Prove2Me
Navigate
DiscoverCollectionsFormalpediaBlogsUsersMomentumMy Missions+
Prove2Me
⌕
Log in
← Formalpedia

Gibbs structure and block Hessian bounds at a minimizer of a regularized entropy functional

Proved
OAI.EntropyPhotonNumber.minimum_all_blocks_hessian_data

by Shuze Chen · Oct 16, 2026 · Mathlib 0df444a (Lean v4.33.1)

entropyoai-mathoptimizationquantum-information

Let F=ℓ2(Nn,C)F=\ell^2(\mathbb{N}^n,\mathbb{C})F=ℓ2(Nn,C) be the nnn-mode Fock space with number basis (∣k⟩)k∈Nn(|k\rangle)_{k\in\mathbb{N}^n}(∣k⟩)k∈Nn​, total number ∣k∣=k1+⋯+kn|k|=k_1+\dots+k_n∣k∣=k1​+⋯+kn​, annihilation operators aja_jaj​, and the diagonal weight N^=1+∑jaj†aj\hat N=1+\sum_ja_j^\dagger a_jN^=1+∑j​aj†​aj​, i.e. N^∣k⟩=(1+∣k∣)∣k⟩\hat N|k\rangle=(1+|k|)|k\rangleN^∣k⟩=(1+∣k∣)∣k⟩. A state is a positive bounded operator on FFF whose number-basis diagonal sums to 111. For a state ρ\rhoρ write S(ρ)=−Tr⁡ρlog⁡ρS(\rho)=-\operatorname{Tr}\rho\log\rhoS(ρ)=−Trρlogρ, E(ρ)=∑k∣k∣ ρkkE(\rho)=\sum_k|k|\,\rho_{kk}E(ρ)=∑k​∣k∣ρkk​ and M4(ρ)=∑k(1+∣k∣)4ρkkM_4(\rho)=\sum_k(1+|k|)^4\rho_{kk}M4​(ρ)=∑k​(1+∣k∣)4ρkk​; ρ\rhoρ has finite mmm-th moment if ∑k(1+∣k∣)mρkk<∞\sum_k(1+|k|)^m\rho_{kk}<\infty∑k​(1+∣k∣)mρkk​<∞.

A Kraus flow consists of bounded operators (Vi)i(V_i)_{i}(Vi​)i​ on FFF with ∑i∥Vix∥2=∥x∥2\sum_i\|V_ix\|^2=\|x\|^2∑i​∥Vi​x∥2=∥x∥2 for all xxx, a constant F0≥0F_0\ge0F0​≥0 with ∑i⟨Vix,n^Vix⟩≤⟨x,n^x⟩+F0∥x∥2\sum_i\langle V_ix,\hat n V_ix\rangle\le\langle x,\hat nx\rangle+F_0\|x\|^2∑i​⟨Vi​x,n^Vi​x⟩≤⟨x,n^x⟩+F0​∥x∥2 (where n^=∑jaj†aj\hat n=\sum_ja_j^\dagger a_jn^=∑j​aj†​aj​, both sides in [0,∞][0,\infty][0,∞]), and a map F\mathcal FF on states such that F(ρ)=∑i,l∣Vivl⟩⟨Vivl∣\mathcal F(\rho)=\sum_{i,l}|V_iv_l\rangle\langle V_iv_l|F(ρ)=∑i,l​∣Vi​vl​⟩⟨Vi​vl​∣ whenever ρ=∑l∣vl⟩⟨vl∣\rho=\sum_l|v_l\rangle\langle v_l|ρ=∑l​∣vl​⟩⟨vl​∣ with ∑l∥vl∥2=1\sum_l\|v_l\|^2=1∑l​∥vl​∥2=1, and F\mathcal FF preserves finite energy. Its Heisenberg dual is F∗(Z)=∑iVi∗ZVi\mathcal F^*(Z)=\sum_iV_i^*ZV_iF∗(Z)=∑i​Vi∗​ZVi​.

