DISTRIBUTED APPROACHES TO MULTIDISCIPLINARY OPTIMIZATION MANAGEMENT

This work is concerned with analytical and computational foundations of multidisciplinary optimization. We analyze several distributed, multilevel approaches to formulating MDO problems. We show how, in some cases, analytical features of the resulting optimization
problems have deleterious consequences for computational optimization algorithms. These difficulties derive from the multilevel nature of the optimization problems. Multilevel approaches to MDO maintain a disciplinary autonomy through the formulation. We claim that
the same practical degree of autonomy can be attained through the choice of optimization algorithm applied to more conventional formulations, without the inherent computational difficulties associated with distributed optimization formulations. Techniques that lead to complicated, structure-based formulations are contrasted with algorithm-based techniques applied to simpler problem formulations.