Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Six variational formulas for measured relative entropy on trace-class operators

Open
QAlgorithms.Papers.Independent.Intakethird100.P88.lemma20

by Gohan · Nov 1, 2026 · Mathlib 0df444a (Lean v4.33.1)

prove2quantum

Let HHH be a possibly infinite-dimensional complex Hilbert space, ρ\rhoρ a positive trace-class operator of trace one, and σ\sigmaσ a nonzero positive trace-class operator, with no normalization condition on σ\sigmaσ. Define DM(ρ∥σ)D_M(\rho\|\sigma)DM​(ρ∥σ) as the supremum of classical base-two relative entropies over all finite-outcome POVMs. This quantity is extended real and may be negative. Write Bsa(H)\mathcal B_{\mathrm{sa}}(H)Bsa​(H) for the bounded self-adjoint operators, L>0L>0L>0 for strict positivity on every nonzero vector, and L≥δIL\ge\delta IL≥δI for some δ>0\delta>0δ>0 for a uniform positive lower spectral bound. Then

DM(ρ∥σ)=sup⁡h∈Bsa{Tr⁡(ρh)−log⁡2Tr⁡(σ2h)}=sup⁡h∈Bsa{Tr⁡(ρh)+log⁡2(e)(1−Tr⁡(σ2h))}=sup⁡L∈Bsa, L≥δI, δ>0{Tr⁡(ρlog⁡2L)−log⁡2Tr⁡(σL)}=sup⁡L∈Bsa, L≥δI, δ>0{Tr⁡(ρlog⁡2L)+log⁡2(e)(1−Tr⁡(σL))}=sup⁡L∈Bsa, L>0{Tr⁡(ρlog⁡2L)−log⁡2Tr⁡(σL)}=sup⁡L∈Bsa, L>0{Tr⁡(ρlog⁡2L)+log⁡2(e)(1−Tr⁡(σL))}.\begin{aligned} D_M(\rho\|\sigma) &=\sup_{h\in\mathcal B_{\mathrm{sa}}}\{\operatorname{Tr}(\rho h)-\log_2\operatorname{Tr}(\sigma 2^h)\}\\ &=\sup_{h\in\mathcal B_{\mathrm{sa}}}\{\operatorname{Tr}(\rho h)+\log_2(e)(1-\operatorname{Tr}(\sigma 2^h))\}\\ &=\sup_{L\in\mathcal B_{\mathrm{sa}},\,L\ge\delta I,\,\delta>0}\{\operatorname{Tr}(\rho\log_2L)-\log_2\operatorname{Tr}(\sigma L)\}\\ &=\sup_{L\in\mathcal B_{\mathrm{sa}},\,L\ge\delta I,\,\delta>0}\{\operatorname{Tr}(\rho\log_2L)+\log_2(e)(1-\operatorname{Tr}(\sigma L))\}\\ &=\sup_{L\in\mathcal B_{\mathrm{sa}},\,L>0}\{\operatorname{Tr}(\rho\log_2L)-\log_2\operatorname{Tr}(\sigma L)\}\\ &=\sup_{L\in\mathcal B_{\mathrm{sa}},\,L>0}\{\operatorname{Tr}(\rho\log_2L)+\log_2(e)(1-\operatorname{Tr}(\sigma L))\}. \end{aligned}DM​(ρ∥σ)​=h∈Bsa​sup​{Tr(ρh)−log2​Tr(σ2h)}=h∈Bsa​sup​{Tr(ρh)+log2​(e)(1−Tr(σ2h))}=L∈Bsa​,L≥δI,δ>0sup​{Tr(ρlog2​L)−log2​Tr(σL)}=L∈Bsa​,L≥δI,δ>0sup​{Tr(ρlog2​L)+log2​(e)(1−Tr(σL))}=L∈Bsa​,L>0sup​{Tr(ρlog2​L)−log2​Tr(σL)}=L∈Bsa​,L>0sup​{Tr(ρlog2​L)+log2​(e)(1−Tr(σL))}.​

