On separably injective Banach spaces (Q1935185): Difference between revisions

A wealth of new information about separable injectivity can be found here; this review will just mention the highlights. A Banach space \(E\), generally infinite dimensional, is called separably injective [respectively, universally separably injective] if, whenever \(Y\) is a subspace of a Banach space \(X\) and \(X\) is separable [respectively, \(Y\) is separable], then every operator \(Y\to E\) admits an extension \(X\to E\). Clearly, injectivity implies universal separable injectivity, which in turns implies separable injectivity; the converses are false. For example, the spaces \(c_0(\Gamma)\), for index sets of any size, are separably injective, and so are any twisted sums of them, since separable injectivity is a 3-space property. However, universally separably injective spaces are characterised by the property that every separable subspace is contained in a subspace isomorphic to \(\ell_\infty\); they must also be Grothendieck spaces. Similarly, separably injective spaces must be \(\mathcal{L}_\infty\) spaces, must contain \(c_0\), and must possess Pełczyński's property \((V)\). Examples of universally separably injective spaces which are not injective include \(\ell_\infty/c_0\) and the space of countably supported functions \(\ell_\infty^c(\Gamma)\). In fact, any quotient of a [universally] separably injective space by a separably injective subspace is [universally] separably injective. Any ultraproduct of \(\mathcal{L}_\infty\) spaces is also universally separably injective.