A Dirichlet-to- Neumann (DtN)Multigrid Algorithm for Locally Conservative Methods Sandia National Laboratories is a multi program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation, for the U.S. Department of Energy's National Nuclear Security Administration under contract DE-AC04-94AL85000. . 1 Mary F. Wheeler The University of Texas at Austin – ICES Tim Wildey Sandia National Labs AM Conference Computational and Mathematical Issues in the Geosciences March 21-24, 2011 Long Beach, CA
53
Embed
A Dirichlet-to-Neumann (DtN)Multigrid Algorithm for Locally Conservative Methods Sandia National Laboratories is a multi program laboratory managed and.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
A Dirichlet-to-Neumann (DtN)Multigrid Algorithm for
Locally Conservative Methods
Sandia National Laboratories is a multi program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation, for the U.S. Department of Energy's National Nuclear Security Administration under contract DE-AC04-94AL85000. .
Mary F. WheelerThe University of Texas at Austin – ICES
Tim WildeySandia National Labs
SIAM Conference Computational and Mathematical Issues in the Geosciences
March 21-24, 2011 Long Beach, CA
Motivation: Multinumerics
Coupling of mixed and DG using mortars – G. Pencheva
General a posteriori error estimation framework: • Vohralik 2007, 2008• Ern, Vohralik 2009, 2010• Pencheva, Vohralik, Wheeler, Wildey 2010
Is there a multilevel solver applicable to both MFEM and DG?
Can it be applied to the case of multinumerics?
Can it be used for other locally conservative methods?
Outline
I. Interface Lagrange Multipliers – Face Centered Schemes
II. A Multilevel Algorithm
III. Multigrid Formulation
IV. Applications
V. Conclusions and Future Work
Mixed methods yield linear systems of the form:
Hybridization of Mixed Methods
Mixed methods yield linear systems of the form:
Hybridization of Mixed Methods
Introduce Lagrange multipliers on the element boundaries:
Hybridization of Mixed Methods
Introduce Lagrange multipliers on the element boundaries:
Hybridization of Mixed Methods
Reduce to Schur complement for Lagrange multipliers:
Hybridization of Mixed Methods
Existing Multilevel Algorithms
Mathematical Formulation
Mathematical Formulation
Assumptions on Local DtN Maps
Defining Coarse Grid Operators
X
A Multilevel Algorithm
A Multilevel Direct Solver
Given a face-centered scheme
A Multilevel Direct Solver
Given a face-centered scheme1. Identify interior DOF
A Multilevel Direct Solver
Given a face-centered scheme1. Identify interior DOF
Eliminate
A Multilevel Direct Solver
Given a face-centered scheme1. Identify interior DOF
Eliminate 2. Identify new interior DOF
A Multilevel Direct Solver
Given a face-centered scheme1. Identify interior DOF
Eliminate 2. Identify new interior DOF
Eliminate
Continue …
Advantages:Only involves Lagrange multipliersNo upscaling of parametersApplicable to hybridized formulations as well as multinumericsCan be performed on unstructured gridsEasily implemented in parallel
Disadvantage: Leads to dense matrices
A Multilevel Direct Solver
An Alternative Multilevel Algorithm
Given a face-centered scheme
Given a face-centered scheme1. Identify interior DOF
An Alternative Multilevel Algorithm
An Alternative Multilevel Algorithm
Given a face-centered scheme1. Identify interior DOF
Coarsen
An Alternative Multilevel Algorithm
Given a face-centered scheme1. Identify interior DOF
Coarsen Eliminate
An Alternative Multilevel Algorithm
Given a face-centered scheme1. Identify interior DOF
Coarsen Eliminate
2. Identify new interior DOF
An Alternative Multilevel Algorithm
Given a face-centered scheme1. Identify interior DOF
Coarsen Eliminate
2. Identify new interior DOF Coarsen
An Alternative Multilevel Algorithm
Given a face-centered scheme1. Identify interior DOF
Developed an optimal multigrid algorithm for mixed, DG, and multinumerics.
No subgrid physics required on coarse grids only local Dirichlet to Neumann maps.
No upscaling of parameters. Only requires solving local problems (of flexible size). Applicable to unstructured meshes. Physics-based projection and restriction operators. Extends easily to systems of equations (smoothers?)? Analysis for nonsymmetric operators/formulations? Algebraic approximation of parameterization