Multivariate Rational Interpolation of Scattered Data

Abstract. Rational data fitting has proved extremely useful in a number of scientific applications. We refer among others to the computation of cell loss probabilities in network traffic [5, 6, 14, 15], to the modelling of electro-magnetic components [20, 12] to model reduction of linear shift-invariant systems [2, 3, 7] and so on. When computing a rational interpolant in one variable, all existing techniques deliver the same rational function, because all rational functions that satisfy the interpolation conditions reduce to the same unique irreducible form. When switching from one to many variables, the situation is entirely different. Not only does one have a large choice of multivariate rational functions, but moreover, different algorithms yield different rational interpolants and apply to different situations. The rational interpolation of function values that are given at a set of points lying on a multidimensional grid, has extensively been dealt with in [10, 9, 8]. The case where the interpolation data are scattered in the multivariate space, is far less discussed and is the subject of this paper. We repeat the multivariate rational interpolation problem under consideration and present a fast solver for the linear block CauchyVandermonde system. In the numerical illustration of the technique, the fast solver is combined with an interval arithmetic verification step.