Path-wise solutions of stochastic differential equations driven by Lévy processes (Q5960829): Difference between revisions

The author studies existence and uniqueness of pathwise solutions to jump-type stochastic integral equations of the form \(Y_t = \int_0^tf(Y_t) dX_t\) where the driving process \(X_t\) is a \(d\)-dimensional Lévy process with a.s.\ finite (strong) \(p\)-variation. The function \(f\) is in a Hölder-Zygmund class of order \(\alpha > p\) and, without loss of generality, \(X_t\) is assumed to have a.s.\ unbounded variation. Since \(dX_t\) is not a well-defined Lebesgue-Stieltjes integrator, the above equation is to be understood in the sense of Young's integral (if \(p < 2\)), resp., using (if \(p>2\)) a càdlàg-modification of \textit{T. J. Lyons'} rough path integral [Rev. Math. Iberoam. 14, No. 2, 215--310 (1998; Zbl 0923.34056)]. Two different types of solutions are considered: the so-called geometric and the forward solution. For the first approach, the jumps of the driving Lévy process are ``interpolated'' by a (reversible!) time-change and the resulting integral equation with the now continuous driving process is then solved by essentially classical ideas and iteration methods [as suggested by \textit{T. Lyons}, Math. Res. Lett. 1, No. 4, 451--464 (1994; Zbl 0835.34004)]; if \(p>2\), the classical integrals have to be replaced by Lyons' rough path methods which use both the path of the process and its Lévy area. The key point here is to show that the antisymmetric Lévy area process \(A^{ij} = \int X^j_- \circ dX^i - X^i_- \circ dX^j\) exists and has a.s.\ finite variation of order \(p/2\). The proof for this is based on \textit{I. Monroe}'s embedding theorem for martingales into a Brownian motion [Ann. Math. Stat. 43, 1293--1311 (1972; Zbl 0267.60050)]. The time-change is finally undone to go back to a càdlàg process. The forward solution consists of solving the original integral equation ``between'' jumps (which is again driven by a continuous process!) and then adding the correct jumps \(\Delta Y = f(\Delta Y) \Delta X\) to the solution at the respective jump times. In an appendix certain technical details of Lyons' rough path approach (which was originally developed for continuous paths) are extended to càdlàg processes. Some of the proofs in the paper are rather sketchy and details are left to the reader. No connection is made between the geometric/forward solutions considered here and the usual Itô and Stratonovich solutions (whenever they exist).