Parallel Computing for Accelerator Simulation

Simulation of the dynamics of beams in particle accelerators is an important tool in the design of new particle accelerators; detailed study and optimization of existing machines; Model based Accelerator Control. In recent years, with new structures and new methods of acceleration being considered, and with the growing importance of high current accelerators, namely machines in which non linear space charge forces are a determining factor, simulation in the full 6-dimension phase space is playing an increasing important role. Accelerator simulation is mostly done by numerically tracking a large number of representative macro particles through the machine lattice (PIC or Water-Bag techniques). Collective space charge forces are applied to each individual particle. These forces are due to the other particles, in the presence of image charges and currents being induced on the walls of the acceleration chamber. Parallel computing is the best answer to the needs of fast simulation. Engineering uid dynamics problems involving the prediction of turbulent ows with direct numerical simulation (DNS) are still beyond the capability of the most state-of-the-art hardware. Large-eddy simulation (LES) is a viable alternative to DNS in which certain portion of the detailed small-scale information resolved in DNS is suppressed. This simpliication makes the computational problem more tractable. However, even with LES, today's challenging problems require accurate schemes with higher grid resolution. In addition to these, shorter turnaround time to approximate the instantaneous ow behavior has become more important. As part of a research eeort towards the large-eddy simulation of diesel combustion engines, widely used KIVA-3 code originally developed by Los Alamos National Laboratory 1] was restructured for parallel execution via message-passing. One-dimensional domain decomposition approach initially proposed & developed on Intel Paragon system by Yasar 2] was successfully implemented into a generic KIVA-3 code from scratch with substantial improvements (Gel 3]). KIVA-3 is a general computational uid dynamics (CFD) code based on the Arbitrary Lagrangian-Eulerian (ALE) method with enhanced features for internal combustion engine applications. The method is typically implemented in three phases. The rst phase is an explicit Lagrangian update of the equations of motion. The second phase is an optional implicit phase that allows sound waves to move many computational cells per time step if the material velocities are smaller than the uid sound speed. The third and nal phase is the remap phase where the solution from the end of phase two is mapped back onto an Eulerian grid …