How poles of orthogonal rational functions affect their Christoffel functions (Q714881): Difference between revisions

The paper under review studies the asymptotic behavior of Christoffel functions associated with orthogonal rational functions (for a thorough treatment of this subject cf. the monograph by \textit{A. Bultheel} et al. [Orthogonal Rational Functions. Cambridge Monographs on Applied and Computational Mathematics. 5. Cambridge: Cambridge University Press (1999; Zbl 0923.42017)]). Assume that a sequence of extended complex numbers, \[ A=\{\alpha_1,\alpha_2,\alpha_3,\dots\}\subset\bar{\mathbb{C}}, \] is given to serve as poles, set \[ \pi_0(x)=1,\;\pi_k(x)=\prod_{j=1}^k\,\left(1-{x\over \alpha_j}\right)\;(k\geq 1). \] \(\mathcal P\) denotes the set of polynomials of degree \(\leq k\) and the nested spaces of rational functions are defined by \[ {\mathcal L}_k={\mathcal L}_k\{\alpha_1,\alpha_2,\dotsc,\alpha_k\}=\left\{{P\over\pi_k}\,:\,\deg(P)\leq k\right\} \] (\({\mathcal L}_{k-1}\subset{\mathcal L}_k\) for \(k\geq 1\); if all \(\alpha_j=\infty\), then \({\mathcal L}_k={\mathcal P}\)). The orthonormal functions \(\varphi_0,\varphi_1,\dots\) with respect to \(\mu\) are defined by \[ \varphi_k\in{\mathcal L}_k\setminus{\mathcal L}_{k-1},\;\int_{-1}^1\,\varphi_j\overline{\varphi}_kd\mu=\delta_{jk}, \] and the rational Christoffel functions are given by \[ \lambda_n^r(d\mu,x)=\left(\sum_{j=0}^{n-1}\,|\varphi_j(x)|^2\right)^{-1}. \] In a recent paper [Anal. Appl., Singap. 10, No. 3, 271--294 (2012; Zbl 1250.42097)], the authors proved a result on the asymptotic behavior of the rational Christoffel functions \[ \lim_{n\rightarrow\infty} n\lambda_n^r(x)=\mu'(x)\pi\sqrt{1-x^2}/\int\operatorname {Re}\left\{{\sqrt{t^2-1}\over t-1}\right\} d\nu(t),\tag{1} \] where the poles have an asymptotic distribution \(\nu\) with support in \(\bar{\mathbb{C}}\setminus [-1,1]\), that arises as the weak \({}^{\ast}\)-limit of the pole counting measures \[ \nu_n={1\over n}\left(\delta_{\infty}+\sum_{j=0}^{n-1}\right)\,\delta_{\alpha_j}\buildrel\ast\over\rightarrow \nu\;\text{as } n\rightarrow\infty, \] under the restriction \[ \operatorname{dist}(\alpha_j,[-1,1])\geq \eta,\;j\geq 1, \] for some \(\eta>0\) (i.e., the poles keep their distance from the interval of orthogonality). After these definitions (and a discussion of the results in the case of ordinary orthogonal polynomials), the authors come to the main results of the paper: results showing that the distance-condition on the poles plays an important role in ascertaining result (1). Theorem 1.2. Let \(\mu\) be the Chebyshev measure of the second kind \[ \mu'(x)=\sqrt{1-x^2},\;x\in (-1,1), \] and let \(\nu\) be a measure supported in \(\bar{\mathbb{C}}\). Then we may choose a sequence of poles \(\{\alpha_j\}\) in \(\bar{\mathbb{C}}\setminus [-1,1]\) that have asymptotic distribution \(\nu\), but such that for all \(x\in (-1,1)\) \[ \liminf_{n\rightarrow\infty}\,n\lambda_n^r(d\mu,x)=0. \] Theorem 1.3. Let \(\mu\) be as in Theorem 1.2 and let \(\{\eta_j\}\) be a sequence of positive numbers with limit \(0\), \(S\) is a countable set in \((-1,1)\). If \(\nu\) is any measure with support in \(\bar{\mathbb{C}}\), we may choose a sequence of poles \(\{\alpha_j\}\) in \(\bar{\mathbb{C}}\setminus [-1,1]\) that have asymptotic distribution \(\nu\), with distance condition \[ \operatorname{dist}(\alpha_j,[-1,1])\geq \eta_j,\;j\geq 1, \] such that for all \(x\in (-1,1)\) \[ \liminf_{n\rightarrow\infty}\,n\lambda_n^r(d\mu,x)=0. \] So, in order to find a result as in \((1)\), it is necessary to impose a `better' restriction on the distance between the poles and the interval \([-1,1]\): Theorem 1.4. Let \(\mu\) be as in Theorem 1.2, and let \(\nu\) be a measure supported in \(\tilde{\mathbf{C}}\) such that \(\nu(\tilde{\mathbf{C}}\setminus [-1,1)>0\). Assume that the poles \(\{\alpha_j\}\) have asymptotic distribution \(\nu\). Fix \(x\in (-1,1)\) and assume that, given \(\varepsilon>0\), there exists a \(\delta>0\) such that \[ \limsup_{n\rightarrow\infty}\,{1\over n}\,\sum_{j\leq n; |\alpha_j-x|\leq\delta}\,{|\operatorname{Im}\alpha_j|\over |x-\alpha_j|^2} <\varepsilon,\tag{2} \] then (1) holds at \(x\). Moreover, it is shown that Theorem 1.4 holds when (2) is replaced by \[ \limsup_{n\rightarrow\infty}\,{1\over n}\,\sum_{j\leq n; |\alpha_j-x|\leq\delta}\,{1\over|\text{Im}\,\alpha_j|}<\varepsilon. \] Finally, the result of Theorem 1.4 is extended to a wider class of measures: {\parindent=6mm \begin{itemize}\item[*] \(g:[-1,1]\rightarrow [0,\infty)\) measurable, \item[*] \(U\) is a polynomial such that \(gU\) and \(g^{-1}U\) are bounded in \([-1,1]\), \item[*] \(\mu\) is an absolutely continuous measure with \(\mu'(t)=g(t)\sqrt{1-t^2},\;t\in (-1,1)\) and \(\mu'\) is integrable, \item[*] the poles \(\{\alpha_j\}\) have asymptotic distribution \(\nu\), \item[*] for fixed \(x\in (-1,1)\), there exists for each \(\varepsilon>0\) a \(\delta>0\), such that (2) holds for infinitely many \(j\), \end{itemize}} then (1) holds.