Bonn 2025 – wissenschaftliches Programm

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

A: Fachverband Atomphysik

A 11: Poster – Cold Atoms and Molecules, Matter Waves (joint session Q/A/MO)

A 11.3: Poster

Dienstag, 11. März 2025, 14:00–16:00, Tent

Machine learning optimized time-averaged potentials — •Max Schlösinger¹, Oliver Anton¹, Victoria Henderson^{1, 3}, Elisa Da Ros¹, Mustafa Gündoğan¹, Simon Kanthak¹, and Markus Krutzik^{1, 2} — ¹Humboldt-Universität zu Berlin, Institut für Physik, Newtonstraße 15, 12489 Berlin, Germany — ²Ferdinand-Braun-Institut (FBH), Gustav-Kirchhoff-Straße 4, 12489 Berlin — ³now: RAL Space, Fermi Ave, Harwell, Didcot OX11 0QX, United Kingdom

Time-averaged potentials (TAPs) are a versatile tool for the generation and manipulation of ultracold atom clouds. Using a CCD-based setup to characterize a 2D acousto-optic deflector (2D-AOD) system, we implement and test machine learning routines to optimize 2D geometries, such as harmonic potentials. This approach allows us to compare different methods, evaluate metrics like homogeneity, and improve the predictability of the resulting potentials.

By employing optimization algorithms such as CMA-ES and various Bayesian optimizers, we compare their performance in terms of speed and efficiency. Additionally, we plan to implement an active learning optimizer to minimize the number of required iterations, which is crucial for future integration into a ⁸⁷Rb Bose-Einstein condensate (BEC) experiment. Ultimately, these advancements will enhance the evaporative cooling routine and improve the performance of a ⁸⁷Rb BEC-based quantum memory [1].

[1] Phys. Rev. Research 5, 033003 (2023)

Keywords: time-averaged potentials; machine learning; acousto-optic deflector; Bose-Einstein condensate; quantum memory