Back to search

Institutes outside Greece

Optimal adaptive wavelet methods for solving first order system least squares

Abstract

dc:description

The ecient numerical approximation of the solution of a partial dierential equation (PDE) requires the use of an approximation space with a local resolution that is adjusted to the local smoothness of the solution. Since a priori this solution is unknown, such a space has to be created adaptively by means of a loop in which increasingly more accurate approximations to the solution are built. Two instances of such adaptive solution methods are adaptive finite element methods (afem) and adaptive wavelet-Galerkin methods (awgm). It is fair to say that the latter methods are more cumbersome to implement, but on the other hand their use is not restrictedto essentially elliptic PDEs. Any well-posed operator equation on a Hilbert space equipped with a Riesz basis has an equivalent formulation as an bi-infinite matrix-vector equation. Given a finite Λ ⊂ N, and having computed a (quasi-) best approximation to the solution by a vector supported on Λ, the norm of the residual vector of this approximation is proportional to the norm of its error. By adding those indices of the residual vector that correspond to its largest entries to the set Λ an extended set is created, and the loop can be repeated. Generally the aforementioned residual vector has infinite support, and therefore has to be approximated. This residual vector is the dierence of the infinite load vector and the bi-infinite stiness matrix applied to the current approximation vector. The standard approach is to approximate both terms separately, the second one using the nonlinear approximate matrix-vector multiplication routine, known as the apply-routine. In contrast to their dierence, being the residual vector, the aforementioned terms do not tend to zero when the iteration proceeds. Therefore, in order to approximate the residual within some fixed relative tolerance, they have to be approximated within a relative tolerance that gets increasingly smaller. In order to improve the quantitative properties of the adaptive wavelet method, Chapter 2 of this thesis is devoted to an approach to approximate the residual vector without splitting it. In order to do so, it is needed to be able to represent both parts of the residual in a common dictionary. In the standard setting of a PDE of 2nd order and continuous piecewise polynomial wavelets, the operator applied to a wavelet is a distribution, and the splitting cannot be avoided. Therefore, we consider a reformulation of such a PDE as a first order system by introducing the flux as a separate unknown. We show that any well-posed semi-linear PDE of 2nd order allowsa reformulation as a well-posed first order system least squares (FOSLS) problem. For such system we design an alternative, more ecient approximate residual evaluation scheme that, in case of a linear PDE, depends linearly on the current approximation. In Chapter 4 we apply this machinery to parabolic evolutionary PDEs in a simultaneously space-time variational formulation. In the recent years, one witnesses a growing interest in such formulations as an alternative for the usual time marching schemes that are inherently sequential, and that are not suited to eciently approximating singularities that are local in both space and time. The arising Hilbert spaces in this setting are (intersections of) Bochner spaces that are most naturally equipped with Riesz bases that are tensor products of temporal and spatial wavelet bases. These tensor product wavelet bases require a modified, more complicated routine for the approximate residual evaluation, since one cannot aort a transformation to a locally single representation. On the other hand, they give the major advantage of an eective dimension reduction known from sparse-grid methods in a non-adaptive setting. We will be able to solve the whole time evolution at a complexity equal to that of solving one stationary problem.For running the awgm, wavelet bases are needed for the generally non-square domains on which the equations are posed. Several constructions of continuous piecewise linear wavelets on general polygonal or polyhedral domains are known. Because of ourFOSLS formulation, we need in addition a wavelet Riesz basis consisting of piecewise quadratics. Such bases are constructed in Chapter 3. Finally, in Chapter 5 details are provided on several aspects of the implementation of the various routines that we have developed.

Degree

thesis:*
Grantor dc:publisher
Institutes outside Greece
Year dc:date
2018

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Rekatsinas, Nikolaos
  • Ρεκατσίνας, Νικόλαος

Subjects

dc:subject × 16

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/55588

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Rekatsinas, Nikolaos; Ρεκατσίνας, Νικόλαος. Optimal adaptive wavelet methods for solving first order system least squares. Institutes outside Greece, 2018. http://hdl.handle.net/10442/hedi/55588