## A posteriori analysis and improved accuracy for an operator decomposition solution of a conjugate heat transfer problem (2006)

Venue: | SINUM, in revision |

Citations: | 7 - 4 self |

### BibTeX

@INPROCEEDINGS{Estep06aposteriori,

author = {D. Estep and S. Tavener and T. Wildey},

title = {A posteriori analysis and improved accuracy for an operator decomposition solution of a conjugate heat transfer problem},

booktitle = {SINUM, in revision},

year = {2006}

}

### OpenURL

### Abstract

Abstract. We consider the accuracy of an operator decomposition finite element method for a conjugate heat transfer problem consisting of two materials coupled through a common boundary. We derive accurate a posteriori error estimates that account for the transfer of error between components of the operator decomposition method as well as the differences between the adjoints of the full problem and the discrete iterative system. We use these estimates to guide adaptive mesh refinement. In addition, we address a loss of order of convergence that results from the decomposition, and show that the approximation order of convergence is limited by the accuracy of the transferred gradient information. We employ a boundary flux recovery method to regain the expected order of accuracy in an efficient manner. Key words. a posteriori error analysis, adaptive mesh refinement, adjoint problem, boundary flux method, conjugate heat transfer, domain decomposition, finite element method, generalized

### Citations

146 |
The superconvergence patch recovery and a posteriori error estimates. Part 2: Error esimates and adaptivity
- Zienkiewicz, Zhu
- 1992
(Show Context)
Citation Context .... (3.8) Other possible approaches to mitigating the loss of order include approximating the boundary flux using a gradient recovery techniques such as the ZienkiewiczZhu (ZZ) patch recovery technique =-=[30]-=- or the polynomial preserving recovery (PPR) method [22], or using higher order polynomials near the interface to improve the accuracy of the finite element flux [2, 20]. We had mixed computational su... |

132 | Introduction to adaptive methods for differential equations, Acta Numerica
- Eriksson, Estep, et al.
- 1995
(Show Context)
Citation Context ...e use the a posteriori error estimate as the basis for adaptivity by employing the standard “optimization framework” after writing the estimate as a sum of element contributions and introducing norms =-=[10, 11, 4, 3]-=-. An element K is marked for refinement when the local error indicator is larger than a local tolerance, usually the global tolerance divided by the current number of elements. Example 4.3. We demonst... |

116 |
An optimal control approach to a posteriori error estimation in finite element methods
- Becker, Rannacher
(Show Context)
Citation Context ...e use the a posteriori error estimate as the basis for adaptivity by employing the standard “optimization framework” after writing the estimate as a sum of element contributions and introducing norms =-=[10, 11, 4, 3]-=-. An element K is marked for refinement when the local error indicator is larger than a local tolerance, usually the global tolerance divided by the current number of elements. Example 4.3. We demonst... |

107 |
Adaptive Finite Element Methods for Differential Equations
- Bangerth, Rannacher
- 2003
(Show Context)
Citation Context ...o guide an adaptive discretization strategy. Our approach is based on the standard techniques using variational analysis, residuals and the generalized Green’s function solution to an adjoint problem =-=[3, 6, 14, 15, 12, 20]-=-, which we modify to account for several new features arising from the operator decomposition. These include the following. • Numerical errors in the solution of each component are propagated through ... |

66 |
The P and H-P Versions of the Finite Element Method, Basic Principles and Properties
- Babuska
- 1993
(Show Context)
Citation Context ...Zhu (ZZ) patch recovery technique [30] or the polynomial preserving recovery (PPR) method [22], or using higher order polynomials near the interface to improve the accuracy of the finite element flux =-=[2, 20]-=-. We had mixed computational success using these other approaches. 4. A posteriori error analysis. To estimate the error of the operator decomposition finite element approximation, we apply a posterio... |

60 |
Domain decomposition
- Smith, Bjørstad, et al.
- 1996
(Show Context)
Citation Context ...gence that is lost due to the operator decomposition. The operator decomposition method we consider is closely related to the nonoverlapping domain decomposition methods as described in, for example, =-=[23, 28, 24, 29]-=-. The major concern of these authors is however the convergence of different iterative methods and the design of effective pre-conditioners or relaxation parameters, while the accuracy of the final fi... |

43 | Finite element methods and their convergence for elliptic and parabolic interface problems
- Chen, Zou
- 1998
(Show Context)
Citation Context ...(ui,vi) = i=1 2� (fi,vi), (2.2) for all vi ∈ H1 0(Ωi) with ai(ui,v) = � Ωi (Ai∇ui · ∇v + ciuiv)dx for i = 1,2. Assuming that each ai(·, ·) is coercive, (2.2) admits a unique weak solution in H1 0(Ωi) =-=[1, 5, 9]-=-. 3. An iterative operator decomposition method. To compute a numerical solution of (2.1), we consider iterative operator decomposition methods. Let u (0) 2 be an initial guess for the Dirichlet data ... |

