Tuning of self-truncated ionization injection with Bayesian optimization

Abstract The complex interconnection between laser and plasma parameters in laser wakefield electron acceleration (LWFA) makes electron beam optimization based on simple models challenging, often non-intuitive, and thus time-consuming. However, for quality-demanding applications such as free-electron lasers (FELs), rapid optimization toward target parameters is essential. In this paper, we demonstrate parameter tuning of an LWFA operated in a tailored self-truncated ionization injection (STII) regime, assisted by a single-objective Bayesian optimization routine, to meet the requirements for the LWFA-driven seeded FEL COXINEL beamline at Helmholtz–Zentrum Dresden-Rossendorf. By using a Gaussian process with an upper confidence bound acquisition function optimized to balance exploration and exploitation, a high spectral charge density of 9.8 pC MeV − 1 centered at 196 MeV is achieved in fewer than 160 shots. This beam initialization, which takes 26 min, is comparatively faster than solely relying on human-assisted tuning taking up to 2 h. Exploiting a total of 2720 shots, a unique correlation between input and output parameters is revealed, advancing the understanding of the complex parameter space of STII-LWFA.