Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Rains-information dispersion converse

Open
QAlgorithms.Papers.Independent.Support50.P04.rains_dispersion_converse

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

prove2quantum

Let N:A→B\mathcal N:A\to BN:A→B be a CPTP channel on finite positive-dimensional complex Hilbert spaces with fixed orthonormal bases. Write R^(n;ε)\widehat R(n;\varepsilon)R(n;ε) and R^cpp(n;ε)\widehat R^{\mathrm{cpp}}(n;\varepsilon)Rcpp(n;ε) for the maximum unassisted and classical-post-processing assisted entanglement-transmission rates. A code has a positive integer message dimension MMM, rate log⁡2M/n\log_2 M/nlog2​M/n, and squared fidelity to the normalized maximally entangled target at least 1−ε1-\varepsilon1−ε. A cpp encoder may retain sender memory; its decoder is LOCC across that memory and the receiver outputs, with its final quantum output at the receiver and no preshared entanglement. Fix 0<ε<10<\varepsilon<10<ε<1. All logarithms are binary and Φ−1\Phi^{-1}Φ−1 is the standard Gaussian quantile. Remainder constants may depend on the fixed channel and error, and are chosen before the block length: there are C>0C>0C>0 and n0≥2n_0\ge2n0​≥2 such that the displayed bound holds for every integer n≥n0n\ge n_0n≥n0​. Let ω=(id⊗N)(ΦA)\omega=(\mathrm{id}\otimes\mathcal N)(\Phi_A)ω=(id⊗N)(ΦA​). The Rains set is PPT′={σ⪰0:∥σTB∥1≤1}\mathrm{PPT}^{\prime}=\{\sigma\succeq0:\|\sigma^{T_B}\|_1\le1\}PPT′={σ⪰0:∥σTB​∥1​≤1}. Put IR(N)=min⁡σ∈PPT′D(ω∥σ)I_R(\mathcal N)=\min_{\sigma\in\mathrm{PPT}^{\prime}}D(\omega\|\sigma)IR​(N)=minσ∈PPT′​D(ω∥σ), where DDD is Umegaki relative entropy in bits, infinite on support failure. Let ΠR\Pi_RΠR​ be its full minimizer set and let VRεV_R^{\varepsilon}VRε​ be the maximum of the relative-information variance V(ω∥σ)V(\omega\|\sigma)V(ω∥σ) over ΠR\Pi_RΠR​ if ε<1/2\varepsilon<1/2ε<1/2, and the minimum otherwise. Assume full input-unitary covariance, with an output unitary representation. The Rains information is finite and

R^cpp(n;ε)≤IR(N)+VRε(N)/n Φ−1(ε)+Clog⁡2n/n.\widehat R^{\mathrm{cpp}}(n;\varepsilon)\le I_R(\mathcal N)+\sqrt{V_R^\varepsilon(\mathcal N)/n}\,\Phi^{-1}(\varepsilon)+C\log_2n/n.Rcpp(n;ε)≤IR​(N)+VRε​(N)/n​Φ−1(ε)+Clog2​n/n.

This supplies a channel-dependent outer bound.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Support50_P04
Formal statement
noncomputable section

open scoped Matrix ComplexConjugate ComplexOrder

open QAlgorithms

namespace QAlgorithms.Papers.Independent.Support50.P04

variable {A B E : Type} [Fintype A] [DecidableEq A]

variable [Fintype B] [DecidableEq B] [Fintype E] [DecidableEq E]

/-- arXiv:1504.04617v2, PDF p.13, Theorem 3, Eqs.(41)–(43).
All Rains minimizers are constructed from N; maximum variance is chosen below 1/2,
minimum at/above 1/2. Constants may depend on fixed N and epsilon. -/
theorem rains_dispersion_converse [Nonempty A] [Nonempty B]
    (N : Superop A B) (hN : IsChannel N) (hcov : FullUnitaryCovariant N)
    (epsilon : ℝ) (he0 : 0 < epsilon) (he1 : epsilon < 1) :
    ∃ a : ℝ, rainsInformation N = (a : EReal) ∧
      dispersionUpper (fun n => cppBoundary N n epsilon) a (rainsVariance N epsilon) epsilon := by sorry

end QAlgorithms.Papers.Independent.Support50.P04

end
Source
Quantum Coding with Finite Resources, arXiv:1504.04617v2, PDF p. 13, Theorem 3, Eqs. (41)–(43)
Read-back

