Bayesian optimization of beam injection and storage in the PSI muEDM Experiment

Abstract: The muEDM experiment at the Paul Scherrer Institute aims to measure the electric dipole moment with an unprecedented sensitivity of $6 \times 10^{-23}\,\mathrm{e}\cdot\mathrm{cm}$. A key aspect of this experiment is the injection and storage of the muon beam, which traverses a long, narrow superconducting channel before entering a solenoid magnet. The muon is then kicked by a pulsed magnetic field into a stable orbit within the solenoid's central region, where the electric dipole moment is measured. To study the beam injection and storage process, we developed a G4beamline simulation to model the dynamics of beam injection and storage, incorporating all relevant electric and magnetic fields. We subsequently employed a Bayesian optimization technique to improve the muon storage efficiency for Phase I of the muEDM experiment. The optimization is demonstrated using data simulated by G4beamline. We have observed an enhancement in the beam injection and storage efficiency, which increased to 0.556\% through the utilization of Bayesian optimization with Gaussian processes, compared to 0.324\% when employing the polynomial chaos expansion. This approach can be applied to adjust actual experimental parameters, aiding in achieving the desired performance for beam injection and storage in the muEDM experiment.