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

Q: Fachverband Quantenoptik und Photonik

Q 68: Quantum Gases: Bosons V

Q 68.1: Vortrag

Freitag, 10. März 2023, 14:30–14:45, B305

Integrating physical intuition into neural networks for potential reconstruction in ultracold atoms — •Miriam Büttner and Axel U. J. Lode — Institute of Physics, Albert-Ludwig University of Freiburg, Hermann-Herder-Strasse 3, 79104 Freiburg, Germany

Ever since the rise of interest in Bose-Einstein Condesates (BECs), the research field of interacting ultracold indistinguishable particles has expanded, both in its experimental realizations as well as in its theoretical descriptions.In this work, we present a physically motivated neural network architecture for the extraction of quantum observables from single-shot measurements of ultracold atoms. The focus of the work is put on the inclusion of physical intuition into such a network architecture. Our proposed architecture utilizes extended pre-processing that exploits the stochastic nature of the measurement results, given that the so called single-shot images consist of samples of the N-body density. As we are extracting an external potential from samples of a density, our loss function takes inspiration from the constrained-search approach to density functional theory. We thus demonstrate, that in a way similar to inverse density functional methods, a Bose-Einstein condensate*s external potential can be reconstructed.