Problem Formulation for MultidisciplinaryOptimizationEvin

This paper is concerned with the optimization of systems of coupled simulations. In computational engineering, this frequently is called multidisciplinary (design) optimization , or MDO. We present an expository introduction to MDO for optimization researchers. We believe the optimization community has much to contribute to this important class of computational engineering problems. In addition, this paper presents a new abstraction for multidisciplinary analysis and design problems as well as new decomposition formulations for these problems. Furthermore, the \individual discipline feasible" (IDF) approaches suggested here make use of existing specialized simulation analysis codes, and they introduce signiicant opportunities for coarse-grained computational parallelism particularly well-suited to heterogeneous computing environments. The key issue in the three fundamental approaches to MDO formulation discussed here is the kind of feasibility that must be maintained at each optimization iteration. In the most familiar \multidisciplinary feasible" (MDF) approach, the multidisciplinary analysis problem is solved multiple times at each optimization iteration, at least once everytime any problem function or constraint or derivative is evaluated. At the other end of the spectrum is the \all-at-once" (AAO) approach where feasibility is guaranteed only at optimization convergence. Between these extremes lie the new IDF formulations that amount to maintaining feasibility of the individual analysis disciplines at each optimization iteration, while allowing the optimizer to drive the computation toward multidisciplinary feasibility as convergence is approached. There are further considerations in choosing a formulation, such as what sensitivities are required and how the optimization is actually done. Our discussion of these and other issues related to MDO problem formulation highlights the trade-oos between reuse of existing software, computational requirements, and probability of success.