On numerical solutions of the stochastic wave equation (Q2505493)

The author considers a stochastic wave equation on an unbounded domain which is of the form \[ \frac{\partial^2 u}{\partial t^2}= \frac{\partial^2 u}{\partial x^2} + f(x,t,u) + g(x,t,u)\dot{W}, \qquad x\in \mathbb{R},\quad t>0, \] with initial conditions \(u(x,0) = u_0(x)\) and \(\frac{\partial u}{\partial t} (x,0)=v_0(x)\) for \(x\in \mathbb{R}\). Here \(\dot{W}\) denotes a standard space-time white noise on \(\mathbb{R}\times \mathbb{R}_+\). He develops a finite-difference method, an adaptation of the leap-frog method, to approximate the solution of the above equation on a certain lattice. The error of this method in the \(L^p\) sense is analysed for points on the lattice and uniformly on the domain. First the author considers the upper bound for the approximation error and finds that this is of order \(\sqrt{h}\), reflecting the fact that the solution paths are Hölder continuous of order \(1/2\). Further the author considers a lower bound on the error. He finds that the rate of convergence of \(O(\sqrt{h})\) is the best possible on the given lattice in the sense that there are Lipschitz functions \(f\) and \(g\) in the problem above such that any numerical scheme using only increments of the driving process for a step-size \(h\) cannot converge in \(L^2\) faster than \(O(\sqrt{h})\). In the last section the author briefly treats the equation above on a bounded domain. This problem can be considered to model a vibrating string subjected to space-time white noise. The numerical analysis for the case of the unbounded domain can be adapted to this setting, yielding the same rates of convergence.