DESC0024245

STATEMENT OF THE PROBLEM Though decades of computational and experimental studies have advanced our understanding of coherent synchrotron radiation, next generation accelerators will produce beams that no longer conform to our existing models. Moreover, high fidelity simulations of coherent synchrotron radiation and space-charge that utilize numerical solutions to the exact point particle potentials are computationally very expensive rendering them impractical for use in the optimization of new accelerators. GENERAL STATEMENT OF HOW THE PROBLEM IS BEING ADDRESSED Our approach is to perform a systematic evaluation of different coherent synchrotron radiation solvers used to model different regimes. We will then build robust machine learning based integrators that will speed up the calculation of coherent synchrotron radiation wakes for used in optimization with conventional particle tracking codes. During Phase II we will explore new physics models that can accurately capture these effects in new regimes. WHAT IS TO BE DONE IN PHASE I? During Phase I we will perform a detailed simulation campaign and benchmark against archive data collected at a representative facility. We will then develop machine learning integrators and build in robustness by developing new domain transfer methods that utilize autoencoders. Finally we will evaluate the efficacy of implementing our solvers with particle tracking codes for use in the modeling and optimization of advanced concept accelerators. COMMERCIAL APPLICATIONS AND OTHER BENEFITS The machine learning integrators developed under this proposal will no doubt improve the ability to optimize novel accelerators but will also extend beyond accelerator technology. Machine learning is a fast growing field, our innovative approach to domain transfer will no doubt have far reaching applications in scientific computing.