R. Ryne, S. Habib, and J. Qiang, Los Alamos National Laboratory
K. Ko, N. Folwell, Z. Li, B. McCandless, C. Ng, and M. Wolf,
Stanford Linear Accelerator Center
G. Golub, W. Mi, M. Saparov, and Y. Sun, Stanford University
V. Decyk, University of California, Los Angeles
J. Ahrens, T. Cleland, J. Cummings, W. Humphrey, S. Karmesin,
P. McCormick, A. McPherson, J. Painter, G. Mark,
Advanced Computing Laboratory, LANL
E. Ng, W. Saphir, NERSC, Lawrence Berkeley National Laboratory

Research Objectives

The goal of this Grand Challenge is to develop a new generation of accelerator modeling tools, targeted to very large scale (terascale) parallel computing platforms, and to apply them to accelerator projects of national importance. The new capability will enable computations for accelerator design and analysis on a scale that is unprecedented in size, accuracy, and resolution. Specific objectives include the development of parallel beam dynamics codes aimed at simulating, from end to end, intense beams through a variety of accelerator systems, and parallel electromagnetics codes for modeling large, complex beamline components and accelerating structures.

Computational Approach

The beam dynamics component of this research uses parallel particle-in-cell (PIC) techniques, particle managers, dynamic load balancing, fast Fourier transform (FFT) based Poisson solvers, and techniques from magnetic optics. Split-operator methods are used to combine magnetic optics and parallel PIC techniques in a single framework and to establish particle advance algorithms. The electromagnetics component utilizes unstructured grid generation, domain decomposition, adaptive mesh refinement, finite element formulation for the eigenmode solver, and the modified Yee algorithm for the time-domain solver. Systems involving particles in electromagnetic structures are treated using hybrid grids, with a structured mesh in the region of the beam and an unstructured grid near the structure boundaries.

Domain decomposition associated with an Omega3P calculation of the Accelerator Production of Tritium coupled cavity linac.

Accomplishments

Three parallel application codes, IMPACT, Omega3P, and Tau3P, have been developed under the Grand Challenge. The following improvements made during FY99 result in significant increases in performance.

The IMPACT beam dynamics code has seen a performance improvement of a factor of 4 due to a replacement of the original charge deposition/field interpolation routines with a parallel particle manager. Other improvements for FY99 include significantly reduced memory overhead, a choice of parallel particle managers (with fixed and variable message buffers), parallel I/O, and restart capabilities. The POOMA version has been modified to improve the performance of FFTs across boxes on the SGI Origin 2000 system. IMPACT was used in the first systematic study of halo formation due to longitudinal/transverse coupling in charged particle beams. IMPACT was also used to model the Accelerator Production of Tritium (APT) and Spallation Neutron Source (SNS) linacs, including the largest simulations to date of the SNS linac, with 500 million particles. The capability to include machine imperfections was added in order to model more realistic accelerators.

The accomplishments in the electromagnetics area include a new, hybrid Jacobi-Davidson algorithm that dramatically accelerates the eigensolver convergence in Omega3P, and the incorporation of a superior mesh distribution preprocessor in Tau3P that greatly improves its parallel efficiency. Using 128 processors on the T3E, Omega3P can calculate the accelerating mode in the Next Linear Collider (NLC) accelerating structure on the order of minutes for a geometry involving 1 million degrees of freedom, and the code achieves close to linear scalability. In addition, progress has been made in developing a complex solver for Omega3P to treat lossy cavities and in implementing a rigid beam in Tau3P to model wakefield effects.

	Volume rendering of phase space output data from an IMPACT simulation of the Spallation Neutron Source linac.

Significance

Publications

A. V. Fedotov, R. L. Gluckstern, S. Kurennoy, and R. Ryne, "Halo formation in 3D bunches with different phase space distributions," Phys. Rev. ST Accel. Beams 2, 014201 (1999).

R. L. Gluckstern, A. Fedotov, S. Kurennoy, and R. Ryne, "Halo formation in three dimensional bunches," Phys. Rev. E 58, 4977 (1998).

W. Humphrey, R. Ryne, T. Cleland, J. Cummings, S. Habib, G. Mark, and J. Qiang, "Particle beam dynamics simulations using the POOMA framework," Lecture Notes in Computer Science 1505 (1998).

http://t8web.lanl.gov/people/salman/capgca/
http://public.lanl.gov/ryne/gca.html