On \(p\)-Dunford integrable functions with values in Banach spaces (Q1748339)

Let \((\Omega,\Sigma,\mu)\) be a complete probability space, \(X\) be a Banach space and \(1\leq p<\infty\). A function \(f:\Omega\to X\) is \(p\)-Dunford integrable if \(x^\ast\circ f\in L^p(\mu)\) for all \(x^\ast\in X^\ast\) and \(p\)-Pettis integrable if \(f\), in addition, is Pettis integrable. Three main topics are considered after an introduction (Section 1) and some preliminaries (Section 2): Compactness of the Dunford operator (Section 3), integrability of \(u\circ f:\Omega\to Y\) where \(u:X\to Y\) is \(p\)-summing (Section 4), and obtaining \(p\)-Dunford integrability from \(x^\ast\circ f\in L^p(\mu)\) for all \(x^\ast\in\Gamma\subset X^\ast\) (Section 5). By the definition of \(p\)-Dunford integrability, there is an operator \(S_f^p:X^\ast\to L^p(\mu)\), given by \(S_f^p(x^\ast)=x^\ast\circ f\). The Dunford operator \(T_f^p\) is the adjoint of \(S_f^p\). When \(f\) is \(p\)-Pettis integrable, the compactness of \(T_f^p\) is equivalent to \(f\) being approximable by simple functions in a natural norm. When \(f\) is just \(p\)-Dunford integrable, compactness of \(T_f^p\) is equivalent to uniform integrability of \(\{|x^\ast\circ f|:x^\ast\in X^\ast\}\) in \(L^1(\mu)\) and \(T_f^1\) being compact. The latter follows from the first if \(f\) is strongly measurable, and also when \(\mu\) is perfect or if \(X\) is WCG. When \(u:X\to Y\) is \(p\)-summing, the basic feeling is that \(u\circ f:\Omega\to Y\) has better integration properties than \(f\) itself. Indeed, essentially due to Joe Diestel, if \(f\) is strongly measurable and \(p\)-Dunford integrable, then the composition \(u\circ f\) is already \(p\)-Bochner integrable. Dropping strong measurability, the authors prove that \(u\circ f\) is still at least scalarly equivalent to a \(p\)-Bochner integrable function and is \(p\)-Bochner integrable as soon as the compact Hausdorff space \((B_{X^\ast},w^\ast)\) has some extra properties (e.g., is Eberlein). The first part of this result, for \(p=1\), goes back to an unpublished manuscript of D. R. Lewis from 1970. Obtaining in general that \(x^\ast\circ f\in L^p(\mu)\) for all \(x^\ast\in X^\ast\) from just \(x^\ast\circ f\in L^p(\mu)\) for \(x^\ast\in \Gamma\subset X^\ast\) must force \(\Gamma\) to be somehow big in \(X^\ast\). The measure of bigness the authors use is the massivity property that exactly forces \(w^\ast\)-continuous linear bounded operators onto \(\Gamma\) to be onto \(X^\ast\). This massivity (\(w^\ast\)-thickness) is not enough in general, but when \(X\) has property \(\mathcal{E}\) of Efremov, in particular, when \(X\) is WCG, it is enough, and \(f\) is even \(p\)-Pettis integrable. The most interesting \(w^\ast\)-thick set is probably the set of \(w^\ast\)-exposed points in \(B_{X^\ast}\) when \(X\) does not contain \(c_0\) (a result of \textit{V. P. Fonf} [Bull. Lond. Math. Soc. 28, No. 1, 51--58 (1996; Zbl 0879.46001)]). The ideas and results of the paper are all the way well motivated and put into a historical perspective, a very important point that, in the reviewer's opinion, unfortunately much too often is neglected by many authors.