Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

§3, Eqs. (17),(18),(22): actual Gaussian relative entropy from CM and EM

Open
QAlgorithms.Papers.Independent.Intakenext100.P46.gaussian_relativeEntropy

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

prove2quantum

Let ρ\rhoρ and σ\sigmaσ be centered Gaussian density operators on a positive finite number of complete bosonic modes. Use any split into two registers; either register may be empty, so single-mode states are included. The input covariance is α\alphaα and may have pure modes. Suppose the covariance of σ\sigmaσ is Sdiag⁡(γ,γ)STS\operatorname{diag}(\gamma,\gamma)S^TSdiag(γ,γ)ST, with SSS symplectic and every γj>1/2\gamma_j>1/2γj​>1/2. Put mj=ln⁡((2γj+1)/(2γj−1))m_j=\ln((2\gamma_j+1)/(2\gamma_j-1))mj​=ln((2γj​+1)/(2γj​−1)), M=S−Tdiag⁡(m,m)S−1M=S^{-T}\operatorname{diag}(m,m)S^{-1}M=S−Tdiag(m,m)S−1, and β=S−1αS−T\beta=S^{-1}\alpha S^{-T}β=S−1αS−T. The input spectral entropy H(ρ)H(\rho)H(ρ) is finite. The reciprocal partition function is

c=∏j2sinh⁡(mj/2)=∏j(γj2−1/4)−1/2>0.c=\prod_j2\sinh(m_j/2)=\prod_j(\gamma_j^2-1/4)^{-1/2}>0.c=j∏​2sinh(mj​/2)=j∏​(γj2​−1/4)−1/2>0.

The independently defined operator divergence satisfies

D(ρ∥σ)=−H(ρ)−ln⁡c+12tr⁡(αM)=−H(ρ)−∑jln⁡(2sinh⁡(mj/2))+12∑j(βjj+βn+j,n+j)mj.D(\rho\Vert\sigma)=-H(\rho)-\ln c+\tfrac12\operatorname{tr}(\alpha M)=-H(\rho)-\sum_j\ln(2\sinh(m_j/2))+\tfrac12\sum_j(\beta_{jj}+\beta_{n+j,n+j})m_j.D(ρ∥σ)=−H(ρ)−lnc+21​tr(αM)=−H(ρ)−j∑​ln(2sinh(mj​/2))+21​j∑​(βjj​+βn+j,n+j​)mj​.

All logarithms are natural. This evaluates the actual divergence rather than defining it through covariances. If νj≥12\nu_j\ge\tfrac12νj​≥21​ are the symplectic eigenvalues of the input covariance, the independently defined spectral entropy also satisfies

H(ρ)=∑jg(νj−12),g(t)=(t+1)ln⁡(t+1)−tln⁡t,g(0)=0.H(\rho)=\sum_j g(\nu_j-\tfrac12),\qquad g(t)=(t+1)\ln(t+1)-t\ln t,\quad g(0)=0.H(ρ)=j∑​g(νj​−21​),g(t)=(t+1)ln(t+1)−tlnt,g(0)=0.

The conclusion identifies the comparison density with its actual normalized Gibbs exponential, rather than assigning an exponential-matrix label only to its covariance.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Intakenext100_P46
Formal statement
noncomputable section

namespace QAlgorithms.Papers.Independent.Intakenext100.P46

open scoped BigOperators ENNReal ComplexOrder Matrix

attribute [local instance] Classical.propDecidable Classical.decEq

open QAlgorithms.RealRegisters (Hilbert)

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

open QAlgorithms.Papers.Independent.Dependency50.P13
  (finiteRelativeEntropy compression entropyWeight)

open QAlgorithms.Papers.Independent.Claimed50.P47
  (Occupation bipartiteCharacteristic bipartiteWeyl gaussianStateCharacteristic)

-- These accepted definitions have the same completed Hilbert-space meaning.

open QAlgorithms.Papers.Independent.Next46.P13 (coordinateBasis)

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

open QAlgorithms.Papers.Independent.Claimed50.P41 (IsDiagonal)

