OctoInformBeta

Relative bicentralizers and modular spectral recovery

Let MM be a von Neumann algebra on a complex Hilbert space HH with a cyclic and separating unit vector ξ\xi, and let ϕ\phi be the vector state of ξ\xi. Let Δ\Delta be the modular operator of ξ\xi. Put D=log⁡ΔD=\log\Delta and Ut=eitDU_t=e^{itD}. The modular automorphism group of ϕ\phi is σt(a)=UtaU−t\sigma_t(a)=U_taU_{-t}. The centralizer MϕM_\phi of ϕ\phi is the algebra of all a∈Ma\in M with ϕ(ab)=ϕ(ba)\phi(ab)=\phi(ba) for every b∈Mb\in M. The state ϕ\phi is ergodic when Mϕ=ℂ1M_\phi=\mathbb C1. For a Borel set I⊆ℝI\subseteq\mathbb R, a vector h∈Hh\in H has DD-spectral support in II when it lies in the range of the spectral projection ED(I)E_D(I) of DD. Fix a free ultrafilter ω\omega on ℕ\mathbb N. The mean of a bounded continuous function ff on ℝ\mathbb R is mtf(t)=lim⁡k→ω12k∫−kkf(t)dtm_tf(t)=\lim_{k\to\omega}\frac{1}{2k}\int_{-k}^{k}f(t)\,dt. Spectral localization in HH does not by itself produce uniformly bounded elements of MM. The question is whether, for T∈B(H)T\in B(H) and s∈ℝs\in\mathbb R, a positive value of limsup⁡nmt∥TUthn∥2\limsup_n m_t\|TU_th_n\|^2 along unit vectors hnh_n with DD-spectral support in intervals shrinking to ss can be recovered as a positive lower bound for ∥Tvjξ∥\|Tv_j\xi\| along uniformly bounded vj∈Mv_j\in M whose vectors vjξv_j\xi have DD-spectral support in intervals shrinking to ss.

For a faithful normal state ϕ\phi on MM and w∈Mw\in M, let wϕw\phi and ϕw\phi w be the normal functionals y↦ϕ(yw)y\mapsto\phi(yw) and y↦ϕ(wy)y\mapsto\phi(wy). The bicentralizer BC(M,ϕ)\mathrm{BC}(M,\phi) is the set of all a∈Ma\in M such that [a,wn]→0[a,w_n]\to0 strongly for every uniformly bounded sequence (wn)(w_n) in MM with ∥wnϕ−ϕwn∥M*→0\|w_n\phi-\phi w_n\|_{M_*}\to0. Connes’ bicentralizer conjecture asserts that BC(M,ϕ)=ℂ1\mathrm{BC}(M,\phi)=\mathbb C1 for every type III1\mathrm{III}_1 factor MM with separable predual. Haagerup proved the conjecture in the amenable case. Houdayer and Marrakchi have announced that BC(M,ϕ)=ℂ1\mathrm{BC}(M,\phi)=\mathbb C1 for every type III1\mathrm{III}_1 factor MM admitting a faithful normal state ϕ\phi, with no separability condition. Marrakchi and Vaes proved that the ergodic faithful normal states form a dense GδG_\delta in the normal state space of every type III1\mathrm{III}_1 factor with separable predual.

The preprint proves bounded recovery, its Theorem 4.1: if Mϕ=ℂ1M_\phi=\mathbb C1, T∈B(H)T\in B(H), s∈ℝs\in\mathbb R, δn>0\delta_n>0 tends to 00, each hnh_n is a unit vector with DD-spectral support in [s−δn,s+δn][s-\delta_n,s+\delta_n], and limsup⁡nmt∥TUthn∥2>0\limsup_n m_t\|TU_th_n\|^2>0, then there are a subsequence (nj)(n_j), elements vj∈Mv_j\in M and constants C<∞C<\infty and η>0\eta>0 such that, for every jj, ∥vj∥≤C\|v_j\|\le C, ∥Tvjξ∥≥η\|Tv_j\xi\|\ge\eta and vjξv_j\xi has DD-spectral support in [s−4δnj,s+4δnj][s-4\delta_{n_j},s+4\delta_{n_j}]. From it the preprint derives a proof that BC(M,ϕ)=ℂ1\mathrm{BC}(M,\phi)=\mathbb C1 for every faithful normal state ϕ\phi on a type III1\mathrm{III}_1 factor MM with separable predual, a derivation that the Lean library does not contain.

The Lean library ModularRecovery proves bounded recovery, for every free ultrafilter ω\omega. It constructs the modular operator and the modular conjugation of ξ\xi. It proves that σt\sigma_t maps MM into MM for every t∈ℝt\in\mathbb R. The preprint takes this fact from the standard modular theory of Takesaki. The preprint defines the centralizer as the fixed-point algebra of σ\sigma. The Lean library proves that this algebra is MϕM_\phi as defined above. The results on these pages are the Lean library’s, each stated as the Lean library states it and linked to the Lean declaration behind it. Lean has checked every one of them, so the Lean library is the reference, and where the preprint states a result differently, the page gives the Lean library’s statement with the preprint’s beside it. Of the preprint’s Lemma 3.2, on smooth spectral localization, the Lean library proves only the part concerning ∫g(t)σt(y)dt\int g(t)\sigma_t(y)\,dt, for any integrable function gg on ℝ\mathbb R and any y∈My\in M. Of the preprint’s Lemma 3.3, on ergodic limits, the Lean library proves only a convex form, in which, when Mϕ=ℂ1M_\phi=\mathbb C1, finite convex combinations of the UtU_t bring aξa\xi arbitrarily close to ϕ(a)ξ\phi(a)\xi for each a∈Ma\in M. The preprint is a guide only: it gives the results their order and their names.