GPGPU Approach: Simulation of the Interaction of Heavy Interstellar Atoms with the Heliosphere

Monday, 15 December 2014
Anthony DeStefano, University of Alabama in Huntsville, Huntsville, AL, United States
Running simulations is an involved process taking many hours of computational time to complete. With the advent of cluster computing and parallel processing, problems may be solved in much less time compared to those run in serial. Specifically, NVIDIA released the parallel computing platform CUDA in 2007 giving researchers and programmers access to the GPU to solve generalized problems, and not those of just images.

In current research, code has previously been developed to study the interaction of the heliosphere and heavy atoms from the local interstellar medium.

Ionized species of hydrogen, helium and other heavy atoms are deflected by the heliosphere where as the neutral species are relatively unimpeded. Charge exchange of these neutral particles may occur between ionized species originating from the solar wind or other populations of pickup ions (PUI) modifying the shape and properties of the heliosphere, compared to one without neutrals. The details of the charge exchange interaction are element dependent and need to be investigated one by one. Current research has studied the interaction of local interstellar hydrogen with the heliosphere quite extensively with theory, simulations and modeling.

Since hydrogen is the most abundant element care must be taken when coupling MHD equations with the charge exchange interactions. Simulation code has been developed to account for this dynamic problem and they have shown that the shape of the heliosphere is affected by this. Interstellar atoms heavier than hydrogen interacting with the heliosphere has been looked at as well, but not nearly with as much detail or sophisticated models as hydrogen. The heavy atom data collected by IBEX has in this sense been under-utilized by models.

Previously, the simulation was computed with the use of MPI (Message Passing Interface) for parallelization. This approach provided a decrease in computational time. However, CUDA enables the programmer to take advantage of the computer's hardware allowing for more code/memory optimization. The results of this work demonstrate a respectable decrease in computation time of the CUDA code compared to the MPI code. In addition, the distribution, temperature, and direction of primary and secondary ions and neutrals were found for that of helium and oxygen in and around the heliosphere.