A Gibbs representation of ρ\rhoρ with quartic coefficient kkk consists of a positive bounded operator BρB_\rhoBρ​, a Hilbert basis (gj)j(g_j)_{j}(gj​)j​ of FFF and numbers uj>0u_j>0uj​>0 with Z=∑je−uj<∞\mathcal Z=\sum_je^{-u_j}<\inftyZ=∑j​e−uj​<∞ such that N^−2(k+N^−3/2BρN^−3/2)−1N^−2gj=uj−1gj\hat N^{-2}\bigl(k+\hat N^{-3/2}B_\rho\hat N^{-3/2}\bigr)^{-1}\hat N^{-2}g_j=u_j^{-1}g_jN^−2(k+N^−3/2Bρ​N^−3/2)−1N^−2gj​=uj−1​gj​ for all jjj and ρ=∑je−ujZ∣gj⟩⟨gj∣\rho=\sum_j\frac{e^{-u_j}}{\mathcal Z}|g_j\rangle\langle g_j|ρ=∑j​Ze−uj​​∣gj​⟩⟨gj​∣ (moreover ρ\rhoρ is injective and has finite seventh moment). Formally ρ=e−H/Z\rho=e^{-H}/\mathcal Zρ=e−H/Z with H=kN^4+N^1/2BρN^1/2H=k\hat N^4+\hat N^{1/2}B_\rho\hat N^{1/2}H=kN^4+N^1/2Bρ​N^1/2. Its quartic logarithm is the bounded operator

Lρ=k Id+N^−3/2BρN^−3/2+(log⁡Z) N^−4,\mathcal L_\rho=k\,\mathrm{Id}+\hat N^{-3/2}B_\rho\hat N^{-3/2}+(\log\mathcal Z)\,\hat N^{-4},Lρ​=kId+N^−3/2Bρ​N^−3/2+(logZ)N^−4,

formally N^−2(−log⁡ρ)N^−2\hat N^{-2}(-\log\rho)\hat N^{-2}N^−2(−logρ)N^−2.

For a Hilbert basis (ej)(e_j)(ej​) and weights wj>0w_j>0wj​>0 with ∑jwj=1\sum_j w_j=1∑j​wj​=1, write the BKM variance of a bounded operator TTT as

Vare,w(T)=∑i,jL(wi,wj) ∣⟨ei,Tej⟩∣2−∣∑jwj⟨ej,Tej⟩∣2,\mathrm{Var}_{e,w}(T)=\sum_{i,j}L(w_i,w_j)\,|\langle e_i,Te_j\rangle|^2-\Bigl|\sum_jw_j\langle e_j,Te_j\rangle\Bigr|^2 ,Vare,w​(T)=i,j∑​L(wi​,wj​)∣⟨ei​,Tej​⟩∣2−​j∑​wj​⟨ej​,Tej​⟩​2,

with LLL the logarithmic mean.

Theorem. Let F\mathcal FF be a Kraus flow, r>0r>0r>0, and τ=τr\tau=\tau_rτ=τr​ the thermal state with mean photon number rrr in each mode. Let κ∈(0,1]\kappa\in(0,1]κ∈(0,1], c>0c>0c>0, a>0a>0a>0, β≥0\beta\ge0β≥0, ζ>0\zeta>0ζ>0. Let ρ\rhoρ be a state with finite fourth moment, and let (dj)j∈J(d_j)_{j\in J}(dj​)j∈J​ be a Hilbert basis and qj>0q_j>0qj​>0, ∑jqj=1\sum_jq_j=1∑j​qj​=1, such that

ρ0:=(1−κ)F(ρ)+κτsatisfiesρ0dj=qjdj  for all j.\rho_0:=(1-\kappa)\mathcal F(\rho)+\kappa\tau\quad\text{satisfies}\quad\rho_0d_j=q_jd_j\ \text{ for all }j .ρ0​:=(1−κ)F(ρ)+κτsatisfiesρ0​dj​=qj​dj​  for all j.

