Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Testing divergence bounded by Umegaki relative entropy

Open
QAlgorithms.Papers.Independent.Support50.P20.testingDivergence_relativeEntropyBound

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

prove2quantum

Let ρ,σ\rho,\sigmaρ,σ be arbitrary normalized positive trace-class operators on the same complex Hilbert space, and let 0≤ϵ<10\le\epsilon<10≤ϵ<1. Write D(ρ∥σ)D(\rho\|\sigma)D(ρ∥σ) for the extended Umegaki relative entropy in bits and Hb(t)=−tlog⁡2t−(1−t)log⁡2(1−t)H_b(t)=-t\log_2t-(1-t)\log_2(1-t)Hb​(t)=−tlog2​t−(1−t)log2​(1−t) for binary entropy, with 0log⁡20=00\log_2 0=00log2​0=0. Then

DHϵ(ρ∥σ)≤D(ρ∥σ)+Hb(ϵ)1−ϵ.D_H^\epsilon(\rho\|\sigma)\le\frac{D(\rho\|\sigma)+H_b(\epsilon)}{1-\epsilon}.DHϵ​(ρ∥σ)≤1−ϵD(ρ∥σ)+Hb​(ϵ)​.

This comparison relates the testing information to ordinary relative entropy. The whole numerator is divided by the positive number 1−ϵ1-\epsilon1−ϵ; no invertibility or finite-relative-entropy condition is imposed. Infinite values are retained.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Support50_P20
Formal statement
noncomputable section

open Filter

open scoped ENNReal

namespace QAlgorithms.Papers.Independent.Support50.P20

attribute [local instance] Classical.propDecidable Classical.decEq

open QAlgorithms.Papers.Independent.Dependency50.P13
  (IsTraceClass IsPositive operatorTrace State)

open QAlgorithms.Papers.Independent.Information100.P63 (logTwo)

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

/-- PDF p.2, Property 3, (2): the WHOLE numerator is divided by 1-epsilon.
Infinite relative entropy is retained; this denominator is strictly positive. -/
theorem testingDivergence_relativeEntropyBound {I : Type*} (b : HilbertBasis I ℂ H)
    (rho sigma : State b) (epsilon : ℝ) (hZero : 0 ≤ epsilon) (hOne : epsilon < 1) :
    hilbertTestingDivergence b epsilon rho.val.val sigma.val.val ≤
      ((traceClassRelativeEntropyBits rho sigma : EReal) + (QAlgorithms.binEntropy2 epsilon : ℝ)) /
        ((1 - epsilon : ℝ) : EReal) := by sorry

end QAlgorithms.Papers.Independent.Support50.P20

end
Source
One-Shot Classical-Quantum Capacity and Hypothesis Testing, arXiv:1007.5456v3, PDF p. 2, Property 3, (2): testing divergence bounded by ordinary relative entropy
Read-back

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

Let HHH be a complete complex inner-product space with Hilbert basis b=(bi)i∈Ib=(b_i)_{i\in I}b=(bi​)i∈I​, for any index type III. Let ρ,σ\rho,\sigmaρ,σ be states: bounded complex-linear operators TTT that are self-adjoint with nonnegative real quadratic forms, satisfy tr⁡bT=∑i⟨bi,Tbi⟩=1\operatorname{tr}_bT=\sum_i\langle b_i,Tb_i\rangle=1trb​T=∑i​⟨bi​,Tbi​⟩=1, and admit representations Tx=∑r=0∞⟨vr,x⟩urTx=\sum_{r=0}^\infty\langle v_r,x\rangle u_rTx=∑r=0∞​⟨vr​,x⟩ur​ with ∑r∥ur∥∥vr∥<∞\sum_r\|u_r\|\|v_r\|<\infty∑r​∥ur​∥∥vr​∥<∞. For 0≤ϵ<10\le\epsilon<10≤ϵ<1, define

