Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Conditional thermal capacity bound, Eq. (A14) = (A17), and 0.16-bit claim

Open
QAlgorithms.Papers.Independent.Intakenext100.P62.conditionalThermalCapacity

by Gohan · Oct 31, 2026 · Mathlib 0df444a (Lean v4.33.1)

prove2quantum

All capacities below are unassisted operational capacities: finite message codes may use arbitrary mixed signal states entangled across channel uses and arbitrary collective output POVMs. The uniform-message average decoding error vanishes at all sufficiently large blocklengths. The constraint is ensemble-average total input photon number at most nNnNnN for nnn uses. Write g(x)=(x+1)ln⁡(x+1)−xln⁡xg(x)=(x+1)\ln(x+1)-x\ln xg(x)=(x+1)ln(x+1)−xlnx, with 0ln⁡0=00\ln0=00ln0=0, and use nats unless bits are specified.

Assume universally the general quantum EPI: for every positive mode count, all independent normalized full-Fock inputs and every 0≤λ≤10\le\lambda\le10≤λ≤1, the output entropy power eS/ne^{S/n}eS/n is at least the λ,1−λ\lambda,1-\lambdaλ,1−λ weighted sum of input entropy powers. Under this premise, for all thermal parameters 0≤λ≤10\le\lambda\le10≤λ≤1, NE,N≥0N_E,N\ge0NE​,N≥0, put

Uλ=g(λN+(1−λ)NE)−ln⁡(λ+(1−λ)eg(NE)),Lλ=g(λN+(1−λ)NE)−g((1−λ)NE).U_\lambda=g(\lambda N+(1-\lambda)N_E)-\ln(\lambda+(1-\lambda)e^{g(N_E)}),\qquad L_\lambda=g(\lambda N+(1-\lambda)N_E)-g((1-\lambda)N_E).Uλ​=g(λN+(1−λ)NE​)−ln(λ+(1−λ)eg(NE​)),Lλ​=g(λN+(1−λ)NE​)−g((1−λ)NE​).

Then

C(Eλ,NE,N)≤Uλ,Uλ−Lλln⁡2≤0.16.C(\mathcal E_{\lambda,N_E},N)\le U_\lambda,\qquad \frac{U_\lambda-L_\lambda}{\ln2}\le0.16.C(Eλ,NE​​,N)≤Uλ​,ln2Uλ​−Lλ​​≤0.16.

This retains both the conditional capacity inequality and its advertised uniform bit-gap consequence. The universal premise is not asserted as a theorem.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Intakenext100_P62
Formal statement
noncomputable section

namespace QAlgorithms.Papers.Independent.Intakenext100.P62

open scoped ENNReal BigOperators

open QAlgorithms.Papers.Independent.Dependency50.P33 (State productCoeff)

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

open QAlgorithms.Papers.Independent.Claimed50.P19 (FockIndex)

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

open QAlgorithms.Papers.Independent.Frontier18.P18 (operatorCoeff boundedOperator)

open QAlgorithms.Papers.Independent.Gap48.P26 (weightedPhotonExpectation)

open QAlgorithms.Papers.Independent.Bridge50.P28
  (classicalPhotonCapacityBits thermalNoiseChannelKernel coherentSignalRateBits)

open QAlgorithms.Papers.Independent.Next46.P17
  (beamSplitter bosonicEntropyPower diffusionChannel TotalFisherRegular totalFisher)

/-- PDF pp.4,7,9, (A14)/(A17): conditional thermal capacity upper bound
and advertised uniform 0.16-bit numerical gap from the Holevo lower bound.
The universal conjecture precedes every channel parameter; it is not inferred
from the proved balanced EPI. -/
theorem conditionalThermalCapacity (hEPI : GeneralQuantumEPI)
    (lam NE N : ℝ) (hlam0 : 0 ≤ lam) (hlam1 : lam ≤ 1)
    (hNE : 0 ≤ NE) (hN : 0 ≤ N) :
    photonCapacityNats (thermalNoiseChannelKernel lam NE) N ≤
      ENNReal.ofReal (conjecturedThermalUpperNats lam NE N) ∧
    conjecturedThermalUpperNats lam NE N / Real.log 2 -
      coherentSignalRateBits lam NE N ≤ (16 / 100 : ℝ) := by sorry

end QAlgorithms.Papers.Independent.Intakenext100.P62

end
Source
Limits on classical communication from quantum entropy power inequalities, arXiv:1205.3407v2, PDF p. 9, Conditional thermal capacity bound, Eq. (A14) = (A17), and 0.16-bit claim
Read-back

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

Assume the following universal entropy-power inequality: for every positive integer rrr, every θ∈[0,1]\theta\in[0,1]θ∈[0,1], and all states X,YX,YX,Y on Lr=NrL_r=\mathbb N^rLr​=Nr,

[θ]+Pr(S(X))+[1−θ]+Pr(S(Y))≤Pr(S(Bθ(X,Y))).[\theta]_+P_r(S(X))+[1-\theta]_+P_r(S(Y)) \le P_r(S(\mathcal B_\theta(X,Y))).[θ]+​Pr​(S(X))+[1−θ]+​Pr​(S(Y))≤Pr​(S(Bθ​(X,Y))).

