The OMFIT GATO module was revamped to simplify production runs over multiple equilibria. The new module enables use of several options in the code, including reading equilibria from a variety of different codes, inclusion of an electron number density profile to construct a mass density profile, and of a fast ion profile for diagnostic purposes. In addition, a preconstructed mesh can be read in to facilitate comparisons to other codes. The GUI permits restarting from a prior failed run, running only the mapping part of the code, or rerunning only the final diagnostics. Standard scans over several parameters are easily run, including mesh convergence scans, wall position scans, toroidal mode number, and eigenvalue number. The new module was designed to be easily modified to enable substitution of other stability codes, and a new module for the MARS-F code is being developed using the GATO module as a template. The ultimate goal is to develop a ‘stability manager’ where stability can be calculated from the same input, using any stability code of choice, and with the output in a uniform format that can be accessed by other external tools (diagnostic, graphics, synthetic diagnostics, etc.) irrespective of the origin the data.

An introductory tutorial to the OMFIT framework was held at GA and broadcasted over the internet. Attendance was high and included SULI students as well as on-site and remote scientists from laboratories across the US, Europe, and Asia. The recordings of the tutorial have been made publicly available at https://goo.gl/MAaiCQ. This is the first in a series of OMFIT tutorials, which are planned to be held every Wednesday at 1:00 PM PST.

Previously, a new set of inherently stable and consistent plasma fluid equations was developed that is specifically conceived for massively parallel computation (see Highlight from November 10 2017 at https://fusion.gat.com/theory/Weekly1117. A manuscript describing this work has been accepted in Physics of Plasmas (F.D.Halpern and R.E.Waltz, “Anti-symmetric plasma moment equations with conservative discrete counterparts”, Phys. Plasmas (2018), in press) and has been chosen as a journal highlight to be featured as an AIP Scilight story. The main point addressed is, how does one arrive to plasma fluid models that translate to reliable, high-quality simulations in the exascale age? The surprising answer is that, there exists a frame of reference where the continuous and discrete equations have analog conservation properties. This reference frame is neither Eulerian (laboratory) nor Lagrangian (particle), in a break with 200 years of tradition in the fluid theory. Most importantly, it is shown that the stability can be a feature of the model, and not of the numerical method. Altogether, one can obtain very robust models with a simple and flexible implementation. In turn, this should result in more reliable and scalable plasma fluid simulation codes. The approach is extensible and has recently been applied to the ideal MHD equations.

Members the GA Exascale and ATOM teams (Candy, Halpern, Kostuk, Sfiligoi) attended a GPU/OpenACC Hackathon last week, sponsored by Nvidia, Google Cloud, Oak Ridge National Laboratories, and UC Boulder. The objective was to port and improve existing high-performance computing codes such as CGYRO, as well as to explore possibilities for GPU-accelerated matrix solvers appropriate for next generation exascale computers. The CGYRO performance was significantly improved by exploiting cutting-edge GPU-to-GPU communication software, which is only available in the latest generation NVIDIA Volta GPUs. This development directly relates to the announcement of the Summit supercomputer, which will feature hardware support for GPU-to-GPU communication capability.

The OMFIT TGLF_scan module has been updated to enable new uncertainty propagation tools and workflows. The module efficiently generates probability density functions (PDFs) of TGLF flux predictions, given specified PDFs of input parameters such as local gradient scale lengths. The new capabilities use rapidly converging and computationally inexpensive surrogate models based upon chaos polynomial expansions. These input parameter PDFs can be user-specified or derived from the underlying profile data measurements and fits. An initial application to transport predictions for a DIII-D ITER baseline plasma, within which the mix of neutral beam injection (NBI) and electron cyclotron heating (ECH) was varied, shows clear differences in the sensitivities and parametric dependencies of the turbulence between the two heating methods. The results indicate a transition of turbulence character from predominantly long-wavelength ion temperature gradient (ITG) modes in the NBI-only case to a multi-scale, mixed-mode regime in the NBI+ECH case, consistent with earlier deterministic modeling studies of the plasma [B. A. Grierson et al, Phys. Plasmas 25 (2018) 022509]. A manuscript detailing these new capabilities is being prepared for publication, and additional applications of the technique are being investigated for problems where traditional Monte Carlo approaches are impractical.