General linear methods for stiff differential equations (Q5942332): Difference between revisions

A general class of numerical methods for stiff initial value problems that contains both the linear multistep and Runge-Kutta methods is considered. The aim of the author is to obtain particular methods that combine the low computational cost shared by the standard backward differential formula (BDF) methods of the class of multistep methods with the stability properties that possess some implicit Runge-Kutta methods. With this idea in mind two types of methods for stiff systems on parallel and sequential implementation are considered. Then after some preliminary results on the so called inherent stability and on the order conditions, he proposes several A-stable methods with \(s\)-stages and order \(p\) where \( s=p+1\) or else \( s=p+2\) so that the implicit equations of all stages are similar to those of a diagonally implicit Runge-Kutta method and therefore its computational cost per step is equivalent to solving \(s\) times a BDF type method with different starting values. Finally some implementation issues that include the estimation of the local truncation error, variation of step size and order for the new methods are considered.