BAIL 2006 Book of Abstracts - Institut für Numerische und ...

BAIL 2006 Book of Abstracts - Institut für Numerische und ... BAIL 2006 Book of Abstracts - Institut für Numerische und ...

num.math.uni.goettingen.de
from num.math.uni.goettingen.de More from this publisher
04.02.2013 Views

M.A. OLSHANSKII: An Augmented Lagrangian based solver for the low-viscosity incompressible flows ✬ ✫ An Augmented Lagrangian based solver for the low-viscosity incompressible flows Maxim A. Olshanskii ∗ We describe an effective solver for the finite element discretization of the Oseen problem −ν∆u + (w · ∇) u + ∇p = f in Ω (1) div u = 0 in Ω (2) u = g on ∂Ω (3) with a known, divergence-free vector function w. Discretization of (1)–(3) using LBB-stable finite elements is based on the following stabilized finite element formulation L(uh, ph; vh, qh) + � στ (−ν∆uh + w·∇uh + ∇ph − f, w·∇vh)τ with τ∈Th = (f, vh) ∀ vh ∈ Vh, qh ∈ Qh (4) L(u, p; v, q) = ν(∇u, ∇v) + ((w · ∇) u, v) − (p, div v) + (q, div u) and a suitable choice of the stabilization parameters στ . With certain assumptions finding FE solution results in solving the linear system of the form � � � � � � A BT u f = . (5) B O p 0 Linear systems of the form (5) are often referred to as generalized saddle point systems. In recent years, a great deal of effort has been invested in solving systems of this form. Most of the work has been aimed at developing effective preconditioning techniques; see [1, 3] for an extensive survey. In spite of these efforts, there is still considerable interest in preconditioning techniques that are truly robust, i.e., techniques which result in convergence rates that are largely independent of problem parameters such as mesh size and viscosity. In this paper we describe a promising approach based on an augmented Lagrangian formulation [4]: � A + γB T W −1 B B T B O � � u p with a positive-definite matrix W and parameter γ > 0. The system (6) has precisely the same solution as the original one (5). Rather than treating (5), a block-triangular preconditioner is constructed for solving (6) with a Krylov subspace iterative method. The success of this method crucially depends on the availability of a robust multigrid solver for the (1,1) block (submatrix) in (6); we develop such a method by building on previous work by Schöberl [5], together with appropriate smoothers for convection-dominated flows. This multigrid iteration will be used to define a block preconditioner for the outer iteration on the ∗ Department of Mechanics and Mathematics, Moscow State M. V. Lomonosov University, Moscow 119899, Russia (Maxim.Olshanskii@mtu-net.ru). The work was supported in part by the Russian Foundation for Basic Research and the Netherlands Organization for Scientific Research grants NWO-RFBR 047.016.008 Speaker: OLSHANSKII, M.A. 128 BAIL 2006 1 � = � f 0 � (6) ✩ ✪

M.A. OLSHANSKII: An Augmented Lagrangian based solver for the low-viscosity incompressible flows ✬ ✫ coupled saddle point system. We will show that this approach is especially appropriate for discretizations based on discontinuous pressure approximations, but can be used to construct preconditioners for other discretizations using continuous pressures. As an example of finite element (FE) method with discontinuous pressures we will use the isoP2-P0 pair. In this case, our numerical experiments demonstrate a robust behavior of the solver with respect to h and ν for some typical wind vector functions w in (1). Further, the isoP2-P1 finite element pair is used for the continuous pressure-based approximation. For this case numerical experiments show an h-independent convergence rates with mild dependence on ν, when the viscosity becomes very small. We note that this approach does not require a sophisticated preconditioner for the pressure Schur complement of (5) or (6). Some spectral estimates for the preconditioned problem are shown; basic components for building appropriate multigrid method are discussed. The role of parameter γ in (6) is addressed. We also present a comparison with one of the best available preconditioning techniques and coupled multigrid methods (Vanka multigrid), showing that our method is quite competitive in terms of convergence rates, robustness, and efficiency. Finally we discuss that using SUPG type stabilization in finite element formulation (4) is vital not only for capturing effects of unresolved subscales in solution to (1)–(3), thus improving accuracy of the discrete solution, but also for developing robust and effective iterative methods. This presentation is based on the collaborative research with M.Benzi [2]. References [1] M. Benzi, G. H. Golub and J. Liesen, Numerical solution of saddle point problems, Acta Numerica, 14 (2005), pp. 1–137. [2] M. Benzi, M.A. Olshanskii, An augmented lagrangian-based approach to the Oseen problem, (submitted), available at www.mathcs.emory.edu/~molshan [3] H. C. Elman, D. J. Silvester and A. J. Wathen, Finite Elements and Fast Iterative Solvers: with Applications in Incompressible Fluid Dynamics, Numerical Mathematics and Scientific Computation, Oxford University Press, Oxford, UK, 2005. [4] M. Fortin and R. Glowinski, Augmented Lagrangian Methods: Applications to the Numerical Solution of Boundary-Value Problems, Studies in Mathematics and its Applications, Vol. 15, North-Holland, Amsterdam/New York/Oxford, 1983. [5] J. Schöberl, Multigrid methods for a parameter dependent problem in primal variables, Numer. Math., 84 (1999), pp. 97–119. 2 Speaker: OLSHANSKII, M.A. 129 BAIL 2006 ✩ ✪

