Dortmund 2021 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

AKBP: Arbeitskreis Beschleunigerphysik

AKBP 11: Diagnostics, Control and Instrumentation III

AKBP 11.8: Talk

Thursday, March 18, 2021, 18:15–18:30, AKBPa

Optimization of laser wakefield accelerators using machine-learning methods — •Faran Irshad, Andreas Döpp, and Stefan Karsch — Ludwig-Maximilians-Universität München, Am Coulombwall 1, 85748 Garching, Germany

With advances in both laser technology and data-acquisition software, the number of parameters that can be tuned to directly influence the behavior of the Laser Wakefield accelerators (LWFA) have increased. One concern with LWFA has been one of stable operation and time spent in optimizing the electron beams. While the former is an engineering challenge, the latter can be improved using techniques from machine learning. Conventionally, optimizing LWFA parameters have been done by human operators on the run. We propose a more systematic approach to optimizing these parameters using sequential optimization on a Gaussian process (GP) model commonly referred to as Bayesian Optimization. Bayesian optimization treats the LWFA as a black-box function and uses GP to model it. The important features of the LWFA such as electron beam energy or peak charge are captured in an objective function defined by the user. This method was applied to FBPIC simulations to optimize plasma parameters such as plasma electron density and laser parameters such as focus position and spectral phase. We show a faster convergence of the laser-plasma interaction to the optimal settings in a computational experiment. In conclusion, the Bayesian optimization can significantly reduce optimization times of a LWFA and help achieve subtle tuning of parameters that otherwise are missed in single-variable scans.