Approximate optimal designs for multivariate polynomial regression (Q1731748)

Starting point of this paper is a classical problem of optimal experimental design for polynomial regression models. The difficulties arise because multivariate polynomial regression is considered and, especially, because the proposed design space is a compact semialgebraic manifold defined by the nonnegativity of suitable polynomials. As in classical theory, the basic tool is an equivalence theorem. To solve the approximate optimal design problem in detail, the moment-sum-of-squares hierarchy of semidefinite programming is used. For some insight in the geometry of designs, duality theory and Christoffel polynomials are involved in the optimality conditions. The authors show further that approximate optimal designs can be computed numerically with their methods. So the paper is illustrated by some interesting more dimensional examples.