Nonsemimartingales: stochastic differential equations and weak Dirichlet processes (Q879256)

Let \(n\geq 2\) and \((X^1,X^2,\dots,X^n)\) be a vector of continuous processes. For any \(\varepsilon>0\) and \(t\) in \([0,1]\) \[ [X^1,X^2, \dots,X^n]_\varepsilon(t)= \frac{1}{\varepsilon}\int^t_0\prod^n_{k=1} (X^k_{s+\varepsilon}-X^k_s)\,ds. \] If \([X^1,X^2,\dots,X^n]_\varepsilon (t)\) converges uniformly in probability (ucp), when \(\varepsilon\to 0\), then the limiting process is called the \(n\)-covariation process of the vector \((X^1,X^2,\dots,X^n)\) and is denoted \([X^1,X^2,\dots,X^n]\). If the processes \(X^k,k=1,2,\dots,n\) are all equal to the real valued process \(X\), then the \(n\)-covariation is denoted by \([X;n]\) and called \(n\)-variation process. If \(n=2\), this process is the quadratic variation (is denoted \([X]\) or \([X,X])\), if \(n=3\), this is the cubic variation. If \(X\) has a quadratic (cubic) variation, \(X\) is called finite quadratic (cubic) variation process. The authors define also strong \(n\)-variation processes. For two continuous processes \(X\) and \(Y\) the symmetric integral is defined as limit ucp (if exists) \[ \lim_{\varepsilon\to 0}\frac{1}{2\varepsilon}\int^t_0Y_s (X_{s+\varepsilon}-X_{s-\varepsilon})\,ds \] and is denoted by \(\int^t_0 Y\,d^0X\). In the present paper the stochastic differential equation \[ d^0 X_t=\sigma(t,X_t)[d^0\xi_t+\beta(t,X_t)\,d^0M_t+\alpha(t,X_t)\,dV_t],\;0\leq t\leq 1,\quad X_0=\eta\tag{1} \] is considered. A solution of (1) is a solution of the integral equation \[ X_t=\eta+\int^t_0\sigma(s,X_s)\,d^0 \xi_s+\int (\sigma\beta)(s,X_s)\,d^0M_s+\int(\sigma\alpha)(s,X_s)\,dV_s. \tag{2} \] Where \(\xi,M, V\) are adapted and, respectively, a strong cubic variation process, a local martingale and a bounded variation process, \(\sigma,\beta:[0,1]\times\mathbb{R}\to \mathbb{R}\) are continuous functions, \(\alpha:[0,1]\times\mathbb{R}\times\Omega\to\mathbb{R}\) is progressively measurable and locally bounded in \(x\), uniformly in \(t\), almost surely and \(\eta\) is given random variable. The authors prove the existence and uniqueness of solutions of (1) when \(\xi\) has the structure \(Q+R\), where \(Q\) is a finite quadratic variation process and \(R\) is strongly predictable (in particular, \(R\) is weak Dirichlet). They discuss also existence and uniqueness in the cases, when \(\xi\) is a Hölder continuous or a fractional Brownian motion.