Subspace correction methods for indefinite problems
Subspace correction methods for indefinite problems
Disciplines
Computer Sciences (40%); Mathematics (60%)
Keywords
-
Subspace Correction Methods,
Multilevel Methods,
Indefinite Systems,
Total Variation Minimization,
Partial Differential Equations,
Nonconforming Finite Elements
In this project we plan to develop and analyze new subspace correction (SC) methods for the numerical solution of coupled systems of partial differential equations (PDE). The focus is on nearly singular symmetric positive definite (SPD) and on indefinite problems. We propose an integrated approach in which it is essential to use discretization techniques that preserve certain conservation laws, and to combine them with an adaptive solution process. In this way, one can design methods that perform optimally with respect to: (i) accurate approximation of the unknown quantities; (ii) obtaining the numerical solution in optimal time; and (iii) scalability with respect to both, problem size and advances in computer hardware. The present project has the following three interrelated Components (C1)--(C3) with a main emphasis on systems with highly oscillatory coefficients: (C1): SC methods for nearly incompressible elasticity and Stokes flow. (C2): SC methods for total variation minimization of discrete functionals arising in sparse data recovery. (C3): Auxiliary space and SC methods for elliptic problems with highly oscillatory coefficients. The primary goal of the proposed research work is to contribute to extending the theory and applicability of subspace correction methods to the above-mentioned classes of problems. Starting point of the research plan is the use and interplay of stable and accurate finite element schemes and of the efficient preconditioning of the related discrete problems. In the present setting nonconforming and in particular discontinuous Galerkin (DG) finite element methods provide adequate discretization tools. Some of their most attractive properties and practical advantages over conforming methods are that a) it is easy to extended DG methods to higher approximation order; b) they are well suited to treat complex geometries in combination with unstructured and hybrid meshes; c) they can be combined with any element type where the grids are also allowed to have hanging nodes; d) they can easily handle adaptive strategies; e) they have favorable properties in view of parallel computing. A main disadvantage of DG discretizations is that they produce an excess of degrees of freedom (as compared to conforming methods of the same approximation order) which in general makes the solution of the arising linear systems more difficult and more time consuming. We therefore put strong efforts on devising new efficient and robust solution methods, covering wider classes of problems (see (C1)--(C3)) that arise from nonconforming and discontinuous Galerkin discretizations. The final aim is to adapt our methods to and to test them on industrial and multiphysics applications, e.g., in reservoir engineering, or in life science. Some of the problems in which we are particularly interested stem from micro-mechanics modeling of heterogeneous media, e.g., the modeling of fluid flow in porous media, the determination of the bio-mechanical properties of bones, or the reconstruction of (medical) images. Typically such problems involve parameters that lead to highly ill-conditioned systems of linear algebraic equations.
The scope of this project was the development and investigation of new subspace correction methods for the numerical solution of coupled systems of partial differential equations (PDE) as they play a key role in industrial applications, e.g., in the area of reservoir engineering, or in medical applications, e.g., in the area of biomechanics, or for the optimal control of flow and filtration processes. Especially challenging are models describing nearly incompressible materials or highly heterogeneous media, which usually lead to extremely ill-conditioned systems of linear algebraic equations. The present project had the following three components (C1)(C3): (C1): Stable discretizations and optimal iterative solution methods for elasticity problems in case of nearly incompressible materials and for the Stokes problem. (C2): Stable discretizations and fast solvers for convection-diffusion problems. (C3): Robust subspace correction methods for elliptic problems with highly oscillatory coefficients. The primary goal of the performed research work was to contribute substantially to the extension of the theory as well as to the applicability of subspace correction methods to the above-mentioned problem classes. The starting point was the utilization and the interplay of problem-oriented finite element methods (stable discretizations) with efficient and robust preconditioning and iterative solution methods for the related discrete problems. In this context nonconforming and discontinuous Galerkin (DG) methods provide attractive discretization tools. The main results in the area of component (C1) of the research plan were the development of robust preconditioners for linear elasticity problems discretized by DG methods as well as a complete theory of optimal multigrid methods for Stokes and Brinkman equations, again for DG discretizations. Regarding component (C2) new a priori error estimates could be derived for a stable monotone discretization scheme. The latter provides the starting point for the construction of fast solvers. In view of component (C3) a novel multigrid method was developed based on the ideas of auxiliary space correction and additive Schur complement approximation. This auxiliary space multigrid (ASMG) method exhibits favorable robustness properties with respect to highly oscillatory coefficients.
