Feature - A fine-grained approach to “cool” simulations

A snapshot of a simulation examining the interactions of six water molecules. Image courtesy of Andrew Schultz.

When deciding how best to design equipment in a chemical plant that produces liquid nitrogen, designers must know the properties of a mixture of nitrogen and oxygen gas at various stages as it cools and compresses. Researchers at the State University of New York at Buffalo have developed a new simulation technique that makes calculating these properties much more efficient.

OSG is “the ideal workhorse”

The new technique involves running simulations on the grid unfeasible on a single computer because they require large amounts of computing power. Andrew Schultz, a SUNY-Buffalo chemical and biological engineering researcher, said Open Science Grid (OSG) is “the ideal workhorse” to run these simulations. Schultz and his team have run more than 60,000 jobs on OSG resources since January—an order of magnitude jump in productivity.

The team’s technique can model any gas or mixture of gases to calculate properties such as pressure and flow rate. After providing a description of the molecules and how they interact, scientists can simulate the stages of cooling to determine how the gas behaves within given temperature ranges. The simulations yield parameters to plug into the equation relating pressure, temperature and density. Engineers can then use these results to determine how to design gas-processing equipment.

Andrew Schultz addressing the audience at Fermilab's Grid Fest on October 3, 2008. Image courtesy of Reidar Hahn, Fermilab.

Bulk is out

The new technique is more accurate and efficient than previous simulation methods, working directly from a description of the actual molecule and operating on a much smaller scale.

The usual “bulk” simulation involves 500 to 1,000 molecules and is run several times, each at a different temperature. The simulation takes long to run on one computer and is usually confined to one processor.

The team’s new “fine-grained” method, however, breaks the hundreds of molecules into chunks of two to eight molecules. Instead of one large simulation, the team runs sets of several smaller independent simulations on the grid at each temperature and averages the results.

Schultz and his team have compared their results against other teams’ calculations derived from experiment, and see excellent agreement.  Comparisons to results gleaned from different simulation techniques also show agreement, at least for density values most typically used.

Harnessing emerging technologies

“Operating on this smaller scale allows us to split up the simulations across many computer processors and harness emerging technologies, such as multi-core CPUs, and grid computing facilities, such as OSG,” Schultz said. “This gives scientists an option to have very accurate descriptions of the properties of a given material without devoting as much time to run a full traditional simulation.”

—Amelia Williamson, iSGTW