Sandia National Laboratories
From Flames to Fusion
Sandia Researchers’ Methods Prove Their Versatility
By Thomas R. O’Donnell
Whether it's the sun, the stars, or the long-sought viable fusion reactor, they all involve plasmas, and researchers at Sandia and Los Alamos national laboratories want to help understand what happens in them.
They’re developing computer algorithms that simulate how ions and electrons move and react in plasmas under the influence of magnetic fields. If they’re successful, it could help physicists better understand the processes fueling the sun and stars, and how to harness that power for energy.
The project builds on nearly 20 years of Sandia’s field-leading research in hardware, computer science, applied math and numerical algorithms to devise efficient methods for massively parallel computing. The results include Aztec, one of the first large-scale, parallel, iterative solver libraries; and Chaco, an early tool to balance the computational workload between parallel processors. A host of complex applications have used those and other codes devised by John Shadid, Ray Tuminaro, Scott Hutchinson, Bruce Hendrickson, Rob Leland and other Sandia researchers.
Contour plot of hydrogen and water concentrations in a fuel cell. Click image for larger version and more information |
But Shadid had his own application in mind. He’s worked with fellow Sandians Roger Pawlowski, Andy Salinger, Karen Devine, Gary Hennigan and Paul Lin to simulate the intricate physics of coupled fluid flow and complex chemistry. These transport/reaction (or chemically reacting flow) systems have a multitude of applications, including simulating combustion, cardiac cell activity, and even the spread of a biological material released in a busy airport.
In one case, Sandia researchers worked with Ford Motor Company researcher Kevin Elwood to simulate a hydrogen solid oxide fuel cell. The researchers’ algorithms and the MPSalsa simulation software helped model flow and reactions in the fuel cell to transform hydrogen and oxygen into water and electricity. The researchers also have collaborated with Dow Chemical Company to optimize a new design for a chemical reactor to efficiently convert ethane gas to liquid ethylene, an important feedstock for plastics.
Stability Control
Such transport/reaction processes couple complex physical processes: Fluid flow, energy transfer, radiation effects, chemical species transport and chemical reactions — both in fluids and on surfaces — and more. They also occur across an expansive range of space and time scales.
“This myriad of coupled phenomena produce a highly nonlinear, multiple timescale behavior,” Shadid says. It’s a huge challenge to find stable, accurate, and efficient numerical methods that simulate these processes and run efficiently on ever-larger computers.
The usual algorithmic approaches have used semi-implicit, operatorsplitting, or explicit time-stepping. In explicit time-stepping, the value of each quantity — density, momentum, energy, species concentration, and others — at the new time is calculated based on the old values. That approach produces a simpler solution method, but it can be unstable, meaning it’s more likely to produce nonsense answers. “You have to go at the smallest time-scale of the individual physics to achieve stability, even if it doesn’t provide extra accuracy in the simulation,” Shadid says.
Shadid and his fellow researchers approach the problem from the other end of the spectrum: using fully implicit methods, which represent all the physics consistently at each new time level. “The advantage is you can take time steps that are associated with the physics you’re interested in,” rather than stability, Shadid says. The desired accuracy dictates the time steps, and fewer time steps can be calculated without losing accuracy. The disadvantage: When the governing partial differential equations (PDEs) are discretized — transformed into algebraic equation systems digital computers can solve — implicit time integration produces a system of strongly coupled, nonlinear equations with tens of millions or hundreds of millions of equations and just as many unknowns. These equations must be solved simultaneously on thousands of computer processors.
