Weak Dirichlet processes with a stochastic control perspective (Q855923)

The weak Dirichlet processes \(D\) are considered. It means that \(D\) can be represented as \(D=M+A\) where \(M\) is a local martingale and \(A\) is a ``weak zero energy process'', i.e. the process orthogonal to every continuous local martingale. The generalized Fukushima-Dirichlet decomposition is derived for functions \(u(s,x)\in C^{0,1}\): \[ u(s,D_s)=u(t,D_t)+\int_t^s\partial_x u(r,D_r)\,dM_r+B^D(u)_s-B^D(u)_r, \] where \(B^D(u)\) is a weak zero energy process. For \(u\in C^{1,2}\), \(B^D(u)\) can be derived by Itô formula. A representation for \(B^D(u)\) is given for \(u\) which solves a partial differential equation. In the stochastic optimal control it can be the Hamilton-Jacobi-Bellman equation. These results are based on the ``calculus via representation'' technique in which the ``forward integral'' is defined as \[ \int_0^s X_td^{-}Y_t=\lim_{\varepsilon\to 0} \int_0^s X_t{Y_{r+\varepsilon}-Y_r\over\varepsilon}\,dr \] and the covariation is \[ [X,Y]_r=\lim_{\varepsilon\to 0}{1\over\varepsilon} \int(X_{r+\varepsilon}-X_r)(Y_{r+\varepsilon}-Y_r)\,dr. \] The orthogonality of \(X\) and \(Y\) means \([X,Y]_t=0\) for all \(t\).