Combustion Modeling

Combustion research is essential for achieving such goals as improving the gas mileage of automobiles and reducing the harmful emissions of diesels. But computational modeling of combustion processes is enormously complex.

For example, many engines are designed to inject fuel and air into the combustion chamber, and mixing of these two streams on the molecular level is necessary for combustion to occur. Near the inflow, the mixing process is dominated by the large-scale structures in the flow; but as the jet develops, the mixing is influenced more by small-scale turbulence. Thus, successful modeling of combustion depends on accurate modeling of turbulence.

Mathematicians in NERSC's Center for Computational Sciences and Engineering and the Applied Numerical Algorithms Group are helping to develop the algorithms and software needed to resolve the complexities of fluid dynamics in general and combustion modeling in particular.

An especially useful tool, adaptive mesh refinement (AMR), is one of our major areas of research. AMR works by dividing a problem into smaller parts, much like putting a mesh over the problem and then looking at each segment. Areas of interest are treated with successively finer meshes to provide more detailed information. The finer the mesh, the higher the accuracy--and the more computing capacity required.

AMR allows scientists to make the most effective use of the available computer power by focusing it on the region of the problem where it is most needed. The end result is better answers at a lower cost.

A recent simulation of turbulent flow demonstrated both the capabilities of AMR and the need for faster computing resources to achieve higher resolution. The AMR simulation achieved a resolution equivalent to 8 million data points per time step using only 2 to 3 million data points per time step for 740 time steps. This simulation required 38 wall-clock hours on 32 processors of NERSC's Cray T3E-900, and generated 50 gigabytes of data.

But simulating a turbulent jet under the conditions typically studied experimentally would require an estimated 1.3 billion grid points, using 8,160 wall clock hours (340 days) on 512 processors of the T3E, and generating 25 terabytes of data. The algorithms being developed today will achieve their full potential on the next generation of high performance computers.