Gaussian collocation via defect correction (Q756383)

The author considers boundary value problems for first order systems of differential equations \(y'(x)=f(x,y(x)),\) \(x\in [0,1]\), \(b(y(0),y(1))=0\) where f: [0,1]\(\times {\mathbb{R}}^ n\to {\mathbb{R}}^ n\) and b: \({\mathbb{R}}^ n\times {\mathbb{R}}^ n\to {\mathbb{R}}^ n\) are smooth functions. For problems of this type collocation schemes based on continuous piecewise polynomials have been extensively studied in the numerical literature. A favourable feature of these schemes is the superconvergence property which is obtained if special classes of collocation nodes are used such as Gaussian or Lobatto points. Since the effort required for the computation of the collocation spline grows rapidly with the order of the scheme, techniques for an efficient solution of high order collocation equations are of interest. This paper presents a defect correction method for the iterative solution of the collocation equations. The main idea to overcome the drawbacks of other authors is to use two different grids, one consisting of the collocation points, the other consisting of equidistant points within each collocation subinterval. The collocation equations can be reformulated in terms of the values on the piecewise equidistant grid. This form of the collocation equations is used for defect evaluation. The trapezoidal scheme on the piecewise equidistant grid is employed as the basic discretization. This method is most attractive if either the order of the collocation scheme or the dimension of the differential system is large. In these cases the amount of computational work required for the solution of the collocation equations is considerably reduced. The difference to other defect correction methods becomes most striking if Gaussian collocation points are used for defect evaluation. In this case the author achieves a final order of accuracy twice as high as standard defect correction methods with a comparable amount of work for each iteration step.