Suppose ρ\rhoρ minimizes

Φ(σ)=c S((1−κ)F(σ)+κτ)−a S(σ)+βE(σ)+ζM4(σ)\Phi(\sigma)=c\,S\bigl((1-\kappa)\mathcal F(\sigma)+\kappa\tau\bigr)-a\,S(\sigma)+\beta E(\sigma)+\zeta M_4(\sigma)Φ(σ)=cS((1−κ)F(σ)+κτ)−aS(σ)+βE(σ)+ζM4​(σ)

among all states σ\sigmaσ with finite fourth moment. Then there is a Gibbs representation (Bρ,g,u)(B_\rho,g,u)(Bρ​,g,u) of ρ\rhoρ with quartic coefficient ζ/a\zeta/aζ/a such that:

  1. for every x∈Fx\in Fx∈F, in [0,∞][0,\infty][0,∞],
⟨x,Bρx⟩=c(1−κ)a∑i∑j(−log⁡qj) ∣⟨dj,ViN^−1/2x⟩∣2+βa∑k∣k∣ ∣(N^−1/2x)k∣2;\langle x,B_\rho x\rangle=\frac{c(1-\kappa)}{a}\sum_i\sum_j(-\log q_j)\,\bigl|\langle d_j,V_i\hat N^{-1/2}x\rangle\bigr|^2+\frac\beta a\sum_k|k|\,\bigl|(\hat N^{-1/2}x)_k\bigr|^2 ;⟨x,Bρ​x⟩=ac(1−κ)​i∑​j∑​(−logqj​)​⟨dj​,Vi​N^−1/2x⟩​2+aβ​k∑​∣k∣​(N^−1/2x)k​​2;
  1. for every finite set s⊆Js\subseteq Js⊆J and every complex matrix Z=(Zii′)i,i′∈sZ=(Z_{ii'})_{i,i'\in s}Z=(Zii′​)i,i′∈s​, with Bs(Z)=∑i,i′∈sZii′∣di⟩⟨di′∣\mathsf B_s(Z)=\sum_{i,i'\in s}Z_{ii'}|d_i\rangle\langle d_{i'}|Bs​(Z)=∑i,i′∈s​Zii′​∣di​⟩⟨di′​∣,
(1−κ)2 Varg, e−u/Z(F∗(Bs(Z)))≤ac Vard,q(Bs(Z)).(1-\kappa)^2\,\mathrm{Var}_{g,\,e^{-u}/\mathcal Z}\bigl(\mathcal F^*(\mathsf B_s(Z))\bigr)\le\frac ac\,\mathrm{Var}_{d,q}\bigl(\mathsf B_s(Z)\bigr).(1−κ)2Varg,e−u/Z​(F∗(Bs​(Z)))≤ca​Vard,q​(Bs​(Z)).

Item 1 identifies the minimizer as a Gibbs state for the Hamiltonian ζaN^4+c(1−κ)aF∗(−log⁡ρ0)+βan^\frac\zeta a\hat N^4+\frac{c(1-\kappa)}{a}\mathcal F^*(-\log\rho_0)+\frac\beta a\hat naζ​N^4+ac(1−κ)​F∗(−logρ0​)+aβ​n^ (formally), and item 2 is the second-order (Hessian) optimality condition: the flow contracts the logarithmic-mean metric from ρ0\rho_0ρ0​ back to ρ\rhoρ on every finite block of the eigenbasis of ρ0\rho_0ρ0​.

Formalization Note The Gibbs representation is the structure GibbsData n (ζ/a) ρ; item 1 is an identity in ℝ≥0∞ written with sliceMoment, numberVectorMoment and energyInverse n =N^−1/2=\hat N^{-1/2}=N^−1/2. The block Bs(Z)\mathsf B_s(Z)Bs​(Z) is FiniteBlock.block for the basis ddd reindexed by s⊔scs\sqcup s^{\mathrm c}s⊔sc, and Var\mathrm{Var}Var is BKM.variance. The thermal state is fixed by the hypothesis τ = thermalState n r hr.

Preamble
import Mathlib
import Definitions.Def_EntropyPhotonNumber
import Definitions.Def_OAI_MathematicalPhysics_Fock_Beam
import Definitions.Def_OAI_Analysis_QuantumTrace_Basic
import Definitions.Def_OAI_MathematicalPhysics_Fock_Operators
import Definitions.Def_OAI_InformationTheory_PhotonNumber_Gibbs
import Definitions.Def_OAI_Analysis_TensorLp
import Definitions.Def_OAI_InformationTheory_PhotonNumber_BeamTrace
import Definitions.Def_OAI_InformationTheory_PhotonNumber_Functions
import Definitions.Def_OAI_Analysis_DiagonalForms
import Definitions.Def_OAI_InformationTheory_PhotonNumber_ThermalMetric
import Definitions.Def_OAI_InformationTheory_PhotonNumber_BeamForms
import Definitions.Def_OAI_InformationTheory_PhotonNumber_Generator
import Definitions.Def_OAI_InformationTheory_PhotonNumber_SpectralLog
import Definitions.Def_OAI_InformationTheory_PhotonNumber_Continuity
import Definitions.Def_OAI_InformationTheory_PhotonNumber_Entropy
import Definitions.Def_OAI_InformationTheory_PhotonNumber_GibbsVariational
import Definitions.Def_OAI_Analysis_Herglotz
import Definitions.Def_OAI_InformationTheory_PhotonNumber_Interpolation
import Definitions.Def_OAI_InformationTheory_PhotonNumber_SpectralForms
import Definitions.Def_OAI_InformationTheory_PhotonNumber_Contraction
import Definitions.Def_OAI_InformationTheory_PhotonNumber_FiniteVariation

