Purity of the Embeddings of Operator Systems into their C$^*$- and Injective Envelopes

TL;DR

This paper investigates when the canonical embeddings of operator systems into their injective and C*-envelopes are pure, revealing that purity relates to the primeness of the C*-algebra and exploring specific cases involving group-generated systems.

Contribution

It establishes a characterization of purity for embeddings into injective envelopes based on the primeness of the C*-envelope and extends the understanding of pure maps to embeddings into injective von Neumann algebras.

Findings

Embedding into injective envelopes is pure iff the C*-envelope is prime.

Identity maps on AW*-factors are pure.

Pure completely positive maps extend to larger operator systems.

Abstract

We study the issue of issue of purity (as a completely positive linear map) for identity maps on operators systems and for their completely isometric embeddings into their C$^*$-envelopes and injective envelopes. Our most general result states that the canonical embedding of an operator system $\mathcal R$ into its injective envelope $\mathcal I(\mathcal R)$ is pure if and only if the C$^*$-envelope of $\mathcal R$ is a prime C$^*$-algebra. To prove this, we also show that the identity map on any AW$^*$-factor is a pure completely positive linear map. For embeddings of operator systems into their C$^*$-envelopes, the issue of purity is seemingly harder to describe in full generality, and so we focus here on operator systems arising from the generators of discrete groups. The question of purity of the identity is quite subtle for operator system that are not C$^*$-algebras, and we have results only for two universal operator systems arising from discrete groups. Lastly, a previously unrecorded feature of pure completely positive linear maps is presented: every pure completely positive linear map on an operator system $\mathcal R$ into an injective von Neumann algebra $\mathcal M$ has a pure completely positive extension to any operator system $\mathcal T$ that contains $\mathcal R$ as an operator subsystem, thereby generalising a result of Arveson for the injective type I factor $\mathcal B(\mathcal H)$.