/-- Chen, PDF pp.4–5, (17),(18),(22): actual operator relative entropy equals
the CM/EM expression, including a single mode (one register may be empty).
rho may have pure normal modes. sigma alone is in the
finite exponential chart. The actual Gibbs density and Gaussian spectral entropy
are identified, and both the matrix trace and the normal-mode sum are given. -/
theorem gaussian_relativeEntropy {A B : Type} [Fintype A] [Fintype B]
    [Nonempty (A ⊕ B)]
    (rho sigma : State (Occupation A × Occupation B))
    (alpha : ModeMatrix (A ⊕ B)) (S : ModeMatrix (A ⊕ B)) (g : (A ⊕ B) → ℝ)
    (hrho : HasGaussian rho 0 alpha) (hS : IsSymplecticMatrix S)
    (hg : ∀ j, 1 / 2 < g j)
    (hsigma : HasGaussian sigma 0 (S * normalDiagonal g * S.transpose)) :
    let M := emFrom S g
    let c := normalization g
    let beta := S⁻¹ * alpha * S.transpose⁻¹
    densityEntropy rho < ⊤ ∧ HasFiniteEM (joinedCoefficients sigma.coefficients) M ∧
      (∃ (Sr : ModeMatrix (A ⊕ B)) (gr : (A ⊕ B) → ℝ),
        IsSymplecticMatrix Sr ∧ (∀ j, 1 / 2 ≤ gr j) ∧
        alpha = Sr * normalDiagonal gr * Sr.transpose ∧
        densityEntropy rho = ENNReal.ofReal (∑ j, bosonic (gr j - 1 / 2))) ∧
      0 < c ∧
      c = ∏ j, (Real.sqrt ((g j) ^ 2 - 1 / 4))⁻¹ ∧
      operatorRelativeEntropy rho sigma = ENNReal.ofReal
        (- (densityEntropy rho).toReal + covarianceObjective alpha M c) ∧
      operatorRelativeEntropy rho sigma = ENNReal.ofReal
        (- (densityEntropy rho).toReal - ∑ j, Real.log (2 * Real.sinh (modeEM (g j) / 2)) +
          (∑ j, (beta (Sum.inl j) (Sum.inl j) +
            beta (Sum.inr j) (Sum.inr j)) * modeEM (g j)) / 2) := by sorry

end QAlgorithms.Papers.Independent.Intakenext100.P46

end
Source
Gaussian relative entropy of entanglement, arXiv:quant-ph/0402109v3, PDF p. 4, §3, Eqs. (17),(18),(22): actual Gaussian relative entropy from CM and EM
Read-back

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

Let A,BA,BA,B be finite types such that I=A⊔BI=A\sqcup BI=A⊔B is nonempty; either AAA or BBB individually may be empty. Write OE=NE\mathcal O_E=\mathbb N^EOE​=NE, HE=ℓ2(E;C)\mathcal H_E=\ell^2(E;\mathbb C)HE​=ℓ2(E;C), J=(01−10)J=\begin{pmatrix}0&1\\-1&0\end{pmatrix}J=(0−1​10​) on I⊔II\sqcup II⊔I, and D(h)=diag⁡(diag⁡h,diag⁡h)D(h)=\operatorname{diag}(\operatorname{diag}h,\operatorname{diag}h)D(h)=diag(diagh,diagh). Let ρ,σ\rho,\sigmaρ,σ be states indexed by OA×OB\mathcal O_A\times\mathcal O_BOA​×OB​. Here a state is a complex coefficient matrix RRR admitting sequences ur,vr∈ℓ2u_r,v_r\in\ell^2ur​,vr​∈ℓ2 with ∑r∥ur∥∥vr∥<∞\sum_r\|u_r\|\|v_r\|<\infty∑r​∥ur​∥∥vr​∥<∞ and Rij=∑rur(i)vr(j)‾R_{ij}=\sum_r u_r(i)\overline{v_r(j)}Rij​=∑r​ur​(i)vr​(j)​, whose every finite quadratic form is real and nonnegative, and whose real diagonal is summable with sum 111. Let α,S\alpha,Sα,S be real matrices indexed by I⊔II\sqcup II⊔I, and let g:I→Rg:I\to\mathbb Rg:I→R. Assume SSS is invertible and SJST=JSJS^{\mathsf T}=JSJST=J, and gj>12g_j>\tfrac12gj​>21​ for every j∈Ij\in Ij∈I. Assume ρ\rhoρ and σ\sigmaσ have centered Gaussian characteristics with covariance matrices α\alphaα and SD(g)STSD(g)S^{\mathsf T}SD(g)ST, respectively: each covariance aaa is symmetric with a−i2Ja-\tfrac i2Ja−2i​J positive semidefinite, and, for every real ξ∈RI⊔I\xi\in\mathbb R^{I\sqcup I}ξ∈RI⊔I, the coefficient sums defining the characteristic are summable, including each inner sum, and satisfy

χR(ξ)=∑u∑vRuvW(ξ)vu=exp⁡ ⁣(−12ξTJTaJξ).\chi_R(\xi)=\sum_u\sum_v R_{uv}W(\xi)_{vu} =\exp\!\left(-\tfrac12\xi^{\mathsf T}J^{\mathsf T}aJ\xi\right).χR​(ξ)=u∑​v∑​Ruv​W(ξ)vu​=exp(−21​ξTJTaJξ).

