Time-dependent Dirichlet Boundary Conditions in Finite Element Discretizations

Similar documents
Finite Element Decompositions for Stable Time Integration of Flow Equations

A note on accurate and efficient higher order Galerkin time stepping schemes for the nonstationary Stokes equations

FEniCS Course. Lecture 0: Introduction to FEM. Contributors Anders Logg, Kent-Andre Mardal

Non-Conforming Finite Element Methods for Nonmatching Grids in Three Dimensions

Affine covariant Semi-smooth Newton in function space

Goal. Robust A Posteriori Error Estimates for Stabilized Finite Element Discretizations of Non-Stationary Convection-Diffusion Problems.

Chapter 2 Finite Element Spaces for Linear Saddle Point Problems

Adaptive methods for control problems with finite-dimensional control space

Newton-Multigrid Least-Squares FEM for S-V-P Formulation of the Navier-Stokes Equations

Solving Distributed Optimal Control Problems for the Unsteady Burgers Equation in COMSOL Multiphysics

Lecture Note III: Least-Squares Method

AMS subject classifications. Primary, 65N15, 65N30, 76D07; Secondary, 35B45, 35J50

PARTITION OF UNITY FOR THE STOKES PROBLEM ON NONMATCHING GRIDS

The Discontinuous Galerkin Finite Element Method

An Iterative Substructuring Method for Mortar Nonconforming Discretization of a Fourth-Order Elliptic Problem in two dimensions

Some recent results for two extended Navier-Stokes systems

Chapter Two: Numerical Methods for Elliptic PDEs. 1 Finite Difference Methods for Elliptic PDEs

Suboptimal Open-loop Control Using POD. Stefan Volkwein

Termination criteria for inexact fixed point methods

Control of Interface Evolution in Multi-Phase Fluid Flows

SUPERCONVERGENCE PROPERTIES FOR OPTIMAL CONTROL PROBLEMS DISCRETIZED BY PIECEWISE LINEAR AND DISCONTINUOUS FUNCTIONS

Priority Programme The Combination of POD Model Reduction with Adaptive Finite Element Methods in the Context of Phase Field Models

Partitioned Methods for Multifield Problems

Overview. A Posteriori Error Estimates for the Biharmonic Equation. Variational Formulation and Discretization. The Biharmonic Equation

Chapter 2. General concepts. 2.1 The Navier-Stokes equations

An interpolation operator for H 1 functions on general quadrilateral and hexahedral meshes with hanging nodes

HIGHER-ORDER LINEARLY IMPLICIT ONE-STEP METHODS FOR THREE-DIMENSIONAL INCOMPRESSIBLE NAVIER-STOKES EQUATIONS

Krylov Subspace Methods for Nonlinear Model Reduction

Scientific Computing I

Hybridized Discontinuous Galerkin Methods

Multilevel Preconditioning of Graph-Laplacians: Polynomial Approximation of the Pivot Blocks Inverses

LECTURE # 0 BASIC NOTATIONS AND CONCEPTS IN THE THEORY OF PARTIAL DIFFERENTIAL EQUATIONS (PDES)

Zonal modelling approach in aerodynamic simulation

Adaptive C1 Macroelements for Fourth Order and Divergence-Free Problems

Space-Time Nonconforming Optimized Schwarz Waveform Relaxation for Heterogeneous Problems and General Geometries

Basic Principles of Weak Galerkin Finite Element Methods for PDEs

Optimal Left and Right Additive Schwarz Preconditioning for Minimal Residual Methods with Euclidean and Energy Norms

Least-Squares Finite Element Methods

20. A Dual-Primal FETI Method for solving Stokes/Navier-Stokes Equations

A Balancing Algorithm for Mortar Methods

A Review of Preconditioning Techniques for Steady Incompressible Flow

Traces and Duality Lemma

BETI for acoustic and electromagnetic scattering

Hybridized DG methods

Time domain boundary elements for dynamic contact problems

Constrained Minimization and Multigrid

Block-Structured Adaptive Mesh Refinement

Adaptive Finite Element Methods Lecture Notes Winter Term 2017/18. R. Verfürth. Fakultät für Mathematik, Ruhr-Universität Bochum

