Fluid Turbulence and Mixing at High Reynolds Number

A 2004 INCITE Project

Led by Professor P. K. Yeung of the Georgia Institute of Technology, this project was allocated 1,200,000 processor hours. Although turbulence is a phenomenon that has applications in a wide range of natural and human activities, it is not well understood and is extremely difficult to model accurately on supercomputers. This group, however, has achieved preliminary results closer to actual turbulent flows by using 2048 processors on NERSC's IBM SP.

A simulation of rotating turbulent flow subject to Coriolis forces in the equation of motion was conducted at 1024³ grid resolution. These preliminary results indicate that the major features of the flow are similar to those previously obtained at lower grid resolutions, but at a higher Reynolds number, which implies increased relevance to actual turbulent flows in engineering applications.

With improved modeling capability of fluid turbulence, scientists will gain greater insight into meteorology, astrophysics, oceanography, environmental quality, combustion and propulsion, among other research areas. Because of the complexity of turbulence, it is difficult for scientists to accurately predict natural phenomena, such as severe storms, and engineering solutions in areas such as aircraft design, internal combustion engines and industrial flows. Improved models could lead to more efficient jet engines and cleaner-running automobiles.

The main target in these computations is a large simulation using 2048³ grid points to resolve the smallest scales in the fluctuations of passive contaminants transported by turbulent flow. This is being carried out in two phases. In Phase I, which is already complete, the researchers ran the simulation with the Eulerian velocity field only in order to produce a statistically steady state. In Phase II, which is now in progress, they include two passive contaminants of different molecular diffusivity, which makes the simulation more time-consuming and memory intensive. Fortunately, NERSC consultants have suggested some timely and easily implemented improvements, such as state-of-the-art communication libraries and a strategy to exploit the memory configuration of 16-task nodes to address the effects of core memory bandwidth on the overall CPU performance.

Professor Yeung has praised the level of support provided by NERSC. "We are very grateful for our INCITE grant and the NERSC scheduling and accounting policies that are very favorable to the progress of our work," he wrote to NERSC staff. "We are excited at the prospect of the great opportunities so uniquely available to us. After we complete the simulations covering a physically adequate period of time we will be able to perform detailed data analyses as well as share the data effectively with members of our research community."

Visualizations of the project's simulations

Additional visualizations