Here occupations are joined across A⊔BA\sqcup BA⊔B, W(ξ)uv=∏j∈Idujvj(−(ξL,j+iξR,j)/2)W(\xi)_{uv}=\prod_{j\in I}d_{u_jv_j}(-(\xi_{L,j}+i\xi_{R,j})/\sqrt2)W(ξ)uv​=∏j∈I​duj​vj​​(−(ξL,j​+iξR,j​)/2​), and the displacement coefficients used in this formula are

dmn(z)=e−∣z∣2/2m! n!∑r=0min⁡(m,n)zm−r(−zˉ)n−rr!(m−r)!(n−r)!.d_{mn}(z)=e^{-|z|^2/2}\sqrt{m!\,n!} \sum_{r=0}^{\min(m,n)} \frac{z^{m-r}(-\bar z)^{n-r}}{r!(m-r)!(n-r)!}.dmn​(z)=e−∣z∣2/2m!n!​r=0∑min(m,n)​r!(m−r)!(n−r)!zm−r(−zˉ)n−r​.

For a coefficient matrix RRR, let TRT_RTR​ be a chosen bounded operator on ℓ2\ell^2ℓ2 having matrix RRR, or the zero operator if no such bounded operator exists. Define its entropy and the relative entropy of two states by

H(R)=∑i[Re⁡⟨ei,(−xlog⁡x)(TR)ei⟩]+,D(R∥Q)=sup⁡r∈Nv1,…,vr orthonormalF ⁣((⟨vi,TRvj⟩)ij,(⟨vi,TQvj⟩)ij),H(R)=\sum_i\left[\operatorname{Re}\langle e_i,(-x\log x)(T_R)e_i\rangle\right]_+,\qquad D(R\Vert Q)=\sup_{\substack{r\in\mathbb N\\v_1,\ldots,v_r\text{ orthonormal}}} F\!\left((\langle v_i,T_Rv_j\rangle)_{ij}, (\langle v_i,T_Qv_j\rangle)_{ij}\right),H(R)=i∑​[Re⟨ei​,(−xlogx)(TR​)ei​⟩]+​,D(R∥Q)=r∈Nv1​,…,vr​ orthonormal​sup​F((⟨vi​,TR​vj​⟩)ij​,(⟨vi​,TQ​vj​⟩)ij​),

where [a]+=max⁡(a,0)[a]_+=\max(a,0)[a]+​=max(a,0) is regarded as an extended nonnegative real, the functions of operators use continuous functional calculus, and

F(X,Y)={[Re⁡tr⁡((xlog⁡x)(X)−X(log⁡)(Y)−X+Y)]+,ker⁡Y⊆ker⁡X,+∞,otherwise.F(X,Y)= \begin{cases} \left[\operatorname{Re}\operatorname{tr} \bigl((x\log x)(X)-X(\log)(Y)-X+Y\bigr)\right]_+, &\ker Y\subseteq\ker X,\\ +\infty,&\text{otherwise}. \end{cases}F(X,Y)={[Retr((xlogx)(X)−X(log)(Y)−X+Y)]+​,+∞,​kerY⊆kerX,otherwise.​

The supremum includes the empty orthonormal family r=0r=0r=0. Put

mj=log⁡2gj+12gj−1,M=S−TD(m)S−1,c=∏j2sinh⁡(mj/2),β=S−1αS−T.m_j=\log\frac{2g_j+1}{2g_j-1},\qquad M=S^{-\mathsf T}D(m)S^{-1},\qquad c=\prod_j2\sinh(m_j/2),\qquad \beta=S^{-1}\alpha S^{-\mathsf T}.mj​=log2gj​−12gj​+1​,M=S−TD(m)S−1,c=j∏​2sinh(mj​/2),β=S−1αS−T.

