Hybrid Krylov methods for nonlinear systems of equations
SIAM Journal on Scientific and Statistical Computing
Using MPI: portable parallel programming with the message-passing interface
Using MPI: portable parallel programming with the message-passing interface
Algorithm for solving tridiagonal matrix problems in parallel
Parallel Computing
Domain decomposition: parallel multilevel methods for elliptic partial differential equations
Domain decomposition: parallel multilevel methods for elliptic partial differential equations
Numerical Initial Value Problems in Ordinary Differential Equations
Numerical Initial Value Problems in Ordinary Differential Equations
Iterative Methods for Sparse Linear Systems
Iterative Methods for Sparse Linear Systems
Jacobian-free Newton-Krylov methods: a survey of approaches and applications
Journal of Computational Physics
SUNDIALS: Suite of nonlinear and differential/algebraic equation solvers
ACM Transactions on Mathematical Software (TOMS) - Special issue on the Advanced CompuTational Software (ACTS) Collection
A fully implicit numerical method for single-fluid resistive magnetohydrodynamics
Journal of Computational Physics
| Hi-index | 31.46 |
A description is given of the parallelization algorithms and results for two codes used extensively to model edge plasmas in magnetic fusion energy devices. The codes are UEDGE, which calculates two-dimensional plasma and neutral gas profiles over long equilibrium time scales, and BOUT, which calculates three-dimensional plasma turbulence using experimental or UEDGE profiles. Both codes describe the plasma behavior using fluid equations. A domain decomposition model is used for parallelization by dividing the global spatial simulation region into a set of domains. This approach allows the use of a recently developed Newton-Krylov numerical solver, PVODE. Results show nearly an order of magnitude speedup in execution time for the plasma transport equations with UEDGE when the time-dependent system is integrated to steady state. A limitation that is identified for UEDGE is the inclusion of the (unmagnetized) fluid gas equations on a highly anisotropic mesh. The speedup of BOUT scales nearly linearly up to 64 processors and gets an additional speedup factor of 3-6 by using the fully implicit Newton-Krylov solver compared to an Adams predictor corrector. The turbulent transport coefficients obtained from BOUT guide the use of anomalous transport models within UEDGE, with the eventual goal of a self-consistent coupling.