OctoInformBeta

OpenAI preprint, September 2026 · §3. Modular localization and averaging

Lemma 3.3: A vector fixed by all UtU_t in their closed convex hull can only be φ(a)ξ\varphi(a)\xi; one average serves aξa\xi and bξb\xi at once

With a trivial centralizer, convex averages of UtaξU_t a\xi approach φ(a)ξ\varphi(a)\xi..

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

Let MM be a von Neumann algebra on a complex Hilbert space HH and ξ\xi a cyclic and separating unit vector for MM, with modular operator Δ\Delta and vector state φ\varphi. For a∈Ma \in M the unitaries Δit\Delta^{it} move aξa\xi along its orbit {Δitaξ:t∈ℝ}\{\Delta^{it} a\xi : t \in \mathbb{R}\}. By von Neumann’s mean ergodic theorem, the averages of the orbit over [−k,k][-k, k] converge, as k→∞k \to \infty, to a vector fixed by every Δit\Delta^{it}. The question is which vector that is when φ\varphi has trivial centralizer, that is, when the only elements of MM fixed by every modular automorphism are the scalar multiples of the identity.

The library answers the question in convex form, with finite convex combinations of the orbit in place of averages over intervals. The closed convex hull co¯{Utx}\overline{\mathrm{co}}\{U_t x\} of the orbit of a vector xx under a family (Ut)t∈ℝ(U_t)_{t \in \mathbb{R}} of operators on a real or complex Hilbert space is the norm closure of the set of finite convex combinations of the vectors UtxU_t x: co¯{Utx}={∑i=1npiUtix:n∈ℕ,ti∈ℝ,pi≥0,∑i=1npi=1}¯.\overline{\mathrm{co}}\{U_t x\} = \overline{\Bigl\{\sum_{i=1}^{n} p_i\, U_{t_i} x : n \in \mathbb{N},\ t_i \in \mathbb{R},\ p_i \ge 0,\ \sum_{i=1}^{n} p_i = 1\Bigr\}}.(1)

Standard Modular Data

A real spectral calculus DD on HH is a nondegenerate **-homomorphism f↦f(D)f \mapsto f(D) from the algebra Cb(ℝ)C_b(\mathbb{R}) of bounded continuous complex functions on ℝ\mathbb{R} into B(H)B(H); nondegenerate means that the vectors f(D)kf(D)k, with ff of compact support and k∈Hk \in H, are dense in HH.✦

The unitary group of a real spectral calculus is the family (Ut)t∈ℝ(U_t)_{t \in \mathbb{R}} of images of the characters et(u)=eitue_t(u) = e^{itu}: Ut=et(D)=eitD.U_t = e_t(D) = e^{itD}.(2)✦

Let DD be a real spectral calculus on HH. The graph Γ\Gamma of eD/2e^{D/2} is the set of pairs (h,k)∈H×H(h, k) \in H \times H with fn(D)h→kf_n(D)h \to k as n→∞n \to \infty, where fn∈Cb(ℝ)f_n \in C_b(\mathbb{R}) is the half exponential cutoff fn(u)=emin⁡(n,u)/2.f_n(u) = e^{\min(n, u)/2}.(3)✦

Standard modular data (M,ξ,D,J)(M, \xi, D, J) on HH consist of a von Neumann algebra MM on HH, a cyclic and separating unit vector ξ\xi for MM, a real spectral calculus DD on HH and an antiunitary involution JJ of HH, such that the polar decomposition S=JΔ1/2S = J\Delta^{1/2} of the Tomita operator holds in the form {(aξ,a*ξ):a∈M}¯={(h,Jk):(h,k)∈Γ},\overline{\{(a\xi, a^*\xi) : a \in M\}} = \{(h, Jk) : (h, k) \in \Gamma\},(4)✦ with DD standing for log⁡Δ\log \Delta and UtU_t for Δit\Delta^{it}.

