For a fixed, continuous, periodic kernel K, an sk-spline is a function of t
he form sk(x) = c(o) + Sigma(i=1)(n) c(i)K(x - x(i)). In this paper we cons
ider a generalization of the univariate sk-spline to the d-dimensional toru
s (d greater than or equal to 2), and give almost optimal error estimates o
f the same order, in power scale, as best trigonometric approximation on So
bolev's classes in L-q. An important component of our method is that the in
terpolation nodes are generated using number theoretic ideas. (C) 1999 Acad
emic Press.