A fast numerical algorithm for the integration of rational functions (Q967583): Difference between revisions

The authors present a new iterative scheme for the high-precision computation of the integral \[ F(a,b) =\int_{-\infty}^{\infty} \frac{b_0 x^{p-2} + \dots + b_{p-3} x + b_{p-2}} {a_0 x^p + \dots + a_{p-1} x + a_p}\;dx \] with \(a = (a_0, a_1, \dots, a_p)\), \(b = (b_0, b_1, \dots, b_{p-2})\). The rational Landen transformations \({\mathcal L}_{m,p}:\, {\mathbb C}^{2p}\to {\mathbb C}^{2p}\) with \((a,b) \mapsto {\mathcal L}_{m,p}(a,b)\), which preserve the value \(F(a,b)\), are constructed. Iteration of \({\mathcal L}_{m,p}\) yields a sequence of coefficients \((a_{n,0}, a_{n,1}, \dots, a_{n,p})\) and \((b_{n,0}, b_{n,1}, \dots, b_{n,p-2})\) for a sequence of rational functions, which converges to \(c\, (1+x^2)^{-1}\) with a constant \(c\). Then the invariance of \(F(a,b)\) under \({\mathcal L}_{m,p}\) yields \[ F(a,b) = \pi \, c = \pi \, \lim_{n\to \infty} \frac{b_{n,0}}{a_{n,0}}\,, \] where the convergence is of order \(m\). Numerical examples are presented. A Mathematica package of this method is available on a website.