Approximating the solution of integro-differential problems via the spectral tau method with filtering (Q2301273)

In this paper Chebyshev-Padé and Legendre-Padé approximants are proposed to filter a ``tau solution'' of some differential problems with slow rate of convergence. The computation is performed using a MATLAB numerical library called Tau Toolbox. The authors describe a new filtering process that is well suited for building rational approximations from approximate polynomial solutions delivered by the the Tau Method. This filtering can improve the accuracy of the spectral tau approximation when working in the vicinity of solutions with singularities. Some algebraic properties of the filtering process are introduced, justifying the good properties of the filtered solutions. Some numerical experiments provided in the paper reveal that it is possible to improve the tau solution approximations using Frobenius-Padé approximations. The noise introduced in the tau coefficients in the numerical computation of Padé approximants yields the occurrence of Froissart doublets for high-order rational approximants.