A probabilistic numerical method for fully nonlinear parabolic PDEs (Q640058): Difference between revisions

The authors consider the backward probabilistic scheme for the approximation of the solution of fully nonlinear partial differential equations introduced by \textit{P. Cheridito, H. M. Soner, N. Touzi} and \textit{N. Victoir} [Commun. Pure Appl. Math. 60, No.~7, 1081--1110 (2007; Zbl 1121.60062)] and observe that this scheme can be introduced naturally without appealing to the notion of backward stochastic differential equation. They decompose the scheme in three steps and assume that the fully nonlinear partial differential equation (PDE) satisfies a comparison result in the sense of viscosity solutions. The convergence of this approximation is established toward the unique viscosity solution and an asymptotic analysis of the approximation error is given. They prove the convergence of the discrete-time approximation for general nonlinear PDEs and provide bounds on the corresponding approximation error for a class of Hamilton-Jacobi-Bellman (HJB) PDEs. Then, they consider the implementable scheme involving a Monte-Carlo error and prove similarly a convergence result. Finally they present numerical results for the approximation of the solution of the mean curvature flow equation in dimensions two and three and for a five-dimensional HJB equation arising in the problem of portfolio optimization in financial mathematics. The results are rather convincing and the paper is clearly written.