Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Capacity from the spread of column entropies

Open
QAlgorithms.Papers.Independent.Support50.P15.proposition13

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

prove2quantum

Let T:X→YT:X\to YT:X→Y be a normalized discrete memoryless channel with nonempty input alphabet and finite output cardinality d=∣Y∣≥2d=|Y|\ge2d=∣Y∣≥2. Capacity is measured in nats as the supremum of single-letter mutual information over finite-support input laws. Put Smin⁡=inf⁡xS(T(⋅∣x))S_{\min}=\inf_x S(T(\cdot|x))Smin​=infx​S(T(⋅∣x)), Smax⁡=sup⁡xS(T(⋅∣x))S_{\max}=\sup_x S(T(\cdot|x))Smax​=supx​S(T(⋅∣x)), a=Smax⁡−Smin⁡a=S_{\max}-S_{\min}a=Smax​−Smin​ and L=log⁡2(d−1)+4L=\log^2(d-1)+4L=log2(d−1)+4. Then

C(T)≥a22L−a33L2.C(T)\ge\frac{a^2}{2L}-\frac{a^3}{3L^2}.C(T)≥2La2​−3L2a3​.

If the fixed-dimensional asymmetry statement of Conjecture 12 holds, also C(T)≥M(−a/2,d)C(T)\ge M(-a/2,d)C(T)≥M(−a/2,d). Finite input alphabets have attained column extrema; supremal and infimal entropies give the same statement for arbitrary nonempty discrete input, without an unmentioned finiteness assumption. This supplies a capacity lower bound from column entropy spread alone.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Support50_P15
Formal statement
noncomputable section

attribute [local instance] Classical.propDecidable

open scoped Matrix ComplexConjugate ENNReal Matrix.Norms.L2Operator

open Filter MeasureTheory

namespace QAlgorithms.Papers.Independent.Support50.P15

open QAlgorithms.Papers.Independent.Information100.P56 (entropy)

/-- Proposition 13, PDF p.15. Arbitrary nonempty discrete input; finite output; capacity in nats.
Supremal/infimal column entropies allow nonattained extrema. -/
theorem proposition13 {X : Type*} [Nonempty X] {d : ℕ} (hd : 2 ≤ d) (c : FiniteOutputChannel X d) :
    let lo := sInf (Set.range (fun x => shannonEntropyNats (c.prob x)))
    let hi := sSup (Set.range (fun x => shannonEntropyNats (c.prob x)))
    redundancyBound d (hi - lo) ≤ finiteOutputCapacityNats c ∧
    (asymmetryConjecture d → tightBound d (-(hi - lo) / 2) ≤ (finiteOutputCapacityNats c : EReal)) := by sorry

end QAlgorithms.Papers.Independent.Support50.P15

end
Source
Tight bound on relative entropy by entropy difference, arXiv:1304.0036v3, PDF p. 15, proposition13
Read-back

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

Let XXX be any nonempty type, with no finiteness assumption, let d≥2d\ge2d≥2 be a natural number, and let a channel assign to each x∈Xx\in Xx∈X numbers px(i)≥0p_x(i)\ge0px​(i)≥0, 0≤i<d0\le i<d0≤i<d, with ∑ipx(i)=1\sum_i p_x(i)=1∑i​px​(i)=1. Using natural logarithms and 0log⁡0=00\log0=00log0=0, define H(p)=−∑ip(i)log⁡p(i)H(p)=-\sum_i p(i)\log p(i)H(p)=−∑i​p(i)logp(i), ℓ=inf⁡x∈XH(px)\ell=\inf_{x\in X}H(p_x)ℓ=infx∈X​H(px​), and h=sup⁡x∈XH(px)h=\sup_{x\in X}H(p_x)h=supx∈X​H(px​). Define its capacity by the real supremum

C=sup⁡{H ⁣(∑j=0m−1qjpxj)−∑j=0m−1qjH(pxj):m∈N, xj∈X, qj≥0, ∑j=0m−1qj=1}.\mathcal C=\sup\left\{ H\!\left(\sum_{j=0}^{m-1}q_jp_{x_j}\right) -\sum_{j=0}^{m-1}q_jH(p_{x_j}): m\in\mathbb N,\ x_j\in X,\ q_j\ge0,\ \sum_{j=0}^{m-1}q_j=1 \right\}.C=sup{H(j=0∑m−1​qj​pxj​​)−j=0∑m−1​qj​H(pxj​​):m∈N, xj​∈X, qj​≥0, j=0∑m−1​qj​=1}.

The allowed m=0m=0m=0 contributes no feasible probability vector. Put Ad=[log⁡(d−1)]2+4A_d=[\log(d-1)]^2+4Ad​=[log(d−1)]2+4 and Ld(z)=z2/(2Ad)−z3/(3Ad2)L_d(z)=z^2/(2A_d)-z^3/(3A_d^2)Ld​(z)=z2/(2Ad​)−z3/(3Ad2​). For H2(u)=−ulog⁡u−(1−u)log⁡(1−u)H_2(u)=-u\log u-(1-u)\log(1-u)H2​(u)=−ulogu−(1−u)log(1−u) and hd(u)=H2(u)+ulog⁡(d−1)h_d(u)=H_2(u)+u\log(d-1)hd​(u)=H2​(u)+ulog(d−1), define

Bd(z)=inf⁡0≤s,r≤(d−1)/dhd(s)−hd(r)=zK(s,r),B_d(z)=\inf_{\substack{0\le s,r\le(d-1)/d\\h_d(s)-h_d(r)=z}}K(s,r),Bd​(z)=0≤s,r≤(d−1)/dhd​(s)−hd​(r)=z​inf​K(s,r),

where K(s,r)=k(s,r)+k(1−s,1−r)K(s,r)=k(s,r)+k(1-s,1-r)K(s,r)=k(s,r)+k(1−s,1−r), k(p,q)=0k(p,q)=0k(p,q)=0 if p=0p=0p=0, k(p,q)=+∞k(p,q)=+\inftyk(p,q)=+∞ if p≠0=qp\ne0=qp=0=q, and k(p,q)=plog⁡(p/q)k(p,q)=p\log(p/q)k(p,q)=plog(p/q) otherwise; an empty infimum is +∞+\infty+∞. Then Ld(h−ℓ)≤CL_d(h-\ell)\le\mathcal CLd​(h−ℓ)≤C. In addition, if Bd(−z)≤Bd(z)B_d(-z)\le B_d(z)Bd​(−z)≤Bd​(z) for every z∈[0,log⁡d]z\in[0,\log d]z∈[0,logd], then Bd(−(h−ℓ)/2)≤CB_d(-(h-\ell)/2)\le\mathcal CBd​(−(h−ℓ)/2)≤C, with this second inequality interpreted in the extended reals.

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