Efficient computation of oscillatory integrals via adaptive multiscale local Fourier bases (Q1577560)

The authors considers the problem of the fast and accurate evaluation for the integral\break \(\int^L_0 K(t,s)f(s) ds\), where \(f\) is a smooth function and \(K(t,s)\) is a highly oscillatory kernel of the type: \(K(t,s)=e^{i\nu\phi(t,s)};\;\nu\) is up to \(2000\). This integral is accurately evaluated with an improved trapezoidal rule and effectively transcribed using a local Fourier basis and an adaptive multiscale local Fourier basis. A representation of the oscillatory kernel in these bases is sparse. The comparison of the sparsities for the coefficient matrices obtained by the ordinary and the multiscale transforms is presented. In the conclusion a further development of this method is discussed.