A numerical scheme to solve nonlinear BSDEs with Lipschitz and non-Lipschitz coefficients (Q2386798): Difference between revisions

The authors consider the following nonlinear backward stochastic differential equation \[ dx(t)=f(x(t),z(t),t)dt -z(t)dW(t),\quad x(T)= \xi,\tag{1} \] where \(x(t) \in R^d\), \(z(t)\in R^{n \times m}\), \(f:R^n \times R^{n\times m}\times [0,T] \rightarrow R^n\), \(\xi \in L_F^2(\Omega,R^n)\) and \(W\) is an \(m\)-dimensional Wiener process. The aim of the paper is the development and convergence analysis of numerical approximations of the solution of (1). The authors prove convergence in the mean-square of the approximations in two cases, i.e., when \(f\) is globally Lipschitz-continuous and when \(f\) satisfies a weaker condition. The ideas behind the construction of the numerical method are first a replacement of the nonlinear function \(f\) by continuous piecewise linear functions and second the exploitation of the relationship between linear backward differential equations and stochastic control. In the last section numerical illustrations are presented.