Ensemble Computations - Present and Future of Engineering Computing

Abstract. Computer simulations are pervasive in science and engineering. Computations, together with theoretical analysis and experiments, constitute the foundation for building knowledge, whether it be to investigate a new physical phenomena or to assess the performance of an innovative device. However, a single computation, despite its sophistication and complexity, can rarely provide sufficient and credible evidence to support a decision. To build confidence in computed outcomes, one typically conducts sensitivity analyses, investigates uncertainty, and explores design variations. All these approaches require an ensemble of computations whose results we combine rigorously via statistical analysis or optimization. In this talk we discuss the use of ensemble of simulations for multidisciplinary design under uncertainty; we have developed novel algorithms to handle ensemble with different fidelity implemented in a new programming environment that enables the efficient use of next generation HPC supercomputers. I will also provide a perspective on the present and the future of computational engineering research at Stanford.