AbstractThis paper presents Iterative Scalable Smoothing (ISS), a new itera-
tive multi-scale method for multivariate interpolation of scattered data. Each iteration
step in the process reduces the residues of the current interpolation result by appli-
cation of a smoothing operator to a piecewise constant function that interpolates the
residues of the current interpolant, and by adding the resulting function to the current
approximation, which is initially set to zero. The convergence of the method is proved
and conditions for the di®erentiability of the convergence result are given. For a uni-
form mesh an e±cient algorithm is constructed, for which the numerical complexity is
estimated.
Several 2D numerical examples illustrate the theoretical results. By a 3D test with
several test-functions and random nodes it is shown that the accuracy of the proposed
method is comparable with the quadratic modiffcation of Shepard's method, which
is known to be more accurate than triangle-based methods. Then, in 1D tests, by
stochastic simulation with random nodes and random functions the ISS method is
compared with the Cubic Splines method, Shepard's method and the Kriging method.
We also compare the stability of these methods with respect to noisy data. For the
special case of regular nodes, properties of the method are verified by comparing its
1D response function with the response function of the cubic spline and the perfect
interpolator (the sinc-function).
Special attention is paid to the e®ect of the tuning parameter