Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

§5 unnumbered: von Neumann entropy approximators

Open
QAlgorithms.Papers.Independent.Support50.P34.entropy_rankApproximation

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

prove2quantum

For a separable complex Hilbert space, the von Neumann entropy H(ρ)=−Tr⁡ρlog⁡ρ∈[0,+∞]H(\rho)=-\operatorname{Tr}\rho\log\rho\in[0,+\infty]H(ρ)=−Trρlogρ∈[0,+∞] is concave, lower semicontinuous and unitarily invariant, and finite continuous on every Sk(H)S_k(\mathcal H)Sk​(H), k≥1k\ge1k≥1. For Borel probabilities μ\muμ with barycenter ρ\rhoρ supported on SkS_kSk​, and their countably atomic subfamily, define

Hk(ρ)=sup⁡{πi,ρi}∑iπiH(ρi)=sup⁡μ∈Pρ(Sk)∫H(σ) dμ.H_k(\rho)=\sup_{\{\pi_i,\rho_i\}}\sum_i\pi_iH(\rho_i)=\sup_{\mu\in\mathcal P_\rho(S_k)}\int H(\sigma)\,d\mu.Hk​(ρ)={πi​,ρi​}sup​i∑​πi​H(ρi​)=μ∈Pρ​(Sk​)sup​∫H(σ)dμ.

The functions HkH_kHk​ are nonnegative concave finite continuous bounded and unitarily invariant, satisfy Hk≤log⁡kH_k\le\log kHk​≤logk, agree with HHH on SkS_kSk​, and increase pointwise to HHH. For the extended relative entropy DDD, both defect formulas hold:

H(ρ)−Hk(ρ)=inf⁡{πi,ρi}∑iπiD(ρi∥ρ)=inf⁡μ∈Pρ(Sk)∫D(σ∥ρ) dμ.H(\rho)-H_k(\rho)=\inf_{\{\pi_i,\rho_i\}}\sum_i\pi_iD(\rho_i\Vert\rho)=\inf_{\mu\in\mathcal P_\rho(S_k)}\int D(\sigma\Vert\rho)\,d\mu.H(ρ)−Hk​(ρ)={πi​,ρi​}inf​i∑​πi​D(ρi​∥ρ)=μ∈Pρ​(Sk​)inf​∫D(σ∥ρ)dμ.

All logarithms are natural. These are the rank entropy approximators and their relative-entropy errors. Formalization Note The valid dimension-independent bound replaces footnote 14's full-range assertion; that correction requires independent moderation.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Support50_P34
Formal statement
noncomputable section

open Filter MeasureTheory

open scoped ENNReal NNReal BigOperators

namespace QAlgorithms.Papers.Independent.Support50.P34

open QAlgorithms.Papers.Independent.Dependency50.P13
  (State PositiveTraceClass traceNormValue stateTraceNormTopology traceNormTopology
   operatorTrace IsPositive IsTraceClass stateEntropy relativeEntropy entropyWeight)

open QAlgorithms.Papers.Independent.Dependency50.P25 (extendedIntegral)

open QAlgorithms.Papers.Independent.Frontier18.P01 (binaryEntropy)

attribute [local instance] Classical.propDecidable Classical.decEq

section Operators

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

variable {I : Type*}

section Statements

variable [hsep : TopologicalSpace.SeparableSpace H] [hI : Countable I]

include hsep hI

/-- Section 5 unnumbered result, arXiv:0904.1963v3, PDF p.16: both entropy roofs
and both relative-entropy defect infima, plus the stated analytical properties.
The footnote's dimension-dependent full-range claim is replaced by the planned
valid upper bound log k; no finite-dimensional state space has all infinite entropies. -/
theorem entropy_rankApproximation (b : HilbertBasis I ℂ H) :
    LowerSemicontinuous (stateEntropyExt b) ∧
    StateConcaveOn b Set.univ (stateEntropyExt b) ∧
    StateUnitaryInvariant b (stateEntropyExt b) ∧
    (∀ k : ℕ, 0 < k → FiniteERealContinuousOn (stateEntropyExt b) (rankStates b k)) ∧
    (∀ k : ℕ, 0 < k →
      rankSigmaRoof b (stateEntropyExt b) k = rankMuRoof b (stateEntropyExt b) k ∧
      StateConcaveOn b Set.univ (rankMuRoof b (stateEntropyExt b) k) ∧
      FiniteERealContinuousOn (rankMuRoof b (stateEntropyExt b) k) Set.univ ∧
      StateUnitaryInvariant b (rankMuRoof b (stateEntropyExt b) k) ∧
      (∀ rho, 0 ≤ rankMuRoof b (stateEntropyExt b) k rho ∧
        rankMuRoof b (stateEntropyExt b) k rho ≤ (Real.log (k : ℝ) : EReal)) ∧
      (∀ rho ∈ rankStates b k, rankMuRoof b (stateEntropyExt b) k rho = stateEntropyExt b rho) ∧
      (∀ rho, stateEntropy b rho - (rankMuRoof b (stateEntropyExt b) k rho).toENNReal =
        stateRelativeDefect b k true rho ∧
        stateEntropy b rho - (rankMuRoof b (stateEntropyExt b) k rho).toENNReal =
        stateRelativeDefect b k false rho)) ∧
    (∀ rho, Monotone (fun n : ℕ => rankMuRoof b (stateEntropyExt b) (n + 1) rho) ∧
      Tendsto (fun n : ℕ => rankMuRoof b (stateEntropyExt b) (n + 1) rho)
        atTop (nhds (stateEntropyExt b rho))) := by sorry

end Statements

end Operators

end QAlgorithms.Papers.Independent.Support50.P34

end
Source
Continuity of the von Neumann entropy, arXiv:0904.1963v3, PDF p. 16, §5 unnumbered: von Neumann entropy approximators
Read-back

