Interpolatory quadrature rules for oscillatory integrals

TL;DR

This paper develops a new class of quadrature rules for oscillatory integrals that adapt to frequency, combining exponential fitting and Filon-type methods for improved accuracy across all frequency ranges.

Contribution

It introduces a novel approach to construct frequency-dependent quadrature nodes using S-shaped functions, enhancing efficiency and accuracy for oscillatory integrals.

Findings

New quadrature rules with optimal error decay for high frequencies

Effective for small, moderate, and large frequencies

Suitable for implementation in automatic software packages

Abstract

In this paper we revisit some quadrature methods for highly oscillatory integrals of the form $\int_{-1}^1f(x)e^{{\rm i}\omega x}dx, \omega>0$. Exponentially Fitted (EF) rules depend on frequency dependent nodes which start off at the Gauss-Legendre nodes when the frequency is zero and end up at the endpoints of the integral when the frequency tends to infinity. This makes the rules well suited for small as well as for large frequencies. However, the computation of the EF nodes is expensive due to iteration and ill-conditioning. This issue can be resolved by making the connection with Filon-type rules. By introducing some $S$-shaped functions, we show how Gauss-type rules with frequency dependent nodes can be constructed, which have an optimal asymptotic rate of decay of the error with increasing frequency and which are effective also for small or moderate frequencies. These frequency-dependent nodes can also be included into Filon-Clenshaw-Curtis rules to form a class of methods which is particularly well suited to be implemented in an automatic software package.