Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma 3: continuity of conditional entropy and mutual information

Proved
QAlgorithms.Papers.Independent.Gap48.P39.continuity_information

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

prove2quantum

Let ρAB,σAB\rho^{AB},\sigma^{AB}ρAB,σAB be positive trace-one density matrices on the same positive-dimensional finite bipartite system. Let 0≤δ≤1/e0\le\delta\le1/e0≤δ≤1/e and suppose their full trace-norm distance is at most δ\deltaδ. Then

∣H(A∣B)ρ−H(A∣B)σ∣≤2H2(δ)+4δlog⁡2∣AB∣,|H(A\mid B)_\rho-H(A\mid B)_\sigma|\le2H_2(\delta)+4\delta\log_2|AB|,∣H(A∣B)ρ​−H(A∣B)σ​∣≤2H2​(δ)+4δlog2​∣AB∣, ∣I(A;B)ρ−I(A;B)σ∣≤3H2(δ)+6δlog⁡2∣AB∣.|I(A;B)_\rho-I(A;B)_\sigma|\le3H_2(\delta)+6\delta\log_2|AB|.∣I(A;B)ρ​−I(A;B)σ​∣≤3H2​(δ)+6δlog2​∣AB∣.

Here H2H_2H2​ is binary entropy, ∣AB∣|AB|∣AB∣ is the product dimension, H(A∣B)=H(AB)−H(B)H(A\mid B)=H(AB)-H(B)H(A∣B)=H(AB)−H(B), and I(A;B)=H(A)+H(B)−H(AB)I(A;B)=H(A)+H(B)-H(AB)I(A;B)=H(A)+H(B)−H(AB), with actual partial traces and base-two von Neumann entropy. Both bounds include δ=0\delta=0δ=0 and singleton systems. These continuity estimates use the full norm, without a factor of one half.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Gap48_P39
Formal statement
noncomputable section

open scoped BigOperators Matrix ComplexConjugate

namespace QAlgorithms.Papers.Independent.Gap48.P39

open QAlgorithms.Papers.Independent.Information100.P21 (Op)

open QAlgorithms.Papers.Independent.Information100.P03 (PureEnsemble)

open QAlgorithms.Papers.Independent.Claimed50.P08 (cqBlocks holevoInfo)

open QAlgorithms.Papers.Independent.Dependency50.P19 (both epr mutualInformation marginalRA marginalRB)

open QAlgorithms.Papers.Independent.Information100.P58 (Distribution)

open QAlgorithms.Papers.Independent.Information100.P67 (squaredFidelity conditionalEntropy groupedChannelPower)

attribute [local instance] Classical.propDecidable Classical.decEq

variable {A B C X T E : Type}

variable [Fintype A] [DecidableEq A] [Fintype B] [DecidableEq B]

variable [Fintype C] [DecidableEq C] [Fintype X] [DecidableEq X]

variable [Fintype T] [DecidableEq T] [Fintype E] [DecidableEq E]

/-- Lemma 3, PDF p.10: absolute differences; full trace norm and base-two logs. -/
theorem continuity_information [Nonempty A] [Nonempty B]
    (rho sigma : Op (A × B)) (hr : IsDensityMatrix rho) (hs : IsDensityMatrix sigma)
    (delta : ℝ) (h0 : 0 ≤ delta) (h1 : delta ≤ 1 / Real.exp 1)
    (hclose : traceNorm (rho - sigma) ≤ delta) :
    |conditionalEntropy rho - conditionalEntropy sigma| ≤
      2 * binEntropy2 delta + 4 * delta * Real.logb 2 (Fintype.card (A × B)) ∧
    |mutualInformation rho - mutualInformation sigma| ≤
      3 * binEntropy2 delta + 6 * delta * Real.logb 2 (Fintype.card (A × B)) := by sorry

end QAlgorithms.Papers.Independent.Gap48.P39

end
Source
Quantum broadcast channels, arXiv:quant-ph/0603098v2, PDF p. 10, Lemma 3: continuity of conditional entropy and mutual information
Read-back

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

For nonempty finite types A,BA,BA,B with decidable equality, let ρ,σ\rho,\sigmaρ,σ be positive semidefinite complex matrices on A×BA\times BA×B, each with trace 111. Define H(M)=−Re⁡tr⁡(f(M))H(M)=-\operatorname{Re}\operatorname{tr}(f(M))H(M)=−Retr(f(M)), with f(u)=ulog⁡2uf(u)=u\log_2uf(u)=ulog2​u applied by continuous functional calculus, H(A∣B)M=H(M)−H(Tr⁡AM)H(A\mid B)_M=H(M)-H(\operatorname{Tr}_A M)H(A∣B)M​=H(M)−H(TrA​M), I(A:B)M=H(Tr⁡BM)+H(Tr⁡AM)−H(M)I(A:B)_M=H(\operatorname{Tr}_B M)+H(\operatorname{Tr}_A M)-H(M)I(A:B)M​=H(TrB​M)+H(TrA​M)−H(M), and ∥Z∥1=Re⁡tr⁡Z†Z\|Z\|_1=\operatorname{Re}\operatorname{tr}\sqrt{Z^\dagger Z}∥Z∥1​=RetrZ†Z​. For every real δ\deltaδ satisfying 0≤δ≤e−10\le\delta\le e^{-1}0≤δ≤e−1 and ∥ρ−σ∥1≤δ\|\rho-\sigma\|_1\le\delta∥ρ−σ∥1​≤δ, both inequalities hold:

∣H(A∣B)ρ−H(A∣B)σ∣≤2h2(δ)+4δlog⁡2(∣A∣∣B∣),∣I(A:B)ρ−I(A:B)σ∣≤3h2(δ)+6δlog⁡2(∣A∣∣B∣),\begin{aligned} |H(A\mid B)_\rho-H(A\mid B)_\sigma| &\le2h_2(\delta)+4\delta\log_2(|A||B|),\\ |I(A:B)_\rho-I(A:B)_\sigma| &\le3h_2(\delta)+6\delta\log_2(|A||B|), \end{aligned}∣H(A∣B)ρ​−H(A∣B)σ​∣∣I(A:B)ρ​−I(A:B)σ​∣​≤2h2​(δ)+4δlog2​(∣A∣∣B∣),≤3h2​(δ)+6δlog2​(∣A∣∣B∣),​

where h2(u)=(−ulog⁡u−(1−u)log⁡(1−u))/log⁡2h_2(u)=(-u\log u-(1-u)\log(1-u))/\log2h2​(u)=(−ulogu−(1−u)log(1−u))/log2, with 0log⁡0=00\log0=00log0=0. The trace-norm hypothesis has no factor of 1/21/21/2, and the endpoint δ=0\delta=0δ=0 is included.

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