M.A. OLSHANSKII: An Augmented Lagrangian based solver for the low-viscosity<br />

incompressible flows<br />

✬<br />

✫<br />

coupled saddle point system. We will show that this approach is especially appropriate for<br />

discretizations based on discontinuous pressure approximations, but can be used to construct<br />

preconditioners for other discretizations using continuous pressures. As an example <strong>of</strong> finite<br />

element (FE) method with discontinuous pressures we will use the isoP2-P0 pair. In this case,<br />

our numerical experiments demonstrate a robust behavior <strong>of</strong> the solver with respect to h and<br />

ν for some typical wind vector functions w in (1). Further, the isoP2-P1 finite element pair<br />

is used for the continuous pressure-based approximation. For this case numerical experiments<br />

show an h-independent convergence rates with mild dependence on ν, when the viscosity becomes<br />

very small. We note that this approach does not require a sophisticated preconditioner for the<br />

pressure Schur complement <strong>of</strong> (5) or (6).<br />

Some spectral estimates for the preconditioned problem are shown; basic components for<br />

building appropriate multigrid method are discussed. The role <strong>of</strong> parameter γ in (6) is addressed.<br />

We also present a comparison with one <strong>of</strong> the best available preconditioning techniques and<br />

coupled multigrid methods (Vanka multigrid), showing that our method is quite competitive in<br />

terms <strong>of</strong> convergence rates, robustness, and efficiency. Finally we discuss that using SUPG type<br />

stabilization in finite element formulation (4) is vital not only for capturing effects <strong>of</strong> unresolved<br />

subscales in solution to (1)–(3), thus improving accuracy <strong>of</strong> the discrete solution, but also for<br />

developing robust and effective iterative methods.<br />

This presentation is based on the collaborative research with M.Benzi [2].<br />

References<br />

[1] M. Benzi, G. H. Golub and J. Liesen, Numerical solution <strong>of</strong> saddle point problems, Acta<br />

Numerica, 14 (2005), pp. 1–137.<br />

[2] M. Benzi, M.A. Olshanskii, An augmented lagrangian-based approach to the Oseen problem,<br />

(submitted), available at www.mathcs.emory.edu/~molshan<br />

[3] H. C. Elman, D. J. Silvester and A. J. Wathen, Finite Elements and Fast Iterative Solvers:<br />

with Applications in Incompressible Fluid Dynamics, Numerical Mathematics and Scientific<br />

Computation, Oxford University Press, Oxford, UK, 2005.<br />

[4] M. Fortin and R. Glowinski, Augmented Lagrangian Methods: Applications to the Numerical<br />

Solution <strong>of</strong> Bo<strong>und</strong>ary-Value Problems, Studies in Mathematics and its Applications,<br />

Vol. 15, North-Holland, Amsterdam/New York/Oxford, 1983.<br />

[5] J. Schöberl, Multigrid methods for a parameter dependent problem in primal variables,<br />

Numer. Math., 84 (1999), pp. 97–119.<br />

2<br />

Speaker: OLSHANSKII, M.A. 129 <strong>BAIL</strong> <strong>2006</strong><br />

✩<br />

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!