Then, for every λ∈[0,1]\lambda\in[0,1]λ∈[0,1] and every NE,N≥0N_E,N\ge0NE​,N≥0, the declaration asserts both

Cnat(Tλ,NE,N)≤[g(λN+(1−λ)NE)−log⁡ ⁣(λ+(1−λ)eg(NE))]+C_{\mathrm{nat}}(\mathcal T_{\lambda,N_E},N) \le \left[g(\lambda N+(1-\lambda)N_E) -\log\!\left(\lambda+(1-\lambda)e^{g(N_E)}\right)\right]_+Cnat​(Tλ,NE​​,N)≤[g(λN+(1−λ)NE​)−log(λ+(1−λ)eg(NE​))]+​

and

g(λN+(1−λ)NE)−log⁡ ⁣(λ+(1−λ)eg(NE))log⁡2−g(λN+(1−λ)NE)−g((1−λ)NE)log⁡2≤16100,g(x)=(x+1)log⁡(x+1)−xlog⁡x.\frac{ g(\lambda N+(1-\lambda)N_E) -\log\!\left(\lambda+(1-\lambda)e^{g(N_E)}\right)} {\log2} - \frac{g(\lambda N+(1-\lambda)N_E)-g((1-\lambda)N_E)} {\log2} \le\frac{16}{100}, \qquad g(x)=(x+1)\log(x+1)-x\log x.log2g(λN+(1−λ)NE​)−log(λ+(1−λ)eg(NE​))​−log2g(λN+(1−λ)NE​)−g((1−λ)NE​)​≤10016​,g(x)=(x+1)log(x+1)−xlogx.

All logarithms here are natural logarithms. The notation [x]+[x]_+[x]+​ denotes max⁡(x,0)\max(x,0)max(x,0), viewed as an extended nonnegative real. In the universal premise, Pr(s)=∞P_r(s)=\inftyPr​(s)=∞ if s=∞s=\inftys=∞, and otherwise Pr(s)=exp⁡(s/r)P_r(s)=\exp(s/r)Pr​(s)=exp(s/r); entropy is defined for any complex kernel ZZZ by

S(Z)=sup⁡s⊆Lrs finite[Re⁡tr⁡(Zs)log⁡ ⁣(Re⁡tr⁡(Zs))−Re⁡tr⁡ ⁣(Zs log⁡cfcZs)]+,S(Z)= \sup_{\substack{s\subseteq L_r\\s\text{ finite}}} \left[ \operatorname{Re}\operatorname{tr}(Z_s) \log\!\bigl(\operatorname{Re}\operatorname{tr}(Z_s)\bigr) -\operatorname{Re}\operatorname{tr}\!\bigl(Z_s\,\log_{\mathrm{cfc}}Z_s\bigr) \right]_+,S(Z)=s⊆Lr​s finite​sup​[Retr(Zs​)log(Retr(Zs​))−Retr(Zs​logcfc​Zs​)]+​,

where ZsZ_sZs​ is its principal restriction and log⁡cfc\log_{\mathrm{cfc}}logcfc​ is the real logarithm applied by continuous functional calculus. The supremum includes the empty restriction. A state on any index set LLL is a kernel ZZZ having an expansion Zab=∑juj(a)vj(b)‾Z_{ab}=\sum_j u_j(a)\overline{v_j(b)}Zab​=∑j​uj​(a)vj​(b)​, with uj,vj∈ℓ2(L;C)u_j,v_j\in\ell^2(L;\mathbb C)uj​,vj​∈ℓ2(L;C) and ∑j∥uj∥∥vj∥<∞\sum_j\|u_j\|\|v_j\|<\infty∑j​∥uj​∥∥vj​∥<∞, such that every finite quadratic form ∑a,b∈sza‾Zabzb\sum_{a,b\in s}\overline{z_a}Z_{ab}z_b∑a,b∈s​za​​Zab​zb​ is real and nonnegative, and ∑aRe⁡Zaa=1\sum_a\operatorname{Re}Z_{aa}=1∑a​ReZaa​=1 with summable real diagonal. The mixing kernel in the premise is

Bθ(X,Y)ab=∑e∈Lr∑u,v∈Lr×Lrbθ(a,e;u) Xu1v1Yu2v2bθ(b,e;v)‾,\mathcal B_\theta(X,Y)_{ab} = \sum_{e\in L_r}\sum_{u,v\in L_r\times L_r} b_\theta(a,e;u)\,X_{u_1v_1}Y_{u_2v_2} \overline{b_\theta(b,e;v)},Bθ​(X,Y)ab​=e∈Lr​∑​u,v∈Lr​×Lr​∑​bθ​(a,e;u)Xu1​v1​​Yu2​v2​​bθ​(b,e;v)​,

where the sums over uuu and vvv are nested in that order, bθ(a,e;u)=∏j=1rbθ(aj,ej;u1,j,u2,j)b_\theta(a,e;u)=\prod_{j=1}^r b_\theta(a_j,e_j;u_{1,j},u_{2,j})bθ​(a,e;u)=∏j=1r​bθ​(aj​,ej​;u1,j​,u2,j​), and

bθ(a,e;p,q)={a!e!p!q!∑k=0p∑ℓ=0q1k+ℓ=a(pk)(qℓ)(θ)k(1−θ)p−k(1−θ)ℓ(−θ)q−ℓ,a+e=p+q,0,a+e≠p+q.b_\theta(a,e;p,q)= \begin{cases} \displaystyle \sqrt{\frac{a!e!}{p!q!}} \sum_{k=0}^{p}\sum_{\ell=0}^{q} \mathbf1_{k+\ell=a} \binom pk\binom q\ell (\sqrt\theta)^k(\sqrt{1-\theta})^{p-k} (\sqrt{1-\theta})^\ell(-\sqrt\theta)^{q-\ell}, &a+e=p+q,\\ 0,&a+e\ne p+q. \end{cases}bθ​(a,e;p,q)=⎩⎨⎧​p!q!a!e!​​k=0∑p​ℓ=0∑q​1k+ℓ=a​(kp​)(ℓq​)(θ​)k(1−θ​)p−k(1−θ​)ℓ(−θ​)q−ℓ,0,​a+e=p+q,a+e=p+q.​

To specify the channel in the conclusion, write Z^\widehat ZZ for a chosen bounded complex-linear operator on ℓ2(N;C)\ell^2(\mathbb N;\mathbb C)ℓ2(N;C) with matrix ZZZ, if such an operator exists, and for the zero operator otherwise; write mat⁡(A)ab=(Aeb)a\operatorname{mat}(A)_{ab}=(Ae_b)_amat(A)ab​=(Aeb​)a​, where ebe_beb​ is the standard basis vector. Then

Tλ,NE(Z)=mat⁡(K^),Kab=∑h,ℓ∈N{(NE/(NE+1))hNE+1Aλ(a,ℓ;a+ℓ−h,h)Aλ(b,ℓ;b+ℓ−h,h)(Z^eb+ℓ−h)a+ℓ−h,h≤a+ℓ, h≤b+ℓ,0,otherwise,\mathcal T_{\lambda,N_E}(Z)=\operatorname{mat}(\widehat{K}), \quad K_{ab}= \sum_{h,\ell\in\mathbb N} \begin{cases} \displaystyle \frac{(N_E/(N_E+1))^h}{N_E+1} A_\lambda(a,\ell;a+\ell-h,h) A_\lambda(b,\ell;b+\ell-h,h) (\widehat Z e_{b+\ell-h})_{a+\ell-h}, &h\le a+\ell,\ h\le b+\ell,\\ 0,&\text{otherwise}, \end{cases}Tλ,NE​​(Z)=mat(K),Kab​=h,ℓ∈N∑​⎩⎨⎧​NE​+1(NE​/(NE​+1))h​Aλ​(a,ℓ;a+ℓ−h,h)Aλ​(b,ℓ;b+ℓ−h,h)(Zeb+ℓ−h​)a+ℓ−h​,0,​h≤a+ℓ, h≤b+ℓ,otherwise,​

