SLAC, Menlo Park, California, USA
LANL, Los Alamos, New Mexico, USA
High-brightness XFELs are demanding for many users, in particular for certain types of imaging applications. Self-seeding XFELs can respond to a heavily tapered undulator more effectively, therefore seeded tapered FELs are considered as a path to high-power FELs in the terawatts level. Due to many effects, including the synchrotron motion, the optimization of the taper profile is intrinsically multi-dimensional and computationally expensive. With an operating XFEL, such as LCLS, the on-line optimization becomes more economical than numerical simulation. Here we report recent on-line taper optimization on LCLS taking full advantages of nonlinear optimizers as well as up-to-date development of artificial intelligence: deep machine learning and neural networks.
