The martingale problem on Banach spaces (Q1189282)

Let \(E\) be a real, separable Banach space and denote by \(\text{Cov}(E)\) the set of all covariance operators of Gaussian measures on \(E\). Suppose that for every \(t\geq 0\) and every \(f\in C([0,t],E)\) there is given an \(a(t,f)\in E\) and a \(Q(t,f)\in\text{Cov}(E)\). The martingale problem for the family \((a(t,f),Q(t,f))_{t\geq 0, f\in C([0,t],E)}\) is essentially the problem to construct a family \((P(t,f))\) of probability measures on \(C(\mathbb{R}_ +,E)\) such that the canonical process \(X\) on \(C(\mathbb{R}_ +,E)\) behaves locally at every \((t,f)\) like the homogeneous, independent increment Gaussian process with expectation \(a(t,f)\) and covariance \(Q(t,f)\). Under certain growth conditions on the family \((a(t,f),Q(t,f))\) and additional smoothness assumptions it is proved that a solution \((P(t,f))\) of the martingale problem exists. The smoothness assumptions are related to the geometry of \(E\) and guarantee the validity of certain Banach space martingale inequalities. If there is a factorization \(Q(t,f)=T(t,f)\circ T(t,f)^ t\) of \(Q(t,f)\) with \(T(t,f)\in L(\ell^ 2,E)\) such that the function \(T\) fulfills a certain Lipschitz condition, then it is proved that the family \((P(t,f))\) is unique. This implies e.g. that the so- called path process associated to the canonical process on \(C(\mathbb{R}_ +,E)\) is a strong Markov process.