Convergence analysis of time-point relaxation iterates for linear systems of differential equations (Q1360166)

A time-point relaxation (TR) sequence for the solution of an ordinary differential equation is a sequence of ordinary differential equations whose solutions converge to that of the given equation. The authors study convergence of continuous Runge-Kutta (CRK) approximations to TR sequences to obtain a Runge-Kutta approximation to the solution of a linear ordinary differential equation with constant coefficients. Three different TR sequences are considered: Jacobi, Gauss-Seidel, and successive overrelaxation. These are obtained by partition of the coefficient matrix into lower, diagonal, and upper matrices as is done in solution of linear equations in linear algebra. It is known that if convergent the CRK approximations to these TR sequences converge to solutions of a new implicit algorithm called a diagonally split Runge-Kutta (DSRK) approximation. The authors prove that the CRK sequences approximating the solution of the above named TR sequences, converge if and only if the solution of the associated DSRK approximation may be obtained by solving it iteratively using corresponding Jacobi, Gauss-Seidel or successive overrelaxation procedures from linear algebra. Sufficient conditions for convergence are presented in case the coefficient matrix of the differential equation is consistently ordered. Regions of convergence using Runge-Kutta algorithms of order 3 and 4 are determined for differential equations whose coefficient matrix is tridiagonal with upper diagonal elements \((-a, -\frac12 a,\dots, -\frac12 a)\), diagonal \((b,\dots,b)\), and lower diagonal \((\frac12 a,\dots, \frac12 a,a)\).