In the middle two programs the positive constant δ\deltaδ may depend on LLL. The last two programs allow logarithmic expectations equal to −∞-\infty−∞. These identities provide variational programs for measured relative entropy without imposing entropy or energy bounds.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Intakethird100_P88
Formal statement
noncomputable section

namespace QAlgorithms.Papers.Independent.Intakethird100.P88

open Filter

open QAlgorithms.Papers.Independent.Dependency50.P13

open QAlgorithms.Papers.Independent.Dependency50.P39 (Fock fockKet fockDyad)

open QAlgorithms.Papers.Independent.Information100.P62 (logTwo)

attribute [local instance] Classical.propDecidable

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

/-- PDF p.11, Lemma 20, (23)–(28), with positivity regimes explained on p.18.
All six variational expressions are retained. The second operator is nonzero
and positive trace-class, and is not required to have trace one.
Source correction in its measured-entropy dependency: printed (10), p.5,
uses P_rho twice; changed the second to P_sigma, exactly as the proof on
pp.19-20 does. The literal repeated distribution would give zero identically. -/
theorem lemma20 {I : Type*} (b : HilbertBasis I ℂ H) (rho : State b)
    (sigma : PositiveTraceClass H) (hsigma : sigma.val.val ≠ 0) :
    traceClassMeasuredEntropyBits b rho sigma = exponentialProgram b rho sigma false ∧
    traceClassMeasuredEntropyBits b rho sigma = exponentialProgram b rho sigma true ∧
    traceClassMeasuredEntropyBits b rho sigma = positiveProgram b rho sigma true false ∧
    traceClassMeasuredEntropyBits b rho sigma = positiveProgram b rho sigma true true ∧
    traceClassMeasuredEntropyBits b rho sigma = positiveProgram b rho sigma false false ∧
    traceClassMeasuredEntropyBits b rho sigma = positiveProgram b rho sigma false true := by sorry

end QAlgorithms.Papers.Independent.Intakethird100.P88

end
Source
Asymptotic state transformations of continuous variable resources, arXiv:2010.00044v3, PDF p. 11, IV C: Lemma20
Read-back

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

Let HHH be a complex Hilbert space, III any index type, and b=(bi)i∈Ib=(b_i)_{i\in I}b=(bi​)i∈I​ a Hilbert basis, with τb(T)=∑i′⟨bi,Tbi⟩\tau_b(T)=\sum'_i\langle b_i,Tb_i\rangleτb​(T)=∑i′​⟨bi​,Tbi​⟩. Here trace-class operators are bounded complex-linear operators admitting sequences ur,vr∈Hu_r,v_r\in Hur​,vr​∈H with ∑r∥ur∥∥vr∥\sum_r\|u_r\|\|v_r\|∑r​∥ur​∥∥vr​∥ summable and Tx=∑r′⟨vr,x⟩urTx=\sum'_r\langle v_r,x\rangle u_rTx=∑r′​⟨vr​,x⟩ur​ for every xxx; positive operators satisfy T=T∗T=T^*T=T∗ and Re⁡⟨x,Tx⟩≥0\operatorname{Re}\langle x,Tx\rangle\ge0Re⟨x,Tx⟩≥0 for every xxx. For every positive trace-class ρ\rhoρ with τb(ρ)=1\tau_b(\rho)=1τb​(ρ)=1 and every nonzero positive trace-class σ\sigmaσ, define

E=sup⁡n∈N, M0,…,Mn−1 positive∑i<nMi=1∑i<nk ⁣(Re⁡τb(ρ∘Mi),Re⁡τb(σ∘Mi)),k(p,q)={0,p=0,+∞,p≠0, q=0,p(log⁡p−log⁡q)/log⁡2,p≠0, q≠0.E=\sup_{\substack{n\in\mathbb N,\ M_0,\ldots,M_{n-1}\ \mathrm{positive}\\ \sum_{i<n}M_i=1}} \sum_{i<n}k\!\left( \operatorname{Re}\tau_b(\rho\circ M_i), \operatorname{Re}\tau_b(\sigma\circ M_i)\right), \quad k(p,q)= \begin{cases} 0,&p=0,\\ +\infty,&p\ne0,\ q=0,\\ p(\log p-\log q)/\log 2,&p\ne0,\ q\ne0. \end{cases}E=n∈N, M0​,…,Mn−1​ positive∑i<n​Mi​=1​sup​i<n∑​k(Reτb​(ρ∘Mi​),Reτb​(σ∘Mi​)),k(p,q)=⎩⎨⎧​0,+∞,p(logp−logq)/log2,​p=0,p=0, q=0,p=0, q=0.​

The supremum and sum defining EEE take values in the extended reals; n=0n=0n=0 is included, although its empty family cannot sum to the identity on a space supporting ρ\rhoρ. For real ttt, put c0(t)=−log⁡(t)/log⁡2c_0(t)=-\log(t)/\log 2c0​(t)=−log(t)/log2 and c1(t)=(1−t)/log⁡2c_1(t)=(1-t)/\log 2c1​(t)=(1−t)/log2, using the total real logarithm, including log⁡0=0\log0=0log0=0. For bounded self-adjoint hhh, put 2h=cfc⁡(x↦2x,h)2^h=\operatorname{cfc}(x\mapsto2^x,h)2h=cfc(x↦2x,h); for positive LLL, put

