A linear numerical scheme for nonlinear BSDEs with uniformly continuous coefficients (Q2570902)

The objective announced by the authors is to study a linear numerical scheme for nonlinear backward stochastic differential equations (for short, BSDE) \[ dY_t=-f(Y_t,Z_t,t)dt+Z_tdW_t,\;t\in[0,T], \quad Y_T=\xi, \] where \(W=(W_t)_{t\in[0,T]}\) is a given \(m\)-dimensional Brownian motion and \(\xi\) a square integrable random variable, measurable with respect to \(W\). Using the restrictive assumption that both solution processes \(Y_t, 0\leq t\leq T,\) and \(Z_t,t\leq T,\) are bounded by some real constants \(M_1\) and \(M_2\) (this assumption has been got by the authors for the general case of a BSDE as consequence of a wrong conclusion), the authors approximate the driver \(f\) of the BSDE uniformly by continuous piecewise linear functions \(f_n\) on the domain of the possible values of the process \((Y_t,Z_t,t),0\leq t\leq T\). Denoting the solution of the BSDE with driver \(f_n\) by \((Y^n,Z^n)\), they then show that, for all \(0\leq t\leq T\), \(Y_t^n\) converges a.s. to \(Y_t\). After, by using recent results by \textit{M. Kohlmann} and \textit{X. Y. Zhou} [SIAM J. Control Optimization 38, No.~5, 1392--1407 (2000; Zbl 0960.60052)], they interpret the piecewise linear BSDE as stochastic control problem with optimal feedback control. In the last paragraph two examples for the numerical treatment are discussed, but the both BSDE have a deterministic driver and a deterministic terminal value so that they reduce to a simple ordinary differential equation which can explicitly be solved.