Bayesian optimization of beam quality of plasma accelerated electron beams
Examensarbete för masterexamen
Laser-plasma acceleration is a promising novel technique for acceleration of charged particles. A challenge with this technique is to maintain a high quality of the accelerated particle bunch. In particular, for accelerated electrons used in Free Electron Lasers, charge and shape of the energy spectrum are important parameters. The aim of this project has been to evaluate and examine the use of Bayesian optimization with respect to these parameters on the LUX laser-plasma accelerator. The focus was to consider how the Bayesian optimization performed under noisy conditions. An important part of Bayesian optimization algorithms is the acquisition function which determines the next point to evaluate in the optimization iteration. In this thesis, two acquisition functions were compared and evaluated from the performance point of view. In order to test and develop the algorithms, Particle-In-Cell (PIC) simulations were used to emulate the LUX experiment. Further, for cheaper evaluation, a model of the target surface was built from a vast amount of PIC simulated data using Gaussian process regression. With this model, different sampling strategies for each parameter set-point could be investigated. Noise was added to the input parameters as well, yielding a more realistic imitation of the system. A significant improvement was seen when the mean value of 20 input parameters and the mean value of corresponding outputs were fed to the Bayesian optimization algorithm.
Gaussian processes , Bayesian optimization , Laser-plasma acceleration , Wakefield acceleration , Noisy Expected Improvement , Expected Improvement