This report presents a Bayesian optimization study of the silicon tracking system of the B0 far-forward detector in ePIC at the Electron-Ion Collider. The optimization explores the longitudinal positioning of four silicon tracking disks together with the distribution of active sensor regions in order to improve the angular reconstruction of charged particles emitted at small polar angles with respect to the outgoing hadron beam line. The study is performed using Geant4-based detector simulations within the ePIC software framework coupled to a Gaussian-process Bayesian optimization procedure. The optimization systematically converges to detector layouts leading to improved angular reconstruction performance by 10-20\% over a realistic range of particles and momenta. A positive correlation is observed between the tracking resolution and the geometric lever-arm parameter, indicating that the optimization is primarily driven by geometric effects. However, the optimized configurations outperform the expectations of an optical model. This suggests that there are additional contributions from particle transport through the magnetic field and the active regions on the disk surfaces. Future work could focus on more realistic track-reconstruction models, including the reconstruction of the vertical angle.
Fraïsse et al. (Mon,) studied this question.