Automatic partitioning in linearly-implicit Runge-Kutta methods (Q1308559): Difference between revisions

Consider the numerical solution of the stiff initial value problem \(y'(t) = f(t,y(t))\), \(y(t_ 0) = y_ 0 \in \mathbb{R}^ n\) by a linearly-implicit Runge-Kutta (LIRK) scheme, e.g. by the \(W\)-method of \textit{T. Steihaug} and \textit{A. Wolfbrandt} [Math. Comput. 33, 521-534 (1979; Zbl 0451.65055)]. If only a few components of the solution \(y\) are stiff then partitioned methods are used leading to lower-dimensional algebraic systems for the stiff components while the nonstiff subsystem is integrated by an explicit scheme [\textit{P. Rentrop}, Numer. Math. 47, 545- 564 (1985; Zbl 0625.65059)]. An automatic procedure for partitioning of stiff and nonstiff subsystems is proposed using the Krylov technique thus avoiding the computation of the Jacobi matrix \(f_ y\) of \(f\). The orthogonal basis of the corresponding Krylov subspace is obtained by Gram-Schmidt orthogonalization. The results of intensive tests are discussed. The conclusion is made that the Krylov technique for partitioning in LIRK methods is an effective tool for treating large partially stiff systems.