Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proposition 5: channel-independent output mutual-information bound

Open
QAlgorithms.Papers.Independent.Support50.P18.output_mutual_information_continuity

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

prove2quantum

Let Φ:A→B\Phi:A\to BΦ:A→B be any CPTP channel on trace class and RRR any separable reference. Suppose HA≥0H_A\ge0HA​≥0 satisfies (22), write E0E_0E0​ for its ground infimum, and fix finite E≥E0E\ge E_0E≥E0​. Put JΦ(ρ)=I(B:R)(Φ⊗IdR)(ρ)J_\Phi(\rho)=I(B:R)_{(\Phi\otimes\mathrm{Id}_R)(\rho)}JΦ​(ρ)=I(B:R)(Φ⊗IdR​)(ρ)​, with mutual information defined by extended relative entropy. This function is finite and uniformly continuous in trace distance on states of ARARAR with input AAA energy at most EEE. The entropy capacity is finite at every finite budget above E0E_0E0​, and

∣JΦ(ρ)−JΦ(σ)∣≤22ε FHA(E/ε)+2g(2ε)|J_\Phi(\rho)-J_\Phi(\sigma)|\le2\sqrt{2\varepsilon}\,F_{H_A}(E/\varepsilon)+2g(\sqrt{2\varepsilon})∣JΦ​(ρ)−JΦ​(σ)∣≤22ε​FHA​​(E/ε)+2g(2ε​)

whenever both input energies are at most EEE and 0<ε≤1/20<\varepsilon\le1/20<ε≤1/2 bounds their half trace distance. For pure joint inputs the modulus parameter may be replaced by ε2/2\varepsilon^2/2ε2/2 under the original distance hypothesis. The bound is independent of the channel and places no energy constraint on its output.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Support50_P18
Formal statement
noncomputable section

open Filter

open scoped ENNReal NNReal BigOperators Topology

attribute [local instance] Classical.propDecidable Classical.decEq

namespace QAlgorithms.Papers.Independent.Support50.P18

open QAlgorithms.Papers.Independent.Dependency50.P33 (State PureState traceSize rankOne)

open QAlgorithms.Papers.Independent.Frontier18.P01 (convexOperatorMixture traceSecond)

open QAlgorithms.Papers.Independent.Frontier18.P07
  (extendedEntropyNats extendedRelativeEntropyNats)

open QAlgorithms.RealRegisters (Hilbert)

open QAlgorithms.Papers.Independent.Claimed50.P39 (g)

open QAlgorithms.Papers.Independent.Renew46.P08
  (coefficientEnergy entropySupremum extendedHalfTraceDistance)

/-- Proposition 5, PDF pp.16--17, (48): arbitrary CPTP channel and reference.
Condition (22) forces discrete spectrum (Appendix p.21), justifying the choice
of an energy eigenbasis on A without a new spectral hypothesis. There is no
output-energy or output-entropy restriction. The pure refinement retains the
original d(rho,sigma)≤eps and substitutes only in the modulus. -/
theorem output_mutual_information_continuity {I J R : Type}
    [Countable I] [Nonempty I] [Countable J] [Nonempty J] [Countable R] [Nonempty R]
    (energy : I → ℝ≥0) (h22 : Condition22 energy)
    (Phi : QAlgorithms.Papers.Independent.Frontier18.P01.Channel I J)
    (E : ℝ) (hE : spectralGround energy ≤ E) :
    (∀ u : ℝ, spectralGround energy ≤ u → entropySupremum energy u < ⊤) ∧
    EnergyContinuity (fun X => coefficientEnergy energy (traceSecond X))
      (@outputMutualInformation I J R Phi) E (entropyModulus (entropySupremum energy) 2 2 E) ∧
    ∀ (rho sigma : PureState (I × R)) (eps : ℝ),
      coefficientEnergy energy (traceSecond rho.val.coefficients) ≤ ENNReal.ofReal E →
      coefficientEnergy energy (traceSecond sigma.val.coefficients) ≤ ENNReal.ofReal E →
      0 < eps → eps ≤ 1 / 2 → nuclearStateDistance rho.val sigma.val ≤ ENNReal.ofReal eps →
      |(outputMutualInformation Phi rho.val.coefficients).toReal -
        (outputMutualInformation Phi sigma.val.coefficients).toReal| ≤
        entropyModulus (entropySupremum energy) 2 2 E (eps ^ 2 / 2) := by sorry

