Local perturbations of conditional expectations (Q1140372)

The paper deals with Dobrushin's theory in the general framework of continuous systems and contains an application to 2-dimensional Euclidean field theory (the proof of the global Markov property under very general conditions on the ``interaction''). The first part deals with the general problem of Dobrushin's theory: find a measure with a preassigned projective family of conditional expectations. The language used is algebraic, but it must be stressed (as done at the end of \S1 in Part II) that this should not be thought as a quantum formulation of Dobrushin's theory because the conditional expectations are norm one projections, and in the quantum case this is not true. In Part II the concept of conditional martingale is introduced and it is shown that the study of Dobrushin's theory is equivalent to the study of conditional martingales. A structure theorem for Markovian conditional martingales is proved which generalizes a result obtained by Averentzev and by Spitzer in the discrete case. In Part III the role of the algebra at infinity in the proof of the global Markov property is isolated and it is shown that the root of the difficulties in proving the global Markov property from the local one lies in the (arbitrary) definition of \(\sigma\)-algebra associated to an unbounded set \(F\). It is shown that, if instead of defining it as the intersection of all \(\sigma\)-algebras associated to open sets containing \(F\), one considers only the open sets with bounded boundary, then the global Markov property is easily recovered. The problem whether the splitting \(\sigma\)-algebra thus obtained is minimal or not is left open.