Abstract: The concept of universal optimality from optimum design theory is introduced into computer experiments, modeled as realizations of stationary Gaussian processes. When the correlation function is a nondecreasing and convex function of a distance measure, it is shown that a design is universally optimal if it is equidistant and of maximum average distance. Examples of universally optimal designs are given with respect to rectangular, Euclidean, Hamming, and Lee distances.
Key words and phrases: Computer experiments, Hamming distance, Lee distance, orthogonal arrays, universally optimal designs.