Embezzling entanglement from a de Sitter observer: sharp entropy-deficit bounds on finite von Neumann algebras
Dylan Z. Liu Liu
Abstract
The type III₁ algebras of quantum field theory are universal embezzlers of entanglement, while semifinite algebras admit no embezzling state. The algebra of a de Sitter static patch with an observer is of type II₁, so gravity excludes exact embezzlement there; we determine how small the error can be. For a state ω on a finite factor and a target pure state ψ with entanglement entropy S(ψ), the smallest one-sided embezzlement error obeys δ ≥ S(ψ)/[D+ln 2D+O(1)], where D is the relative entropy of ω to the tracial state. If the density of ω is bounded by e^(Λ), then δ ≥ 2S(ψ)/[Λ+ln Λ+O(1)]. Both bounds are attained to this order, the first by an observer with uniformly distributed energy and the second by the van Dam–Hayden states. In the observer algebra of Chandrasekaran, Longo, Penington and Witten, D is the deficit of generalized entropy below that of empty de Sitter space. An observer whose energy distribution is not sharper than the de Sitter temperature T_dS therefore needs mean energy ⟨E⟩ ≳ T_dS S(ψ)/δ. If the generalized entropy is nonnegative, every state has δ ≥ S(ψ)/S_dS up to logarithmic corrections, an error of first order in Newton's constant.