Remarks on Hilbert-space-valued multimartingale measures (Q1324886)

The subject of the paper is ``square-integrable, Hilbert space-valued multimartingales measures''. These objects are defined as follows. One fixes a ``multidimensional interval'' \(D = (0,z^*]\). One then considers the algebra \({\mathcal A}\) generated by the semiring of ``boxes'' within \(D\). One can then define processes indexed by \({\mathcal A}\), and taking values in a separable Hilbert space, with the following properties: -- the square of the norm of the process has finite expectation, -- the process is additive with respect to its index, -- when the index is a ``box'', the process is adapted to a pre-existing filtration, and has a martingale property. These processes can in turn be used to define measures on \({\mathcal A}\), and the processes for which these measures are \(\sigma\)-additive are distinguished: they serve later in the paper to define the notion of stable subspaces in the sense of the usual martingale theory. Furthermore, these ``multimartingales'' lead to a notion of stochastic integral for properly defined predictable random functions with values in spaces of Hilbert space operators. The thrust of the paper is the use of this integral to produce orthogonal decompositions in terms of stochastic integrals and stable subspaces, as in the usual real case. Much of the paper is devoted to technical details too complicated to be stated, but necessary to the proper definition of the objects being considered.