{\operatorname{Fin}d_a}\to H_{I^{a+1}}V:HFinda​​→HIa+1​ whose coefficient matrix Kij=(Vej)iK_{ij}=(Ve_j)_iKij​=(Vej​)i​ satisfies E(X)ij=∑uKiu∑vXuvKjv‾E(X)_{ij}=\sum_uK_{iu}\sum_vX_{uv}\overline{K_{jv}}E(X)ij​=∑u​Kiu​∑v​Xuv​Kjv​​ for every trace-class XXX. Let Qfwd(T)Q_{\mathrm{fwd}}(T)Qfwd​(T) allow at each length an arbitrary positive finite number of encoder branches EjE_jEj​, each trace-class preserving and complex linear on trace-class inputs and satisfying (Ej⊗idq)(ρ)=pσ(E_j\otimes\mathrm{id}_q)(\rho)=p\sigma(Ej​⊗idq​)(ρ)=pσ for every finite-ancilla density ρ\rhoρ, for some 0≤p≤10\le p\le10≤p≤1 and density σ\sigmaσ; require ∑jEj\sum_jE_j∑j​Ej​ to be a channel, give each branch its own channel decoder DjD_jDj​, and require the sum of the fourth fidelity values of DjT⊗bEjD_jT^{\otimes b}E_jDj​T⊗bEj​ to tend to 111, with the same dimension-rate condition. Finally, Qϵ(T)Q_\epsilon(T)Qϵ​(T) replaces the vanishing-error requirement in Q(T)Q(T)Q(T) by eventual eT(na,ma)≤max⁡(ϵ,0)e_T(n_a,m_a)\le\max(\epsilon,0)eT​(na​,ma​)≤max(ϵ,0). With this notation the theorem asserts

∀c,Fc(T)=Q(T),Qsp(T)=Qord(T)=Qexp⁡(T)=Qsrc,QAEP(T)=Qsrc,exponential dimension(T)=Qgen(T)=Qiso(T)=Qfwd(T)=Q(T),∀R>0,R<Q(T) ⟹ R is exponentially reliable,lim⁡ϵ↓0Qϵ(T)=Q(T).\begin{gathered} \forall c,\quad F_c(T)=Q(T),\\ Q_{\mathrm{sp}}(T)=Q_{\mathrm{ord}}(T)=Q_{\exp}(T) =Q_{\mathrm{src},\mathrm{QAEP}}(T) =Q_{\mathrm{src},\mathrm{exponential\ dimension}}(T) =Q_{\mathrm{gen}}(T)=Q_{\mathrm{iso}}(T)=Q_{\mathrm{fwd}}(T)=Q(T),\\ \forall R>0,\quad R<Q(T)\ \Longrightarrow\ R\text{ is exponentially reliable},\\ \lim_{\epsilon\downarrow0}Q_\epsilon(T)=Q(T). \end{gathered}∀c,Fc​(T)=Q(T),Qsp​(T)=Qord​(T)=Qexp​(T)=Qsrc,QAEP​(T)=Qsrc,exponential dimension​(T)=Qgen​(T)=Qiso​(T)=Qfwd​(T)=Q(T),∀R>0,R<Q(T) ⟹ R is exponentially reliable,ϵ↓0lim​Qϵ​(T)=Q(T).​

It also asserts that Q(T⊗S)=Q(T)Q(T\otimes S)=Q(T)Q(T⊗S)=Q(T) for every pair of nonempty types K,LK,LK,L and every channel S:K→LS:K\to LS:K→L having the following measure-and-prepare representation: there exist a measurable outcome space XXX, bounded positive operator effects M(A)M(A)M(A) with M(X)=1M(X)=1M(X)=1, finite measures μu\mu_uμu​ satisfying μu(A)=max⁡(Re⁡⟨u,M(A)u⟩,0)\mu_u(A)=\max(\operatorname{Re}\langle u,M(A)u\rangle,0)μu​(A)=max(Re⟨u,M(A)u⟩,0) on measurable AAA, probability laws PρP_\rhoPρ​ for density inputs satisfying Pρ(A)=max⁡(Re⁡∑i∑jρij(M(A))ji,0)P_\rho(A)=\max(\operatorname{Re}\sum_i\sum_j\rho_{ij}(M(A))_{ji},0)Pρ​(A)=max(Re∑i​∑j​ρij​(M(A))ji​,0), and output densities σx\sigma_xσx​ with measurable coefficients, such that every coefficient of S(ρ)S(\rho)S(ρ) is the integrable average ∫(σx)ij dPρ(x)\int(\sigma_x)_{ij}\,dP_\rho(x)∫(σx​)ij​dPρ​(x). Here the tensor map is literally (T⊗S)(X)(j,l),(j′,l′)=T((S((X(i,k),(i′,k′))k,k′)l,l′)i,i′)j,j′(T\otimes S)(X)_{(j,l),(j',l')}=T((S((X_{(i,k),(i',k')})_{k,k'})_{l,l'})_{i,i'})_{j,j'}(T⊗S)(X)(j,l),(j′,l′)​=T((S((X(i,k),(i′,k′)​)k,k′​)l,l′​)i,i′​)j,j′​. No countability assumptions are imposed on the nonempty index types; code dimensions may equal 111, capacities may be infinite, and the final limit uses strictly positive ϵ\epsilonϵ.

Human review
  • Endorsed by Shuze Chen · Oct 22, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Goku · Oct 22, 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