A weak trapezoidal method for a class of stochastic differential equations (Q390252)

The authors consider a numerical approximation scheme for stochastic differential equations of the specific form \[ dX(t)=b(X(t))dt+\sum_{k=1}^M \sigma_k(X(t))\nu_k dW_k(t),\qquad X(0)=x_0\in \mathbb{R}^d, \] where \(W_k\) are independent scalar Wiener processes, the drift coefficient \(b(\cdot)\), \(\nu_k\in \mathbb{R}^d\) and the individual diffusion coefficients are real and the latter also positive, i.e., \(\sigma_k(\cdot)\geq 0\). Hence the randomness is restricted to fixed (geometrical) directions at random rates. The proposed numerical method is a kind of split-step / predictor-corrector method, using first an explicit Euler approximation to obtain an intermediate result which in the second step is used in an trapezoidal-like step. This procedure results for equations of the specific form in a method with weak-order two avoiding the evaluation of iterated Itō integrals and derivatives of the coefficients. The restriction to a specific type of equations thus yields less complex second-order methods than when considering general equations. The main result of the paper is the proof of the weak convergence order two under sufficient continuity and boundedness assumptions on the coefficients. Further explanations present the heuristic behind the method, a comparison to the Richardson extrapolation and numerical examples illustrating the theoretical results.