E(s2)E(s2)-Optimal supersaturated designs with good minimax properties

An improved E(s2)E(s2) lower bound is derived for two-level supersaturated designs. This improved bound is used to prove E(s2)E(s2)-optimality of the best designs obtained via algorithmic search in all cases with N=10N=10, 12, 14, and 16 runs (except the N=14N=14 run, m=16m=16 factor case). New exchange algorithms which generalize the NOA algorithm of Nguyen [1996. An algorithmic approach to constructing supersaturated designs. Technometrics 38, 69–73] and which tend to find E(s2)E(s2)-optimal designs with better minimax properties are proposed. Row swapping algorithms are used to find E(s2)E(s2)-optimal designs when the number of factors is large. E(s2)E(s2)-optimal designs found via algorithmic search are compared to cyclicly constructed E(s2)E(s2)-optimal designs using the minimax criterion.