Solving PDEs with Multigrid Methods p.1

Optimal control problems with PDE constraints

Konstantinos Chrysafinos 1 and L. Steven Hou Introduction

On Multigrid for Phase Field

Priority Program 1253

Splitting methods with boundary corrections

Discontinuous Galerkin Methods

Discontinuous Galerkin Time Discretization Methods for Parabolic Problems with Linear Constraints

PREPRINT 2010:23. A nonconforming rotated Q 1 approximation on tetrahedra PETER HANSBO

Numerical methods for PDEs FEM convergence, error estimates, piecewise polynomials

Algorithms for Scientific Computing

Numerical methods for PDEs FEM convergence, error estimates, piecewise polynomials

FEniCS Course. Lecture 6: Incompressible Navier Stokes. Contributors Anders Logg André Massing

Numerical approximation for optimal control problems via MPC and HJB. Giulia Fabrini

PDE Constrained Optimization selected Proofs

Proper Orthogonal Decomposition for Optimal Control Problems with Mixed Control-State Constraints

A Ginzburg-Landau Type Problem for Nematics with Highly Anisotropic Elastic Term

Glowinski Pironneau method for the 3D ω-ψ equations

Optimal Control of Partial Differential Equations I+II

Nonlinear, Transient Conduction Heat Transfer Using A Discontinuous Galerkin Hierarchical Finite Element Method

2 FORMULATIONS 2.1 The H-J formulation Let Ω be a domain consisting of a conducting region R and a non-conducting region S. Here assume that Ω, R and

A posteriori error estimates applied to flow in a channel with corners

A Finite Element Method Using Singular Functions for Poisson Equations: Mixed Boundary Conditions

arxiv: v1 [math.na] 29 Feb 2016

Minimal periods of semilinear evolution equations with Lipschitz nonlinearity

Large Scale Fluid-Structure Interaction by coupling OpenFOAM with external codes

WELL POSEDNESS OF PROBLEMS I

Exponential integrators for semilinear parabolic problems

Efficient smoothers for all-at-once multigrid methods for Poisson and Stokes control problems

Finite Element Solver for Flux-Source Equations

1. Introduction. The Stokes problem seeks unknown functions u and p satisfying

From Completing the Squares and Orthogonal Projection to Finite Element Methods

Trust-Region SQP Methods with Inexact Linear System Solves for Large-Scale Optimization

Efficient FEM-multigrid solver for granular material

A Two-Grid Stabilization Method for Solving the Steady-State Navier-Stokes Equations

Numerical Solutions to Partial Differential Equations

Weak Galerkin Finite Element Scheme and Its Applications

Weak Galerkin Finite Element Methods and Applications

Towards pressure-robust mixed methods for the incompressible Navier Stokes equations

IMPACTS OF SIGMA COORDINATES ON THE EULER AND NAVIER-STOKES EQUATIONS USING CONTINUOUS/DISCONTINUOUS GALERKIN METHODS

Global Carleman inequalities and theoretical and numerical control results for systems governed by PDEs

From the Boundary Element Domain Decomposition Methods to Local Trefftz Finite Element Methods on Polyhedral Meshes

On Friedrichs inequality, Helmholtz decomposition, vector potentials, and the div-curl lemma. Ben Schweizer 1

Scientific Computing WS 2018/2019. Lecture 15. Jürgen Fuhrmann Lecture 15 Slide 1

Coupling Stokes and Darcy equations: modeling and numerical methods

A posteriori analysis of a discontinuous Galerkin scheme for a diffuse interface model

Numerical methods for the Navier- Stokes equations

FINITE ELEMENT APPROXIMATION OF ELLIPTIC DIRICHLET OPTIMAL CONTROL PROBLEMS

A Robust Preconditioner for the Hessian System in Elliptic Optimal Control Problems

