• P. Messmer, T.M. Austin, J.R. Cary
  Tech-X, Boulder, Colorado

Funding: This project was in part supported by DOE Office of Advanced Scientific Computing Research SBIR Phase II grant #DE-FG02-07ER84731, SciDAC Grant #DE-FC02-07ER41499, and Tech-X Corporation.

In order to meet the design and budget constraints of next generation particle accelerators, individual components have to be optimized using numerical simulations. Among the optimizations are the geometric shape of RF cavities and the placement of coupler and dampers, requiring large numbers of simulations. It is therefore desirable to accelerate individual cavity simulations. Finite-Difference Time-Domain (FDTD) is a widely used algorithm for modeling electromagnetic fields. While being a time-domain algorithm, it can also be used to determine cavity modes and their frequencies. Weak scaling of parallel FDTD yields good results due to the algorithm locality, but the maximum speedup is determined by the usually small problem size. Graphics Processing Units (GPUs) offer a huge amount of processing power and memory bandwidth, well suited for accelerating FDTD simulations. We therefore developed an FDTD solver on GPUs and incorporated it into the plasma simulation code VORPAL. We will present GPU accelerated VORPAL simulations, provide speedup figures and address the effect of running these simulations in single precision.

• B.M. Cowan, D.L. Bruhwiler, P. Messmer, K. Paul
  Tech-X, Boulder, Colorado
• E. Cormier-Michel, E. Esarey, C.G.R. Geddes
  LBNL, Berkeley, California

Funding: Work supported by Department of Energy contracts DE-AC02-05CH11231 (LBNL), DE-FC02-07ER41499 (SciDAC), and DE-FG02-04ER84097 (SBIR).

Simulation of laser wakefield accelerator (LWFA) experiments is computationally highly intensive due to the disparate length scales involved. Current experiments extend hundreds of laser wavelengths transversely and many thousands in the propagation direction, making explicit PIC simulations enormously expensive. We can substantially improve the performance of LWFA simulations by modeling the envelope modulation of the laser field rather than the field itself. This allows for much coarser grids, since we need only resolve the plasma wavelength and not the laser wavelength, and this also allows larger timesteps. Thus an envelope model can result in savings of several orders of magnitude in computational resources. By propagating the laser envelope in a Galilean frame moving at the speed of light, dispersive errors can be avoided and simulations over long distances become possible. Here we describe the model and its implementation. We show rigorous studies of convergence and discretization error, as well as benchmarks against explicit PIC. We also demonstrate efficient, fully 3D simulations of downramp injection and meter-scale acceleration stages.