Initiative in Innovative Computing (IIC) Harvard University

For the better part of the past 50 years, advances in algorithms and the increasing computing power have made computational fluid dynamics, CFD, a mature discipline. What are some remaining frontiers of the discipline? One of the challenges in fluid simulation continues to be the need to straddle many scales.

New computational methods still need to be developed that are able to adapt to the many scales of a problem. Another frontier recently opened is the development of hardware-aware software. Multi-core computers are on everyone’s desktop nowadays, and a growing trend in using graphics cards and other specialized hardware is buzzing.

On both these frontiers, there is great potential for meshfree methods. Particle-type formulations for CFD offer an alternative which is low in numerical diffusion, devoid of numerical dispersion and stability constraints. Meshfree methods offer a natural adaptivity in situations where mesh generation is a large burden. And meshfree methods could be especially well suited to exploit the new hardware technologies entering the scene.

I will present an overview of a particle-type formulation for fluid dynamics, the vortex method.  This method requires an N-body solver within it, for which we use the fast multipole method.  In our goal of obtaining hardware acceleration for this method, I’ll describe our progress with the fast multipole method, where we currently achieve 480 gigaflops for a hundred-fold speedup on a GPU card.

Lorena Barba obtained her Ph.D. in aeronautics from California Institute of Technology in 2004.  She then joined the Department of Mathematics at the University of Bristol, UK. There, she was the leader of an EC-funded international project titled "Scientific Computing Advanced Training" involving 10 institutions in Europe and Latin America. The project allowed more than 30 young aspiring scientists to spend an extended period immersed in a research group in Europe. In the fall of 2008, she started a new position as Assistant Professor of Mechanical Engineering in Boston University. Her research as a computational scientist and a fluid dynamicist covers particle methods used for fluid simulation, the development of fast and efficient algorithms, the use of novel computer architectures, as well as fundamental and applied aspects of fluid dynamics.