Machine Learning Optimization of E-Beam Transport for a Superradiant FEL
We present an optimization procedure using machine learning (ML) libraries for optimization of electron beam transport for maximal bunch compression and optimal operation of a bunched-beam Superradiant FEL. This is exemplified for the parameters of the 6MeV ORGAD Accelerator at Ariel University that is driving a THz Superradiant waveguide FEL. For superradiant emission (proportionally to the number of electrons squared), the bunch duration ${\sigma}_t$ at the undulator should be shorter than the optical period (2${\pi}/{\omega}$) of the radiation. Also, the beam transport optimization must confine the transverse dimensions of the beam to enter the undulator waveguide. The variables of the ML Bayesian optimization are the RF parameters and the currents of the coils and quads along the beamline. The beam dimensions and duration are provided from full 3D GPT simulations that are automatically driven by the ML exploration and exploitation algorithms. Twenty-five simulation iterations sufficed to arrive to an optimal beam transport design.