Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Gaussian conditional-input optimization is exact without a conjecture

Open
QAlgorithms.Papers.Independent.Gap48.P44.gaussian_input_optimality

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

prove2quantum

Write gn(x)=(x+1)ln⁡(x+1)−xln⁡xg_n(x)=(x+1)\ln(x+1)-x\ln xgn​(x)=(x+1)ln(x+1)−xlnx and gb(x)=gn(x)/ln⁡2g_b(x)=g_n(x)/\ln2gb​(x)=gn​(x)/ln2 for x≥0x\ge0x≥0, with 0ln⁡0=00\ln0=00ln0=0. Fix η∈[1/2,1]\eta\in[1/2,1]η∈[1/2,1] and NS≥0N_S\ge0NS​≥0. For every n≥1n\ge1n≥1 and normalized finite ensemble of Gaussian nnn-mode conditional states with unconditional total photon energy at most nNSnN_SnNS​, all four common-parameter inequalities (60)–(63) hold for one λ∈[0,1]\lambda\in[0,1]λ∈[0,1], without the unrestricted minimum-output-entropy conjecture. Intra-block Gaussian correlations are permitted, and the unconditional mixture need not be Gaussian.

Let Gη,NS\mathcal G_{\eta,N_S}Gη,NS​​ be the closure of the union, over all positive blocklengths and all such finite ensembles, of the per-use entropy regions (3)–(5). Then

Gη,NS=⋃0≤λ≤1{(C,Q,E)∈R3:C+2Q≤gb(λNS)+gb(ηNS)−gb((1−η)λNS),Q+E≤gb(ηλNS)−gb((1−η)λNS),C+Q+E≤gb(ηNS)−gb((1−η)λNS)}.\mathcal G_{\eta,N_S}=\bigcup_{0\le\lambda\le1}\left\{(C,Q,E)\in\mathbb R^3:\begin{aligned}C+2Q&\le g_b(\lambda N_S)+g_b(\eta N_S)-g_b((1-\eta)\lambda N_S),\\Q+E&\le g_b(\eta\lambda N_S)-g_b((1-\eta)\lambda N_S),\\C+Q+E&\le g_b(\eta N_S)-g_b((1-\eta)\lambda N_S)\end{aligned}\right\}.Gη,NS​​=0≤λ≤1⋃​⎩⎨⎧​(C,Q,E)∈R3:C+2QQ+EC+Q+E​≤gb​(λNS​)+gb​(ηNS​)−gb​((1−η)λNS​),≤gb​(ηλNS​)−gb​((1−η)λNS​),≤gb​(ηNS​)−gb​((1−η)λNS​)​⎭⎬⎫​.

The rates are signed net generated resources per use. The restriction concerns the input-state optimization; it does not define a class of codes using only Gaussian gates. Closure retains the displaced-thermal continuum through finite-alphabet approximation.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Gap48_P44
Formal statement
noncomputable section

namespace QAlgorithms.Papers.Independent.Gap48.P44

open Filter

open scoped ENNReal Topology

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

open QAlgorithms.Papers.Independent.Information100.P54 (thermalEntropy)

/-- PDF pp.11-13, Section III G: Gaussian conditional inputs need no
unrestricted minimum-output-entropy premise. The same lambda satisfies all four
announced block bounds, and the regularized Gaussian entropy optimization is
exactly the complete three-face family. Gaussian correlations are retained and
no Gaussian assumption is imposed on the unconditional mixture. -/
theorem gaussian_input_optimality (eta budget : ℝ)
    (heta0 : (1 / 2 : ℝ) ≤ eta) (heta1 : eta ≤ 1) (hbudget : 0 ≤ budget) :
    (∀ (n : ℕ), 0 < n → ∀ {X : Type} [Fintype X] (e : BlockEnsemble X n),
      (∀ x, Fock.Gaussian (e.state x).coefficients) →
      e.energy ≤ ENNReal.ofReal ((n : ℝ) * budget) →
      ∃ share ∈ Set.Icc (0 : ℝ) 1,
        e.entropyAverage id ≤ ENNReal.ofReal ((n : ℝ) * thermalEntropy (share * budget)) ∧
        blockEntropy (Fock.pureLoss n eta e.average) ≤
          ENNReal.ofReal ((n : ℝ) * thermalEntropy (eta * budget)) ∧
        e.entropyAverage (Fock.pureLoss n eta) ≤
          ENNReal.ofReal ((n : ℝ) * thermalEntropy (eta * share * budget)) ∧
        ENNReal.ofReal ((n : ℝ) * thermalEntropy ((1 - eta) * share * budget)) ≤
          e.entropyAverage (lossComplement n eta)) ∧
    gaussianDynamicOptimization eta budget = lossQuantumFamily eta budget := by sorry

