An adaptive method for the numerical solution of Fredholm integral equations of the second kind. I. Regular kernels (Q1094121)

The authors develop an adaptive method based on the trapezoidal rule for obtaining the numerical solutions of Fredholm integral equations of the second kind. The idea is to start with a given number of equally spaced points (or a given mesh) and obtain an approximate solution by solving a linear system of algebraic equations. The program then decides if the mesh needs to be refined and where. The choice of mesh points is made automatically so as to equidistribute both the change in the discrete solution and its gradient. The matrix at each stage is not recomputed, but only the necessary rows and columns are computed. The new system is solved by the Gauss-Seidel iterative method. The initial guess at the previous nodes is taken to be equal to the solution in the previous stage. At the new nodes the initial guess is taken to be the average of the previous solution at the neighboring points. Some numerical experiments with this method are presented.