Stochastic Control of Magnetization Dynamics

DSpace Repositorium (Manakin basiert)


Dateien:

Zitierfähiger Link (URI): http://hdl.handle.net/10900/71846
http://nbn-resolving.de/urn:nbn:de:bsz:21-dspace-718467
http://dx.doi.org/10.15496/publikation-13258
Dokumentart: Dissertation
Erscheinungsdatum: 2016
Sprache: Englisch
Fakultät: 7 Mathematisch-Naturwissenschaftliche Fakultät
7 Mathematisch-Naturwissenschaftliche Fakultät
Fachbereich: Mathematik
Gutachter: Prohl, Andreas (Prof. Dr.)
Tag der mündl. Prüfung: 2016-06-14
DDC-Klassifikation: 510 - Mathematik
Schlagworte: Optimierung , Stochastische partielle Differentialgleichung , Finite-Elemente-Methode , Magnetismus
Lizenz: 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
Zur Langanzeige

Inhaltszusammenfassung:

Ziel dieser Arbeit ist es, numerische Approximationen verschiedener Problemstellungen, die sich mit der stochastischen Landau-Lifshitz-Gilbert-Gleichung (kurz SLLG) befassen, zu untersuchen. Die SLLG ist eine nichtlineare stochastische partielle Differentialgleichung mit einer nichtkonvexen Zwangsbedingung (Sphärenbedingung) und wird bei der Modellierung ferromagentischer Dynamiken eingesetzt. Die Dissertation ist in drei wesentliche Teile untergliedert. Der erste Teil untersucht das Konvergenzverhalten einer auf dem Mittelpunktverfahren basierenden Zeitdiskretisierung der SLLG in einer Raumdimension, welche die Sphärenbedingung erhält. Für diese Zeitdiskretisierung wird Ratenkonvergenz in Wahrscheinlichkeit mit Ordnung 1/2 bewiesen. Zu den Hauptschwierigkeiten, die zu überwinden sind, gehören sowohl die analytische, als auch die numerische Behandlung der Nichtlinearitäten, sowie des stochastischen Integralterms. Durchgeführte Simulationen unterstützen die nachgewiesene Konvergenzordnung. Die beiden weiteren Teile der vorliegenden Dissertation befassen sich mit der Approximation stochastischer optimaler Steuerungsprobleme. Im zweiten Teil wird ein System stochastischer partieller Differentialgleichungen (kurz FBSHE) untersucht, welches aus einer vorwärtsgerichteten, sowie einer rückwärtsgerichteten stochastischen Wärmeleitungsgleichung (kurz BSHE) besteht, und aus den Bedingungen erster Ordnung eines stochastischen optimalen Steuerungsproblems motiviert ist. Ratenkonvergenz mit optimaler Ordnung für eine Raumdiskretisierung basierend auf P1 -finiten Elementen wird für die BSHE sowie das System FBSHE nachgewiesen. Anschließend wird eine Volldiskretisierung der FBSHE vorgeschlagen, welche zusätzlich auf dem impliziten Eulerverfahren basiert. Dabei auftretende bedingte Erwartungen werden mit der least squares Monte-Carlo method approximiert. Für die Simulation der FBSHE wird neben einer Picard-Iteration ein stochastisches Gradientenverfahren vorgeschlagen, welches sich in durchgeführten Simulationen als flexibler erweist. Im dritten Teil werden die Methoden des zweiten Teils mit der SLLG verbunden. Ziel ist es, die optimale Kontrolle der Dynamik einer fixierten Anzahl ferromagentischer Partikel bei erhöhten Temperaturen zu untersuchen. Das stochastische optimale Steuerungsproblem setzt sich dabei aus der Minimierung eines quadratischen Funktionals unter der Nebenbedingung, dass die SLLG erfüllt ist zusammen. Existenz eines solchen Minimums wird nachgewiesen. Damit verbundene Bedingungen erster Ordnung welche aus einem System von gekoppelten vorwarts-rückwärtsgerichteten stochastischen Differentialgleichungen besteht werden unter Verwendung einer strukturerhaltenden Zeitdiskretisierung, der least squares Monte-Carlo method, sowie des stochastischen Gradientenverfahrens diskretisiert sowie simuliert. In computergestützten Experimenten wird der Einfluss der optimalen stochastischen Kontrolle im Falle des Zusammenwirkens von Anisotropie, Streufeld, Austauschenergie und Rauschen untersucht.

Abstract:

This thesis is concerned with the approximation of various problems related to the stochastic Landau-Lifshitz-Gilbert equation (SLLG), which models the dynamics of a ferromagnetic body at elevated temperatures. The SLLG is a nonlinear stochastic partial differential equation which possesses an inherent non-convex side constraint. Firstly, the time discretization of the stochastic partial differential equation is addressed, where we study the convergence behavior of a structure-preserving discretization. Secondly, the approximation and simulation of the stochastic optimal control problem subject to the SLLG is studied by means of the necessary first order optimality conditions. The thesis is split into three parts. In the first part we focus on the time discretization of the SLLG. We show convergence in probability with rate of order 1/2 for a time discretized scheme which is based on the midpoint rule and preserves the sphere constraint. Main difficulties were the analytical and numerical treatment of the nonlinear and stochastic terms. Computational studies carried out in this part support this convergence rate. The second and the third part are contributed to the stochastic optimal control problem. In the second part, we prove strong convergence with optimal rates for a spatial discretization of the forward-backward stochastic heat equation which describes the stochastic optimal control problem subject to the stochastic heat equation. As an intermediate step, we show optimal rates for a spatial discretization of the backward stochastic heat equation. A full discretization which is based on the implicit Euler method for a temporal discretization and a least squares Monte-Carlo method is then proposed. Next to an iterative solution strategy which is based on a well-known Picard-type algorithm, the new stochastic gradient method turns out to be much more flexible. Concluding computational experiments compare the efficiency of different discretization approaches. The third part combines the methodology of the second part with the SLLG. Here, we control the dynamics of a fixed number of ferromagnetic spins at elevated temperatures by minimizing a quadratic functional subject to the SLLG. Existence of a minimum of the stochastic optimal control problem with control constraints is shown. The related first order optimality conditions consist of a coupled forward-backward SDE system, which is numerically solved by a structure-inheriting discretization, the least squares Monte-Carlo method to approximate related conditional expectations, and the stochastic gradient method. Computational experiments are reported which motivate optimal controls in the case of interacting anisotropy, stray field, exchange energies, and acting noise.

Das Dokument erscheint in: