Definition and uniqueness of integral approximants (Q805955)

The authors study a generalization of the ordinary Padé approximant in the following setting. Consider a functional element f(z)\(\not\equiv 0\) (a partial representation of a locally analytic function, here the Taylor series about 0 with positive radius of convergence), non-negative integers k, L and a \((k+1)\)-tuple \(\vec m=(m_ 0,m_ 1,...,m_ k)\) of integers \(\geq -1\). The problem to find polynomials \(P_ L\), \(\vec Q_ L=(Q_{0,L},Q_{1,L},...,Q_{k,L})\) of degrees at most L, \(\vec m\) respectively, satisfying \[ \sum^{k}_{j=0}f^{(j)}(z)Q_{j,L}(z)- P_ L(z)={\mathcal O}(z^{L+M+1}),\quad M=\sum^{k}_{j=0}(m_ j+1)-1. \] There always exists a non-trivial solution [the so called Latin polynomials; cf. \textit{K. Mahler}, Compositio Math. 19, 95-166 (1968; Zbl 0168.313)]; the authors coin the phrase ``integral polynomials of type (L,\(\vec m)''\). The integral approximant of type (L,\(\vec m)\) is then the solution y(z) of \(\sum^{k}_{j=0}y^{(j)}(z)Q_{j,L}(z)-P_ L(z)=0,\) subject to boundary conditions (for instance \(y^{(j)}(0)=f^{(j)}(0)\), \(j=0,1,...,k)\). The authors give a thorough treatment of this type of approximation and their results include uniqueness of the polynomials (if the degrees of the Q's are minimized, starting with \(Q_{k,L}\) and lowering the index k), existence of infinite subsequences of minimal or reduced minimal (common factors of the form \(z^ j\) divided out) diagonal integral polynomials (i.e. \(L=N\), \(m_ j=N)\), the existence of an infinite subsequence of minimal or reduced minimal integral polynomials for any connected increasing sequence of indices (a natural generalization of staircase sequences, including diagonal and partial diagonal sequences) and accurate-through- order properties for the integral approximants. This paper once and for all puts the subject on firm ground; most importantly it shows the existence of ``enough'' integral approximants to make convergence investigations worthwile.