Measures admitting extremal extensions (Q1088013)

For \({\mathfrak A}\subset {\mathfrak B}\) two \(\sigma\)-algebras of subsets of some set \(\Omega\) and \(\mu\) a finite positive measure on \({\mathfrak A}\) we investigate the convex set \(M({\mathfrak B};\mu)\) of measures \(\nu\) on \({\mathfrak B}\) extending \(\mu\). If \(X\subset \Omega\) is an \({\mathfrak A}\)-atom for \({\mathfrak B}\) of full \(\mu\)-measure (i.e \(X\cap {\mathfrak A}=X\cap {\mathfrak B}\) and \(\mu _ *(\complement X)=0),\) it is shown that \(\nu: \nu (B)=\mu ^ *(X\cap B)\) defines an extremal point of \(M({\mathfrak B};\mu.)\) (This is an obvious generalization of a Dirac measure \(\delta _ a\), viz. \(\delta _ a=\mu ^ *(\{a\}\cap \cdot)\) for \(\mu\) the trivial probability on the trivial \(\sigma\)-algebra \({\mathfrak A}=\{\emptyset,\Omega \}.)\) If \({\mathfrak B}\) is a countable extension of \({\mathfrak A}\), each \(\nu \in ex M({\mathfrak B};\mu)\) arises in this way, but in general this is not true. On the other hand it is shown that, in a certain ''dilation'' of the measure space, as constructed in the author's paper in Arch. Math. 43, 434-439 (1984; Zbl 0538.28002), each extremal extension of \(\mu\) is of the above form. By this description it is proved that the set of measures \(\mu\) on \({\mathfrak A}\) admitting extremal extensions generates a band in the vector lattice \(M({\mathfrak A})-M({\mathfrak A})\) of signed measures. \({\mathfrak A}\)-atoms are also used to construct extremal extensions of a given \(\mu\) on \({\mathfrak A}\) to \({\mathfrak B}\) via extremal extensions to intermediate \(\sigma\)-algebras \({\mathfrak B}':\) \({\mathfrak A}\subset {\mathfrak B}'\subset {\mathfrak B}\).