Two three-parallel and three-processor SDIRK methods for stiff initial-value problems (Q1379025)

The authors construct numerical methods based on parallel diagonal iteration of implicit Runge-Kutta methods for the numerical integration of stiff initial-value problems of the form \[ y'(t)= f(t,y(t)),\quad t\in[t_0, T],\quad y(t_0)= y_0,\tag{\(*\)} \] where \(y_0\) is a given \(N\)-dimensional real vector, \(y(t)\in \mathbb{R}^N\) is unknown and \(f\) denotes a given mapping from \([t_0, T]\times V\) to \(\mathbb{R}^N\) where \(V\) is some subset of \(\mathbb{R}^N\). Problems of this type often arise in modelling of mechanical and electrical engineering systems or in the solution of semidiscretized convection-diffusion problems associated to time-dependent parabolic partial differential equations. The stiffness of these problems require that the numerical methods to be used should be unconditionally stable, preferable A-stable or L-stable and therefore implicit. So, implicit Runge-Kutta methods are excellent candidates in order to integrate stiff initial value problems. Because of the phenomenon of order reduction in many stiff problems methods with high stage order provide more accurate results. From a computational point of view the diagonal implicit Runge-Kutta methods are attractive methods since they have suitable stability properties and their implementation can be carried out with lower computational cost than fully Runge-Kutta methods. The disadvantage of these methods is their low stage order. The so-called parallel diagonally iterated Runge-Kutta methods for the parallel numerical integration of problem \((*)\) have important computational advantages when compared with fully implicit Runge-Kutta methods. Especially for solving semidiscretized partial differential equations where one needs relatively low-accurate results at a low price there were proposed in 1994 L-stable parallel singly diagonally implicit Runge-Kutta (PSDIRK) methods. Singly diagonal means that the diagonal matrix \(D\) of the iteration process is of the type \(D= dI\) (\(I\) identity-matrix). The authors construct two three-parallel and three-processor SDIRK methods based on the concept and ideas of a paper of \textit{Nguyen huu Cong} [J. Comput. Appl. Math. 54, No. 1, 121-127 (1994; Zbl 0819.65109)].