Then H(ρ)<∞H(\rho)<\inftyH(ρ)<∞, and the joined coefficient matrix σ^uv=σ(u∣A,u∣B),(v∣A,v∣B)\widehat\sigma_{uv}=\sigma_{(u|_A,u|_B),(v|_A,v|_B)}σuv​=σ(u∣A​,u∣B​),(v∣A​,v∣B​)​ has the following finite exponential-matrix representation: it is a density matrix in the coefficient sense above, and there exist an invertible real S′S'S′ with S′JS′T=JS'JS'^{\mathsf T}=JS′JS′T=J, positive numbers mj′m'_jmj′​, and a surjective complex linear isometry UUU of ℓ2(OI)\ell^2(\mathcal O_I)ℓ2(OI​), such that

CU−1(W(ξ))=W(−J(S′)TJξ)for every ξ,σ^=CU(Qm′),M=(S′)−TD(m′)(S′)−1,\mathcal C_{U^{-1}}(W(\xi))=W(-J(S')^{\mathsf T}J\xi)\quad\text{for every }\xi,\qquad \widehat\sigma=\mathcal C_U(Q_{m'}),\qquad M=(S')^{-\mathsf T}D(m')(S')^{-1},CU−1​(W(ξ))=W(−J(S′)TJξ)for every ξ,σ=CU​(Qm′​),M=(S′)−TD(m′)(S′)−1,

where Qm′(u,v)=1u=v∏j(1−e−mj′)e−mj′ujQ_{m'}(u,v)=\mathbf1_{u=v}\prod_j(1-e^{-m'_j})e^{-m'_ju_j}Qm′​(u,v)=1u=v​∏j​(1−e−mj′​)e−mj′​uj​, and, writing Uia=(Uea)iU_{ia}=(Ue_a)_iUia​=(Uea​)i​, the conjugation here literally means CU(X)ij=∑aUia∑bXabUjb‾\mathcal C_U(X)_{ij}=\sum_aU_{ia}\sum_bX_{ab}\overline{U_{jb}}CU​(X)ij​=∑a​Uia​∑b​Xab​Ujb​​. Moreover, there exist an invertible real SρS_\rhoSρ​ and gρ:I→Rg_\rho:I\to\mathbb Rgρ​:I→R satisfying

SρJSρT=J,(gρ)j≥12,α=SρD(gρ)SρT,H(ρ)=[∑jb((gρ)j−12)]+,b(x)=(x+1)log⁡(x+1)−xlog⁡x.S_\rho JS_\rho^{\mathsf T}=J,\qquad (g_\rho)_j\ge\tfrac12,\qquad \alpha=S_\rho D(g_\rho)S_\rho^{\mathsf T},\qquad H(\rho)=\left[\sum_j b((g_\rho)_j-\tfrac12)\right]_+, \quad b(x)=(x+1)\log(x+1)-x\log x.Sρ​JSρT​=J,(gρ​)j​≥21​,α=Sρ​D(gρ​)SρT​,H(ρ)=[j∑​b((gρ​)j​−21​)]+​,b(x)=(x+1)log(x+1)−xlogx.

Finally, all of the following hold:

c>0,c=∏j(gj2−14)−1/2,D(ρ∥σ)=[−H(ρ)R−log⁡c+12tr⁡(αM)]+=[−H(ρ)R−∑jlog⁡(2sinh⁡(mj/2))+12∑j(βLj,Lj+βRj,Rj)mj]+.\begin{gathered} c>0,\qquad c=\prod_j(g_j^2-\tfrac14)^{-1/2},\\ D(\rho\Vert\sigma) =\left[-H(\rho)_{\mathbb R}-\log c+\tfrac12\operatorname{tr}(\alpha M)\right]_+\\ =\left[-H(\rho)_{\mathbb R} -\sum_j\log(2\sinh(m_j/2)) +\tfrac12\sum_j(\beta_{L j,L j}+\beta_{R j,R j})m_j\right]_+. \end{gathered}c>0,c=j∏​(gj2​−41​)−1/2,D(ρ∥σ)=[−H(ρ)R​−logc+21​tr(αM)]+​=[−H(ρ)R​−j∑​log(2sinh(mj​/2))+21​j∑​(βLj,Lj​+βRj,Rj​)mj​]+​.​

Here H(ρ)RH(\rho)_{\mathbb R}H(ρ)R​ is the real value of the finite extended entropy, log⁡\loglog is the real natural logarithm, and its total-function convention includes log⁡0=0\log0=0log0=0, so the entropy formula allows (gρ)j=12(g_\rho)_j=\tfrac12(gρ​)j​=21​.

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