Numerical procedures in multiobjective optimization with variable ordering structures (Q467448)

Multiobjective optimization problems were usually considered with the linear space \(\mathbb{R}^{m}\) being partially ordered by a convex cone. Replacing this partial ordering by other binary relations allows the treatment of multiobjective optimization problems where the cone of preferred or dominated directions depends on the actual element. This is mathematically modeled by a set-valued map which associates with each element a pointed and convex cone. Based on this map, which is called the ordering map, two binary relations can be defined depending on the choice of the cone w.r.t. which the elements are compared.\newline The author shows that both binary relations are in general not transitive without additional assumptions. The loss of transitivity is the main drawback when trying to generalize known numerical methods developed in partially ordered spaces to multiobjective optimization problems with a variable ordering structure. The author presents a necessary condition for optimality as well as scalarization results which can be combined to a numerical algorithm for continuous multiobjective optimization problems with a variable ordering structure. Additionally, the author adapts an algorithm for discrete problems in partially ordered spaces, the Jahn-Graef-Younes method, to variable ordering structures. The algorithms are applied to some test instances derived from known test instances for multiobjective optimization problems in partially ordered spaces.