43 | Computational Differential Equations
- ERIKSSON, ESTEP, et al.
- 1996
(Show Context)
Citation Context ...e use the a posteriori error estimate as the basis for adaptivity by employing the standard “optimization framework” after writing the estimate as a sum of element contributions and introducing norms =-=[10, 11, 4, 3]-=-. An element K is marked for refinement when the local error indicator is larger than a local tolerance, usually the global tolerance divided by the current number of elements. Example 4.3. We demonst... |

42 |
Estimating the error of numerical solutions of systems of rea ction- diffusion equations, Memoirs
- Estep, Larson, et al.
- 2000
(Show Context)
Citation Context ...o guide an adaptive discretization strategy. Our approach is based on the standard techniques using variational analysis, residuals and the generalized Green’s function solution to an adjoint problem =-=[3, 6, 14, 15, 12, 20]-=-, which we modify to account for several new features arising from the operator decomposition. These include the following. • Numerical errors in the solution of each component are propagated through ... |

41 |
A posteriori error bounds and global error control for approximations of ordinary differential equations
- Estep
- 1995
(Show Context)
Citation Context ...o guide an adaptive discretization strategy. Our approach is based on the standard techniques using variational analysis, residuals and the generalized Green’s function solution to an adjoint problem =-=[3, 6, 14, 15, 12, 20]-=-, which we modify to account for several new features arising from the operator decomposition. These include the following. • Numerical errors in the solution of each component are propagated through ... |

32 |
Linear Differential Operators
- Lanczos
- 1961
(Show Context)
Citation Context ...ures the numerical error in solving the iterative problem. It is helpful to remark that in the conventional approach, the adjoint is defined in such a way as to make the formal bilinear identity hold =-=[21, 13]-=-, i.e., to make the boundary terms arising from integration by parts equal to zero. In the analysis below, we find it convenient to consider the values passed between components as quantities of inter... |

26 |
A finite element method for interface problems in domains with smooth boundary and interfaces
- Bramble, King
- 1996
(Show Context)
Citation Context ...(ui,vi) = i=1 2� (fi,vi), (2.2) for all vi ∈ H1 0(Ωi) with ai(ui,v) = � Ωi (Ai∇ui · ∇v + ciuiv)dx for i = 1,2. Assuming that each ai(·, ·) is coercive, (2.2) admits a unique weak solution in H1 0(Ωi) =-=[1, 5, 9]-=-. 3. An iterative operator decomposition method. To compute a numerical solution of (2.1), we consider iterative operator decomposition methods. Let u (0) 2 be an initial guess for the Dirichlet data ... |

24 | Adaptive error control for finite element approximations of the lift and drag coefficients in viscous flow
- Giles, Larson, et al.
- 1997
(Show Context)
Citation Context ...oximation to the normal flux on the boundary which is relatively inexpensive to compute. In general, the accuracy of the flux approximation depends on the regularity of an associated Green’s function =-=[17, 27]-=-, but in many cases this post-processed flux is O(h 2 ) rather than the standard O(h) for the normal flux of a piecewise linear finite element approximation. It turns out that a fortunate cancellation... |

21 | A new finite element gradient recovery method: superconvergence property
- Zhang, Naga
- 2005
(Show Context)
Citation Context ...the loss of order include approximating the boundary flux by using a gradient recovery technique such as the Zienkiewicz–Zhu patch recovery technique [33] or the polynomial-preserving recovery method =-=[32]-=- or by using higher order polynomials near the interface to improve the accuracy of the finite element flux [1, 22]. We had mixed computational success using these other approaches. Copyright © by SIA... |

20 |
The finite element method for elliptic equations with discontinuous coefficients
- Babuska
- 1970
(Show Context)
Citation Context ...(ui,vi) = i=1 2� (fi,vi), (2.2) for all vi ∈ H1 0(Ωi) with ai(ui,v) = � Ωi (Ai∇ui · ∇v + ciuiv)dx for i = 1,2. Assuming that each ai(·, ·) is coercive, (2.2) admits a unique weak solution in H1 0(Ωi) =-=[1, 5, 9]-=-. 3. An iterative operator decomposition method. To compute a numerical solution of (2.1), we consider iterative operator decomposition methods. Let u (0) 2 be an initial guess for the Dirichlet data ... |

16 | Duality-based adaptivity in the hp-finite element method
- Heuveline, Rannacher
- 2003
(Show Context)
Citation Context |

