Disjunctive Aspects in Generalized Semi-infinite Programming

In this thesis the close relationship between generalized semi-infinite problems (GSIP) and disjunctive problems (DP) is considered. We start with the description of some optimization problems from timber industry and illustrate how GSIPs and DPs arise naturally in that field. Three different applications are reviewed. Next, theory and solution methods for both types of problems are examined. We describe a new possibility to model disjunctive optimization problems as generalized semi-infinite programs. Applying existing lower level reformulations for the obtained semi-infinite program we derive conjunctive nonlinear problems without any logical expressions, which can be locally solved by standard nonlinear solvers. In addition to this local solution procedure we propose a new branch-and-bound framework for global optimization of disjunctive programs. In contrast to the widely used reformulation as a mixed-integer program, we compute the lower bounds and evaluate the logical expression in one step. Thus, we reduce the size of the problem and work exclusively with continuous variables, which is computationally advantageous. In contrast to existing methods in disjunctive programming, none of our approaches expects any special formulation of the underlying logical expression. Where applicable, under slightly stronger assumptions, even the use of negations and implications is allowed. Our preliminary numerical results show that both procedures, the reformulation technique as well as the branch-and-bound algorithm, are reasonable methods to solve disjunctive optimization problems locally and globally, respectively. In the last part of this thesis we propose a new branch-and-bound algorithm for global minimization of box-constrained generalized semi-infinite programs. It treats the inherent disjunctive structure of these problems by tailored lower bounding procedures. Three different possibilities are examined. The first one relies on standard lower bounding procedures from conjunctive global optimization. The second and the third alternative are based on linearization techniques by which we derive linear disjunctive relaxations of the considered sub-problems. Solving these by either mixed-integer linear reformulations or, alternatively, by disjunctive linear programming techniques yields two additional possibilities. Our numerical results on standard test problems with these three lower bounding procedures show the merits of our approach.