Zur näherungsweisen Berechnung spezieller Integrale. (Approximate computation of special integrals) (Q1094087)

The author proposes an algorithm for numerical quadrature \(I(f)=\int^{b}_{a}f(x)dx\), especially suitable when the integrand has poles \(z_ 1,...,z_ m\) close to the interval [a,b]. His idea is based on a rational approximation of the integrand, and the algorithms are as follows. First take an ordinary integration formula \(Q_ n(f)=\sum^{n}_{k=0}d_ kf(x_ k),\) with nodes \(a\leq x_ 0<x_ 1<...<x_ n\leq b\). Let \(\mu_ j\) be the order of the pole \(z_ j\) (1\(\leq j\leq m)\), and put \(\mu =\mu_ 1+...+\mu_ m\). Put \(\omega (x)=\prod^{n}_{i=0}(x-x_ i)/\prod^{m}_{j=1}(x-z_ j)^{\mu_ j},\quad q_ i(x)=x^ i\omega (x)\quad (0\leq i\leq \mu -1),\) and compute \(\gamma_ i=\int^{b}_{a}q_ i(x)dx\) analytically. Take a suitable positive integer s and additional nodes \(x_{n+1},...,x_{n+s+\mu}\), sufficiently close to the poles \(z_ j's.\) Then compute the quotient-difference \(\lambda_{ik}=D^{n+k+s+1}q_ i(x_{n+s+k+1})\), where \(D^ i\) is defined recursively by \[ D^ 0f(x)=f(x);\quad D^{i+1}f(x)=(D^ if(x)-D^ if(x_ i))/(x-x_ i). \] Finally solve the linear equations \(\sum^{\mu - 1}_{k=0}\lambda_{ik}\alpha_ k=\gamma_ i\) (0\(\leq i\leq \mu -1)\) with respect to \(\alpha_ k\) and modify the previous value \(Q_ n(f)\) by adding \(\sum^{\mu -1}_{k=0}\alpha_ kD^{n+s+k+1}f(x_{n+s+k+1}).\) He gives an error estimation and a simple numerical example.