Time Integration in Quantum Dynamics using Tree Tensor Networks

DSpace Repository


Dateien:

URI: http://hdl.handle.net/10900/161566
http://nbn-resolving.org/urn:nbn:de:bsz:21-dspace-1615662
http://dx.doi.org/10.15496/publikation-102898
Dokumentart: PhDThesis
Date: 2025-02-03
Language: English
Faculty: 7 Mathematisch-Naturwissenschaftliche Fakultät
Department: Mathematik
Advisor: Lubich, Christian (Prof. Dr.)
Day of Oral Examination: 2024-12-13
DDC Classifikation: 510 - Mathematics
Keywords: Numerische Mathematik , Differentialgleichung
Other Keywords: Dynamische niedrigrang-Approximation
Zeitintegration
Quantendynamik
Tensor Methoden
Tree Tensor Network
Time Integration
Quantum Dynamics
Dynamical low-rank approximation
License: http://tobias-lib.uni-tuebingen.de/doku/lic_ohne_pod.php?la=de http://tobias-lib.uni-tuebingen.de/doku/lic_ohne_pod.php?la=en
Show full item record

Inhaltszusammenfassung:

Die vorliegende Arbeit befasst sich mit der numerischen Lösung von hochdimensionalen Tensor-Differentialgleichungen. Die enormen Rechenkosten und Speicheranforderungen bei der numerischen Simulation solcher Gleichungen werden oft als ”Fluch der Dimensionalität” bezeichnet. Solche hochdimensionalen Differentialgleichungen treten in vielen Anwendungsgebieten wie Plasmaphysik, maschinelles Lernen, Strahlentransport oder Quantenphysik auf. Ein vielversprechender Ansatz zur Überwindung dieses Fluchs ist die dynamische Niedrigrangapproximation, bei der hochdimensionale Tensoren in einem niedrigrangig Format dargestellt werden und dann eine projizierte Differentialgleichung auf einer Niedrigrang-Mannigfaltigkeit gel¨ost wird. Die in dieser Arbeit betrachtete Niedrigrang-Mannigfaltigkeit ist die Mannigfaltigkeit von Baum-Tensornetzwerken. Die Zeitintegration von Baum-Tensornetzwerken erfordert die Zeitentwicklung jedes niederrang Faktors. In dieser Arbeit werden mehrere numerische Verfahren zur Berechnung dieser Zeitintegration vorgeschlagen. Alle diese Verfahren gehören zur Klasse der Basis-Update- und Galerkin-Integratoren (BUG), bei denen alle Basismatrizen durch eine kleine Matrixdifferentialgleichung und alle Kerntensoren durch einen Galerkin-Schritt entwickelt werden. Wir stellen eine rigorose Fehleranalyse aller Integrationsverfahren vor, die Robustheit gegenüber kleinen Singulärwerten aufweist. Dies ist wichtig, da kleine Singulärwerte zu numerischen Instabilitäten f¨uhren können, da sie hohen Krümmungen in der Niedrigrang-Mannigfaltigkeit entsprechen. Es werden bemerkenswerte Eigenschaften wie Rangadaptivität, Parallelität, Norm- und Energieerhaltung und Energieverringerung in Gradientensystemen diskutiert. Des Weiteren wird die Darstellung der rechten Seite einer Differentialgleichung in Form eines Baum-Tensornetzwerks für eine Klasse von Hamiltonoperatoren mit langreichweitigen Wechselwirkungen diskutiert. Effiziente Konstruktionen dieser Baum-Tensornetzwerk Operatoren werden gegeben und Schranken für die maximalen Ränge für eine exakte und approximierte Darstellung des Operators werden bewiesen. Numerische Experimente für mehrere Probleme aus der Quantenphysik verifizieren die theoretischen Ergebnisse und untersuchen die Anwendbarkeit der dynamischen Niedrigrangapproximation auf Vielteilchen-Quantensysteme im Detail.

Abstract:

This thesis studies the numerical solution of high-dimensional tensor differential equations. The prohibitive computational cost and memory requirements of numerically simulating such equations is often referred to as the curse of dimensionality. Such prohibitively large differential equations arise in many fields of applications such as plasma physics, machine learning, radiation transport or quantum physics. Dynamical low-rank approximation offers a promising ansatz to overcome the curse by representing the highdimensional tensors in a low-rank format and solving a projected differential equation on a low-rank manifold. The low-rank manifold considered in this thesis is the manifold of tree tensor networks. The time integration of tree tensor networks requires the update of each low-rank factor. Several numerical schemes to compute this time integration are proposed in this thesis. All of those methods fall into the class of Basis Update and Galerkin (BUG) integrators, where all basis matrices are evolved through a small matrix differential equation and all core tensors by a Galerkin step. We present a rigorous error analysis of all integration schemes that show robustness with respect to small singular values. This is important since small singular values can lead to numerical instabilities as they correspond to high curvatures in the corresponding low-rank manifold. Remarkable properties like rank-adaptivity, parallelism, norm and energy preservation and diminishing of energy in gradient systems are discussed. Further, the representation of the right-hand side of a differential equation in tree tensor network format is discussed for a class of long-range interacting Hamiltonians. Efficient constructions of these tree tensor network operators are given and bounds on the maximal tree rank for an exact and approximated representation of the operator are proven. Numerical experiments for several problems from quantum physics verify theoretical results and investigate the applicability of dynamical low-rank approximation to manybody quantum systems in detail.

This item appears in the following Collection(s)