Computing Enables New Insights into Generating Power Like the Sun

If humans could harness nuclear fusion, the process that powers stars like our Sun, the world could have an inexhaustible energy source. In theory, scientists could produce a steady stream of fusion energy on Earth by heating up two types of hydrogen atoms—deuterium and tritium—to more than 100 million degrees centigrade until they become a gaseous stew of electrically charged particles, called plasma. Then use powerful magnets to compress these particles until they fuse together, releasing energy in the process.

Although magnetic fusion has been achieved on Earth, researchers still do not understand the behavior of plasma well enough to effectively confine it to generate a sustainable flow of energy. That's where the U.S. Department of Energy (DOE) comes in. Over the past 50 years the Energy Department has invested significantly in fusion research and even established a supercomputing center to simulate the behavior of plasma in a fusion reactor. This facility eventually evolved into the National Energy Research Scientific Computing Center (NERSC), which is currently managed by the Lawrence Berkeley National Laboratory and serves about 3,000 researchers annually.

The DOE's Scientific Discovery Through Advanced Computing (SciDAC) Center for Extended Magnetohydrodynamic Modeling has also played a vital role in developing computer codes for fusion modeling. Recently, the collaboration created an extension to the Multilevel 3D (M3D) computer code that allows researchers to simulate what happens when charged particles are ejected from a hot plasma stew and splatter on the walls of a "tokamak" devices, the doughnut-shaped "pot" used to magnetically contain plasmas. These simulations are vital for ensuring the safety of future plasma containment devices to achieve sustainable fusion for electricity.

Using NERSC computers, the team used the M3D extension to simulate the development of the magnetic fields in a tokamak fusion experiment; identify and dissect points where the plasma becomes chaotic; and map the trajectory of high-energy plasma eruptions from these so called X-points to determine how these particles affect the tokamak walls.

According to Linda Sugiyama of the Massachusetts Institute of Technology, who led the code development, these findings will be critical for ensuring the safety and success of upcoming magnetic confinement fusion experiments like ITER, which is currently under construction in southern France and will attempt to create 500 megawatts of fusion power for several-minute stretches.

Modeling Fusion to Ensure Safety

In this image, the torus of plasma is tilted upward to show the bottom. The magnetic fields within the plasma are represented as white lines and approximately follow a contour of constant plasma temperature as it circles around the torus. The temperature is shown as blue/yellow surface.

Because hot plasma is extremely dangerous, it is imperative that researchers understand plasma instability so that they can properly confine it. For fusion to occur, plasmas must be heated to more than 100 million degrees centigrade. At these searing temperatures, the material is too hot to contain with most earthly materials; the charged particles can only be held in place with magnetic fields generated by devices like the tokamak.

In the tokamak, plasma is confined to a torus, or doughnut shape. However, Sugiyama's NERSC simulations show that that instabilities at the plasma edge can alter the constraining magnetic fields generated by the tokamak, allowing blobs of plasma to break out of the torus shape. Instead of a single crash, this phenomenon can cause multiple pulses of instability, each ejecting large blobs of plasma onto the tokamak walls. The number of unstable pulses affects how much plasma is thrown onto the walls, which in turn determines the extent of wall damage.

"Studies of nonlinear dynamics show that if you have X-points on the plasma boundary, which is standard for high temperature fusion plasmas, particles can fly out and hit the walls. But no one had ever seen what actually happens if you let the plasma go," says Sugiyama. "This is the first time that we have seen a stochastic magnetic tangle, a structure well known from Hamiltonian chaos theory, generated by the plasma itself. Its existence also means that we will have to rethink some of our basic ideas about confined plasmas."

She notes that the primary causes of plasma instability vary in the different plasma regions of the tokamak, including the core, edge and surrounding vacuum. To understand the most dangerous of these instabilities, computational physicists model each plasma region on short time-and-space scales. The plasma regions are strongly coupled through the magnetic and electric fields, and researchers must carry out an integrated simulation that shows how all the components interact in the big picture, over long time scales.

"Modeling plasma instabilities is computationally challenging because these processes occur on widely differing scales and have unique complexities," says Sugiyama. "This phenomenon couldn't be investigated without computers. Compared to other research areas, these simulations are not computationally large. The beauty of using NERSC is that I can run my medium-size jobs for a long time until I generate all the time-steps I need to see the entire process accurately."

She notes that the longest job ran on 360 processors for 300 hours on NERSC's Cray XT4 "Franklin" system. However, she also ran numerous other jobs on the facility's Franklin and DaVinci systems using anywhere from 432 to 768 processors for about 200 CPU hours.

"I greatly appreciate the NERSC policy of supporting the work required to scale up from small jobs to large jobs, such as generating input files and visualizing the results. The center's user consultants and analytics group were crucial to getting these results," says Sugiyama.

A paper based on her NERSC results was recently accepted by Physics of Plasmas and will appear in print later this year. Sugiyama will also present her findings at the SciDAC 2010 conference in Chattanooga, Tenn. in July. Steve Jardin of the Princeton Plasma Physics Laboratory is the principal investigator of the SciDAC Center for Extended Magnetohydrodynamic Modeling, which supported this work.