Speaker
Description
Modern accelerator design requires extensive beam dynamics simulations and efficient exploration of large parameter spaces. Complex nonlinear dynamics and collective effects make these tasks computationally demanding. Artificial intelligence (AI) offers opportunities to accelerate beam dynamics simulation and improve the efficiency of accelerator design optimization. This talk revisits how AI approaches can connect computationally efficient surrogate models with effective design optimization. Based on our previous work, we discuss strategies for constructing fast surrogate models for nonlinear beam dynamics and collective effects, and demonstrate how machine learning can enhance conventional multi-objective optimization algorithms. Representative examples show the surrogate models can achieve percent-level accuracy while reduce computing time by orders of magnitude, enabling broader exploration of accelerator design space within limited computing time and/or resources.