Nu-FuSE: An Exascale software project
Posted: 16 Nov 2014 | 23:11
The Nu-FuSE (Nuclear Fusion Simulations at Exascale) project was a 3-year, G8 funded, international research project to investigate the challenges and requirements for fusion simulations at Exascale levels. The project’s aim was to significantly improve computational modelling capabilities for fusion, and fusion-related sciences, enhancing the predictive capabilities needed to address key physics challenges of a new generation of fusion systems.
The project focused on three specific scientific areas - fusion plasmas; radiation damage in materials; and the plasma edge - that require extreme-scale computing across a range of simulation codes, and that would benefit from interdisciplinary research engaging Applied Mathematicians, Computer Scientists, and Science Specialists. It had a range of scientific and technical successes, which can be explored in more detail on the Nu-FuSE website, however two particularly important scientific outcomes are highlighted here.
For over a decade, both experimental observations and theoretical simulations of turbulent losses of fusion-grade tokamak plasmas have indicated that energy confinement degrades as the size of the tokamak increases in the so-called “Bohm regime.” However, such simulations have also predicted that for sufficiently large tokamaks there will be a turnover point into a “Gyro-Bohm regime,” where the losses become independent of system size. For the next generation of fusion reactors it is of key importance that systems can operate in this favourable regime.
During Nu-FuSE a number of plasma simulation codes have been optimised and improved to enable them to scale to very large numbers of cores, and therefore undertake longer, larger, and more detailed plasma simulations. One such code, GTC-P, has been scaled to over 1.5 million cores. Using this code it has been possible to undertake large simulations that show the magnitude of turbulent losses in the Gyro-Bohm regime can be up to 50% lower than indicated by earlier, much lower-resolution, simulations and that the Bohm to Gyro-Bohm transition is much more gradual as the plasma size increases. This finding was made possible only after going to the high-resolution, long time-scale simulations needed to achieve the physics fidelity enabled by computing at extreme scales.
Figure 1: Nanoparticles of Yttria self-assemble in steel matrix. Shown by Kimura group to reduce radiation damage. Simulation shows this is due to catalytic healing.
Commercial fusion reactors will produce energy in the form of very high-energy neutrons, which can cause enormous radiation damage to the structure of such systems. Since this damage is inevitable, what is needed are a new generation of materials that can self-heal. However, different materials can have hugely different responses to neutron irradiation, and there is no macroscopic theory for these differences. Since experiments with irradiated samples are costly, screening via computer simulations is an essential part of the process. Moreover, with millions of atoms involved, codes that can exploit exascale systems are essential. One class of materials which show good radiation resistance are the so-called "ODS steels". Through simulations undertaken in the project we have been able to identify that this is because of catalytic recombination of topological defects at the interface between iron and yttrium oxide particles. Now that the mechanism is understood, a systematic pathway to improving on existing materials can be devised.
Figure 2: Ion density in kinetic PEPC simulation of magnetized plasma-vacuum interface (Steinbusch, Gibbon, Sydora, 2014).
EPCC was heavily involved in Nu-FuSE, particularly in working on optimising plasma simulation codes, and on porting materials simulation codes to GPU systems and optimising the performance on the GPU hardware. Part of the Nu-FuSE work undertaken by EPCC won a recent HPC Innovation Award (see this blog post for more details). For further information on this work please visit the Nu-FuSE website.
Image at top of post shows electrostatic potential fluctuations in an annular region at midradius in the MAST tokamak, from a gyrokinetic simulation of the saturated turbulence using the GS2 code. A wedge of plasma has been removed from the visualisation to show the nature of the fluctuations inside the annulus.