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

Let HHH be a complex Hilbert space with Hilbert basis b=(bi)i∈Ib=(b_i)_{i\in I}b=(bi​)i∈I​. Here trace-class operators admit Tx=∑r⟨vr,x⟩urTx=\sum_r\langle v_r,x\rangle u_rTx=∑r​⟨vr​,x⟩ur​ with ∑r∥ur∥∥vr∥<∞\sum_r\|u_r\|\|v_r\|<\infty∑r​∥ur​∥∥vr​∥<∞; denote the infimum of these sums by ν(T)\nu(T)ν(T). Let Sb\mathcal S_bSb​ be the self-adjoint trace-class operators with nonnegative quadratic forms and basis trace 111, topologized by all functions T↦ν(T−S)T\mapsto\nu(T-S)T↦ν(T−S) for trace-class SSS, and let Sb,k\mathcal S_{b,k}Sb,k​ consist of states whose ranges lie in the span of some kkk vectors. Put h(t)=−tlog⁡th(t)=-t\log th(t)=−tlogt, using the natural logarithm, and define

Sb(ρ)=∑imax⁡ ⁣{Re⁡⟨bi,h(ρ)bi⟩,0}∈[0,∞],S_b(\rho)=\sum_i \max\!\left\{\operatorname{Re}\langle b_i,h(\rho)b_i\rangle,0\right\} \in[0,\infty],Sb​(ρ)=i∑​max{Re⟨bi​,h(ρ)bi​⟩,0}∈[0,∞],

where h(ρ)h(\rho)h(ρ) is supplied by continuous functional calculus; regard this also as an extended-real-valued function. Let Fk(ρ)F_k(\rho)Fk​(ρ) be the supremum of the extended integrals of SbS_bSb​ over probability measures on Sb,k\mathcal S_{b,k}Sb,k​ with barycenter ρ\rhoρ, where barycenter requires integrability and equality of every matrix element, and let FkatF_k^{\mathrm{at}}Fkat​ restrict to measures concentrated on countable sets. For trace-class operators X,YX,YX,Y, define D(X,Y)D(X,Y)D(X,Y) as the supremum over all natural nnn, including 000, and all orthonormal families v:{0,…,n−1}→Hv:\{0,\ldots,n-1\}\to Hv:{0,…,n−1}→H, of the following value: for the compressed matrices Aij=⟨vi,Xvj⟩A_{ij}=\langle v_i,Xv_j\rangleAij​=⟨vi​,Xvj​⟩, Bij=⟨vi,Yvj⟩B_{ij}=\langle v_i,Yv_j\rangleBij​=⟨vi​,Yvj​⟩, use +∞+\infty+∞ if ker⁡B⊈ker⁡A\ker B\not\subseteq\ker AkerB⊆kerA, and otherwise use

max⁡ ⁣{Re⁡Tr⁡(Alog⁡A−Alog⁡B−A+B),0},\max\!\left\{\operatorname{Re}\operatorname{Tr} \bigl(A\log A-A\log B-A+B\bigr),0\right\},max{ReTr(AlogA−AlogB−A+B),0},

with the matrix functions supplied by continuous functional calculus. The declaration asserts that SbS_bSb​ is lower semicontinuous, concave in operator mixtures, and invariant under every bounded complex-linear UUU satisfying U∗U=UU∗=1U^*U=UU^*=1U∗U=UU∗=1, in the sense that σ=UρU∗\sigma=U\rho U^*σ=UρU∗ implies equal values. It is finite and real-continuous on each Sb,k\mathcal S_{b,k}Sb,k​, k>0k>0k>0. For every such kkk, Fkat=FkF_k^{\mathrm{at}}=F_kFkat​=Fk​; FkF_kFk​ is concave, finite, real-continuous on all states, and invariant under those same unitary conjugations; and

0≤Fk(ρ)≤log⁡k,Fk(ρ)=Sb(ρ)(ρ∈Sb,k).0\le F_k(\rho)\le\log k,\qquad F_k(\rho)=S_b(\rho)\quad(\rho\in\mathcal S_{b,k}).0≤Fk​(ρ)≤logk,Fk​(ρ)=Sb​(ρ)(ρ∈Sb,k​).

Here concavity means convexity of the operator image of the domain together with aF(ρ)+cF(σ)≤F(τ)aF(\rho)+cF(\sigma)\le F(\tau)aF(ρ)+cF(σ)≤F(τ) whenever τ=aρ+cσ\tau=a\rho+c\sigmaτ=aρ+cσ, a,c≥0a,c\ge0a,c≥0, and a+c=1a+c=1a+c=1. Furthermore, the difference Sb(ρ)−(Fk(ρ))+S_b(\rho)-(F_k(\rho))_+Sb​(ρ)−(Fk​(ρ))+​, taken in [0,∞][0,\infty][0,∞], equals both the infimum of ∫D(σ,ρ) dμ(σ)\int D(\sigma,\rho)\,d\mu(\sigma)∫D(σ,ρ)dμ(σ) over all probability measures on Sb,k\mathcal S_{b,k}Sb,k​ with barycenter ρ\rhoρ, and the infimum over only those measures concentrated on countable sets. Finally, for every ρ\rhoρ, Fn+1(ρ)F_{n+1}(\rho)Fn+1​(ρ) is nondecreasing and converges in the extended-real topology to Sb(ρ)S_b(\rho)Sb​(ρ), which may be +∞+\infty+∞; in particular, the bound at k=1k=1k=1 forces F1(ρ)=0F_1(\rho)=0F1​(ρ)=0.

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