14 | Validation of the a posteriori error estimator based on polynomial preserving recovery for linear elements
- Zhang, Naga
- 2004
(Show Context)
Citation Context ... of order include approximating the boundary flux using a gradient recovery techniques such as the ZienkiewiczZhu (ZZ) patch recovery technique [30] or the polynomial preserving recovery (PPR) method =-=[22]-=-, or using higher order polynomials near the interface to improve the accuracy of the finite element flux [2, 20]. We had mixed computational success using these other approaches. 4. A posteriori erro... |

13 | Generalized Green’s functions and the effective domain of influence
- Estep, Holst, et al.
(Show Context)
Citation Context |

12 | Interface relaxation methods for elliptic differential equations
- Rice, Tsompanopoulus, et al.
(Show Context)
Citation Context ...gence that is lost due to the operator decomposition. The operator decomposition method we consider is closely related to the nonoverlapping domain decomposition methods as described in, for example, =-=[23, 28, 24, 29]-=-. The major concern of these authors is however the convergence of different iterative methods and the design of effective pre-conditioners or relaxation parameters, while the accuracy of the final fi... |

9 | A posteriori analysis and adaptive error control for multiscale operator decomposition solution of elliptic systems I: triangular systems
- Carey, Estep, et al.
(Show Context)
Citation Context ...are distinct operators whose differences must be recognized when seeking to obtain accurate error estimates. These kinds of effects are characteristic of operator decomposition discretizations, e.g., =-=[9, 10, 14]-=-, and generally require extensions of the usual a posteriori analysis techniques. In addition to obtaining accurate estimates, we seek to improve the accuracy of the operator decomposition method in a... |

8 |
Derivative calculation from finite element solutions
- Carey
- 1982
(Show Context)
Citation Context ... estimates, we seek to improve the accuracy of the operator decomposition method in an efficient way. In particular, we adapt the “boundary element flux” technique developed by Wheeler [25] and Carey =-=[16, 7]-=- to compute normal derivatives on a boundary, and show that this can be used to improve accuracy, and in particular, restore the order of convergence that is lost due to the operator decomposition. Th... |

8 |
A Galerkin procedure for estimating the flux for two-point boundary-value problems using continuous piecewise-polynomial spaces
- Wheeler
- 1974
(Show Context)
Citation Context ...aining accurate estimates, we seek to improve the accuracy of the operator decomposition method in an efficient way. In particular, we adapt the “boundary element flux” technique developed by Wheeler =-=[25]-=- and Carey [16, 7] to compute normal derivatives on a boundary, and show that this can be used to improve accuracy, and in particular, restore the order of convergence that is lost due to the operator... |

8 | An a posteriori-a priori analysis of multiscale operator splitting
- Estep, Ginting, et al.
(Show Context)
Citation Context ...are distinct operators whose differences must be recognized when seeking to obtain accurate error estimates. These kinds of effects are characteristic of operator decomposition discretizations, e.g., =-=[9, 10, 14]-=-, and generally require extensions of the usual a posteriori analysis techniques. In addition to obtaining accurate estimates, we seek to improve the accuracy of the operator decomposition method in a... |

7 | A short course on duality, adjoint operators, Green’s functions, and a posteriori error analysis, Sandia National Laboratories
- Estep
- 2004
(Show Context)
Citation Context ...ures the numerical error in solving the iterative problem. It is helpful to remark that in the conventional approach, the adjoint is defined in such a way as to make the formal bilinear identity hold =-=[21, 13]-=-, i.e., to make the boundary terms arising from integration by parts equal to zero. In the analysis below, we find it convenient to consider the values passed between components as quantities of inter... |

7 | Interface solvers and preconditioners of domain decomposition type for multiphase flow in multiblock porous media
- Yotov
- 2001
(Show Context)
Citation Context ... for A1 = A2, but optimal relaxation parameters are often difficult to find for the general problem. Recognition of this difficulty has prompted the development of, for example, Newton–Krylov methods =-=[31, 16]-=-. We do not pursue this issue further but rather choose to focus on analyzing the loss accuracy in the case that the iterations can be forced to converge. Copyright © by SIAM. Unauthorized reproductio... |

6 |
Approximate boundary-flux calculations
- Carey, Chow, et al.
- 1985
(Show Context)
Citation Context ... estimates, we seek to improve the accuracy of the operator decomposition method in an efficient way. In particular, we adapt the “boundary element flux” technique developed by Wheeler [25] and Carey =-=[16, 7]-=- to compute normal derivatives on a boundary, and show that this can be used to improve accuracy, and in particular, restore the order of convergence that is lost due to the operator decomposition. Th... |

