TU Darmstadt / ULB / TUbiblio

Pointwise Approximation of Coupled Ornstein-Uhlenbeck Processes

Henkel, Daniel (2012)
Pointwise Approximation of Coupled Ornstein-Uhlenbeck Processes.
Technische Universität Darmstadt
Ph.D. Thesis, Primary publication

Abstract

We consider a stochastic evolution equation on the spatial domain D=(0,1)^d, driven by an additive nuclear or space-time white noise, so that the solution is given by an infinite-dimensional Ornstein-Uhlenbeck process. We study algorithms that approximate the mild solution of the equation, which takes values in the Hilbert space H=L_2(D), at a fixed point in time. The error of an algorithm is defined by the average distance between the solution and its approximation in H. The cost of an algorithm is defined by the total number of evaluations of one-dimensional components of the driving H-valued Wiener process W at arbitrary time nodes. We construct algorithms with an asymptotically optimal relation between error and cost. Furthermore, we determine the asymptotic behaviour of the corresponding minimal errors. We show how the minimal errors depend on the spatial dimension d, on the smoothing effect of the semigroup generated by the drift term, on the coupling of the infinite-dimensional system of scalar Ornstein-Uhlenbeck processes, which is specified by the diffusion term, and on the decay of the eigenvalues of W in case of nuclear noise. Asymptotic optimality is achieved by drift-implicit Euler-Maruyama schemes together with non-uniform time discretizations. This optimality cannot necessarily be achieved by uniform time discretizations, which are frequently analyzed in the literature. We complement our theoretical results by numerical studies.

Item Type: Ph.D. Thesis
Erschienen: 2012
Creators: Henkel, Daniel
Type of entry: Primary publication
Title: Pointwise Approximation of Coupled Ornstein-Uhlenbeck Processes
Language: English
Referees: Ritter, Prof. Dr. Klaus ; Lang, Prof. Dr. Jens
Date: 30 July 2012
Refereed: 4 May 2012
URL / URN: urn:nbn:de:tuda-tuprints-30650
Abstract:

We consider a stochastic evolution equation on the spatial domain D=(0,1)^d, driven by an additive nuclear or space-time white noise, so that the solution is given by an infinite-dimensional Ornstein-Uhlenbeck process. We study algorithms that approximate the mild solution of the equation, which takes values in the Hilbert space H=L_2(D), at a fixed point in time. The error of an algorithm is defined by the average distance between the solution and its approximation in H. The cost of an algorithm is defined by the total number of evaluations of one-dimensional components of the driving H-valued Wiener process W at arbitrary time nodes. We construct algorithms with an asymptotically optimal relation between error and cost. Furthermore, we determine the asymptotic behaviour of the corresponding minimal errors. We show how the minimal errors depend on the spatial dimension d, on the smoothing effect of the semigroup generated by the drift term, on the coupling of the infinite-dimensional system of scalar Ornstein-Uhlenbeck processes, which is specified by the diffusion term, and on the decay of the eigenvalues of W in case of nuclear noise. Asymptotic optimality is achieved by drift-implicit Euler-Maruyama schemes together with non-uniform time discretizations. This optimality cannot necessarily be achieved by uniform time discretizations, which are frequently analyzed in the literature. We complement our theoretical results by numerical studies.

Alternative Abstract:
Alternative abstract Language

Wir betrachten eine stochastische Evolutionsgleichung auf dem räumlichen Bereich D=(0,1)^d, getrieben entweder von einem additiven nuklearen oder einem additiven Raum-Zeit weißen Rauschen, so daß die Lösung durch einen unendlichdimensionalen Ornstein-Uhlenbeck-Prozeß gegeben ist. Wir untersuchen Algorithmen zur Approximation der milden Lösung dieser Gleichung, die Werte in dem Hilbertraum H=L_2(D) annimmt, zu einem festen Zeitpunkt. Der Fehler eines Algorithmus ist definiert durch den mittleren Abstand zwischen der Lösung und ihrer Approximation in H. Die Kosten eines Algorithmus sind definiert durch die Gesamtanzahl der Auswertungen der eindimensionalen Komponenten des treibenden H-wertigen Wiener-Prozesses W an beliebigen Zeitpunkten. Wir konstruieren Algorithmen mit einer asymptotischen optimalen Beziehung zwischen Fehler und Kosten. Desweiteren bestimmen wir das asymptotische Verhalten der entsprechenden minimalen Fehler. Wir zeigen die Abhängigkeit der minimalen Fehler von der räumlichen Dimension d, vom Glättungseffekt der vom Driftterm erzeugten Halbgruppe, von der durch den Diffusionsterm festgelegten Kopplung des unendlichdimensionalen Systems skalarer Ornstein-Uhlenbeck-Prozesse und von dem Zerfall der Eigenwerte von W im Falle nuklearen Rauschens. Asymptotische Optimalität wird erreicht durch implizite Euler-Maruyama-Verfahren, versehen mit nicht-uniformen Zeitdiskretisierungen. Diese Optimalität kann nicht notwendigerweise durch uniforme Zeitdiskretisierungen erreicht werden, welche häufig in der Literatur verwendet werden. Wir ergänzen unsere theoretischen Resultate durch numerische Untersuchungen.

German
Classification DDC: 500 Science and mathematics > 510 Mathematics
Divisions: 04 Department of Mathematics > Stochastik
04 Department of Mathematics
Date Deposited: 14 Aug 2012 10:48
Last Modified: 05 Mar 2013 10:02
PPN:
Referees: Ritter, Prof. Dr. Klaus ; Lang, Prof. Dr. Jens
Refereed / Verteidigung / mdl. Prüfung: 4 May 2012
Export:
Suche nach Titel in: TUfind oder in Google
Send an inquiry Send an inquiry

Options (only for editors)
Show editorial Details Show editorial Details