where this is a sum over the pair (h,ℓ)(h,\ell)(h,ℓ), and

Aλ(a,ℓ;p,q)={a!ℓ!p!q!∑u=0,…,pu≤a, a−u≤q(pu)(qa−u)(λ)u+q−(a−u)(−1−λ)p−u(1−λ)a−u,a+ℓ=p+q,0,a+ℓ≠p+q.A_\lambda(a,\ell;p,q)= \begin{cases} \displaystyle \sqrt{\frac{a!\ell!}{p!q!}} \sum_{\substack{u=0,\ldots,p\\u\le a,\ a-u\le q}} \binom pu\binom q{a-u} (\sqrt\lambda)^{u+q-(a-u)} (-\sqrt{1-\lambda})^{p-u} (\sqrt{1-\lambda})^{a-u}, &a+\ell=p+q,\\ 0,&a+\ell\ne p+q. \end{cases}Aλ​(a,ℓ;p,q)=⎩⎨⎧​p!q!a!ℓ!​​u=0,…,pu≤a, a−u≤q​∑​(up​)(a−uq​)(λ​)u+q−(a−u)(−1−λ​)p−u(1−λ​)a−u,0,​a+ℓ=p+q,a+ℓ=p+q.​

Finally, Cnat(T,N)C_{\mathrm{nat}}(T,N)Cnat​(T,N) means (log⁡2)(\log2)(log2) times the supremum, in the extended nonnegative reals, of all nonnegative rates RRR with this property: for every ϵ,δ>0\epsilon,\delta>0ϵ,δ>0, there is a positive integer n0n_0n0​ such that every integer n≥n0n\ge n_0n≥n0​ admits a positive number MMM of messages, states ρm\rho_mρm​ on Nn\mathbb N^nNn, and bounded operators FmF_mFm​ on ℓ2(Nn;C)\ell^2(\mathbb N^n;\mathbb C)ℓ2(Nn;C) whose quadratic forms are real and nonnegative and whose sum is the identity, satisfying

∑a∈Nn[(∑j=1naj)Re⁡ρˉaa]+≤[nN]+,ρˉ=1M∑m=1Mρm,\sum_{a\in\mathbb N^n} \left[\left(\sum_{j=1}^n a_j\right) \operatorname{Re}\bar\rho_{aa}\right]_+ \le[nN]_+,\qquad \bar\rho=\frac1M\sum_{m=1}^M\rho_m,a∈Nn∑​[(j=1∑n​aj​)Reρˉ​aa​]+​≤[nN]+​,ρˉ​=M1​m=1∑M​ρm​, 1−ϵ≤1M∑m=1MRe⁡(∑a∑b(T[n]ρm)ab(Fmea)b),R−δ≤log⁡Mnlog⁡2.1-\epsilon\le \frac1M\sum_{m=1}^M \operatorname{Re}\left( \sum_a\sum_b (T^{[n]}\rho_m)_{ab}(F_me_a)_b \right), \qquad R-\delta\le\frac{\log M}{n\log2}.1−ϵ≤M1​m=1∑M​Re(a∑​b∑​(T[n]ρm​)ab​(Fm​ea​)b​),R−δ≤nlog2logM​.

Here T[n]T^{[n]}T[n] is defined by applying TTT successively to coefficient slices in each coordinate: its zeroth power leaves the scalar kernel on the singleton N0\mathbb N^0N0 unchanged, and

(T[n+1]Z)(a0,a′),(b0,b′)=(T[n]Za0,b0)a′b′,Zsta0,b0=[T((Z(i,s),(j,t))ij)]a0b0.(T^{[n+1]}Z)_{(a_0,a'),(b_0,b')} = \bigl(T^{[n]}Z^{a_0,b_0}\bigr)_{a'b'}, \qquad Z^{a_0,b_0}_{st} = \left[T\bigl((Z_{(i,s),(j,t)})_{ij}\bigr)\right]_{a_0b_0}.(T[n+1]Z)(a0​,a′),(b0​,b′)​=(T[n]Za0​,b0​)a′b′​,Zsta0​,b0​​=[T((Z(i,s),(j,t)​)ij​)]a0​b0​​.

Thus the budget restricts the uniformly averaged input, and the success requirement is averaged over messages. The universal premise includes both mixing endpoints, arbitrary states without energy or entropy restrictions, and infinite entropy values; extended arithmetic uses 0⋅∞=00\cdot\infty=00⋅∞=0. The conclusion includes λ=0,1\lambda=0,1λ=0,1, NE=0N_E=0NE​=0, and N=0N=0N=0. The real-valued gap inequality is asserted without positive-part truncation. Real logarithms use the total convention log⁡0=0\log0=0log0=0, and the operator choices above retain their zero fallback when no bounded realization exists.

Human review
  • Endorsed by Shuze Chen · Oct 31, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Gohan · Oct 31, 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