Implementing Bayesian optimization to compare injection regimes and pulse propagation in two stage LWFA

Predicting the properties of electrons accelerated to relativistic energies via laser wakefield acceleration is dependent on a wide range of parameters that are difficult to efficiently optimize. Such problems would benefit from the use of Bayesian optimization schemes, which if implemented correctly may efficiently find optimums in complex parameter spaces. By separating the laser wakefield mechanism into two stages, an initially higher density plasma of high Z gases may be used to inject higher charge while a low density homogeneous plasma can be used to efficiently accelerate electrons with the same laser pulse. However, the nonlinear relationship between the two stages makes the determination of optimal beam parameters difficult. We show that through the use of a Bayesian optimization process on 2.5D3V particle-in-cell simulations we are able to efficiently map out regions of interest across a wide variety of physical regimes while keeping down the computational costs of the parameter scan. Numerically modeling the two stage system, we varied the injection stage density profile and length and examined the impact on electron beam charge, energy spectrum, and emittance. We also examined the impact on the laser pulse itself with regards to the temporal pulse duration. Our analysis identified specific regions in the parameter space where variations in pulse propagation and injection processes significantly influenced the characteristics of the resulting electron beam. This approach enabled us to efficiently identify numerous physical regimes with options on how to tune the electron beam for various applications.