6 |
Adaptive Error Control for Finite Element
- Giles, Larson, et al.
- 1997
(Show Context)
Citation Context ...rmal flux on the boundary which is relatively inexpensive to compute. In general, the accuracy of the recovered boundary flux approximation depends on the regularity of an associated Green’s function =-=[20, 27]-=-. In some cases, though perhaps not all, the postprocessed flux is O(h2 ) rather than the standard O(h) for the normal flux of a piecewise linear finite element approximation. However, we show that th... |

5 | Stability analysis of numerical interface boundary conditions in fluid–structure thermal analysis
- Giles
- 1997
(Show Context)
Citation Context ...xed iterations. Unfortunately, the simple iterative (3.3)-(3.4) may not converge. In particular, the convergence depends on the values of A1 and A2 along the interface and the geometry of each region =-=[18, 23, 28]-=-. As an alternative, we consider two “relaxed” iteration schemes. (1) Relaxed Dirichlet values We choose α ∈ [0,1] and update the Dirichlet interface values with U (k) 1 = αU(k−1) 1 + (1 − α)π1U (k−1)... |

4 |
Some useful techniques for pointwise and local error estimates of the quantities of interest in the finite element approximation
- Cao, Kelly, et al.
(Show Context)
Citation Context |

4 |
A posteriori analysis of operator decomposition for interface problems
- Wildey
- 2007
(Show Context)
Citation Context ... account for the discontinuity. For the application to more general boundary conditions, such as a Neumann or Robin condition on one segment, or an interface condition on both, we refer the reader to =-=[26]-=-. Now, as an alternative to (3.4), we may solve � a2(U (k) 2 ,v) = (f2,v)Ω2 − (σ(k) ,v)Γ, for all v ∈ S2,0, U (k) 2 = 0, x ∈ ∂Ω2\Γ. (3.8) Other possible approaches to mitigating the loss of order incl... |

4 | A parallel nonoverlapping Schwarz domain decomposition method for elliptic interface problems
- Yang
- 1996
(Show Context)
Citation Context ...gence that is lost due to the operator decomposition. The operator decomposition method we consider is closely related to the nonoverlapping domain decomposition methods as described in, for example, =-=[23, 28, 24, 29]-=-. The major concern of these authors is however the convergence of different iterative methods and the design of effective pre-conditioners or relaxation parameters, while the accuracy of the final fi... |

4 | A multilevel newton-krylov interface solver for multiphysics couplings of flow in porous media
- Yotov
- 2001
(Show Context)
Citation Context |

3 | A posteriori error estimation of approximate boundary fluxes
- Wildey, Estep, et al.
(Show Context)
Citation Context ...oximation to the normal flux on the boundary which is relatively inexpensive to compute. In general, the accuracy of the flux approximation depends on the regularity of an associated Green’s function =-=[17, 27]-=-, but in many cases this post-processed flux is O(h 2 ) rather than the standard O(h) for the normal flux of a piecewise linear finite element approximation. It turns out that a fortunate cancellation... |

1 |
A-posteriori error control of one-way coupled elliptic systems
- Carey, Estep, et al.
- 2006
(Show Context)
Citation Context ... are distinct operators whose differences must be recognized when seeking to obtain accurate error estimates. These kinds of effects are characteristic of operator decomposition discretizations, e.g. =-=[8, 19]-=-, and generally require extensions of the usual a posteriori analysis techniques. In addition to obtaining accurate estimates, we seek to improve the accuracy of the operator decomposition method in a... |

1 |
A-posteriori analysis of operator splitting for ordinary differential equations
- Ginting, Estep, et al.
- 2006
(Show Context)
Citation Context ... are distinct operators whose differences must be recognized when seeking to obtain accurate error estimates. These kinds of effects are characteristic of operator decomposition discretizations, e.g. =-=[8, 19]-=-, and generally require extensions of the usual a posteriori analysis techniques. In addition to obtaining accurate estimates, we seek to improve the accuracy of the operator decomposition method in a... |

1 |
Iterative Techniques for the Solution of Coupled Multiphysics Problems, preprint
- Estep, Tavener, et al.
- 2008
(Show Context)
Citation Context ... for A1 = A2, but optimal relaxation parameters are often difficult to find for the general problem. Recognition of this difficulty has prompted the development of, for example, Newton–Krylov methods =-=[31, 16]-=-. We do not pursue this issue further but rather choose to focus on analyzing the loss accuracy in the case that the iterations can be forced to converge. Copyright © by SIAM. Unauthorized reproductio... |

1 |
A Short Course on Duality
- Estep
- 2004
(Show Context)
Citation Context ... two estimates are nearly identical if c {k} ≈ 0. It is helpful to remark that, in the conventional approach, the adjoint problem is defined in such a way as to make the formal bilinear identity hold =-=[23, 19]-=-, i.e., to make the boundary terms arising from integration by parts equal to zero. In the analysis below, we find it convenient to consider the values passed between components as quantities of inter... |