An algorithm for the computation of zeros of a special class of entire functions (Q1062726): Difference between revisions

The authors discuss an algorithm for finding numerically zeros \(\alpha_ k\) of a class of entire functions \[ f(z)=e^{Q(z)}\prod^{\infty}_{k=1}(1-z/\alpha_ k)\exp (z/\alpha_ k+z^ 2/2\alpha^ 2_ k+...+z^ p/p\alpha^ p_ k)=\sum^{\infty}_{h=0}(-1)^ ha_ hz^ h \] where the \(\alpha_ k\) are all real, \(0<\alpha_ 1\leq \alpha_ 2\leq \alpha_ 3\leq...,\sum^{\infty}_{j=1}\alpha_ j^{-p-1}\) is convergent, f(z) is real for real z, with \(f(0)=1\), and Q(z) is a polyonomial of degree \(\leq p\) with real coefficients, with \(Q(0)=0\). The coefficients \(a_ h\) are then all real, with \(a_ 0=1\). Based on an algorithm developed by \textit{G. Pólya} [Numer. Math. 11, 315-319 (1968; Zbl 0191.159)] using Graeffe's method, they construct a sequence of upper and lower bounds for the zeros \(\alpha_ k\). In view of the considerable numerical difficulties involved in this method, a modification of the algorithm proposed by \textit{A. A. Grau} [J. Assoc. Comput. Machin. 10, 538-544 (1963; Zbl 0163.387)] is introduced and described in detail. The bounds so obtained are used as starting values for a so-called Laguerre type iteration using a cubic polynomial. The corresponding computer program is given in a non-standard FORTRAN 77 language, together with some organizational details concerning precision and storage. Computational results are discussed extensively, and numerical tables of the first 30 zeros of \(f(x)=J_ 2(x)\) (the Bessel function) and of the first 20 zeros of \(f(x)=\cos x \cosh x-1\) are given, with 14 and 17 digits, respectively.