GPU Acceleration of Spin Tracking in Colliding Beam Accelerators
Small Business Information
5621 Arapahoe Ave, Boulder, CO, 80303-1379
AbstractTo elucidate the mysterious origins of nuclear spin, the Nuclear Science Advisory Committee (NSAC) has identied the science of electron-ion colliders, and specically the proposed polarized electron and ion collider, as absolutely central to U.S. science. These machines, estimated to cost as much as 500M 1B, will require highly polarized particle beams. To reduce the risk associated with building these machines, scientists need accurate simulations of spin dynamics in colliding beam accelerators. However current spin tracking technology takes many days to track an ensemble of particles across a single resonance in RHIC. We plan to develop a spin-orbit code capable of simulating a realistic distribution of particles through the typical acceleration cycle on a manageable compute time scale. This will be accomplished in a well-maintained and stable software development environment. This will put into the hands of DoE scientists an essential and maintainable tool necessary for performing fast and accurate simulations of spin-polarized particle beams. We are developing algorithms and code that take advantage of the highly-parallel and relatively inexpensive compute power of modern graphics processing units (GPUs). In Phase I we benchmarked, corrected and MPI parallelized the existing TEAPOT-SPINK code. We also developed a self-contained spin-orbit transport version of this code for the GPU capable of tracking 1000s of particles 50 -100 times faster at double precision than serial execution on a 2.4 GHz CPU. We studied the magnitude of element slicing necessary as function of energy and resonance proximity. As well we developed an approach to handle dynamic changes in optics. The approach developed in Phase I to simulate dynamic changes in the lattice under acceleration and beta squeeze needs to implemented. Realistic helical dipole spin rotator and snake models will be developed parameterized by energy and /or current. Additionally the UAL infrastructure stands in need of upgrading via the development of a better build system, tool chains, lattice parsers and establishment of regression tests to ensure the stability of this software tool for use by future accelerator physicists. Commercial Applications and Other Benets: The proposed software development will directly benet scientists working to design high-luminosity polarized beam accelerators required for fundamental advances in experimental nuclear physics. Computational scientists working on this project will develop expertise in this area, creating opportunities for future contract work. In addition, the development of new computational algorithms for graphics processing units will provide additional technology and opportunities for applications and consulting.
* information listed above is at the time of submission.