Some computational methods for the Fokker-Planck equation (Q2101335)

The Fokker-Planck equation is studied using three finite difference based computational methods, and the full discretization of the models considered using the proposed methods is presented. For the linear problem, an implicit forward difference scheme and the Crank-Nicolson scheme are applied, while the implicit forward difference scheme is used for the linear backward Kolmogorov equation. The semi-implicit scheme is used for the nonlinear case, so that nonlinear terms are converted to linear, resulting in a linear discrete model without linearization, and thus preserving all the properties of original equation. Unconditional von-Neumann stability is established for the linear models and numerical experiments conducted using three test problems with known solutions. The proposed methods are compared with a cubic B-spline scaling method, which is known to be ``easy to implement and produces very accurate results''. Comparison of the resulting schemes show the proposed methods to be much simpler and hence easier to implement, and simulation results tabulated for various times and solutions displayed in figures reveal their accuracy competitiveness. The maximum absolute errors seen in the tables show that the numerical and exact solutions are extremely close, although the numerical accuracy is seen to decrease with time.