On the convergence of two-point partial Padé approximants for meromorphic functions of Stieltjes type (Q1775614)

This paper is to a certain extent a continuation of a paper by the same authors in [Numer. Math. 72, 295--312 (1996; Zbl 0854.41018)]. Let \(\mu\) be a complex measure supported on \( {\mathbb R}^{+}=[0,\infty)\), with all moments finite (\(\int x^kd| \mu| (x)<\infty,\;k\in {\mathbb Z}\)) and let its Stieltjes-transform be given by \[ \widehat{\mu}(z)=\int_0^{\infty}\,{d\mu(x)\over z-x}, \] holomorphic on \({\mathbb D}={\mathbb C}\setminus{\mathbb R}^{+}\). Let \(r=P/Q\) be a given rational function with denominator \(Q\) of exact degree \(m\) and numerator \(P\) of degree \(m-1\) at most (coprime); denote the set of poles of \(r\) by \(\mathcal P\) and assume \({\mathcal P}\subset {\mathbb D}\). The function to be approximated is defined by \[ f(z)=\widehat{\mu}(z)+r(z),\quad z\in {\mathbb C}\setminus (\mathbb R^{+}\cup{\mathcal P}). \] Padé approximation and convergence of this type of function has been studied by \textit{A. A. Gonchar} [Math. USSR Sb. 26, 555--575 (1975); translation from Mat. Sb., n. Ser. 97(139), 607--629 (1975; Zbl 0341.30029)]. The aim of the paper is to give a de Montessus de Ballore type theorem for so called two-point partial Padé approximants (2PA for short), to be denoted by \(R_n(f)\) and defined as follows. Let \(\{m(n)\}\) and \(\{p(n)\}\) (with \(n\in {\mathbb N}\)) be two sequences of non-negative integers satisfying \(0\leq m(n)\leq n,\;0\leq p(n)\leq n+m(n)\) and let \(\{T_k\}\) be a given sequence of polynomials with \(T_k\) of exact degree \(k\) and \(T_k(0)\not= 0\). Then for each positive integer \(n\) there exist two polynomials \(Q_{m(n)}\) and \(P_{n-1}\) satisfying \[ Q_{m(n)}\in {\mathbb P}_{m(n)},\;Q_{m(n)}\not\equiv 0\quad \text{and}\quad P_{n-1}\in {\mathbb P}_{n-1}, \] \[ {Q_{m(n)}(z)T_{n-m(n)}(z)f(z)-P_{n-1}(z)\over z^{p(n)}}=\begin{cases} O(1),&z\rightarrow 0,\;z<0,\\O\left({1\over z^{m(n)+1}}\right), &z\rightarrow \infty,\;z<0. \end{cases} \] The 2PA are then the rational functions \[ R_n(f)={P_{n-1}(z)\over Q_{m(n)}(z)T_{n-m(n)}}, \] with preassigned poles at the zeros of the sequence \(\{T_{n-m(n)}\}\); \({\mathbb P}_k\) denotes the set of the polynomials of degree at most \(k\). The main result is a three quarter page theorem with a half page corollary on the situation where the prescribed poles are constructed from the orthonormal polynomials \(L_k\) connected with \[ w(t)={e^{-\delta t^{\gamma}}\over \sqrt{t}}\;(\delta>0,\,\gamma>\tfrac12), \] with \[ \int_0^{\infty}\,t^Ne^{\delta(t^{\gamma}+1/t^{\gamma})}\,{d| \mu| (t)\over 1+t^d}<\infty\quad (d\;\text{a positive integer}), \] for some \(N\in {\mathbb Z}\). With an extra condition \[ \lim_{n\rightarrow\infty}\,{p(n)\over n}=\theta\in [0,1], \] and definitions \[ \begin{aligned} k(n)&=p(n)-p(m),\quad 0\leq k(n)\leq n-m, \\ \ell (n)&=n-m-k(n),\quad n>m, \end{aligned} \] the poles follow from \[ T_{n-m}=\widetilde{L}_{k(n)} L_{\ell (n)}\quad (\widetilde{L}_{k(n)}(z)=z^{k(n)} L_{k(n)}(1/z)). \] The results for functions \(f\) having \(m\) poles in \({\mathbb D}\) are of the form 1. Each pole of \(f\) in \({\mathbb D}\) attracts as many poles of \(R_n(f)\) as its order for \(n\rightarrow\infty\), 2. On compact subsets \(S\subset{\mathbb C}\setminus (\mathbb {R}^{+}\cup{\mathcal P})\) \[ \limsup_{n\rightarrow\infty}\,\| f-R_n(f)\| _S^{1/(2n)^{1-1/2\gamma}}\leq e^{-\| \Phi\| _S}, \] with \(\Phi\) given explicitly. The proofs use results due to Rakhmanov, Levin-Lubinsky, Lubinsky-Mhaskar-Saff and Gonchar.