Abstract: have been widely used in various fields. Traditionally they were generated by so-called good lattice point (glp) sets. In this paper, we propose a new approach based on cyclic Latin squares (CLS). The UDs generated have much smaller discrepancies than those proposed by Wang and Fang (1981). A threshold accepting algorithm is employed for finding the CLS with smallest -discrepancy.
Key words and phrases: Experimental design, Latin squares, -discrepancy, threshold accepting algorithm, uniform design, uniformity.