A numerical scheme for BSDEs (Q1431562)

The author considers the decoupled system of forward-backward stochastic differential equations \[ \begin{aligned} X_t&=x + \int_0^t b(s,X_s) \,ds + \int_0^t \sigma(s,X_s)\,dW_s,\\ Y_t&= \Phi(X)+\int_t^Tf(s,X_s,Y_s,Z_s)\,ds-\int_t^TZ_s\,dW_s, \end{aligned} \tag{1} \] where \(b,\,\sigma\) and \(f\) are deterministic functions and \(W\) is a standard Brownian motion. The deterministic functional \(\Phi\) is assumed to be a so-called \(L^\infty\)-Lipschitz or a \(L^1\)-Lipschitz functional or takes the form \(g(X_T)\). Based on results of \textit{J. Ma} and \textit{J. Zhang} [Probab. Theory Relat. Fields 122, No. 2, 163--190 (2002; Zbl 1014.60060)] he proves \(L^2\)-regularity of the martingale integrand \(Z\). A numerical scheme is proposed, consisting of the Euler method for the forward diffusion \(X\) and an approximation via step processes for the pair \((Y,Z)\). Convergence in the mean-square sense of the method is proved and convergence rates are obtained. The latter depend in particular on the above stated smoothness properties of \(\Phi\).