Abstract: In nonparametric regression estimation the best possible rate of convergence is attained when the design points are equidistributed. Whole sequences of equidistributed designs can be generated from vectors with algebraically independent components. Generators are determined which perform well already for moderate numbers of observations. The proposed design sequences are nearly optimal in classical settings and simultaneously appropriate for the nonparametric approach.
Key words and phrases: Additive regression, equidistributed sequence, lattice design, nonparametric regression, optimal design.