Nikishin systems are perfect (Q662811)

The notion of a `perfect system' goes back to \textit{K. Mahler} [Compos. Math. 19, 95--166 (1968; Zbl 0168.31303)]; it can be formulated in the `point of development at \(\infty\) approach' as follows. Let \({\mathbf f}=(f_0,\ldots,f_m)\) be a family of analytic functions in some domain \(D\) of the extended complex plane containing \(\infty\). Fix a nonzero multi-index \({\mathbf n}=(n_0,\ldots,n_m)\in{\mathbf Z}_{+}^{m+1}\), \(|{\mathbf n}|=n_0+\cdots+n_m\). There exist polynomials \(a_{{\mathbf n},0},\ldots,a_{{\mathbf n},m}\) ( type I polynomials), not all equal to zero, such that \begin{align*} &\operatorname{deg}a_{{\mathbf n},j}\leq n_j-1,\quad j=0,\ldots,m\quad (\operatorname{deg}a_{{\mathbf n},j}\leq -1\text{ means that } a_{{\mathbf n},j}\equiv 0),\tag{i}\\ &\sum_{j=0}^m\,a_{{\mathbf n},j}f_j(z)-d_{\mathbf n}(z)=O\bigg(\frac{1}{z^{|{\mathbf n}|}}\bigg),\quad z\rightarrow\infty,\tag{ii} \end{align*} for some polynomial \(d_{\mathbf n}\). Analogously, there exist \(Q_{\mathbf n}\) ( type II polynomials), not identically equal to zero, such that \begin{align*} &\operatorname{deg}Q_{\mathbf n}\leq|{\mathbf n}|,\tag{i}\\ &Q_{\mathbf n}(z)f_j(z)-P_{{\mathbf n},j}(z)=O\bigg(\frac{1}{z^{n_j+1}}\bigg),\quad z\rightarrow\infty,\quad j=0,\ldots,m,\tag{ii} \end{align*} Definition 1.1. A multi-index \(\mathbf n\) is said to be normal for the system \(\mathbf f\) for type I approximation (respectively, for type II) if \(\operatorname{deg}a_{{\mathbf n},j}=n_j-1\), \( j=0,\ldots,m\), (respectively, \(\operatorname{deg}Q_{\mathbf n}=|{\mathbf n}|\)). A system \(\mathbf f\) is said to be perfect if all multi-indices are normal. There are not many systems to be known perfect for both types of approximation (see the paper under review for some historical remarks), and the authors add a large class of new systems that are perfect with respect to both types: Nikishin systems (introduced by \textit{E. M. Nikishin}, [Mat. Sb., N. Ser. 113(155), 499--519 (1980; Zbl 0456.30009) Russian; english translation in [Math. USSR, Sb. 41, 409--425 (1982; Zbl 0478.30007)]). Definition 1.2. Take a collection \(\Delta_j\), \(j=0,\ldots m\), of intervals such that \[ \Delta_j\cap \Delta_{j+1}=\emptyset,\quad j=0,\ldots,m-1. \] Let \((\sigma_0,\ldots,\sigma_m)\) be a system of measures such that \(\operatorname{supp} \sigma_j=\Delta_j\), \(\sigma_j\in{\mathcal M}(\Delta_j)\), \( j=0,\ldots,m\). We say that \((s_0,\ldots,s_m)={\mathcal N}(\sigma_0,\ldots,\sigma_m)\), where \[ s_0=\sigma_0,\quad s_1=\big\langle\sigma_0,\quad \sigma_1\big\rangle,\;\ldots\;,s_m=\big\langle\sigma_0,\langle\sigma_1,\ldots,\sigma_m\rangle\big\rangle \] is the Nikishin system of measures generated by \((\sigma_0,\ldots,\sigma_m)\). The Nikishin system of functions is given by \((\widehat s_0,\ldots,\widehat s_m)\), where \(\widehat{s}_j\) is the Stieltjes transform \[ \widehat s_j(z)=\int\,{ds(x)\over z-x},\quad j=0,\ldots,m. \] Moreover, the construction \(\langle\cdot,\cdot\rangle\) is defined by (in differential form) \[ d\langle\sigma_{\alpha},\sigma_{\beta}\rangle(x)=\widehat{\sigma}_{\beta}\,d\sigma_{\alpha}(x), \] i.e. the first measure is multiplied by the Cauchy transform of the second measure. The main result of the paper is now that Nikishin systems are type I and type II perfect (follows from a slightly more general result on AT-systems and mixed type systems in a matrix of Markov-type functions). Moreover, results are given on convergence of simultaneous quadrature rules, logarithmic asymptotics of sequences of multiple orthogonal polynomials and ratio asymptotics of mixed-type multiple orthogonal polynomials.