Parametric modeling of transverse phase space of an rf photoinjector

Abstract

High brightness electron beam sources such as rf photo- injectors as proposed for SASE FELs must consistently produce the desired beam quality. We report the results of a study in which a combined neural network (NN) and first- principles (FP) model is used to model the transverse phase space of the beam as a function of quadrupole strength, while beam charge, solenoid field, accelerator gradient, and linac voltage and phase are kept constant. The parametric transport matrix between the exit of the linac section and the spectrometer screen constitutes the FP component of the combined model. The NN block provides the parameters of the transport matrix as functions of quad current. Using real data from SLAC Gun Test Facility, we will highlight the significance of the constrained training of the NN block and show that the phase space of the beam is accurately modeled by the combined NN and FP model, while variations of beam matrix parameters with the quad current are correctly captured. We plan to extend the combined model in the future to capture the effects of variations in beam charge, solenoid field, and accelerator voltage and phase.