Objective: The objective of this activity is to develop a suite of computational fluid dynamics (CFD) codes using three types of numerical algorithms and test their applicability on different massively parallel processor (MPP) architectures. The three algorithms are: Fourier transform pseudospectral codes, finite volume/difference codes, and particle codes. These codes are being written to run on Single Instruction Multiple Data (SIMD) and Multiple Instruction Multiple Data (MIMD) architectures using data parallel and message passing paradigms. These codes will be applied to the physical problem of developing a greater understanding of the mechanisms that control the behavior of the solar heliosphere; they lead to a predictive capability for solar activity.
Approach: In this project we started with a suite of codes that were optimized to run on single vector processors (e.g., CRAY Y-MP) and have recoded them for the SIMD or MIMD architecture as appropriate. The goal is to develop codes that scale with the number of processors to take advantage of larger machines as they become available. Since the new MPP machines are potentially faster than previously available machines, this approach has allowed us to become more ambitious in the problems we address and to include increasingly more complex physics packages in these codes as they are being developed.We are concurrently applying these codes to a variety of scientifically important questions concerning the dynamics of the solar heliosphere, in particular, to evolve our understanding of the nature of solar coronal heating and magnetospheric dynamics.
Accomplishments: Four codes have been developed utilizing five machines: the Thinking Machines (TMC) CM-200 (SIMD), the TMC CM-5 (SIMD/MIMD), the Intel iPSC 860 (MIMD), the Intel Paragon (MIMD), and the CRAY T3D (SIMD/MIMD). We have increased the resolution of our 3D magnetohydrodynamics (MHD) code from 64 x 64 x 64 at the start of the project to 256 x 256 x 256, an increase of over two orders of magnitude in computational intensity. This improvement has enabled us to resolve the magnetic reconnection down to the dissipative scale. The figure to the right shows the remarkable result of one such simulation where two orthogonal flux tubes have passed through one other and reconnected in the original configuration. This phenomena was never discovered with coarser resolution.
Increased computing power has permitted order-of-magnitude longer simulations and more than double the resolution in our 2.5-dimensional flux-corrected transport MHD (FCT-MHD) code simulating driven magnetic reconnection in solar magnetic structures, a model for chromospheric eruptions observed in the Sun's atmosphere. The increased resolution has permitted the study of asymmetric cases where a new phenomena of jetting along field lines is seen reminiscent of coronal eruptions. The figure below depicts the evolution of a typical magnetic reconnection phenomena.
These codes have demonstrated linear scaling with a peak performance of over 4 GFLOPS on the 256-node Naval Research Laboratory CM-5. We have also developed a new particle-particle, particle-mesh (PPPM) code combining the Monotonic Lagrangian grid (MLG) and particle-in-cell (PIC)-type codes, taking advantage of the efficiency on parallel machines of the MLG and the long-scale accuracy of the PIC codes. This code also scales linearly; it has achieved 2.6 GFLOPS on a 256-node CM-5 and has been used to investigate a heretofore unknown quasistatic equilibria in 3D.
Significance: Modeling the solar corona involves many different scales of phenomena, from the smallest dissipation scales to the largest ideal MHD scales. This task requires very-large-resolution codes in 3D. Present-day computers do not have the speed or memory to resolve all the relevant scales. The MPP architecture potentially with teraFLOPS speeds will provide for the first time sufficient power to perform these simulations. The present codes are gearing up to take advantage of these potential speeds. In the meantime, with the rapidly advancing computing power we are still able to perform less ambitious but relevant scientific studies.
Status/Plans: Most of our effort has revolved around the CM-5. We have begun work on the CRAY T3D and IBM SP-2 and are planning to put more effort into MIMD machines. We plan to continue code development while taking advantage of the available codes to perform scientifically exciting simulations of solar activity.
Point of Contact:
John H. Gardner
Naval Research Laboratory
gardner@lcp.nrl.navy.mil
202-767-6582