A novel approach to construct numerical methods for stochastic differential equations (Q2453472): Difference between revisions

The author presents an approach to numerically solve a scalar stochastic differential equation of the form \[ x_t = x_0 +\int_0^t a(x_s)\,ds+\int_0^t b(x_s) \,dW_s. \] The method works by finding a semi-discretisation \[ y_t = y_{t_k}+\int_{t_k}^t f(y_s,y_{t_k})\,ds +\int_{t_k}^t g(y_s,y_{t_k})\,dWs \] for \(t\in [t_k,t_{k+1}]\) on a grid \(0=t_0<t_1 \dots t_n=T\) such that \(f(x,x) = a(x)\) and \(g(x,x)=b(x)\) and that the semi-discretization possesses an explicit solution. Methods of this form are explicit and contain, e.g., the forward Euler method for \(f(y,x)=a(x)\) and \(g(y,x)=b(x)\). Then, under local Lipschitz conditions on \(f\) and \(g\) and square-integrability of the exact solution and the semi-discretization, it is proven that the semi-discretization converges to the exact solution in the mean-square sense. Furthermore, an explicit example for a method is presented and also a numerical example (a super linear stochastic differential equation) is considered comparing the new method with the tamed Euler scheme.