Borel and continuous systems of measures (Q470391)

Systems of measures are taken to be families of measures \(\lambda^{\bullet}=\{ \lambda^y\}_{y \in Y}\) associated with mappings \(\pi:X \to Y\) such that each \(\lambda^{y}\) has support \(\pi^{-1}(y)\). The authors consider the families of measures as fibre systems on base \(Y\). Categories specified are, for instance, topological spaces with a continuous \(\pi\) and Borel spaces with a Boolean algebra of subsets and \(\pi\) Borel. These correspond respectively to topological and measured groupoids (i.e., like groups but without an everywhere defined binary operation); G. Mackey's virtual groups are equivalence classes of `ergodic' measured groupoids. The formulation facilitates to express the relationship between measured groupoids and von Neumann algebras. The ideas are as described in the survey by \textit{C. Anantharaman-Delaroche} and \textit{J. Renault} [Amenable groupoids. Monographies de l'Enseignement Mathématique. 36. Genève: L'Enseignement Mathématique; Université de Genève (2000; Zbl 0960.43003)]. The survey deals with (topological) groupoids having systems of left Haar measures (defined for measured groupoids by \textit{P. Hahn} [Trans. Am. Math. Soc. 242, 1--33 (1978; Zbl 0343.43003)]). In order to multiply two elements of a groupoid one uses the fibre structure of the groupoid over its base space consisting of its (left and right) unit elements. One may multiply the two elements if the source of right hand element is the same as the range (or target) of the left hand element. The definition in the article under review of a Haar system for a topological groupoid is that \(\lambda^{\bullet}\) is continuous, left-invariant and positive on open sets. This is equivalent to the definition given in the survey. A system of measures \(\lambda^{\bullet}\) is called \textit{locally finite} if there is a neighbourhood \(U_{x} \subset X\) such that \(\lambda^{y}(U_{x}) < \infty\) for all \(y \in Y\). Continuous systems of measures are always locally finite and if \(\mu,\nu\) are \(\sigma\)-finite measures then equality of the measures on all open sets implies equality for all Borel sets. In the article the authors describe the basic constructions for systems of measures; results are usually restricted to locally finite systems: For a Borel system \(X \to Y \to Z\), \({\beta \circ \alpha}^{\bullet}\) is the family \({\alpha}^{y}(E)d\beta^{z}:y \in Y\) for all Borel \(E \subset X\). A \textit{lifting}, as referred to in Appendix A.2 of the survey, is constructed from a commutative square pullback diagram and is usually denoted \((X \times_Z Y) \to X\) to distinguish it from the product \(X \times Y\) which does not involve any \(Z\). It is called a pullback of \(Y \to Z\) by \(f:X \to Z\) or as a pullback of \(Y\) by \(f\). If it exists it would be universal, i.e., unique up to isomorphism. The authors denote the lifting of a locally finite system \(\alpha ^{\bullet}\) of measures on \(Y\), from a pullback diagram \(X \rightarrow Z \leftarrow Y\) to a system of measures \(X * Y\) by \((q^{*}\alpha)^{y} = \alpha^{q(y)} \times \delta_{y}\) where the \(q:Y \to Z\). For \textit{weak pullbacks} (Definition 2.8) for Haar (topological) systems the corresponding square does not commute. The \(\pi\)-systems for groupoids of the survey are weak pullbacks, each endowed with a Haar system; a quasi-invariant measure on its space of units is continuous, left invariant and positive on open sets. The authors show this definition of weak pullback to be equivalent to that defined in the survey. \textit{Fibered products} (\S 1.3 in the survey) of locally finite systems \(\gamma^{\bullet}_{X}\) and \(\gamma ^{\bullet}_{Y}\) are from pairs of diagrams \(X_{i} \rightarrow Z \leftarrow Y_{i}\) pulling back to \(X_{1} * Y_{1}\) and \(X_{2} * Y_{2}\) by using a commutative cube instead of a commutative square. The Borel mapping \(f * g:X_{1} * Y_{1} \to X_2 * Y_2\) has \(f*g (x_{1},y_{1}) =f(x_{1})g(y_{1})\) restricting \(f \times g\) to \(X_{1} * Y_{1}\). The fibres of \(f*g\) are products of the form \((f*g)^{-1}(x_2,y_2) = f^{-1} (x_2) g^{-1}(y_2)\). The origin of modern disintegration comes from J. von Neumann in 1931--32 decomposing an operator on a separable Hilbert space into a direct integral of operator algebras. The progress of decomposition theory for measures gained mainly due to studies by \textit{P. R. Halmos} [Proc. Natl. Acad. Sci. USA 35, 38--42 (1949; Zbl 0031.40701)]. The basic formulation of a decomposition of \(\sigma\)-finite measure spaces \((X,\mu)\) and \((Y,\nu)\) is as follows: given a positive Borel mapping \(\phi: X \to Y\) such that \(\phi_{*}\) and \(\mu\) have the same null sets, then a system \(\gamma^{\bullet}\) such that \(\nu(E) = \int_{Y}{\gamma^{y}}(E)d\nu(y)\) for every Borel \(E \subset X\) leads to a \textit{disintegration} of \(\mu\) over fibres of \(\phi:X\to Y\). The authors use a considerably restricted version due to \textit{R. C. Fabec} [Fundamentals of infinite dimensional representation theory. Monographs and Surveys in Pure and Applied Mathematics. 114. Boca Raton, FL: Chapman \& Hall/CRC (2000; Zbl 0964.22001)] assuming that \(X\) is a standard Borel space and that \(Y\) is countably separated, and so assumes that \(X\) is a Cantor set \((0,1)^{\mathbf{N}}\); its Boolean algebra is made up from the clopren sets. His proof also requires the mapping \(X \to Y\) to be continuous so the authors have assumed for their result further that \(X\) and \(Y\) are locally compact. For a positive \(\phi: X \to Y\) one has \(\gamma^{\bullet}\) as a system of Radon-Nikodym type derivatives. The disintegration is unique in the following sense that for any disintegrations \(\gamma_{1}, \gamma_{2}\) the \(\gamma^{y}_{i}\) are equal on the fibres \(y \to \mu^{y}(A)\) a.e. in \(Y\), where the \(A\) are finite sets in the corresponding Boolean algebra. The disintegration gives rise to a locally finite Borel system if for every compact \( K \subset X\) there is a boundedness condition that \(K \cap \phi^{-1}(E) \leq C_{K}\nu{E}\) for some \(C_{k}\) and all Borel sets \(E \subset Y\).