Multi-particle physics simulations for Accelerator Design

The ASTeC/ Cockcroft Institute have a requirement to confirm the performance of many aspects of their accelerator designs through multi-particle simulations. These simulations are generally computationally expensive and require significant computing resource, and are also typically of the type requiring large numbers of processors but not needing significant communication between them. An example would be performing a start-to-end calculation many times with different starting conditions, to derive statistical information about accelerator tolerances. There are some applications where more parallelism would be desirable.

A typical simulation, and one which we have an initial requirement for, is to perform error analysis for the 4GLS start-to-end accelerator simulation; these are 'inherently parallel' simulations. A number (say 100) of initial conditions are randomly seeded: each initial condition takes the form of a macro-particle distribution of up to 1M particles, which are then transported through an electro-magnetic system under the action of self-forces such as space charge and coherent radiation. The different initial seeds are tracked using the same simulation code; the results can then be analysed statistically to determine the positional tolerances required for each magnetic element in the accelerator.

Initially, we would like to run the following codes:

• ASTRA; Linux, space-charge code
• Elegant; Linux, CSR transport code

Both codes take a set of input files and produce a set of output files in a simple batch-like way:

(input files) - code - (output files)

Presently, we do custom pre- and post-processing of the data with a variety of tools (Python, SDDS, Mathematica), and control/ host the siumulation desks using Eclipst/ Subversion.

Update 7/6/2009

Parallel Elegant v22.0.1 is now installed on the DL1 cluster