Dϵ(ρ,σ)=L ⁣(inf⁡Q an effectRe⁡tr⁡b(Qρ)≥1−ϵmax⁡(Re⁡tr⁡b(Qσ),0)),D_\epsilon(\rho,\sigma)= L\!\left(\inf_{\substack{Q\text{ an effect}\\ \operatorname{Re}\operatorname{tr}_b(Q\rho)\ge1-\epsilon}} \max(\operatorname{Re}\operatorname{tr}_b(Q\sigma),0)\right),Dϵ​(ρ,σ)=L​Q an effectRetrb​(Qρ)≥1−ϵ​inf​max(Retrb​(Qσ),0)​,

where an effect is a bounded operator for which QQQ and 1−Q1-Q1−Q are self-adjoint with nonnegative real quadratic forms, the infimum lies in [0,∞][0,\infty][0,∞], and L(0)=+∞L(0)=+\inftyL(0)=+∞, L(+∞)=−∞L(+\infty)=-\inftyL(+∞)=−∞, L(z)=−log⁡2zL(z)=-\log_2zL(z)=−log2​z for positive finite zzz. For each natural number nnn and each orthonormal family v0,…,vn−1v_0,\ldots,v_{n-1}v0​,…,vn−1​ in HHH, form matrices Rij=⟨vi,ρvj⟩R_{ij}=\langle v_i,\rho v_j\rangleRij​=⟨vi​,ρvj​⟩ and Sij=⟨vi,σvj⟩S_{ij}=\langle v_i,\sigma v_j\rangleSij​=⟨vi​,σvj​⟩. Define

F(R,S)={max⁡ ⁣(Re⁡Tr⁡ ⁣(f(R)−R g(S)−R+S),0),ker⁡S⊆ker⁡R,+∞,otherwise,f(x)=xln⁡x,g(x)=ln⁡x,F(R,S)= \begin{cases} \max\!\left(\operatorname{Re}\operatorname{Tr}\!\left(f(R)-R\,g(S)-R+S\right),0\right), &\ker S\subseteq\ker R,\\ +\infty,&\text{otherwise}, \end{cases} \qquad f(x)=x\ln x,\quad g(x)=\ln x,F(R,S)={max(ReTr(f(R)−Rg(S)−R+S),0),+∞,​kerS⊆kerR,otherwise,​f(x)=xlnx,g(x)=lnx,

where f(R)f(R)f(R) and g(S)g(S)g(S) use continuous functional calculus, with the real logarithm taking value 000 at 000, and let Drelbits(ρ,σ)=(sup⁡n,vF(R,S))/ln⁡2D_{\mathrm{rel}}^{\mathrm{bits}}(\rho,\sigma)=(\sup_{n,v}F(R,S))/\ln2Drelbits​(ρ,σ)=(supn,v​F(R,S))/ln2, with the supremum in [0,∞][0,\infty][0,∞] and the empty orthonormal family n=0n=0n=0 included. Set h2(ϵ)=[−ϵln⁡ϵ−(1−ϵ)ln⁡(1−ϵ)]/ln⁡2h_2(\epsilon)=[-\epsilon\ln\epsilon-(1-\epsilon)\ln(1-\epsilon)]/\ln2h2​(ϵ)=[−ϵlnϵ−(1−ϵ)ln(1−ϵ)]/ln2, using 0ln⁡0=00\ln0=00ln0=0. The declaration asserts

Dϵ(ρ,σ)≤Drelbits(ρ,σ)+h2(ϵ)1−ϵD_\epsilon(\rho,\sigma)\le \frac{D_{\mathrm{rel}}^{\mathrm{bits}}(\rho,\sigma)+h_2(\epsilon)}{1-\epsilon}Dϵ​(ρ,σ)≤1−ϵDrelbits​(ρ,σ)+h2​(ϵ)​

in the extended reals. The denominator is strictly positive; ϵ=0\epsilon=0ϵ=0 and infinite relative entropy are 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).

  • Endorsed by Gohan · Oct 29, 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