Dresden 2020 – wissenschaftliches Programm

Die DPG-Frühjahrstagung in Dresden musste abgesagt werden! Lesen Sie mehr ...

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

TT: Fachverband Tiefe Temperaturen

TT 36: Many-body Systems: Equilibration, Chaos and Localization II (joint session DY/TT)

TT 36.12: Vortrag

Mittwoch, 18. März 2020, 12:30–12:45, HÜL 186

Learning many-body localization indicators directly from the Hamiltonian — •Alexander Gresch, Lennart Bittel, and Martin Kliesch — Heinrich-Heine University, Düsseldorf, Germany

Many-body localization (MBL) captures the phenomenon that the propagation of correlations in disordered quantum systems can be strongly suppressed due to interference effects. Quantum systems undergoing MBL do practically not equilibrate but instead preserve local signatures of their initial conditions for arbitrarily long times. This property makes such systems potential candidates for storage devices in quantum computation. However, a full analytical understanding has not been achieved and numerical approaches have to deal with the exponential growth of the Hilbert space dimension. Hence, approximate methods have been proposed. A recent approach uses artificial neural networks for distinguishing MBL states from non-localized ones, which allows to calculate the phase diagram of the transition. The already proposed deep learning schemes require an expensive preprocessing, e.g. the eigenstates of the Hamiltonian or their entanglement spectrum, as inputs.

In this work, we investigate the Heisenberg spin chain with random local magnetic field. We demonstrate that an MBL-prediction is possible from the given disorder parameters alone without any preprocessing. We guide the learning process via several different indicators for MBL that have previously served as a basis for numerical studies. Here, we provide new insights in their predictive capabilities from a machine learning perspective.