Spectral collocation methods and polar coordinate singularities (Q808197)

The paper considers the spectral collocation method for the solution of elliptic differential equations on the unit disk. Difficulties arise here since the polar coordinates behave singular at the origin and hence, some of the trial functions are not in the differential operator's domain of definition in the classical sense. The problem is solved by replacing the collocation equations at the origin by a collocation equation for the smooth part of the numerical solution and a continuity condition. Another possibility is to use the smooth trial functions only and a least squares method for the resulting overdetermined system of equations. For separable differential equations the numerical solution can be computed within \({\mathcal O}(N^ 2\log N)\) operations. Numerical results for the Poisson and Helmholtz equations are presented which indicate exponential decay of the considered discrete error.