A comparison of alternative input models for synthetic optimization problems

We analyze two strategies for randomly generating optimization test problems with two types of coefficients. One strategy is to generate test problems with independent coefficients; the other strategy is to generate test problems with induced correlation between the coefficient types. We discuss the likely effect of test problem size, i.e., the number of decision variables, on the sample correlations among the test problem coefficients generated with each strategy. We also propose some guidelines for experimenters based on our analysis.