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

Let A,BA,BA,B be nonempty finite sets with decidable equality, let N:MA(C)→MB(C)\mathcal N:M_A(\mathbb C)\to M_B(\mathbb C)N:MA​(C)→MB​(C) be a channel with a finite complete Kraus family, and suppose there is a group homomorphism V:U(A)→U(B)V:U(A)\to U(B)V:U(A)→U(B) satisfying N(UXU†)=V(U)N(X)V(U)†\mathcal N(UXU^\dagger)=V(U)\mathcal N(X)V(U)^\daggerN(UXU†)=V(U)N(X)V(U)† for every unitary UUU and matrix XXX. For every 0<ϵ<10<\epsilon<10<ϵ<1, there exists a real number aaa such that

a=inf⁡σ≥0∥σTB∥1≤1D(ωN∥σ)a=\inf_{\substack{\sigma\ge0\\\|\sigma^{T_B}\|_1\le1}} D(\omega_{\mathcal N}\|\sigma)a=σ≥0∥σTB​∥1​≤1​inf​D(ωN​∥σ)

as an equality in the extended reals, and there are C>0C>0C>0 and an integer n0≥2n_0\ge2n0​≥2 such that, for every n≥n0n\ge n_0n≥n0​,

Ccpp(N,n,ϵ)≤a+vϵn Φ−1(ϵ)+Clog⁡2nn.C_{\mathrm{cpp}}(\mathcal N,n,\epsilon) \le a+\sqrt{\frac{v_\epsilon}{n}}\,\Phi^{-1}(\epsilon) +C\frac{\log_2n}{n}.Ccpp​(N,n,ϵ)≤a+nvϵ​​​Φ−1(ϵ)+Cnlog2​n​.

Here ωN=JN/∣A∣\omega_{\mathcal N}=J_{\mathcal N}/|A|ωN​=JN​/∣A∣, with JN,(a,b),(c,d)=N(∣a⟩⟨c∣)bdJ_{\mathcal N,(a,b),(c,d)}=\mathcal N(|a\rangle\langle c|)_{bd}JN,(a,b),(c,d)​=N(∣a⟩⟨c∣)bd​; σ(a,b),(c,d)TB=σ(a,d),(c,b)\sigma^{T_B}_{(a,b),(c,d)}=\sigma_{(a,d),(c,b)}σ(a,b),(c,d)TB​​=σ(a,d),(c,b)​; and ∥X∥1=Re⁡Tr⁡X†X\|X\|_1=\operatorname{Re}\operatorname{Tr}\sqrt{X^\dagger X}∥X∥1​=ReTrX†X​. The relative entropy is +∞+\infty+∞ unless ker⁡σ⊆ker⁡ωN\ker\sigma\subseteq\ker\omega_{\mathcal N}kerσ⊆kerωN​; when that inclusion holds it is

D(ω∥σ)=Re⁡Tr⁡[ωln⁡ω−ωln⁡σ]ln⁡2.D(\omega\|\sigma) =\frac{\operatorname{Re}\operatorname{Tr}[\omega\ln\omega-\omega\ln\sigma]}{\ln2}.D(ω∥σ)=ln2ReTr[ωlnω−ωlnσ]​.

The number vϵv_\epsilonvϵ​ is the supremum, when ϵ<1/2\epsilon<1/2ϵ<1/2, and the infimum, when ϵ≥1/2\epsilon\ge1/2ϵ≥1/2, of

Re⁡Tr⁡ ⁣[ωN(log⁡2ωN−log⁡2σ)2]−D(ωN∥σ)2\operatorname{Re}\operatorname{Tr}\!\left[ \omega_{\mathcal N}(\log_2\omega_{\mathcal N}-\log_2\sigma)^2\right] -D(\omega_{\mathcal N}\|\sigma)^2ReTr[ωN​(log2​ωN​−log2​σ)2]−D(ωN​∥σ)2

over precisely those feasible σ\sigmaσ attaining the displayed relative-entropy infimum; no optimizer-attainment assertion is separately included. Matrix logarithms use functional calculus with the total scalar logarithm, including ln⁡0=0\ln0=0ln0=0. These variance extrema are real infima or suprema, with the formal real-order default value 000 for an empty set. The normal quantile is Φ−1(ϵ)=inf⁡{z:ϵ≤∫−∞ze−x2/2/2π dx}\Phi^{-1}(\epsilon)=\inf\{z:\epsilon\le\int_{-\infty}^z e^{-x^2/2}/\sqrt{2\pi}\,dx\}Φ−1(ϵ)=inf{z:ϵ≤∫−∞z​e−x2/2/2π​dx}. The quantity CcppC_{\mathrm{cpp}}Ccpp​ is the extended-real supremum of rates log⁡2M/n\log_2M/nlog2​M/n achievable to entanglement fidelity at least 1−ϵ1-\epsilon1−ϵ by a channel encoder from MMM dimensions to AnA^nAn and a positive finite memory, followed by N⊗n\mathcal N^{\otimes n}N⊗n and a channel decoder acting jointly on that memory and BnB^nBn. Its decoder must factor through separate isometric embeddings, a two-factor map L\mathcal LL, tracing out the first factor, and a channel on the second. The one-outcome map L\mathcal LL must be the stabilized trace-norm limit of total maps of a continuing sequence of finite positive-round protocols: at each history one party uses a countably indexed completely positive instrument and the other outcome-dependent channels, and each stage extends every history by one round; branch maps are summable on every input and preserve trace in total. The stabilized norm takes the supremum over density inputs with an equally dimensional reference. Entanglement fidelity is overlap with the normalized maximally entangled state after applying the effective channel to its second half.

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