Locally linearized methods for the simulation of stochastic oscillators driven by random forces (Q512850)

Locally linearized methods (LLM) for approximating the solution of the system of stochastic differential equations \[ dx(t)= a(t,x(t))\,dt+ \sum^m_{j=1} b_j(t) dw^j_t,\quad x(t_0)= x_0,\quad t\in [t_0,T], \] used to model a stochastic oscillator are shown to preserve important known properties of the solution such as linear growth of energy along the paths, oscillatory behavior, and phase flock that retains the symplectic structure of the equivalent stochastic Hamiltonian system. A sharper bound for the mean-square error of LLM is derived. Results of numerical experiments are presented that provide computational evidence and demonstrate the advantages in these respects of LLM over other conventional methods.