Participants in the March 2009 workshop “Scientific Grand Challenges in Fusion Energy Sciences (FES) and the role of Computing at the Extreme Scale” considered five FES/plasma science areas that would benefit from access to exascale computing.
Burning Plasma/ITER. Sustaining a burning plasma is the next major step in making fusion energy a reality, says Princeton’s William Tang, workshop co-chair. But “you’re entering a regime that has yet to be achieved,” with highly energetic particles, long-lasting plasmas and tremendous heat loads. Workshop participants estimated that simulations 10 times more detailed than today’s are needed to help realistically model this complex behavior.
For UC-Irvine physicist Zhihong Lin and his colleagues, exascale computing would provide the capacity to track a larger number of particles and use a finer computational mesh – a grid of points spread through the physical domain of the fusion reactor. More points make a more accurate simulation but increase the demand for computer power.
Properly used, exascale computing can be expected to help computational scientists incorporate more accurate and complex physics in their simulations. “If we have access to more powerful computers we could develop and apply a more complete physical model,” Lin says.
Advanced Physics Integration. Scientists typically simulate the many different kinds of physics governing a fusion system separately, focusing on single aspects like magnetohydrodynamics for large-scale plasma stability, radio frequency energy for plasma heating, and microturbulence for efficient confinement.
“All these different physical processes are currently simulated by different codes,” Lin says, but for a realistic model they must be integrated – at great cost in computer power, especially since each model spans multiple scales of space and time.
Some integrated models are beginning to successfully couple two or more kinds of physical processes, Tang says, but research is needed to join them all into a clear picture of an entire reactor or ignition facility.
Plasma-Material Interaction. “You have to understand how the plasma-facing components in a reactor are going to hold up and what the best material to use will be,” Tang says. The metal blankets lining the reactor would suffer serious damage if large disruptions can’t be controlled.
Scientists also need improved models to understand how plasma-material interactions (PMI) affect key features of the plasma, including confinement quality, edge temperature and density, and others. More powerful computers can help drive progress, but the fundamental challenge is capturing essential elements needed for more realistic PMI models that are relevant for ITER and for DEMO, a future demonstration fusion reactor.
Laser-Plasma Interactions and High Energy Density Laboratory Physics (HEDLP).Inertial confinement fusion (ICF) uses powerful lasers, particle beams or magnetic fields to create plasmas with energy densities that are unprecedented in the laboratory, helping investigate prospects for alternate pathways to fusion power. ICF devices like the National Ignition Facility (NIF) at Lawrence Livermore National Laboratory in California and the Z machine at Sandia National Laboratories in New Mexico also probe the fundamental nature of matter and energy.
Basic Plasma Science/Magnetic Reconnection Physics. Magnetic reconnection is the spontaneous breaking and rearrangement of magnetic field lines that occurs as a plasma seeks to relax to a lower energy state. As field lines break, they release large amounts of energy. When “everything heals back up in a magnetic fusion device like a tokamak, it can look the same geometrically as it did initially,” Tang says, except the plasma is cooler, possibly inhibiting sustained fusion reactions.
More significantly, magnetic reconnection also represents an important class of space and astrophysical plasma phenomena, such as solar flares, Tang says. How reconnection dynamics evolve is of great interest for those fields as well as for fusion energy research. Addressing these issues will require algorithms able to run well on increasingly powerful computers.