A quadrature-based approach to improving the collocation method (Q1118377)

The author considers the numerical solution of strongly elliptic boundary integral equations in the plane. Compared to the (Petrov-) Galerkin method, the collocation method using odd-degree splines, suffers from a lack of optimal negative norm error estimates, which come into play when smooth functionals of the solution need to be computed. To remedy this ailment, the author considers the Petrov-Galerkin method with odd-degree splines as trial functions and trigonometric polynomials as test functions, and proposes a quadrature rule for the resulting inner products consisting of 2n quadrature points when the dimensions of the test and trial spaces equals n. (Typically, the quadrature points are the knots and the midpoints defining the splines). The resulting method has the flavor of a collocation method. The author analyzes in detail a class of operator equations which includes the single layer potential on the unit disk, with a quadrature rule that assigns the same weight to all the knots, and a different weight to all the midpoints of a uniform mesh, that also determines the splines. Optimal error estimates are proved for Sobolev norms from a (larger) negative scale than is possible for the collocation method.