New formulas for approximation of \(\pi\) and other transcendental numbers (Q1587024)

Ways of producing iterative methods of given order for solving equations have been around for a long time [see e.g. pp. 333-335 in \textit{A. Ralston}, A first course in numerical analysis (McGraw-Hill) (1965; Zbl 0139.31603)]. The present author uses, to solve \(f(x)=0\), determinants \(D_m(x)\) of \(m\times m\) matrices \(d_{ij}\) where \(d_{ij}= f^{(j+1-i)}(x)/ (j+1-i)\) for \(j\geq i-1\) and \(d_{ij}=0\) if \(j< i-1\). (A formula is given that enables the \(D_m(x)\) to be calculated recursively.) Writing \(B_m(x)= x-f(x) D_{m-1}(x)/ D_{m-2}(x)\) we have that \(x_{k+1}= B_m(x_k)\) is an iterative procedure of order \(m\). The case \(m=2\) is Newton's method, and \(m=3\) is Halley's method. Suitable choices of \(f\) and \(x_0\) enable approximations to be made in various fields; e.g. \(f(x)= s\) in \(x- \sqrt{3/2}\), \(x_0= \pi/4\) gives approximations to \(\pi\) in \(\mathbb{Q}(\sqrt{3})\). It is necessary in such a case to evaluate sines to more and more places at each iteration, which, as the author writes, ``may not be desirable''. An experiment with Maple is reported that gives the number of places obtained by using the procedure with various values of \(k\) and unspecified \(m\). However no run-times are given and so one cannot judge how effective the procedure is compared with earlier ones.