Quadrature formulas on the unit circle with prescribed nodes and maximal domain of validity (Q2271980)

Let \(\mu\) be a finite positive Borel measure on \([-\pi,\pi]\). The paper studies quadrature formulas of the form \[ \int_{-\pi}^\pi f(e^{i\theta})d\mu(\theta)\approx \sum_{j=1}^m A_jf(x_j)+\sum_{k=1}^n \lambda_k f(z_k), \] where \(x_j,z_k\in \mathbb{T}:=\{z\in \mathbb{C}:|z|=1\}\), \(1\leq j\leq m\), \(1\leq k\leq n\). The nodes \(x_j\) are fixed beforehand, whereas the choice of the \(n\) points \(z_k\) is free. One expects that this kind of formula must be exact in a space \(\Lambda_{-p,p}\) of Laurent polynomials, where \(p=p(n)\) is a nonnegative integer such that \(n+E[(m-1)/2]\leq p \leq n+m-1\), and \(E[x]\) is the largest integer \(\leq x\). The paper provides a characterization for quadrature formulas which fulfil the above condition and investigates the problem of finding \(p\) as large as possible when \(m=1,2\), the two cases which correspond to Szegő-Radau and Szegő-Lobatto formulas, respectively. For the case \(m=1\), it is proved that such a formula really exists when \(p=n\) and the points \(\{z_k\}_{k=1}^n\cup\{x_1\}\) are the zeros of a para-orthogonal polynomial with respect to \(\mu\). The authors show that the case \(m=2\) can be technically reduced to a problem concerning the classical Szegő-Lobatto formulas, already solved by \textit{C. Jagels} and \textit{L. Reichel} in [J. Comput. Appl. Math. 200, No.~1, 116--126 (2007; Zbl 1109.65027)]. Several results about convergence are obtained for these rules when the integrand is continuous or analytic on \(\mathbb{T}\). Numerical examples are given to illustrate the numerical performance of these procedures.