STATEMENT OF THE PROBLEM The high energy physics community has identified four grand challenges in accelerator beam physics, the fourth one being to advance simulation capabilities by orders of magnitude in speed and fidelity, enabling virtual particle accelerators for wide-ranging applications. Success requires the development of simulation and design tools ranging from first-principles electromagnetic algorithms executing on supercomputers to reduced-model algorithms that execute in seconds on a single GPU. Effective use of limited resources dictates close collaboration between national research laboratories, universities, and industry. GENERAL STATEMENT OF HOW THE PROBLEM IS BEING ADDRESSED A unique reduced-model algorithm, which achieves correct physics for high-current traveling wave linear accelerators, executing 1,000 times faster than other codes, will be ported to a state-of-the-art parallel computing framework. With an expected 100x speedup on a GPU, as compared to serial CPU execution, this new code will join an emerging suite of massively parallel accelerator physics codes. In addition to addressing grand challenge problems, the code will be made available in the cloud with a browser-based GUI for industrial and medical designers. This open source effort will be coordinated with the community. WHAT IS TO BE DONE IN PHASE I? The core algorithms of the serial C++ code will be ported to a modern computing framework that supports heterogeneous parallel computing with a single code base. Parallel performance will be documented and tested. An idealized photocathode electron gun with solenoid focusing and a traveling wave linac will be simulated in regimes of unprecedentedly high phase space densities. An open beta testing program will be conducted, including use of the browser-based GUI to engage both industrial designers and lab scientists. Generalizations of the underlying equations of motion will be explored, including a strong collision model. COMMERCIAL APPLICATIONS AND OTHER BENEFITS The parallel code to be developed here, when executed on one or more GPUs, will be five orders of magnitude faster than expensive commercial codes in use now by industrial linear accelerator designers. This capability, together with an easy-to-use online GUI, will revolutionize industrial linac design, empowering small-to-medium sized US enterprises to innovate and effectively compete with the dominant international hardware vendors. An open-source business model will maximize benefits to the scientific community, as the proposed code becomes an important member of the emerging suite of massively parallel particle accelerator simulation tools.
