Surrogate Models studies for laser-plasma accelerator electron source design through numerical optimisation

• Published in: arXiv.org 2024-08
• Main Authors: Kane, G, Drobniak, P, Kazamias, S, Kubytskyi, V, Lenivenko, M, Lucas, B, Serhal, J, Cassou, K, Beck, A, Specka, A, Massimo, F
• Format: Article
• Language: English
• Subjects: Bayesian analysis; Decision trees; Design optimization; Design parameters; Electron beams; Gaussian beams (optics); Gaussian process; Injectors; Lasers; Machine learning; Multilayer perceptrons; Plasma density
• ISSN: 2331-8422

The optimisation of the plasma target design for high quality beam laser-driven plasma injector electron source relies on numerical parametric studies using Particle in Cell (PIC) codes. The common input parameters to explore are laser characteristics and plasma density profiles extracted from computational fluid dynamic studies compatible with experimental measurements of target plasma density profiles. We demonstrate the construction of surrogate models using machine learning technique for a laser-plasma injector (LPI) electron source based on more than 12000 simulations of a laser wakefield acceleration performed for sparsely spaced input parameters [1]. Surrogate models are very interesting for LPI design and optimisation because they are much faster than PIC simulations. We develop and compare the performance of three surrogate models, namely, Gaussian processes (GP), multilayer perceptron (MLP), and decision trees (DT). We then use the best surrogate model to quickly find optimal working points to get a selected electron beam energy, charge and energy spread using different methods, namely random search, Bayesian optimisation and multi-objective Bayesian optimisation