Center for the Study of Plasma Microturbulence
Plasma microturbulence, the dominant mechanism for the loss of heat from tokamaks, will determine the fusion gain that can be achieved in ITER. While plasma microturbulence has been studied since the 1960s [Mikhailovskii, 1974], the magnetic fusion community has yet to develop a complete predictive understanding of the turbulent transport of heat, momentum and particles across magnetic surfaces. The development of such a predictive understanding has been identified as a major goal for the US fusion program [Baker, 2002], and achieving it requires the development and utilization of new, more powerful computational tools.
The most significant recent developments in this context are (1) the evolving capacity to perform high-fidelity computer simulations of plasma microturbulence with comprehensive physics models, and (2) the amazing advances in computer power. Since the early 1990s, nonlinear turbulence simulations have provided important insights into the essential characteristics of plasma transport, such as the importance of linear marginal stability of the plasma to ion temperature gradient modes in determining the ion temperature profile [Kotschenreuther, 1995]. The realization of a predictive description of plasma transport requires that we develop more accurate, comprehensive models of microturbulence through increased fidelity of gyrokinetic simulation codes. Among the effects which need to be included in simulations to yield a predictive understanding of transport are kinetic electrons, multiple gyrokinetic ion species, a full electromagnetic description of the turbulent fields, and a sufficiently accurate model of flux-surface shape for actual (or projected) tokamak discharges.
The three highest-fidelity gyrokinetic turbulence codes available within the US magnetic fusion program are GYRO, GS2, and GEM. In addition to their outstanding physics fidelity, these codes are computationally efficient and able to effectively utilize the largest parallel computers supported by the Office of Science. Gyrokinetic turbulence is an ideal application for petascale computing platforms, as the 5-dimensional nature of the problem guarantees that high-resolution simulations involve a huge number of grid points, with the potential for excellent scaling to a very large number of processors. For example, the GYRO code has explicitly demonstrated almost perfectly linear weak-scaling (i.e, as the problem size is made larger) up to (and presumably, beyond) 16,384 cores of the 120 TeraFLOP Cray XT4.
The Center for the Study of Plasma Microturbulence (CSPM) uses all three of these codes to address the US fusion program goal of developing a predictive understanding of plasma transport. This will be accomplished by seeking further improvements to the efficiency and fidelity of each code, by providing tools to facilitate code validation and leading code verification and validation exercises, and by maintaining a gyrokinetic simulation database for use by the MFE community together with appropriate tools for data analysis and visualization. The CSPM is a collaboration between scientists at LLNL, General Atomics, PPPL, the University of Maryland (UMD), the University of Colorado (CU), and MIT. In addition, we have submitted proposals for Scientific Application Partnerships (SAPPs) with the SciDAC TOPS, SDM, and VACET Centers for Enabling Technology.
Plasma microturbulence simulations are based on implementations of the gyrokinetic-Maxwell system of equations. The complexity of gyrokinetic codes increases with increasing code capability, leading to the possibility of errors in programming or in the application of the underlying equations. Our 3-code collaboration provides a unique opportunity to perform numerous cross-code comparisons through benchmarking exercises involving GA, UMD, MIT, CU and LLNL. All three codes have third-party users. Hence, validation efforts will be assisted by collaborations with members of the MFE modeling and experimental communities at PPPL, MIT, GA, UT-Austin, and CU who are supported on other budgets, in addition to focused efforts under this proposal.
Our collaborative efforts at further improvements in code fidelity and efficiency on today's leadership class supercomputers will focus on GYRO. Workers at MIT, PPPL, and the SAP with the TOPS Center for Enabling Technology will work to improve the realism and computational performance of the GYRO collision operator. Scientists from the University of Maryland Institute for Advanced Computer Studies with work with UMD, MIT and CU physicists to develop and document the performance of new gyrokinetic algorithms for heterogeneous multicore supercomputers, members of the next generation of American leadership class supercomputers. LLNL will work with CDM to develop a modern gyrokinetic database, together with advanced data analysis tools, which will be made available to the MFE community consistent with the OFES data sharing policy. Improved visualization tools will be coupled to this improved data analysis capability through the efforts of MIT, LLNL, UMD and the SAP with VACETs. These code, database, and data analysis improvements will be employed in focused studies of Trapped Electron Mode turbulence (MIT), the behavior of plasma turbulence at high plasma pressure (GA, UMD), and energy flow in plasma microturbulence (LLNL). Finally, all of these capabilities will be employed in simulations aimed at projecting the steady state performance of ITER burning plasmas.
.