end QAlgorithms.Papers.Independent.Support50.P18

end
Source
Adaptation of the Alicki-Fannes-Winter method for the set of states with bounded energy and its use, arXiv:1609.07044v5, PDF p. 17, Proposition 5
Read-back

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

Let I,J,RI,J,RI,J,R be countable nonempty index types, let e:I→[0,∞)e:I\to[0,\infty)e:I→[0,∞) satisfy (∑ie−λei)λ→1\left(\sum_i e^{-\lambda e_i}\right)^\lambda\to1(∑i​e−λei​)λ→1 in [0,∞][0,\infty][0,∞] as λ↓0\lambda\downarrow0λ↓0, and let E≥e0:=inf⁡ieiE\ge e_0:=\inf_i e_iE≥e0​:=infi​ei​. For any index set LLL, an expansion of a matrix XXX means Xij=∑nun(i)vn(j)‾X_{ij}=\sum_nu_n(i)\overline{v_n(j)}Xij​=∑n​un​(i)vn​(j)​ with un,vn∈ℓ2(L;C)u_n,v_n\in\ell^2(L;\mathbb C)un​,vn​∈ℓ2(L;C) and ∑n∥un∥∥vn∥<∞\sum_n\|u_n\|\|v_n\|<\infty∑n​∥un​∥∥vn​∥<∞; call such a matrix trace class, and let τ(X)\tau(X)τ(X) be the infimum of these costs, with empty infimum +∞+\infty+∞. A state is a trace class matrix whose quadratic forms on every finite coordinate set are real and nonnegative and whose real diagonal is summable with sum 111. Let Φ\PhiΦ be a map from III-indexed matrices to JJJ-indexed matrices that preserves trace class, is additive and complex homogeneous on trace class matrices, and, for every natural qqq and every state XXX on I×{0,…,q−1}I\times\{0,\ldots,q-1\}I×{0,…,q−1}, sends its blocks to a state: the resulting entries are Φ(i,j↦X(i,a),(j,b))uv\Phi(i,j\mapsto X_{(i,a),(j,b)})_{uv}Φ(i,j↦X(i,a),(j,b)​)uv​. This finite ancillary condition includes q=0q=0q=0, where there is no input state. For any bipartite matrix XXX, define its two marginals by summing the diagonal in the other coordinate. Define HHH and DDD by

H(A)=sup⁡s[tslog⁡ts−Re⁡tr⁡(Aslog⁡As)]+,D(A∥B)=sup⁡s{[Re⁡tr⁡(Aslog⁡As−Aslog⁡Bs+Bs−As)]+,ker⁡Bs⊆ker⁡As,+∞,otherwise,H(A)=\sup_s[t_s\log t_s-\operatorname{Re}\operatorname{tr}(A_s\log A_s)]_+,\qquad D(A\|B)=\sup_s \begin{cases} [\operatorname{Re}\operatorname{tr}(A_s\log A_s-A_s\log B_s+B_s-A_s)]_+, &\ker B_s\subseteq\ker A_s,\\ +\infty,&\text{otherwise}, \end{cases}H(A)=ssup​[ts​logts​−Retr(As​logAs​)]+​,D(A∥B)=ssup​{[Retr(As​logAs​−As​logBs​+Bs​−As​)]+​,+∞,​kerBs​⊆kerAs​,otherwise,​

