Some schemes for the implementation of implicit Runge-Kutta methods (Q688620): Difference between revisions

The implementation of implicit Runge-Kutta methods leads to some nonlinear equations, which may be solved by a modified Newton iteration, but alternative iteration schemes have been suggested to reduce the linear algebra costs. The authors study a linear iteration scheme. For the scalar differential equation \(x'=qx\), the convergence rate of the scheme depends on the spectral radius \(\rho[M(z)]\) of the iteration matrix \(M\), a function of \(z=hq\) where \(h\) is the steplength. A lower bound for \(\rho[M(z)]\) is established and new schemes are obtained for the two-stage Gauss method by minimizing the supremum of this lower bound over regions of the complex plane. Some numerical experiments are reported on.