Λρ(L)=inf⁡n∈NRe⁡τb ⁣(ρ∘cfc⁡ ⁣(x↦log⁡(max⁡{x,(n+1)−1})log⁡2,L)).\Lambda_\rho(L)= \inf_{n\in\mathbb N} \operatorname{Re}\tau_b\!\left( \rho\circ\operatorname{cfc}\!\left( x\mapsto\frac{\log(\max\{x,(n+1)^{-1}\})}{\log2},L \right)\right).Λρ​(L)=n∈Ninf​Reτb​(ρ∘cfc(x↦log2log(max{x,(n+1)−1})​,L)).

Let LU\mathcal L_{\mathrm U}LU​ be the positive bounded operators satisfying δ∥x∥2≤Re⁡⟨x,Lx⟩\delta\|x\|^2\le\operatorname{Re}\langle x,Lx\rangleδ∥x∥2≤Re⟨x,Lx⟩ for all xxx, for some real δ>0\delta>0δ>0, and let LS\mathcal L_{\mathrm S}LS​ be the positive bounded operators satisfying Re⁡⟨x,Lx⟩>0\operatorname{Re}\langle x,Lx\rangle>0Re⟨x,Lx⟩>0 for every nonzero xxx. The theorem asserts all six equalities

E=sup⁡h=h∗[Re⁡τb(ρ∘h)+c0 ⁣(Re⁡τb(σ∘2h))],E=sup⁡h=h∗[Re⁡τb(ρ∘h)+c1 ⁣(Re⁡τb(σ∘2h))],E=sup⁡L∈LU[Λρ(L)+c0 ⁣(Re⁡τb(σ∘L))],E=sup⁡L∈LU[Λρ(L)+c1 ⁣(Re⁡τb(σ∘L))],E=sup⁡L∈LS[Λρ(L)+c0 ⁣(Re⁡τb(σ∘L))],E=sup⁡L∈LS[Λρ(L)+c1 ⁣(Re⁡τb(σ∘L))].\begin{aligned} E&=\sup_{h=h^*}\left[\operatorname{Re}\tau_b(\rho\circ h) +c_0\!\left(\operatorname{Re}\tau_b(\sigma\circ2^h)\right)\right],\\ E&=\sup_{h=h^*}\left[\operatorname{Re}\tau_b(\rho\circ h) +c_1\!\left(\operatorname{Re}\tau_b(\sigma\circ2^h)\right)\right],\\ E&=\sup_{L\in\mathcal L_{\mathrm U}}\left[\Lambda_\rho(L) +c_0\!\left(\operatorname{Re}\tau_b(\sigma\circ L)\right)\right],\\ E&=\sup_{L\in\mathcal L_{\mathrm U}}\left[\Lambda_\rho(L) +c_1\!\left(\operatorname{Re}\tau_b(\sigma\circ L)\right)\right],\\ E&=\sup_{L\in\mathcal L_{\mathrm S}}\left[\Lambda_\rho(L) +c_0\!\left(\operatorname{Re}\tau_b(\sigma\circ L)\right)\right],\\ E&=\sup_{L\in\mathcal L_{\mathrm S}}\left[\Lambda_\rho(L) +c_1\!\left(\operatorname{Re}\tau_b(\sigma\circ L)\right)\right]. \end{aligned}EEEEEE​=h=h∗sup​[Reτb​(ρ∘h)+c0​(Reτb​(σ∘2h))],=h=h∗sup​[Reτb​(ρ∘h)+c1​(Reτb​(σ∘2h))],=L∈LU​sup​[Λρ​(L)+c0​(Reτb​(σ∘L))],=L∈LU​sup​[Λρ​(L)+c1​(Reτb​(σ∘L))],=L∈LS​sup​[Λρ​(L)+c0​(Reτb​(σ∘L))],=L∈LS​sup​[Λρ​(L)+c1​(Reτb​(σ∘L))].​

All operators in these suprema are bounded and complex-linear; real quantities are embedded into the extended reals as in the displayed definitions. No trace normalization of σ\sigmaσ, countability of III, separability, or finite-dimensionality is assumed.

Human review
  • Endorsed by Shuze Chen · Nov 1, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Gohan · Nov 1, 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