where sss ranges over all finite coordinate sets, ts=Re⁡tr⁡Ast_s=\operatorname{Re}\operatorname{tr}A_sts​=RetrAs​, and matrix logarithms use continuous functional calculus with log⁡0=0\log0=0log0=0. Put Ee(A)=∑iei[(Aii).re]+\mathcal E_e(A)=\sum_i e_i[(A_{ii}).\mathrm{re}]_+Ee​(A)=∑i​ei​[(Aii​).re]+​, F(u)=sup⁡{H(A):A is a state on I, Ee(A)≤[u]+}F(u)=\sup\{H(A):A\text{ is a state on }I,\ \mathcal E_e(A)\le[u]_+\}F(u)=sup{H(A):A is a state on I, Ee​(A)≤[u]+​}, and g(x)=(x+1)log⁡(x+1)−xlog⁡xg(x)=(x+1)\log(x+1)-x\log xg(x)=(x+1)log(x+1)−xlogx; empty suprema are 000. For a state XXX on I×RI\times RI×R, let Y(j,r),(k,s)=Φ(i,l↦X(i,r),(l,s))jkY_{(j,r),(k,s)}=\Phi(i,l\mapsto X_{(i,r),(l,s)})_{jk}Y(j,r),(k,s)​=Φ(i,l↦X(i,r),(l,s)​)jk​, and set MΦ(X)=D(Y∥YJ⊗YR)M_\Phi(X)=D(Y\|Y_J\otimes Y_R)MΦ​(X)=D(Y∥YJ​⊗YR​). Then F(u)<+∞F(u)<+\inftyF(u)<+∞ for all real u≥e0u\ge e_0u≥e0​, and on the set A={X:Ee(XI)≤E}\mathcal A=\{X:\mathcal E_e(X_I)\le E\}A={X:Ee​(XI​)≤E} of states, MΦM_\PhiMΦ​ is finite and its real value is uniformly continuous for d(X,Z)=τ(X−Z)/2d(X,Z)=\tau(X-Z)/2d(X,Z)=τ(X−Z)/2: for every η>0\eta>0η>0, some δ>0\delta>0δ>0 makes d(X,Z)<δd(X,Z)<\deltad(X,Z)<δ imply ∣MΦ(X)♭−MΦ(Z)♭∣<η|M_\Phi(X)^\flat-M_\Phi(Z)^\flat|<\eta∣MΦ​(X)♭−MΦ​(Z)♭∣<η for all X,Z∈AX,Z\in\mathcal AX,Z∈A. For all X,Z∈AX,Z\in\mathcal AX,Z∈A and 0<ε≤120<\varepsilon\le\tfrac120<ε≤21​ with d(X,Z)≤εd(X,Z)\le\varepsilond(X,Z)≤ε,

∣MΦ(X)♭−MΦ(Z)♭∣≤22ε F(E/ε)♭+2g(2ε).|M_\Phi(X)^\flat-M_\Phi(Z)^\flat| \le 2\sqrt{2\varepsilon}\,F(E/\varepsilon)^\flat+2g(\sqrt{2\varepsilon}).∣MΦ​(X)♭−MΦ​(Z)♭∣≤22ε​F(E/ε)♭+2g(2ε​).

In addition, if X,ZX,ZX,Z are pure states, meaning Xab=uaub‾X_{ab}=u_a\overline{u_b}Xab​=ua​ub​​ and Zab=vavb‾Z_{ab}=v_a\overline{v_b}Zab​=va​vb​​ for unit vectors u,v∈ℓ2(I×R;C)u,v\in\ell^2(I\times R;\mathbb C)u,v∈ℓ2(I×R;C), the same energy and distance hypotheses give

∣MΦ(X)♭−MΦ(Z)♭∣≤2ε F(2E/ε2)♭+2g(ε).|M_\Phi(X)^\flat-M_\Phi(Z)^\flat| \le 2\varepsilon\,F(2E/\varepsilon^2)^\flat+2g(\varepsilon).∣MΦ​(X)♭−MΦ​(Z)♭∣≤2εF(2E/ε2)♭+2g(ε).

Here a♭a^\flata♭ converts finite extended nonnegative values to real numbers and sends +∞+\infty+∞ to 000; the asserted finiteness applies to all information values in these bounds, and zero distance is allowed.

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