Error estimate for a corrected Clenshaw-Curtis quadrature rule

From MaRDI portal
Publication:2340901





Let \(I(f)=\int_{-1}^1 f(t)dt\). This paper presents a new quadrature formula \(I(f)-Q_n(f)=E_n(f)\) that represents an improved version of the well-known Clenshaw-Curtis (C-C) rule ( cf. [\textit{C. W. Clenshaw} and \textit{A. R. Curtis}, Numer. Math. 2, 197--205 (1960; Zbl 0093.14006)]). The connection between the new rule and the C-C rule can be appraised by comparing the respective node polynomials. The \(n\)th node polynomial corresponding to the C-C rule is \((x^2-1)U_{n-2}(x)\), while the proposed rule (named endpoint-corrected rule) is related to the node polynomial \((x^2-\beta^2)(x^2-1)U_{n-4}\), where in both cases \(U_k\) is the Chebyshev polynomial of the second kind with degree \(k\), and \(\pm \beta=\pm \cos(\pi/(2n-6))\). For the latter a family of nested rules is obtained by considering the subsequence \(\{Q_{3+2^k}(f)\}\). The endpoint-corrected rule is also compared with the Fejér rules of the first and second kind and the Basu rule. The following result refers to its asymptotic behavior. If \(f\) is analytic on a region \(V\supset [-1,1]\) and \(M_1\), \(M_2\) stand for constants, then the quadrature error \(E_n(f)\) is bounded by \(M_1C(r)n^{-5}r^n+\mathcal{O}(n^{-6}r^n)\), as \(n\to \infty\), where \(0<r<1\) and \(C(r)\leq M_2 /(1-r^2)^4\). Indeed, an explicit formula for \(C(r)\) is given. The weights of the endpoint-corrected rule are explicitly found for \(n\geq 6\), and it is also proved that they are asymptotically positive. As for the convergence rate, a conclusion is that the new endpoint-corrected rule is superior to all rules above mentioned.











This page was built for publication: Error estimate for a corrected Clenshaw-Curtis quadrature rule

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2340901)