Nonlinear Eigenvalue Problems: An Introduction

Short title: Total FETI. Corresponding author: Zdenek Dostal, VŠB-Technical University of Ostrava, 17 listopadu 15, CZ Ostrava, Czech Republic

On discontinuity capturing methods for convection diffusion equations

Transcription:

Time-dependent Dirichlet Boundary Conditions in Finite Element Discretizations Peter Benner and Jan Heiland November 5, 2015 Seminar Talk at Uni Konstanz

Introduction Motivation A controlled physical processes are controlled via interfaces is modelled via PDEs for the state υ with boundary control u F( υ, υ; u) and approximated through, e.g., finite elements F ( v, v; u). For further treatment like optimal control or model reduction, we want to have a standard state space form of F : v = f (v; u) or v = Av + Bu. Heiland Time-dependent Boundary Conditions in FEM 2/23

Introduction Or a bit more concrete The PDE υ = Aυ, υ = u in the domain, at the boundary is spatially discretized to give an ODE with a constraint M v = Av, v = u at the inner nodes, at the boundary nodes however, elimination like v = v i + v u, where v u = G 1 u fulfills the boundary values, does lead to M v i = Av i + AG 1 u MG 1 u, which is not of the standard form. Heiland Time-dependent Boundary Conditions in FEM 3/23

Introduction Functional Analytical Framework Ω R d, d {2, 3} a bounded and smooth domain with boundary Γ. V := W 1,2 (Ω) and H := L 2 (Ω) the state spaces V the dual space of V with respect to the embedding V H, the trace spaces Q = W 1 2,2 (Γ) and Q = Q := L(Q, R), and the trace operator γ : V Q. Heiland Time-dependent Boundary Conditions in FEM 4/23

Introduction A generic PDE Problem Let T > 0 and consider A: V V. For F L 2 (0, T ; V ), for υ 0 H, and U L 2 (0, T ; Q ), find υ with υ(t) V and υ(t) V, a.e. on (0, T ), so that υ(t) Aυ(t) = F(t), γυ(t) = U(t), holds for almost all t (0, T ), and so that υ(0) = υ 0 in H. We will not discuss time regularity here. For illustration purposes, we consider only the linear case. Heiland Time-dependent Boundary Conditions in FEM 5/23

Introduction Convection-diffusion Example Problem Given a diffusion parameter ν, a convection wind β, and a function g prescribing the boundary conditions, find a function ρ that satisfies ρ(t) + β ρ(t) ν ρ(t) = 0, ρ Γ (t) = g(t). In standard weak formulations, the convection diffusion problem resembles the generic PDE with υ(t) W 1,2 (Ω) and, e.g., A defined via ( ) ( ) β υ(t), φ +ν υ(t), φ dω Aυ(t), φ V,V = for all φ V := W 1,2 (Ω) and with n Ω Γ ν ( υ n (t), φ) dγ, denoting the normal derivative. Heiland Time-dependent Boundary Conditions in FEM 6/23

Introduction Finite Element Discretization Finite element space V = span{ψ i } nv i=1 V Assume that the basis {ψ i } nv i=1 is a nodal basis, i.e. the basis functions are associated with nodes of a mesh and they have local support. We consider the decomposition V = V I V Γ, with dimensions n v = n I + n d where V I = span{ψ i } n I of the inner i=1 is the space spanned by the basis functions and V Γ V is the space of the basis functions {ψ i } nv the boundary i=n I +1 that live on Heiland Time-dependent Boundary Conditions in FEM 7/23

Introduction Finite Element Discretization The splitting of the FEM space V = V I V Γ, is applied for ansatz and test spaces. Accordingly the mass matrix M := [( ψ i, ψ j )H ] i,j=1,n v is split with respect to the test space: M = [ MI and, once more, with respect to the trial space, M I = [ M II M I Γ ], i.e. M II and M I Γ are the part associated with the inner dofs and the part relates to the boundary dofs tested against the inner nodes, respectively. M Γ ], Heiland Time-dependent Boundary Conditions in FEM 8/23

