Numerical methods for systems with measurable coefficients (Q2277774)

The initial value problem for the n-dimensional system \(dy/dt=f(t,y)\), \(y(t_ 0)=y_ 0\), where f is smooth in y and bounded and measurable in t, is considered. A family of numerical algorithms to solve this problem, which is loosely akin to the Runge-Kutta method, is given. Firstly a discretization with respect to y is done and the dependence on t is retained only through mean properties. Mean values of m-fold self- substitution of f are allowed at m different values of t. By estimating the resulting means with Monte Carlo simulation one obtains actual numerical procedures, named Runge-Kutta Monte Carlo (RKMC) methods, which simulate estimates in the statistical sense for the solution. One parameter families of second and third order RKMC methods are emphasized and it is stated that these algorithms give stable, qualitative correct answers with a small number of steps, which can be insufficient for a Runge-Kutta method of the same order to give meaningful results.