Composite schemes for multivariate blending rational interpolation (Q1612371)

For bivariate rational interpolation, a blending interpolant for a number of grid points can be obtained that is either a Newton interpolating polynomial in one variable with coefficients that are Thiele fractions in the other variable (TN interpolant) or that is a Thiele fraction in one variable with coefficients that are Newton polynomials in the other variable (NT approximant). If a square grid of data points is given, one can take a TN interpolant for the lower triangular part and an NT interpolant for the upper triangle. An interpolant for the lower triangular part is combined with an approximant for the upper triangular part with appropriate polynomial coefficients to form an interpolant for the whole grid. For the latter rational interpolants, the degree structure and an error bound is derived. By using the Samelson inverse of a vector, these techniques can be extended to the case of vector valued rational interpolation and by vectorization, it is further extended to matrix valued functions. The latter generalizations are however not worked out in any detail.