Section Subsection similarly for the system matrix A approximating A is decomposed [ ] AI A = and A I = [ ] A II A I Γ as is a right hand side. A Γ Also, we write the FEM solution as n I v = v I + v Γ = v i ψ i + i=1 n v i=n I +1 and freely assign it with the coefficient vector v = v i ψ i, [ vi v Γ ]. Heiland Time-dependent Boundary Conditions in FEM 9/23

Introduction Finite Element Discretization To assign the boundary values, we simply assign the dofs associated with the corresponding boundaries via Gv = u, where G = [ 0 I ] R nv n I,n v Eventually, the discrete equations read: Problem Find v ( [0, T ] R nv ) so that M I v(t) A I v(t) = f (t), v(0) = α, Gv(t) = u(t), where α is a given initial value and u is the control variable. Heiland Time-dependent Boundary Conditions in FEM 10/23

Possible Approaches The direct approach Note that we cannot simply eliminate the boundary nodes in M I v A I v = 0, Gv = v Γ = u. with the partitioning of M I and A I with respect to the inner and boundary nodes, we obtain [ ] [ ] v MII M I I Γ = A v II v I + A I Γ v Γ Γ and, having inserted v Γ = u, M II v I = A II v I + A I Γ u M I Γ u. Heiland Time-dependent Boundary Conditions in FEM 11/23

Possible Approaches Lifting Define a function that fulfills the boundary values for all time: ] [ṽi (t) ṽ(t) =. u(t) Then the difference to the actual solution ˆv I = v I ṽ I solves M II ˆv I = A II ˆv I + Bu, ˆv I (0) = α I + M 1 II M I Γ u(0), with B = [A II M 1 I M I Γ A I Γ ]. the obtained system is independent choice of the lifting ṽ I some choices of ṽ I are readily extended to nonlinear equations Heiland Time-dependent Boundary Conditions in FEM 12/23

Possible Approaches Multipliers If the constraint is incorporated via a multiplier: one can apply a projection scheme: M v(t) Av(t) G T λ(t) = 0, Gv(t) = u(t), Define P := I M 1 G T S 1 G, where S := GM 1 G T, consider v = Pv + (I P)v =: v i + v g, and find that v g = M 1 G T S 1 u and with B := AM 1 G T S 1. M v i P T Av i = P T Bu, Heiland Time-dependent Boundary Conditions in FEM 13/23

Possible Approaches Ultra Weak Formulations Let Φ = W 2,2 (Ω) W 1,2 0 (Ω) and consider the pure diffusion equation. υ a is called very weak solution if Ω for all φ Φ. ( υ, φ ) dω Ω ν ( υ, φ ) dω = Γ ν ( u, φ ) dγ n The abstract equations indicate that a spatial discretization may lead to a standard system The difficulty however, lies in the definition of matching test functions of high regularity with zero boundary values and suitable ansatz functions. Heiland Time-dependent Boundary Conditions in FEM 14/23

Possible Approaches Ultra Weak Formulations With the nonconforming ansatz spaces V W 1,2 0 (Ω), the ultra weak solution is approximated ( ) v, φ dω + ν ( v, φ ) ( ) ( φ) dω = f, φ dω ν u, dγ n Ω for all φ V. Ω which is doable by in standard FEM packages. Note that the solution is approximated by a functions with a zero trace, i.e. v Γ = 0. Ω Γ Heiland Time-dependent Boundary Conditions in FEM 15/23

Possible Approaches Robin Relaxation The Dirichlet conditions can be approximated by a Robin type condition v α v n + v = g or v n 1 (g v) on Γ, α with a parameter α that is intended to go to zero which are incorporated naturally in the weak formulation Heiland Time-dependent Boundary Conditions in FEM 16/23

Numerical Studies Test Setup For β 1 (x) = 1 10 [ x0 + 1 (x 1 + 1) ] and ν 1 = 0.1, x1 1 1 1 1 Γ 0 Γ 1 x0-1 Γ 3-1 Γ 2 1 find approximations to the scalar function ρ satisfying ρ(t) + β 1 ρ(t) ν 1 ρ(t) = 0, ρ Γ0 (t) = u(t), ρ Γ1 (t) = 0, Γ 2 ρ ν Γ3 (t) = 0, on given discretizations N h and N τ of the spatial domain Ω = [ 1, 1] 2 and of the time interval [0, 0.2]. Heiland Time-dependent Boundary Conditions in FEM 17/23

