Probabilistic evolution of the error of numerical method for linear stochastic differential equation (2304.01602v2)

Abstract: The quantitative characterization of the evolution of the error distribution (as the step-size tends to zero) is a fundamental problem in the analysis of stochastic numerical method. In this paper, we answer this problem by proving that the error of numerical method for linear stochastic differential equation satisfies the limit theorems and large deviation principle. To the best of our knowledge, this is the first result on the quantitative characterization of the evolution of the error distribution of stochastic numerical method. As an application, we provide a new perspective to explain the superiority of symplectic methods for stochastic Hamiltonian systems in the long-time computation. To be specific, by taking the linear stochastic oscillator as the test equation, we show that in the long-time computation, the probability that the error deviates from the typical value is smaller for the symplectic methods than that for the non-symplectic methods, which reveals that the stochastic symplectic methods are more stable than non-symplectic methods.