On the efficient use of the Galerkin-method to solve Fredholm integral equations (Q1316219)

Let \(\Gamma\) be a piecewise smooth two-dimensional manifold in \(\mathbb{R}^ 3\). The authors consider boundary integral equations of the form \[ \lambda (x)u(x)+Ku(x)=f(x),\quad x \in \Gamma, \tag{1} \] where \[ Ku(x):=\int_ \Gamma k(x,y,y-x) u(y)d \Gamma(y),\;k(x,y,y-x)={1 \over | y-x |^{s+t}} \sum_{| \alpha | \geq t} c_ \alpha (x,y)(y-x)^ \alpha, \] with \(s+t\) odd, \(t \in N_ 0\) and piecewise smooth functions \(c_ \alpha\). This covers a wide class of integral equations (including hypersingular kernels) on piecewise smooth surfaces, approximated by splines of arbitrary degree. Using suitable coordinate transformations, the computation of the singular and nearly singular integrals arising in the discretization of (1) via the Galerkin method is discussed. Furthermore, the panel-clustering method introduced by the first author and \textit{Z. P. Nowak} [Numer. Math. 54, No. 4, 463-491 (1989; Zbl 0652.65031)] is developed for Galerkin discretizations and the corresponding error analysis is presented, yielding similar asymptotic estimates for the expense as shown in that paper for the collocation method. Numerical test calculations show that the reduction of the panel- clustering method is stronger for the Galerkin procedure than for the collocation method.