An optimal multivariate spline method of recovery of twice differentiable functions (Q639967)

The problem of optimal recovery of a function based on incomplete information is addressed. In the case of a univariate function, the optimal recovery problem is solved by univariate polynomial splines. For the multivariate case, there are studied some multivariate polynomial spline spaces over triangulations. For example, given the function values at the vertices of a multidimensional simplex, a solution to the problem of optimal recovery of twice differentiable functions is given by a linear polynomial which interpolates the function at the vertices of the simplex. In this paper, the authors find a spline-based method for the optimal recovery problem for the class of functions defined on a convex \(d\)-dimensional polytope whose second derivatives in any direction are uniformly bounded and for the periodic analogue of this class over a given full-rank lattice. The information consists in the values and gradients of function at some set of nodes in the real \(d\)-dimensional space. The method of optimal recovery obtained by the authors is given by a continuous quadratic interpolating spline over a Delaunay triangulation of the nodes sets in \(d\)-dimensional real space. In the end of the paper, the particular case of bivariate functions and the proposed spline defined over the hexagonal partition of bidimensional real space is also presented.