Description
Laser wakefield acceleration (LWFA) has proven highly effective in producing high-energy electron beams and bright femtosecond X-ray pulses. However, achieving high conversion efficiency remains a key challenge for widespread applications. In this work, we numerically investigate the accelerator–radiator scheme -- a method that combines a low-density acceleration stage with a higher-density radiator to amplify transverse electron oscillations and boost X-ray yield. To navigate the complex parameter space, we employ Bayesian optimization (BO) within multidimensional particle-in-cell simulations, varying the radiator's position and density to maximize the laser-to-X-ray energy conversion efficiency, $\eta = W_{\mathrm{X-ray}} / W_{\mathrm{laser}}$. The optimization successfully identifies a high-yield regime where the refocusing of the nearly depleted laser pulse pairs with electron rephasing in the radiator gradient, driving higher electron energies and enhanced X-ray emission. These results demonstrate that BO provides a powerful, computationally efficient framework for maximizing LWFA-based radiation sources and uncovering the underlying nonlinear laser-plasma dynamics.
