Pseudospectral Fourier reconstruction with the modified inverse polynomial reconstruction method (Q2655682)

The inverse polynomial reconstruction method (IPRM) was introduced by \textit{J.-H. Jung} and \textit{B. D. Shizgal} [J. Comput. Appl. Math. 172, No.~1, 131--151 (2004; Zbl 1053.65102)] in order to remedy the Gibbs phenomenon. Let \(f\) be a piecewise polynomial function defined on \([-1,\,1]\) and let \(m,\,n\in \mathbb N\) with \(m\geq n\) be given. In this paper, a modified IPRM is proposed that approximates \(f\) by a polynomial \(p(x) = \sum_{l=0}^{n-1} a_l\,P_l(x)\) with \(x\in [-1,\,1]\) such that \[ \sum_{k=-\lfloor (m-1)/2\rfloor}^{\lfloor m/2\rfloor} |{\hat f}(k) - {\hat p}(k)|^2 \] is minimal, where \(P_l\) are the normalized Legendre polynomials and \({\hat f}(k)\) are the Fourier coefficients of \(f\). The modified IPRM finds a truncated Legendre series of the given function \(f\) from its truncated Fourier series by solving a rectangular least squares problem. If \(m\geq n^2\), the authors show that the condition number of this least squares problem is small and that the convergence rate for an analytic function \(f\) is root exponential on \([-1,\,1]\). Numerical stability and accuracy of the proposed IPRM algorithm are validated experimentally.