New versions of the Radon-Nikodým theorem (Q2490660)

Let \(\alpha,\beta:{\mathfrak A}\to [0,\infty]\) be measures on the \(\sigma\)-algebra \(\mathfrak A\) over the set \(X\). If \(\alpha\) is finite then a standard version of the Radon-Nikodým theorem says that \(\beta\ll \alpha\) if and only if \(\beta\) has a density function w.r.t. \(\alpha\), and then this density function is unique up to \(\alpha\)-null sets. If, however, \(\beta\) is finite and \(\alpha\) arbitrary, then a density function for \(\beta\) w.r.t. \(\alpha\) need not exist if \(\beta\ll \alpha\). To give an idea of the contents of the present paper we quote Theorem 1.1: If \(\beta\) is finite, then \(\beta\ll\alpha\) if and only if there exists a measurable function \(F:X\to[0,\infty]\) such that \[ \beta(A)=\int_A F\,d\alpha\text{ for all }A\in{\mathfrak A}\text{ with }\alpha(A)<\infty \] and \[ \beta(A)\geq \int_A F\,d\alpha\text{ for all }A\in{\mathfrak A}. \] The density function \(F\) then is unique up to \(\alpha\)-null sets. Similar versions of this theorem hold for Fremlin's notion of ``\(\beta\) truly continuous \(\alpha\)'' and in the case where the lattice minimum \(\alpha\wedge \beta\) is finite or where the measure space is of decomposition type.