Until recently, all sparse linear algebra operations in the neoclassical transport code NEO were carried out using the UMFPACK library, which is bundled with GACODE and also provides sparse linear algebra support for GYRO. Now, support for the SuperLU library has been added to the GACODE build system. SuperLU is a general purpose library for the direct solution of large, sparse, nonsymmetric systems of linear equations. Importantly, it can also make use of distributed memory (via MPI), threading (via OpenMP) and GPUs (via CUDA). We have updated the NEO module to implement the SuperLU library and, using the parallel computing capabilities of the library, preliminary tests show significant speedups over the current sequential UMFPACK library routines with good parallel efficiency across multi-core processors. This will provide a significant enhancement for integrated modeling applications with NEO, particularly in highly-shaped edge geometry and with many ion species. The next step is to integrate the GPU support in this library which will enable us to utilize large machines such as Titan at ORNL. As part of our future plans, we plan to replace UMFPACK with SuperLU in the rest of the GACODE modules, such as GYRO, that require it.
A new set of inherently stable and consistent plasma fluid equations was developed that is specifically conceived for massively parallel computation. A longstanding weakness of fluid computations is the lack of manifestly consistent equations, as well as stable and efficient numerical schemes. For instance, it has been known since the 1950s that, naïve computations of turbulent flows lead to artifacts and blow-ups because of numerical aliasing. Discretized conservation laws do not trivially retain the physical properties of their analytic counterparts. We have overcome these obstacles by exploiting symmetries in the plasma dynamics to construct “symbiotic” models with physical properties that translate one-to-one to supercomputers. The new model can be interpreted as an alternative to the traditional Eulerian and Lagrangian approaches, one in which plasma flows constitute infinitesimal rotations of fluid quantities. The procedure is general, and can be applied to many models, such as ideal and resistive MHD, the drift-ordered fluid models, and the Navier-Stokes equations. Ongoing work concentrates on obtaining “symbiotic” variants of these models for many applications in plasma physics and beyond.
Alan Turnbull returned from the MHD meeting in Madison WI where he presented ideas on detection of incipient MHD instabilities using MHD Spectroscopy and ideas on how to interpret the Spectroscopy results in terms of ‘disruptability’, using criteria based on mode amplitudes and widths, and differential damping rates.
A new version of OMFIT, v0.23.0, which includes contributions from many different contributors, was released. In addition to a considerable number of bug fixes and tweaks, upgrades to the physics modules for SOLPS, TRANSP and OMFITprofiles, among others, were included. The GYRO_scan module was updated to allow parallel GYRO_scan and execution, and an enhanced TRIP3DGPU version was released, with new GUI, scripts, and a smart batch management system. A new CHEASE module and OMFITchease class for parsing the EXPEQ.OUT file is now available and a new MARS module is under development with a new OMFITmars class for parsing the output of the MARS code. The CHEASE and MARS modules will greatly streamline the application of the MARS code to DIII-D equilibria. Finally, the new release also includes an improved GUI visualization of commands being executed.
These highlights are reports of research work in progress and are accordingly subject to change or modification