end QAlgorithms.Papers.Independent.Gap48.P44

end
Source
Quantum trade-off coding for bosonic communication, arXiv:1105.0119v4, PDF p. 11, §III G: Gaussian-input restricted optimum without conjecture
Read-back

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

Put [x]+=max⁡(x,0)[x]_+=\max(x,0)[x]+​=max(x,0), 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, and H(x)=G(x)/ln⁡2H(x)=G(x)/\ln2H(x)=G(x)/ln2, using ln⁡0=0\ln0=0ln0=0. To specify all kernels in the assertion, for any number of modes mmm write Im=NmI_m=\mathbb N^mIm​=Nm; a density kernel ρ\rhoρ on ImI_mIm​ has a coefficient expansion ρab=∑j′uj(a)vj(b)‾\rho_{ab}=\sum'_ju_j(a)\overline{v_j(b)}ρab​=∑j′​uj​(a)vj​(b)​ with uj,vj∈ℓ2(Im;C)u_j,v_j\in\ell^2(I_m;\mathbb C)uj​,vj​∈ℓ2(Im​;C) and ∑j∥uj∥∥vj∥<∞\sum_j\|u_j\|\|v_j\|<\infty∑j​∥uj​∥∥vj​∥<∞, all its finite quadratic forms are real and nonnegative, and its real diagonal is summable with sum 111. Let S(ρ)S(\rho)S(ρ) be the infimum of ∑j′[−wjln⁡wj]+\sum'_j[-w_j\ln w_j]_+∑j′​[−wj​lnwj​]+​ over all coefficient expansions ρab=∑j′wjvj(a)vj(b)‾\rho_{ab}=\sum'_jw_jv_j(a)\overline{v_j(b)}ρab​=∑j′​wj​vj​(a)vj​(b)​ with summable nonnegative weights of sum 111, unit vectors at positive weights, and orthogonal vectors at distinct positive weights; zero-weight vectors are unrestricted, and the infimum is +∞+\infty+∞ if there is no such expansion. Write h(ρ)=S(ρ)/ln⁡2h(\rho)=S(\rho)/\ln2h(ρ)=S(ρ)/ln2. Define

aα(k)=e−∣α∣2/2αkk!,Qρ(α)=1πmRe⁡(∑a′∑b′∏iaαi(ai)‾ ρab ∏iaαi(bi)).a_\alpha(k)=e^{-|\alpha|^2/2}\frac{\alpha^k}{\sqrt{k!}}, \qquad Q_\rho(\alpha)=\frac1{\pi^m}\operatorname{Re} \left(\sum'_a\sum'_b \overline{\prod_i a_{\alpha_i}(a_i)} \,\rho_{ab}\, \prod_i a_{\alpha_i}(b_i)\right).aα​(k)=e−∣α∣2/2k!​αk​,Qρ​(α)=πm1​Re(a∑′​b∑′​i∏​aαi​​(ai​)​ρab​i∏​aαi​​(bi​)).

A Gaussian kernel means a density kernel ρ\rhoρ for which there exist μ∈Cm\mu\in\mathbb C^mμ∈Cm, a symmetric positive-definite real matrix AAA indexed by {1,…,m}×{0,1}\{1,\ldots,m\}\times\{0,1\}{1,…,m}×{0,1}, and c>0c>0c>0 such that, for every α∈Cm\alpha\in\mathbb C^mα∈Cm,

Qρ(α)=cexp⁡ ⁣(−∑i,jqi(α−μ)Aijqj(α−μ)),q(i,0)(α)=Re⁡αi,q(i,1)(α)=Im⁡αi.Q_\rho(\alpha)=c\exp\!\left(-\sum_{i,j}q_i(\alpha-\mu)A_{ij}q_j(\alpha-\mu)\right), \quad q_{(i,0)}(\alpha)=\operatorname{Re}\alpha_i,\quad q_{(i,1)}(\alpha)=\operatorname{Im}\alpha_i.Qρ​(α)=cexp(−i,j∑​qi​(α−μ)Aij​qj​(α−μ)),q(i,0)​(α)=Reαi​,q(i,1)​(α)=Imαi​.

