Solution of integral equations using function-valued Padé approximants. II (Q688119)

The authors describe some numerical experiments in the approximate determination, in a case in which exact solutions are known in simple closed form, of the eigenvalues and associated eigenfunctions of an integral equation \(\varphi (x) = 1+ \lambda \int K(x,y) \varphi (y)dy\) \((-1 \leq y \leq 1)\). Four methods are considered. The first involves certain rational functions of \(\lambda\) derived from series of the form \(\sum c(j | x) \lambda^ j\) \((0 \leq j \leq \infty)\) as described in part I [\textit{P. R. Graves-Morris}, J. Comp. Appl. Math. 32, 117-124 (1990; Zbl 0714.65098)]. The second involves locating the zeros of a truncated version of the resolvent kernel expressed in ascending powers of \(\lambda\). The third involves minimisation by use of polynomial approximations to the eigenfunctions. The fourth involves determination of the roots of the denominators of certain approximating fractions derived from a Neumann series in ascending powers of \(\lambda,x\) being suitably fixed. It is found that, with regard to the selected example, of these four methods the first is the most effective.