Developments of the NIMROD MHD code for next-generation fusion-device modeling and computational hardware*

Incorporation of atomic physics associated with multiple species is required to study next-generation fusion device topics such as integration of MHD modeling of resonant magnetic perturbations or edge-harmonic oscillations with advanced edge solutions. We present an operator-splitting formulation of the atomic interactions using a Strang-splitting technique to naturally break equations into constituent ODE and PDE parts and preserve the structure exploited by the semi-implicit leapfrog. By testing on a battery of cases, we show that a second-order-in-time Douglas-Rachford inspired coupling between the ODE and PDE advances is effective in reducing the time-discretization error to be comparable to that of Crank-Nicholson with Newton iteration of the nonlinear terms. Since all of the nonlinear atomic interaction is handled by a local ODE solver using an Adams-Bashforth method, no nonlinear iteration is required and each spatial point can be treated independently and in parallel. This parallelism is advantageous for exploiting GPUs. We use OpenACC with a modern Fortran implementation using a continuous-integration development cycle to port NIMROD algorithms to the GPU architecture. The performance of the ported finite-element and matrix preconditioning kernels is reported.