Let Vm(a,b)=1V_m(a,b)=1Vm​(a,b)=1 when a=b=0a=b=0a=b=0 and 000 otherwise, and define

cτ(r,l;p,q)={r! l!p! q!∑0≤u≤pu≤r, r−u≤q(pu)(qr−u)(τ)u(1−τ)p−u+(r−u)(−τ)q−(r−u),r+l=p+q,0,r+l≠p+q.c_\tau(r,l;p,q)= \begin{cases} \displaystyle \sqrt{\frac{r!\,l!}{p!\,q!}} \sum_{\substack{0\le u\le p\\u\le r,\ r-u\le q}} \binom pu\binom q{r-u} (\sqrt\tau)^u (\sqrt{1-\tau})^{p-u+(r-u)} (-\sqrt\tau)^{q-(r-u)},&r+l=p+q,\\ 0,&r+l\ne p+q. \end{cases}cτ​(r,l;p,q)=⎩⎨⎧​p!q!r!l!​​0≤u≤pu≤r, r−u≤q​∑​(up​)(r−uq​)(τ​)u(1−τ​)p−u+(r−u)(−τ​)q−(r−u),0,​r+l=p+q,r+l=p+q.​

Natural-number subtraction truncates at zero, and every zeroth power is 111. For a,b∈Im2a,b\in I_m^2a,b∈Im2​, put

Bτ(a,b)=∏icτ(ai(1),ai(2);bi(1),bi(2)),Jτ(U,V)u,v=∑a′∑b′Bτ(u,a)Ua(1),b(1)Va(2),b(2)Bτ(v,b)‾,Pτ(U)a,b=∑e′Jτ(U,Vm)(a,e),(b,e),Lτ(U)a,b=(−1)∑iai(∑e′Jτ(U,Vm)(e,a),(e,b))(−1)∑ibi.\begin{aligned} B_\tau(a,b)&=\prod_i c_\tau(a_i^{(1)},a_i^{(2)};b_i^{(1)},b_i^{(2)}),\\ J_\tau(U,V)_{u,v} &=\sum'_a\sum'_bB_\tau(u,a) U_{a^{(1)},b^{(1)}}V_{a^{(2)},b^{(2)}} \overline{B_\tau(v,b)},\\ P_\tau(U)_{a,b}&=\sum'_eJ_\tau(U,V_m)_{(a,e),(b,e)},\\ L_\tau(U)_{a,b}&=(-1)^{\sum_i a_i} \left(\sum'_eJ_\tau(U,V_m)_{(e,a),(e,b)}\right) (-1)^{\sum_i b_i}. \end{aligned}Bτ​(a,b)Jτ​(U,V)u,v​Pτ​(U)a,b​Lτ​(U)a,b​​=i∏​cτ​(ai(1)​,ai(2)​;bi(1)​,bi(2)​),=a∑′​b∑′​Bτ​(u,a)Ua(1),b(1)​Va(2),b(2)​Bτ​(v,b)​,=e∑′​Jτ​(U,Vm​)(a,e),(b,e)​,=(−1)∑i​ai​(e∑′​Jτ​(U,Vm​)(e,a),(e,b)​)(−1)∑i​bi​.​

The complex infinite sums are total sums, assigned zero when the relevant family is not summable. For a finite ensemble e=(px,ρx)e=(p_x,\rho_x)e=(px​,ρx​) of density kernels on ImI_mIm​, where ppp is a probability mass function, define

ρˉab=∑xpxρx,ab,Ee=∑xpx∑i∑a′[aiRe⁡ρx,aa]+,Ae(C)=∑xpxh(C(ρx)).\bar\rho_{ab}=\sum_xp_x\rho_{x,ab},\qquad \mathcal E_e=\sum_xp_x\sum_i\sum'_a[a_i\operatorname{Re}\rho_{x,aa}]_+, \qquad A_e(C)=\sum_xp_xh(C(\rho_x)).ρˉ​ab​=x∑​px​ρx,ab​,Ee​=x∑​px​i∑​a∑′​[ai​Reρx,aa​]+​,Ae​(C)=x∑​px​h(C(ρx​)).