The modular automorphism group of standard modular data is the family of maps σt\sigma_t on B(H)B(H) given by σt(a)=UtaU−t.\sigma_t(a) = U_t\, a\, U_{-t}.(5)✦ The vector state of standard modular data is the functional φ(a)=⟨ξ,aξ⟩\varphi(a) = \langle \xi, a\xi \rangle on B(H)B(H).✦ The centralizer of φ\varphi is the fixed-point algebra Mφ={a∈M:σt(a)=a for all t∈ℝ}M_\varphi = \{a \in M : \sigma_t(a) = a \text{ for all } t \in \mathbb{R}\}. The state φ\varphi has trivial centralizer when Mφ=ℂ1M_\varphi = \mathbb{C}1, that is, when every element of MM fixed by every σt\sigma_t is a scalar multiple of the identity.✦

The Claim

Theorem 1. (i) Let EE be a real Hilbert space and (Ut)t∈ℝ(U_t)_{t \in \mathbb{R}} a family of real-linear isometries of EE with Ut+r=UtUrU_{t+r} = U_t U_r for all t,r∈ℝt, r \in \mathbb{R}. Then for every x∈Ex \in E there is a vector vv with v∈co¯{Utx}andUtv=v for all t∈ℝ.v \in \overline{\mathrm{co}}\{U_t x\} \quad\text{and}\quad U_t v = v \ \text{ for all } t \in \mathbb{R}.(6)✦ Let (M,ξ,D,J)(M, \xi, D, J) be standard modular data on HH whose vector state φ\varphi has trivial centralizer, Mφ=ℂ1M_\varphi = \mathbb{C}1, and let a,b∈Ma, b \in M.

(ii) Every vector v∈co¯{Utaξ}v \in \overline{\mathrm{co}}\{U_t a\xi\} with Utv=vU_t v = v for all t∈ℝt \in \mathbb{R} is v=φ(a)ξ.v = \varphi(a)\xi.(7)✦

(iii) Moreover, φ(a)ξ∈co¯{Utaξ}.\varphi(a)\xi \in \overline{\mathrm{co}}\{U_t a\xi\}.(8)✦

(iv) For every ε>0\varepsilon > 0 there are n∈ℕn \in \mathbb{N}, weights p1,…,pn≥0p_1, \dots, p_n \ge 0 with ∑ipi=1\sum_i p_i = 1 and times t1,…,tn∈ℝt_1, \dots, t_n \in \mathbb{R} such that ∥∑i=1npiUtiaξ−φ(a)ξ∥<εand∥∑i=1npiUtibξ−φ(b)ξ∥<ε.\Bigl\| \sum_{i=1}^{n} p_i\, U_{t_i} a\xi - \varphi(a)\xi \Bigr\| < \varepsilon \quad\text{and}\quad \Bigl\| \sum_{i=1}^{n} p_i\, U_{t_i} b\xi - \varphi(b)\xi \Bigr\| < \varepsilon.(9)✦

In the paper, the result reads: if Mφ=ℂ1M_\varphi = \mathbb{C}1 and μk\mu_k is the uniform probability measure on [−k,k][-k, k], then for every y∈My \in M and every ψ\psi in the predual M*M_* of MM, ∫σt(y)ξdμk(t)→φ(y)ξ\int \sigma_t(y)\xi \, d\mu_k(t) \to \varphi(y)\xi in HH and ∫ψ∘σ−tdμk(t)→ψ(1)φ\int \psi \circ \sigma_{-t} \, d\mu_k(t) \to \psi(1)\varphi in M*M_* as k→∞k \to \infty.

What Is Checked

The claim is stated over standard modular data on a complex Hilbert space, with the unitary group Ut=eitDU_t = e^{itD} of its real spectral calculus, the modular automorphism group σt\sigma_t, the vector state φ\varphi and the hypothesis Mφ=ℂ1M_\varphi = \mathbb{C}1, and, in its first part, over a group of real-linear isometries of a real Hilbert space.