Quantitative asymptotic error analysis and Richardson extrapolation for exponential Euler approximations of stochastic wave equations

Abstract

We establish quantitative weak convergence for the joint terminal law of the solution to a stochastic wave equation and the normalized error of its exponential Euler approximation. For multiplicative trace-class noise and spatial regularity $α\in(1,2]$, the normalized error converges to its limiting distribution at rate $O(m^{-(α-1)})$, where $m$ is the number of time steps. This estimate identifies the leading weak bias and yields a first-order weak-error expansion with remainder $O(m^{-α})$. Moreover, by canceling the leading weak-error term, we construct a Richardson extrapolation scheme with weak convergence order $α$ for $α\in(1,2]$ and show that this rate is sharp. To overcome the lack of parabolic smoothing and control the remainders amplified by normalization, we combine dimension-uniform regularity estimates for the joint Kolmogorov equation with a conditional weak expansion and backward transport of coefficient directions.