This research program will address the high-performance parallel simulation of multiphysics problems that couple dynamic models in fluid mechanics, heat transfer, structures and control. This program requires the close collaboration of researchers in engineering, computer science and computational mathematics. The activities in Engineering are driven by the simulation, on high performance parallel computers, of problems involving the dynamic interaction of physical models from disciplines of importance in aerospace, civil and mechanical engineering. Four specific applications of this class of problems will be considered: the mitigation of turbulent vorticity in high-temperature engines; the study of high turbulent flow in pipe bends, the use of active and semi-active control to safeguard composite structures in civil and aerospace engineering; and the implementation of topology/shape structural design and optimization for aeroelastic design. These application are unified by their multidisciplinary nature which involves at least two mechanical fields, use of a common basis in modeling, visualization, problem decomposition, parallel simulation methodology, and software/hardware infrastructure. The research in Computer Science will focus on the development of a problem Solving Environment (PSE) for the applications problems which is suitable for the parallel implementation of large-scale multiphysics simulations. The primary goal of the PSE is to provide transparent, portable and scalable access to all of the resources needed for execution of large parallel applications on heterogeneous and/or distributed systems, including interprocessor communication, access to disk, visualization, debugging and performance monitoring. The following application s of the PSE will be investigated: network-based distributed object systems, distributed shared memory systems, parallel persistent I/O, and computational steering and debugging. In addition we propose to integrate quality of service protocols in high-speed networks with our distributed object system to provide a true real-time distributed environment, and to investigate numerical algorithms for optimal memory access using the PSE to facilitate distributed performance analysis. The application problems are interconnected in modeling, computational and software integration aspects, facilitating the common use of tools such as domain decomposers and response visualization. This project will provide unique opportunities for the training of students and postdocs in an interdisciplinary setting. Regular weekly meetings of the full research group will introduce all researchers to the breadth of issues involved and will allow students to locate suitable thesis topics. Students will be encouraged to have advisors from two disciplinary areas represented by the research team, and each student will be paired with a postdoctoral student as a mentor.
