Runge-Kutta time semidiscretizations of semilinear PDEs with non-smooth data (Q315719)

The one-dimensional semilinear wave equation and the nonlinear Schrödinger equation with periodic, Neumann and Dirichlet boundary conditions can be reformulated as an abstract semilinear evolution equation of the form \(dU/dt = A U + B(U)\) with non-smooth initial data \(U(0) = U^0\) in a Hilbert space \(\mathcal{Y}\). Here \(A\) is a normal operator generating a strongly continuous semigroup and \(B\) is a smooth nonlinearity on a scale of Hilbert spaces \(\mathcal{Y}_{\ell} = D(A^{\ell}) \subseteq \mathcal{Y}\), \(\ell \geq 0\). In the paper under review, the abstract evolution equation is formally discretized in time by an A-stable Runge-Kutta method, so that the numerical method (now considered as a nonlinear operator on the infinite dimensional space \(\mathcal{Y}\)) is well-defined, and the effects of non-smooth data on the order of convergence of the method are analyzed. When an A-stable Runge-Kutta method of classical order \(p\) is applied to the problem at hand with initial data \(U^0 \in \mathcal{Y}_{\ell}\), it is shown that the order of convergence is \(\mathcal{O}(h^q)\), with \(q= p\) \(\ell / (p+1)\) for \(0 \leq \ell < p +1\). The order reduction taking place when \(\ell < p + 1\) is caused by the presence of unbounded operators in the expansion of the local error of the method. The approach followed here consists in applying a spectral Galerkin approximation to the exact semiflow of the abstract differential equation and then discretize the projected equation in time, so as to bound the local error coefficients in terms of the accuracy of the projection. The numerical example reported in the paper seems to indicate that the obtained estimate is sharp.