Abstract
The algebraic model (ALG) proposed by the authors has sufficiently high accuracy in calculating the motion of a test particle with all the field particles at rest. When all the field particles are moving, however, the ALG has relatively poor prediction ability on the motion of the test particle initially at rest. Nonetheless, the ALG approximation gives a good results for the statistical quantities, such as variance of velocity changes or the scattering cross section, for a sufficiently large number of Monte Carlo trials. We have implemented a graphics processing unit (GPU) using NVIDIA's CUDA architecture into the ALG scheme for Coulomb multibody problems. For N=28-body problem, the ALG calculations on the GPU is several times faster than on a typical CPU. The achieved speedup ratios on an NVIDIA GTX-285 are 10.5 and 2500 against the ALG-CPU and the DIM-CPU, respectively both on an Intel Celeron @3.06 GHz.
