A new convergence theorem for successive overrelaxation iterations (Q1300637)

The authors propose convergence criteria (sufficient conditions) on the successive overrelaxation iterations for systems of linear real equation. Tight bounds for the relaxation parameter are given which guarantee convergence of the method. These bounds depend on the positive real roots of an \(n\)th degree polynomial. To get simple estimates to the roots, quadratic or linear approximations to that polynomial around zero may be used. A sequence of convergence tests in ascending order of computational complexity is presented. All these theoretical results, combined with reordering equations techniques to accelerate the convergence, are demonstrated by numerical examples.