For every real η,B\eta,Bη,B with 12≤η≤1\tfrac12\le\eta\le121​≤η≤1 and B≥0B\ge0B≥0, two assertions hold. First, for every natural m>0m>0m>0, every finite label type XXX, and every such ensemble indexed by XXX, if every ρx\rho_xρx​ is Gaussian, including states of probability zero, and Ee≤[mB]+\mathcal E_e\le[mB]_+Ee​≤[mB]+​, there exists s∈[0,1]s\in[0,1]s∈[0,1] satisfying all four inequalities

Ae(id)≤[mH(sB)]+,h(Pη(ρˉ))≤[mH(ηB)]+,Ae(Pη)≤[mH(ηsB)]+,[mH((1−η)sB)]+≤Ae(Lη).A_e(\mathrm{id})\le[mH(sB)]_+,\quad h(P_\eta(\bar\rho))\le[mH(\eta B)]_+,\quad A_e(P_\eta)\le[mH(\eta sB)]_+,\quad [mH((1-\eta)sB)]_+\le A_e(L_\eta).Ae​(id)≤[mH(sB)]+​,h(Pη​(ρˉ​))≤[mH(ηB)]+​,Ae​(Pη​)≤[mH(ηsB)]+​,[mH((1−η)sB)]+​≤Ae​(Lη​).

Second, define the real numbers I=Ae(id)realI=A_e(\mathrm{id})^{\mathrm{real}}I=Ae​(id)real, O=h(Pη(ρˉ))realO=h(P_\eta(\bar\rho))^{\mathrm{real}}O=h(Pη​(ρˉ​))real, C=Ae(Pη)realC=A_e(P_\eta)^{\mathrm{real}}C=Ae​(Pη​)real, and E=Ae(Lη)realE=A_e(L_\eta)^{\mathrm{real}}E=Ae​(Lη​)real, where conversion from [0,∞][0,\infty][0,∞] to R\mathbb RR sends +∞+\infty+∞ to zero, and let

De(η)={r∈R3:m(r0+2r1)≤I+O−E,m(r1+r2)≤C−E,m(r0+r1+r2)≤O−E}.D_e(\eta)= \left\{r\in\mathbb R^3: \begin{aligned} m(r_0+2r_1)&\le I+O-E,\\ m(r_1+r_2)&\le C-E,\\ m(r_0+r_1+r_2)&\le O-E \end{aligned} \right\}.De​(η)=⎩⎨⎧​r∈R3:m(r0​+2r1​)m(r1​+r2​)m(r0​+r1​+r2​)​≤I+O−E,≤C−E,≤O−E​⎭⎬⎫​.

The Euclidean closure of all rrr belonging to some De(η)D_e(\eta)De​(η) for some m>0m>0m>0, some natural k>0k>0k>0, and some ensemble indexed by {0,…,k−1}\{0,\ldots,k-1\}{0,…,k−1} consisting entirely of Gaussian kernels with Ee≤[mB]+\mathcal E_e\le[mB]_+Ee​≤[mB]+​ is exactly

{r∈R3:∃s∈[0,1], r0+2r1≤H(sB)+H(ηB)−H((1−η)sB),r1+r2≤H(ηsB)−H((1−η)sB),r0+r1+r2≤H(ηB)−H((1−η)sB)}.\left\{r\in\mathbb R^3:\exists s\in[0,1],\ \begin{aligned} r_0+2r_1&\le H(sB)+H(\eta B)-H((1-\eta)sB),\\ r_1+r_2&\le H(\eta sB)-H((1-\eta)sB),\\ r_0+r_1+r_2&\le H(\eta B)-H((1-\eta)sB) \end{aligned} \right\}.⎩⎨⎧​r∈R3:∃s∈[0,1], r0​+2r1​r1​+r2​r0​+r1​+r2​​≤H(sB)+H(ηB)−H((1−η)sB),≤H(ηsB)−H((1−η)sB),≤H(ηB)−H((1−η)sB)​⎭⎬⎫​.

Neither of these sets of triples imposes nonnegativity on the coordinates of rrr. The parameter ranges include η=12\eta=\tfrac12η=21​, η=1\eta=1η=1, B=0B=0B=0, and both endpoints for sss.

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