Several additional capabilities have recently been implemented in the M3D-C1 nonlinear extended MHD code to aid future non-axisymmetric response calculations. First, non-axisymmetric field data generated by the PROBE_G code using actual coil geometries may now be directly imported into M3D-C1. Second, M3D-C1 can now optionally represent fields generated from external currents separately from fields generated from plasma currents. The current densities associated with the external fields are excluded from the model equations. This allows non-axisymmetric coils to be present within the computational domain without causing numerical difficulties, which is necessary when using actual vacuum vessel shapes. Finally, a number of numerical methods within M3D-C1 have been reformulated, resulting in a reduction in time-to-solution of up to 20% in some cases.
GYRO and TGYRO have recently been modified to enable operation in so-called hybrid mode, which combines distributed-memory parallelization using MPI with shared-memory parallelization using OpenMP. Initial timing tests are very favorable, showing performance improvements up to the maximum number of OpenMP threads per NUMA node (the smallest group of cores that shares a common cache). The user scripts and platform files have also been modified to aid the user in setting up and submitting jobs, which is very complex at the system level. Currently, detailed performance analyses are being carried out and significant additional performance improvements are expected to be realized over the coming months.
A recently completed NIMROD simulation of massive gas injection into an Alcator C-Mod inner wall limited (IWL) plasma was compared with a similar lower single null (LSN) C-Mod simulation, as well as with simulations of DIII-D in both configurations. C-Mod experiments with IWL plasmas indicate better runaway electron (RE) confinement following the thermal quench phase of a disruption than for LSN plasmas, which is consistent with DIII-D observations. The IWL simulations for both devices showed the retention of discrete islands at all integer rational surfaces when the n=1 mode reached saturation. This is in contrast to LSN plasmas, in which more complete island overlap produced fully stochastic fields outside the q=1 surface. The improved RE confinement in IWL plasmas is explained by the reduction in island overlap.
These highlights are reports of research work in progress and are accordingly subject to change or modification