Runge-Kutta convolution quadrature for operators arising in wave propagation (Q639371): Difference between revisions

Convolution quadrature is a numerical scheme to evaluate \[ u(t) = \int_0^t k(t-\tau)\, g(\tau)\, dt \quad (0\leq t \leq T) \] for given \(g\) and \(k\). The numerical realization is based on properties of the Laplace transform (transfer function) \(K(s)\) of \(k\). In this paper \(K\) satisfies, besides the standard assumptions of analyticity in a half-plane \({\mathrm{Re}}\,s > \sigma_0\) and a polynomial bound \({\mathcal O}(|s|^{\mu_1})\) there, the stronger polynomial bound \({\mathcal O}(|s|^{\mu_2})\) with \(\mu_2 \leq \mu_1\) in convex sectors \(|\arg\,s| \leq \pi/2 - \theta\). The authors show that the order of convergence of the Runge-Kutta convolution quadrature is determined by \(\mu_2\) and the underlying Runge-Kutta method, but is independent of \(\mu_1\). Time domain boundary integral operators for wave propagation problems have Laplace transforms with these properties. Numerical examples from acoustic scattering illustrate the results, where the full classical order of the Runge-Kutta method is attained away from the scattering boundary.