A variational Henstock integral characterization of the Radon-Nikodým property (Q2267689)

It is shown that the Radon-Nikodým property (RNP) works also for the finitely additive case, using Henstock integrals: A real Banach space \(X\) has the RNP if and only if one of the following conditions holds: {\parindent7mm \begin{itemize}\item[(a)] If \(\Phi: A\to X\) is additive and BVG, then \(\Phi\) is differentiable \(\lambda\)-a.e.\ in \(I_0:= [0,1]\), i.e., \(\lim_{h\to 0} (\Phi([0, t+ h])- \Phi([0, t])))/h\) exists \(\in X\), for almost all \(t\in I_0\). \item[(b)] If \(\Phi: A\to X\) is additive with \(V_\Phi\ll\) Lebesgue measure \(\lambda\), then \(\Phi'\) exists \(\lambda\)-a.e.\ in \(I_0\). \item[(c)] If \(\Phi: A\to X\) is additive with \(V_\Phi\ll\lambda\), then there exists a \(f\in H(I_0,X)\) with \(\Phi(I)= H-\int_I f\,dt\) for \(I\in A\). \end{itemize}} Then one can take \(f=\Phi'\) in (c), even \(\Phi'\in VH(I_0,X)\). Here, \(A\) denotes the set of all closed intervalls \(\subset I_0\) (mod endpoints a semiring). \(\Phi\) BVG means there exist finitely many sets \(E_1,\dots, E_q\subset I_0\) with \(I_0= \bigcup^q_1 E_k\) so that \(\sup\{\sum^n_{j=1} \sup\{\|\Phi(u)- \Phi(z)\|: u,z\in I_j\}: n\in\mathbb{N}\), \(I_1,\dots, I_n\) nonoverlapping \(\in A\), endpoints for all \(I_j\) in \(E_k\}<\infty\), \(1\leq k\leq q\). For any \(E\subset I_0\), \(V_\Phi(E):=\inf\{\sup\{\sum^p_{j=1} \|\Phi(I_j)\|:P= (I_1,t_1),\dots, (I_p, t_p)\) \(\delta\)-fine partition with all \(t_j\in E\}\): all \(\delta: E\to (0,\infty)\}\), where \(P\) partition means \(I_1,\dots, I_p\) nonoverlapping \(\in A\) and all \(t_j\in I_j\); \(P\) \(\delta\)-fine means \(I_j\subset(t_j- \delta(t_j), t_j+\delta(t_j))\), \(1\leq j\leq p\). \(V_\Phi\ll\lambda\) means \(V_\Phi(E)= 0\) if \(E\) \(\lambda\)-null set \(\subset I_0\). \(f\in VH(I_0, X)\) means there exists an additive \(\Phi: A\to X\) so that for each \(\varepsilon> 0\) there exists a \(\delta: I_0(0,\infty)\) with \[ \sum^p_1\| f(t_j)\lambda(I_j)- \Phi(I_j)\|< \varepsilon\tag{1} \] for each \(\delta\)-fine partition \((I_1,t_1),\dots, (I_p, t_p)\) with \(I_0= \bigcup^p_1 I_j\); replacing (1) by \[ \Biggl\|\sum^p_1 f(t_j)\lambda(I_j)- \Phi(I_0)\Biggr\|< \varepsilon \] defines \(f\in H(I_0, X)\), \(f\) Henstock integrable; \(VH-\int_I fd\lambda:= \Phi(I)\), \(I\in A\), resp. \(H- \int_{I_0} fd\lambda:= \Phi(I_0)\). With the above equivalences, a gap in the proof of Theorem 2 of [\textit{V.\,A.\thinspace Skvortsov} and \textit{A.\,P.\thinspace Solodov}, Mosc.\ Univ.\ Math.\ Bull.\ 57, No.\,3, 36--39 (2002); translation from Vestn.\ Mosk.\ Univ., Ser.\,I 2002, No.\,3, 57--60 (2002; Zbl 1075.28500)] is closed. Corollary. If \(f: I_0\to X\) is strongly \(\lambda\)-measurable and Pettis integrable, \(\Phi(I):= \text{Pettis-}\int_I fd\lambda\), then \(f\in VH(I_0,X)\) if and only if \(V_\Phi\ll\lambda\).