scientific article; zbMATH DE number 1029271
Let \(X\) be an \(L^2\)-bounded martingale over a given (commutative) filtration, \[ {\mathcal W}=(\Omega,W,(W_t)_{t\in R_+},P),\text{ where }R_+= [0,+\infty[\text{ and } W=\sigma\{W_t\mid t\in R_+\},\tag{1} \] then the Itô-isometry maps each \(L^2\)-bounded integrand \(Y\in L^2_{\text{loc}}(\langle X\rangle)\) into a non-definite stochastic integral \(Y*X=\{Y* X(t,.)\mid t\in R_+\}\). It is known that \(Y*X\) turns out to be an \(L^2\)-bounded martingale over \(\mathcal W\), hence it is complete, by the following identities, (2) \(Y*X(t)= \text{Pr}_t\{Y*X(\infty,.)\}\) for each \(t\in R_+\), where Pr\(_t\) is the orthogonal projection from \(L^2(W)\) onto \(L^2(W_t)\), and \(*X(\infty,.)= \lim_{t\uparrow+\infty} Y*X(t)\) a.s. \(P\). The authors investigate the same problem for their (stochastic) belated integrals over a non-commutative filtration, \[ \Sigma= (H,A,(A_t)_{t\in R_+},\varphi),\tag{3} \] and raise the question whether there is a non-trivial set of integrands commun to each \(L^2\)-bounded martingale. The notion of belated integral is given in a previous paper of the second author with I. F. Wilde; let us recall that (a) in (3) above \(H\) denotes a real Hilbert space, \((A_t)_{t\in R_+}\) a right-continuous, increasing family of finite von Neumann algebras over \(H\), \(A=[\bigcup A_t\mid t\in R_+]\), and \(\varphi\) a normalized faithful normal trace on \(A\). (b) A ``process \(X=\{x_t\in A_t\mid t\in R_+\}\) is called an \(L^p\)-martingale provided that \(M_s(x_t)= x_s\) for \(0\leq s\leq t<+\infty\), where \(M_t\) denotes the \(\varphi\)-invariant conditional expectation from \(L^p(A)\) onto \(L^p(A)\) \((1<p<+\infty)\). (c) \(L^2(A)\) is supposed to be a separable Hilbert space. The ``belated semivariations of the stochastic differential ``\(dX\) generated by the martingale \(X\) is given as follows: Let \(dX([s,t])= X_{t+}- X_{s-}\) (resp. \(dX((s,t])= X_{t+}- X_{s+}\) and so on\({\dots)}\) and \(E\subset R_+\) be a Borel set, then \[ \begin{multlined} |E|^{bX}= \sup\Bigl|\sum x_i dX(E_i)\Bigr|,\text{ where }(E_i)\text{ is a discrete partition of }E\text{ into}\\ \text{disjoint intervals in }R_+\text{ and }x_i\in X_{\inf E_i}.\end{multlined}\tag{4} \] The infiniteness of \(|.|^{bX}\) assures the existence of ``\(dX\) as a finitely additive function on the field generated by left-open, right-closed intervals in \(R_+\). The \(dX\)-integrand processes, say \(Y= \{y_t\in A_t\mid t\in R_+\}\), are controlled by the bounded monotone function \(t\to\varphi(|x_t|^2)\). Thus the Itô-isometry over ``simple processes could be defined similarly as in the commutative case, namely as follows \[ I_X: f\to \int fdX= \Bigl(\sum (f_{t_i})(X_{t_i+1}- X_{t_i})\Bigr), \] and so is the extension of \(I_X\) over the class of \(dX\)-integrable processes. The authors study the belated semivariations of the family of projections \((M_t)_{t\in R_+}\) in replacing ``\(dX((s,t])\) and so on\dots in (4) by 11\(\mu([s,t])= M_{t+}- M_{s-}\) (resp. \(\mu((s,t])= M_{t+}- M_{s+}\) and so on\dots) and obtain then a belated semivariation denoted by \(\mu^{bI}\). The answers to the question raised in the beginning of this review for the non-commutative case (see (3)) are given in Theorems 9-10 (Section 3) and Corollary 5 (Section 4), where it is stated: If \(Y\) is a \(\mu^{bI}\)-essentially bounded process, then \(Y\) is \(|.|^{bX}\)-essentially bounded (see (4)) for each \(L^2\)-bounded martingale \(X\), and hence such integrand \(Y\) is a priori ``\(dX\)-integrable for each \(L^2\)-bounded martingale \(X\).
- scientific article; zbMATH DE number 3905242
- scientific article; zbMATH DE number 22272
- Sur l'intégration stochastique par rapport à une martingale hilbertienne de carré intégrable. (On stochastic integration with respect to a square-integrable Hilbert-valued martingale)
- scientific article; zbMATH DE number 4069945
This page was built for publication:
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4343076)