Surrogate Modelling for Boosting Research of Electron Acceleration Processes

Recent studies of laser plasma acceleration processes feature increasing requirements to the
technical equipment and time consumption in both numerical and experimental research.
This rising demand on statistical and mathematical methods for inversion of the system
state, comprehension of measurement data and quantification of data stability can only be
met by a comprehensive machine learning based surrogate model for Laser-driven Plasma
Accelerators (LPA). This surrogate potentially accelerates theoretical comprehension of the
system, novel means for design space exploration and promises reliable in-situ analysis of
experimental data which leads to novel guidance mechanisms for future LPA experiments.
The main aim of our work is to elaborate a surrogate model for electron acceleration
processes by virtue of that one could unveil beam dynamics on the scope of collected
diagnostic. Recently achieved results on laser-wakefield electron acceleration, demonstrate
the capability to learn an approximation of the data-dependent posterior distribution by
conditional inventible neural networks. The further derived model is able to describe an
electron bunch transformation in the simulated beam transport in terms of phase space
particle distribution based on its initial parameters: divergence and size. This step opens a
perspective to a potential elaborated model that could use obtained diagnostics for
reconstructions at any point in the electron beamline consisting of conventional magnetic
elements.