Computational Accelerator Physics Grand Challenge: Accomplishments

Accomplishments Update

The latest information on Grand Challenge progress is available as a PDF file.

Accomplishments

In 1997 we developed a parallel version of a code called LINAC, which is being used to support the Accelerator Production of Tritium (APT) project. This is now the primary code for performing large-scale beam halo simulations for the project. One of our main accomplishments this year was to achieve a 5-fold increase in the performance of the space charge algorithm over our initial implementation. In addition to LINAC, we developed a new code called IMPACT, which is based on split-operator techniques. IMPACT (Integrated-Map and Particle Accelerator Tracking code) has an especially accurate and efficient treatment of radio frequency accelerating gaps, obtained by numerical integration of the gap transfer map rather than integrating single particle trajectories. The code is especially useful for modeling superconducting proton linear accelerators, where there are only a few types of accelerating cavities. A third code, called HALO, has been developed specifically for beam halo studies. Beam loss is known to be associated with the very low-density distribution of charge far from the beam core (the halo). This is a major issue for future high intensity linacs, since the particle loss can lead to activation of accelerator components, thereby hindering or preventing hands-on maintenance. HALO was developed in collaboration with University of Maryland. It includes a new three-dimensional beam equilibrium model, based on analytical work of the Maryland group, which helps isolate beam halo growth mechanisms.

The huge amount of data in a high-resolution beam dynamics simulation, coupled with the fact that we are often interested in the very small fraction of the particles in the halo, necessitates the use of internal data analysis in our codes prior to storing simulation results. We have developed and implemented algorithms on the T3E that drastically reduce the amount of data needed to visualize the halo. An example is shown in the figure, which is the result of a HALO simulation. The system being modeled is a spheroidal bunch, initially a stationary solution of the Vlasov/Poisson equations, which develops a halo due to improper matching into the beamline. The simulation used 25 million particles and a 256x256x256 grid for the Poisson solver. The figure shows the longitudinal phase space, (z,pz), after the halo has formed. It contains approximately 100,000 particles color-coded according to density. In the blue region (the core), the density equals 1 at the center and 1/10 at the blue-green boundary. It decreases through the green, yellow, and red regions, and equals 1/10,000 at the edge of the red region. If we had simply plotted the same number of particles chosen randomly from the 25 million in the simulation, the halo would barely be visible and show almost no structure.

Two and three dimensional eigenmode solvers for computing normal modes in accelerator cavities have been developed using quadratic elements with mesh refinement capabilities. A parallel version of the two-dimensional module has been used to calculate the dipole wakefields in a tapered accelerator section (Detuned Structure) consisting of 206 individual cavities. The figure at right shows some of the modes in the section. These are the detuned modes which lead to reduced transverse wakefields due to decoherence. Work is in progress to model the Damped Detuned Structure (DDS) which incorporates damping by coupling each cell to external waveguides. This is a three-dimensional structure, and the figure below displays the wall loss due to the accelerating mode in one cell. The simulation of a complete section of many DDS cells will require a parallel 3D solver that utilizes substantial high performance computing resources.

The utilization of object-oriented methodologies is one of the features of this Grand Challenge. In order to improve the usability and maintenance of our codes we are in the process of implementing both C++-based and F90/HPF-based paradigms. A 2D linac code has recently been rewritten utilizing the POOMA framework. An extended abstract describing this work is available here.

Back to main page.