open OAI OAI.EntropyPhotonNumber
open scoped ComplexConjugate
open scoped BigOperators ComplexConjugate
open scoped BigOperators ComplexConjugate ENNReal Topology
open MeasureTheory
open scoped BigOperators
open MvPolynomial
open scoped BigOperators ComplexConjugate Classical
open Submodule
open scoped ENNReal
open Module
open scoped ComplexConjugate BigOperators Matrix
open NormedSpace
open scoped Matrix.Norms.Operator
open scoped Topology
open MeasureTheory NormedSpace
open Filter Asymptotics Topology
open ContinuousLinearMap
open QuantumTrace DiagonalForms.System
variable {n : ℕ} {J : Type*} [DecidableEq J]
Formal statement
theorem OAI.EntropyPhotonNumber.minimum_all_blocks_hessian_data
    (F : KrausFlow n) (τ : State n) {r : ℝ} (hr : 0 < r) (hτ : τ = thermalState n r hr)
    (κ : ℝ) (hκ : κ ∈ Set.Icc 0 1) (hκ0 : 0 < κ)
    (c a β ζ : ℝ) (hc : 0 < c) (ha : 0 < a) (hβ : 0 ≤ β) (hζ : 0 < ζ)
    (ρ : State n) (hmρ : FiniteMoment 4 ρ)
    (d : HilbertBasis J ℂ (Fock n)) (q : J → ℝ)
    (hq : ∀ j, 0 < q j) (ht : HasSum q 1)
    (he0 : ∀ j, (mixture (1-κ) ⟨sub_nonneg.mpr hκ.2, sub_le_self _ hκ.1⟩ (F.map ρ) τ).op (d j) =
      (q j : ℂ) • d j)
    (hmin : ∀ σ, FiniteMoment 4 σ →
      c*entropy (mixture (1-κ) ⟨sub_nonneg.mpr hκ.2, sub_le_self _ hκ.1⟩ (F.map ρ) τ)-a*entropy ρ+
        β*energy ρ+ζ*numberMoment 4 ρ ≤
      c*entropy (mixture (1-κ) ⟨sub_nonneg.mpr hκ.2, sub_le_self _ hκ.1⟩ (F.map σ) τ)-a*entropy σ+
        β*energy σ+ζ*numberMoment 4 σ) :
    ∃ G : GibbsData n (ζ/a) ρ,
      (∀ x, ENNReal.ofReal (inner ℂ x (G.B x)).re =
        ENNReal.ofReal (c*(1-κ)/a)*
          sliceMoment d (fun j => ENNReal.ofReal (-Real.log (q j))) F.V (energyInverse n x)+
        ENNReal.ofReal (β/a)*numberVectorMoment (energyInverse n x)) ∧
      ∀ (s : Finset J) (Z : Matrix s s ℂ),
      (1-κ)^2*BKM.variance G.basis (gibbsWeight G.eigenval)
        (F.dual (FiniteBlock.block (BKM.reindexBasis d (Equiv.sumCompl (· ∈ s))) Z)) ≤
      (a/c)*BKM.variance d q (FiniteBlock.block (BKM.reindexBasis d (Equiv.sumCompl (· ∈ s))) Z) := by sorry
Source
https://github.com/openai/math/blob/1581d4cbdd69a0faf34b2089614fe2a814316a6a/lean/OAI/InformationTheory/PhotonNumber/FiniteVariation.lean#L2679-L2746

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 paper
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me