Freiburg 2024 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

A: Fachverband Atomphysik

A 35: Poster VIII

A 35.10: Poster

Donnerstag, 14. März 2024, 17:00–19:00, Tent C

Applying machine learning optimization to a transfer beamline for highly charged ions — •Elwin A. Dijck, Vera M. Schäfer, Stepan Kokh, Lukas F. Storz, Christian Warnecke, Thomas Pfeifer, and José R. Crespo López-Urrutia — Max Planck Institute for Nuclear Physics, Heidelberg

We optimize the production and transport of highly charged ions (HCIs) through a low-energy beamline that serves to decelerate and inject HCIs produced by an electron beam ion trap (EBIT) into a cryogenic radiofrequency trap for precision spectroscopy experiments [1]. The parameters to be optimized include EBIT settings, several dozen electrode voltages of electrostatic ion optics, as well as the timing of voltage pulses for deceleration, charge state selection and re-capture of HCI bunches. The online optimization is implemented using the open-source software package M-LOOP, which includes the machine learning methods of Gaussian process regression and a gradient-based approximator using an artificial neural network. The automated procedure allows faster optimization, as well as the investigation of apparatus stability over time. We discuss defining appropriate cost functions and the results obtained.

[1] Dijck et al., Rev. Sci. Instrum. 94, 083203 (2023)

Keywords: Ion optics; Machine learning; M-LOOP; Electron beam ion trap