Numerical Studies Test Setup We consider the following schemes lift lifting of the boundary conditions via split mass matrix proj incorporation of the constraint via Lagrange multiplier and projections ncul nonconforming approximation of ultra week solutions pero relaxation via Robin approximation and for a varying space discretization N h, time discretization N τ, and polynomial degrees cg, and a numerically computed reference solution, we investigate the errors e pcg hn h,τn τ := ρ pcg hn h,τn τ ρ ref measured in a numerical approximation of the L 2 (0, 1; L 2 ([ 1, 1] 2 )) norm. Heiland Time-dependent Boundary Conditions in FEM 18/23

Numerical Studies Estimated Order of Convergence e p1 hnh,τ240 10 4 10 6 e p2 hnh,τ240 10 4 10 6 10 0 10 1 10 2 N h proj lift ncul 10 0 10 1 10 2 Convergence tests for the consistent implementations the dashed lines indicate the slope of a quadratic convergence the dotted lines indicate a convergence of order 2.5. ++ Heiland Time-dependent Boundary Conditions in FEM 19/23 N h

Numerical Studies Penalty versus Accuracy 10 0 e p1 hnh,τ120 10 2 10 6 10 2 10 2 α 10 4 10 6 10 2 10 2 α N h = 6 N h = 12 N h = 24 N h = 48 Errors for pero for different values of the penalization parameter α for exact system solves (right) and approximate system solves (left) Heiland Time-dependent Boundary Conditions in FEM 20/23

Numerical Studies Peformance in GMRES av.#its e p1,tol1e 7 h48,τ120 proj 10.6 1.2 10 5 lift 10.6 1.2 10 5 pero 14.2 7.8 10 6 ncul 10.6 1.1 10 5 Performance of GMRES within the various formulations for N h = 48, N τ = 120, and linear elements (cg = 1) The averaged number of iterations per time-step av.#its and the approximation error in the case that the resulting linear equations are solved using GMRES up to a relative residual of tol = 10 7. The colored cells contain the lowest measured values. Heiland Time-dependent Boundary Conditions in FEM 21/23

Conclusion and an outlook There are numerous approaches that approximate a Dirichlet boundary control problem into a standard state space system Penalization schemes are easy to implement but require a wise choice of the parameter Consistent schemes need but a little extra implementation efforts Future work: Investigate the performance of the schemes in optimal control or model reduction setups Analyse the consistency of the reformulations with the infinite dimensional PDE Heiland Time-dependent Boundary Conditions in FEM 22/23

Conclusion and an outlook There are numerous approaches that approximate a Dirichlet boundary control problem into a standard state space system Penalization schemes are easy to implement but require a wise choice of the parameter Consistent schemes need but a little extra implementation efforts Future work: Investigate the performance of the schemes in optimal control or model reduction setups Analyse the consistency of the reformulations with the infinite dimensional PDE Thank you for your attention. Heiland Time-dependent Boundary Conditions in FEM 22/23

Literature F. Ben Belgacem, H. El Fekih, and J.-P. Raymond, A penalized Robin approach for solving a parabolic equation with nonsmooth Dirichlet boundary conditions, Asymptotic Anal. 34 (2003), no. 2, 121 136. Peter Benner and Jan Heiland, Time-dependent dirichlet conditions in finite element discretizations, ScienceOpen Research (2015). V. Girault and P.-A. Raviart, Finite Element Methods for Navier-Stokes Equations. Theory and Algorithms, Springer, Berlin, Germany, 1986. M. Hinze, R. Pinnau, M. Ulbrich, and S. Ulbrich, Optimization with PDE Constraints, Springer, Dordrecht, Netherlands, 2009. Heiland Time-dependent Boundary Conditions in FEM 23/23