next up previous
Next: About this document ...

Edward Phillips
Block Preconditioners for Fully Coupled MHD with Lagrange Multipliers

Mathematics Building
University of Maryland
College Park
MD 20742
egphillips@math.umd.edu
Howard Elman
Eric Cyr
John Shadid

Magnetohydrodynamics (MHD) models the flow of electrically conducting fluids by coupling the Navier-Stokes to Maxwell's equations. In this setting Maxwell's equations are often posed in mixed form to explicitly enforce the solenoidal condition $ \nabla\cdot\vec{B} = 0$ , which requires adding a Lagrange multiplier $ r$ to the induction equation governing the magnetic field. The resulting equations form a set of two coupled saddle point problems. Discretizing the fully coupled system results in a sequence of linear systems with the block form

$\displaystyle \left(\begin{array}{cccc} A & D^t & -Z^t & 0 \\ D & 0 & 0 & 0 \\ ...
...n{array}{c}\mathbf{g}\\ \mathbf{0}\\ \mathbf{f}\\ \mathbf{0}\end{array}\right).$    

If we define the blocks

$\displaystyle \mathcal{M} = \left(\begin{array}{cc} A & D^t \\ D & 0 \end{array}\right), \quad \mathcal{Z} = \left(\begin{array}{cc} Z & 0 \end{array}\right),$    

then a block LU decomposition of the discrete MHD system suggests a preconditioner of the form

$\displaystyle \mathcal{P} = \left(\begin{array}{ccc} \hat{\mathcal{M}} & -\mathcal{Z}^t & 0 \\ 0 & \hat{X} & B^t \\ 0 & 0 & \hat{Y} \end{array}\right),$    

where $ \hat{\mathcal{M}}$ is an approximation to $ \mathcal{M}$ , and $ \hat{X}$ and $ \hat{Y}$ are approximations to the Schur complements

$\displaystyle X = F + \mathcal{Z}\mathcal{M}^{-1}\mathcal{Z}^t,\quad Y = -BX^{-1}B^t.$    

We consider expressions for $ \hat{\mathcal{M}}$ proposed for Maxwell's equations in mixed form and use these to develop the approximations $ \hat{X}$ and $ \hat{Y}$ . We test the performance of the resulting preconditioners for both stable and stabilized finite element discretizations, demonstrating their parallel scalability as well as their robustness over physical parameters on a set of two- and three-dimensional test problems.




next up previous
Next: About this document ...
Copper Mountain 2014-02-23