ITER predictions with the TGLF-EP+Alpha critical gradient model of energetic particle (EP) transport by Alfvén eigenmodes (AEs) have confirmed that a steady-state, 7.5 MA current scenario can actually exhibit less AE radial re-distribution than the higher 15 MA current base scenario, despite having a generally destabilizing higher safety factor q. Exploring different levels of current penetration through analytically varied q profiles, the study has identified an optimal distribution of the steady-state current that redistributes roughly 15% of alphas and 29% of NBI ions from the inner core to larger radii. In comparison, in the base scenario, 23% of alphas and 37% of NBI ions are re-distributed. The optimized steady-state case has generally lower magnetic shear, particularly in the regions of strongest EP source. As has been predicted before, AE-induced EP transport falls well short of the edge in all cases.

Active tailoring of the plasma toroidal flow profile can be beneficial for both MHD mode control and improving plasma confinement. In tokamak plasmas, magnetic coils designed for error field or ELM control can also be utilized to apply neoclassical toroidal viscosity (NTV) torque. MARS-Q modeling performed for MAST double-null (DN) and single-null (SN) discharges, with two rows of coils above and below the midplane, showed that for DN plasmas in the even parity configuration, the NTV torque damps the plasma flow in the core. In contrast, in the odd parity configuration, the edge flow is damped by the electromagnetic torque. For the SN plasma, the NTV torque still provides the largest core flow damping in even parity. The largest edge flow damping, however, is induced by the electromagnetic torque with 270 degree coil phasing. By combining all torques, the coil phasing scan can be used as a very effective technique to optimize the torque distributions in the plasma core and edge regions in MAST/MAST-U plasmas.

Topological bifurcation of magnetic islands was observed in M3D-C1 linear resistive MHD calculations of model NSTX-U equilibria during the application of small 3D magnetic perturbation fields. Comparison to a DIII-D ELM suppression case suggests that a stronger kink response in the NSTX-U plasma may be the primary cause of the bifurcations. Island evolution and the bifurcation from increasing perturbation current in both cases, also suggests a new hypothesis that the magnetic island topology bifurcation results in improved particle confinement, which results in a rise in the density inside the islands. This was examined by implementing a random kick algorithm in the field line integration code TRIP3D-GPU to approximate electron collisional transport. Although full consideration of all forces between electrons requires 3D large scale kinetic simulation in toroidal geometry, the positive result of this approximate transport simulation does provide first step confirmation towards fully validating the hypothesis.

Two 3D fluid models have been implemented using the Arbitrarily-Large Moment Anti-Symmetric (ALMA) toolkit developed at GA (see Highlight for June 15 2018 at https://fusion.gat.com/theory/Weekly0618). ALMA utilizes a reference frame that is neither Eulerian nor Lagrangian in which the continuous and discrete equations have the same conservation properties, so that the numerical stability is guaranteed when the system is physically stable. The first implementation is an extended MHD code in slab geometry. A simple test case based on the Orzsag-Tang vortex – a high beta case that naturally forms narrow current sheets as well as super-Alfvénic shocks – has been verified, and its conservation and stability properties studied. We have also confirmed its strong scaling on NERSC’s Cori platform. A second code, implementing Braginskii’s two-fluid equations with a very simple closure has been developed and tested in our local compute nodes. The short development time of two complex, scalable, and GPU-ready codes demonstrates the practicality of the toolkit. The Braginskii code incorporates a geometric multigrid solver for Gauss’ law, developed in-house. This solver is significantly faster than the standard hypre-package on both CPU and GPU, and is capable of MPI-aware GPU directives developed for NVIDIA’s Nvlink hardware. Recent optimizations of the ALMA multigrid solver involved various techniques to decrease the amount of communication between processes, reduce memory usage, and split part of the computation between the CPU and the GPU. The optimizations have significantly sped up computation in both CPU and GPU, indeed leading to a further 2x speedup.