Probabilistic Numerical Methods for Fully Non-linear Non-local Parabolic PDEs
Résumé
We introduce a probabilistic numerical method for the approximation of the solutions of fully non--linear parabolic non--local PDEs. The method is the generalization of the method in \cite{ftw} for fully non--linear parabolic PDEs. As an independent result, we also introduce a Monte Carlo Quadrature method to approximate the integral with respect to Lévy measure which may appear inside the scheme. We consider the equations whose non--linearity is of the Hamilton--Jacobi--Belman type. We avoid the difficulties of infinite Lévy measures by truncation of the Lévy integral by some $\kappa>0$ near $0$. The first result provides the convergence of the scheme for general parabolic non--linearities. The second result provides bounds on the rate of convergence for concave non--linearities from above and below. For both results, it is crucial to choose $\kappa$ appropriately dependent on $h$.
Domaines
Probabilités [math.PR]
Origine : Fichiers produits par l'(les) auteur(s)