Hybrid collocation methods for Fredholm integral equations with weakly singular kernels (Q881477)

Let \(C(I)\) denote the space of all continuous functions defined on \(I := [0, 1]\). For a given parameter \(\alpha\in (0, 1]\), introduce a kernel \(G_\alpha\) by \(G_\alpha(s,t):=g_\alpha(| s-t| )m(s,t)\), \(s,t\in I\), \(s\neq t\), where \(m\in C(I\times I)\) and \[ g_\alpha(| s-t| )=\left\{ \begin{aligned} &\log| s-t| ,\quad \alpha=1,\\ &| s-t| ^{\alpha-1},\quad0<\alpha<1. \end{aligned}\right. \] Define a Fredholm integral operator \(\mathcal G_\alpha : C(I)\to C(I)\) by \[ (\mathcal G_\alpha y)(s):=\int_0^1\mathcal G_\alpha(s,t)y(t)\,dt, \quad s\in I, \] and consider the weakly singular Fredholm integral equations of the second kind \[ y(s)-(\mathcal G_\alpha y)(s)= f(s),\quad s\in I,\tag{1} \] where \(f\in C(I)\) is a given function and \(y\in C(I)\) is a unknown function to be determined. Assume that \(y\) is not an eigenvalue of \(\mathcal G_\alpha\), and, consequently, (1) has a unique solution in \(C(I)\). In order to approximate functions with mild singularities, many numerical analysts used graded meshes to develop numerical methods with an optimal order of convergence. The main disadvantage of the graded mesh methods is that subintervals near the singular point in the graded mesh have very small lengths and thus may cause serious round-off error problems for small \(\alpha\) when high order polynomials are used for the approximate solution. To overcome such a difficulty, the authors use the hybrid collocation method to solve (1). It is shown that the optimal order of convergence can be achieved. Also, the authors apply the results for Fredholm integral equations of the second kind to Fredholm integro-differential equations and investigate the superconvergence property of the iterated solution for Fredholm integro-differential equations with weakly singular kernels. Numerical examples to illustrate the theoretical results are provided.