- Svetozar Margenov, Bulgarian Academy of Sciences - Bulgaria
- Blanca Ayuso, Technische Universität Hamburg - Germany
- Peter Arbenz, Eidgenössische Technische Hochschule Zürich - Switzerland
- Panayot Vassilevski, Lawrence Livermore National Laboratory - USA
- Ludmil Zikatanov, The Pennsylvania State University - USA
Research Output
- 147 Citations
- 17 Publications
-
2015
Title On the stable solution of transient convection–diffusion equations DOI 10.1016/j.cam.2014.12.001 Type Journal Article Author Bayramov N Journal Journal of Computational and Applied Mathematics Pages 275-293 Link Publication -
2014
Title Auxiliary space multigrid method based on additive Schur complement approximation DOI 10.1002/nla.1959 Type Journal Article Author Kraus J Journal Numerical Linear Algebra with Applications Pages 965-986 Link Publication -
2016
Title Uniformly Stable Discontinuous Galerkin Discretization and Robust Iterative Solution Methods for the Brinkman Problem DOI 10.1137/14099810x Type Journal Article Author Hong Q Journal SIAM Journal on Numerical Analysis Pages 2750-2774 -
2013
Title A subspace correction method for discontinuous Galerkin discretizations of linear elasticity equations DOI 10.1051/m2an/2013070 Type Journal Article Author De Dios B Journal ESAIM: Mathematical Modelling and Numerical Analysis Pages 1315-1333 Link Publication -
2015
Title A robust multigrid method for discontinuous Galerkin discretizations of Stokes and linear elasticity equations DOI 10.1007/s00211-015-0712-y Type Journal Article Author Hong Q Journal Numerische Mathematik Pages 23-49 Link Publication -
0
Title Preconditioning of weighted H(div)-norm and applications to numerical simulation of highly heterogeneous media. Type Other Author Kraus J -
2012
Title Preconditioning of Elasticity Problems with Discontinuous Material Parameters DOI 10.1007/978-3-642-33134-3_80 Type Book Chapter Author Georgiev I Publisher Springer Nature Pages 761-769 -
2012
Title Multilevel preconditioning of graph-Laplacians: Polynomial approximation of the pivot blocks inverses DOI 10.1016/j.matcom.2012.06.009 Type Journal Article Author Boyanova P Journal Mathematics and Computers in Simulation Pages 1964-1971 -
2012
Title Additive Schur Complement Approximation and Application to Multilevel Preconditioning DOI 10.1137/110845082 Type Journal Article Author Kraus J Journal SIAM Journal on Scientific Computing Link Publication -
2011
Title Algebraic multilevel iteration method for lowest order Raviart–Thomas space and applications DOI 10.1002/nme.3103 Type Journal Article Author Kraus J Journal International Journal for Numerical Methods in Engineering Pages 1175-1196 -
2012
Title Polynomial of Best Uniform Approximation to 1/x and Smoothing in Two-level Methods DOI 10.2478/cmam-2012-0026 Type Journal Article Author Kraus J Journal Computational Methods in Applied Mathematics Pages 448-468 Link Publication -
2013
Title Robust Algebraic Multilevel Preconditioners for Anisotropic Problems DOI 10.1007/978-1-4614-7172-1_12 Type Book Chapter Author Kraus J Publisher Springer Nature Pages 217-245 -
2013
Title Algebraic Multilevel Preconditioners for the Graph Laplacian Based on Matching in Graphs DOI 10.1137/120876083 Type Journal Article Author Brannick J Journal SIAM Journal on Numerical Analysis Pages 1805-1827 Link Publication -
2013
Title Robust multilevel methods for quadratic finite element anisotropic elliptic problems DOI 10.1002/nla.1876 Type Journal Article Author Kraus J Journal Numerical Linear Algebra with Applications Pages 375-398 Link Publication -
2011
Title Preconditioning of elasticity problems with discontinuous material properties. Type Book Chapter Author Georgiev I -
2017
Title Multigrid methods for convection–diffusion problems discretized by a monotone scheme DOI 10.1016/j.cma.2017.01.004 Type Journal Article Author Bayramov N Journal Computer Methods in Applied Mechanics and Engineering Pages 723-745 -
2018
Title Incomplete factorization by local exact factorization (ILUE) DOI 10.1016/j.matcom.2017.10.007 Type Journal Article Author Kraus J Journal Mathematics and Computers in Simulation Pages 50-61