Phenomenological model of a free-electron laser using machine learning

Abstract

Free electron lasers (FELs) are used in various fields of scientific research. Programs and methods are created for their design and calibration. The development of machine learning has opened up opportunities for new methods of research and data analysis. This paper presents a technique for building a neural network for analyzing FEL parameters. We collected numerical simulation data of about 2000 configurations, found the optimal architecture and trained a neural network that can analyze several FEL configurations depending on the undulator and the electron beam parameters in a short time. This technique is capable of simulating more complex systems (FEL with helical undulators, etc.) and can be applied to facilities for their optimization.

Three-dimensional, time-dependent simulation of tapered EUV FELs with phase shifters

“…The proposed design is based on shortperiod undulators (2 cm and 2.4 cm) and 440-484 MeV electron beam energy. The simulations were carried out with the 3D time-dependent program [16][17][18] in the SASE mode [19] and amplifier mode.…”