The global GYRO gyrokinetic eigenvalue solver GKEIGEN has been successfully applied to global simulations of a beam-heated, shear-reversed DIII-D discharge, #142111 for a time slice near t=725ms. In contrast to the previous initial-value solver, which can only find the most unstable mode, the GKEIGEN solver can simultaneously determine the dominant and subdominant eigenmodes. In the associated eigenvalue problem for this three-species global simulation, of electrons, thermal ions, and fast neutral beam ions, the leading modes of a massive, 1.15 million × 1.15 million matrix are found using 57,600 cores on NERSC’s Hopper supercomputer. Beam ions are approximated using an isotropic thermal distribution with a high effective temperature, varying across flux surfaces. The results confirm the presence of at least three distinct unstable eigenmodes in the Alfvén branch, namely a reversed shear Alfven eigenmode (RSAE), a toroidal Alfven eigenmode (TAE), and an energetic particle mode (EPM). With the previous initial value solver, there were hints that the three modes were present in a scan of qmin; this is shown in the attached figure where the three modes each become the dominant mode for different qmin. In addition, the global eigenvalue solver reveals a number of high frequency modes in the MHz range that were completely missed by the implicit electron time stepping algorithm used in the initial value solver. These as-yet unidentified modes propagate in the electron diamagnetic direction and are apparently absent in local simulations.
Recent work showed that the chaotic radial diffusion of electrons is the cause of the well-established “flutter” transport. Electromagnetic plasma microturbulence breaks magnetic surfaces in tokamaks resulting in chaotic, or “tangled”, magnetic field lines. This occurs at values of the plasma pressure (beta) that are well below the limit for so-called ideal MHD instabilities, a limit that was previously thought to be the precursor to magnetic chaos. The resulting flutter transport is generally well approximated by following the motion of individual electrons in the chaotic magnetic field. Even for values of the plasma beta for which the electron flutter transport is relatively strong, and the small-scale destruction of magnetic surfaces is complete, the ion transport is weakly affected and remains close to its value at zero beta. Two mechanisms for breaking magnetic surfaces have been proposed: linearly unstable microtearing modes, and the nonlinear excitation of damped modes with tearing parity.
Alan Turnbull attended the Stochasticity in Fusion Plasmas Workshop in Julich April 11-14 where he presented the GA Theory work on the plasma response to external non-axisymmetric perturbations on behalf of the group.
The new 216-core GA theory cluster is now in operation. The dual hex-core configuration is similar to the ORNL XT5, and preliminary tests indicate that the system is slightly faster per-processor on GYRO simulations than the XT5. This performance was obtained after some customization of the MVAPICH environment. Coupled with the new Linux cluster, VENUS (see April 01 2011 highlight at Theory Weekly Highlights for April 2011), this will significantly enhance our computational capabilities.
Motivated by recent stiffness results (see Theory Weekly Highlights for March 2011 March 25 Highlight) well as the general increase in model validation studies, a new suite of turbulent transport validation metrics is being developed, in order to better assess the fidelity of various models to experiment. These metrics are based upon using ensembles of turbulence calculations to self-consistently quantify uncertainties in predicted fluxes, profiles, and fluctuation characteristics. These ensemble statistics are then used to formulate more robust comparisons of model predictions to experiment. Current work is focused on examining optimal methods for generating the ensembles and mathematical forms of the metrics. Initial results were presented this week at the 2011 TTF workshop in San Diego.
The new 64-bit computational Linux cluster, VENUS has been released to DIII-D researchers. VENUS has nine computational nodes plus two head nodes, and each node has 16 CPU cores and 48GB memory. Key features of the new cluster are single sign-on, automatic load balancing, and high-availability. The deployment of the VENUS cluster is expected to significantly increase the computing power, which is critical for scientific data analysis activities at DIII-D.
These highlights are reports of research work in progress and are accordingly subject to change or modification