On the computation of Patterson-type quadrature rules (Q2667127)

The Gauss-Kronrod formula is a \((2n+1)\)-point interpolating quadrature formula for \(I[f]=\int_{-1}^1 f(x)d\mu(x)\). It has \(n\) Gauss nodes (zeros of orthogonal polynomial \(p_n\)), and the remaining \(n+1\) nodes are zeros of a the Stieltjes polynomial \(Q_{n+1}\) (orthogonal to \(p_n(x) x^k\), \(k=0,\ldots,n\)) [\textit{D. Calvetti} et al., Math. Comput. 69, No. 231, 1035--1052 (2000; Zbl 0947.65022)]. In Patterson quadrature, \(r\) of the latter nodes are fixed as zeros of a given polynomial \(q_r\), and the remaining ones are zeros of \(Q_{n+1-r}\) orthogonal to \(x^k q_r(x)p_n(x)\), \(k=0,\ldots,n-r\). The authors analyze the location of the zeros and the positivity of the weights for Radau and Lobatto versions of Patterson quadrature, i.e., when \(q_r\) has one or two zeros at the endpoints \(\pm1\) and when \(d\mu(x)=g(x)(x+1)^\alpha(x-1)^\beta dx\), \(\alpha,\beta\in\{\pm1/2\}\) with a positive \(g\in\mathrm{Lip}_s([-1,1])\), for \(s\in(0,1)\). The properties of the nodes and the numerical computation of the quadrature is essentially based on an analysis of the Jacobi matrix for these nodal polynomials.