Abstract
This paper presents a methodology for computing upper and lower bounds for both the algebraic and total errors in the context of the conforming finite element discretization of the Poisson model problem and an arbitrary iterative algebraic solver. The derived bounds do not contain any unspecified constants and allow estimating the local distribution of both errors over the computational domain. Combining these bounds, we also obtain guaranteed upper and lower bounds on the discretization error. This allows to propose novel mathematically justified stopping criteria for iterative algebraic solvers ensuring that the algebraic error will lie below the discretization one. Our upper algebraic and total error bounds are based on locally reconstructed fluxes in \({\mathbf {H}}(\mathrm{div},\varOmega )\), whereas the lower algebraic and total error bounds rely on locally constructed \(H^1_0(\varOmega )\)-liftings of the algebraic and total residuals. We prove global and local efficiency of the upper bound on the total error and its robustness with respect to the approximation polynomial degree. Relationships to the previously published estimates on the algebraic error are discussed. Theoretical results are illustrated on numerical experiments for higher-order finite element approximations and the preconditioned conjugate gradient method. They in particular witness that the proposed methodology yields a tight estimate on the local distribution of the algebraic and total errors over the computational domain and illustrate the associated cost.
Similar content being viewed by others
Notes
For example, for a square domain \(\varOmega \subset \mathbb {R}^2\) we can take \(C_{\mathrm {F}}= 1/(2\pi )\), corresponding to the smallest eigenvalue of the Laplace operator; see, e.g., [51, relation (18.48) on p. 196]
References
Ainsworth, M.: Robust a posteriori error estimation for nonconforming finite element approximation. SIAM J. Numer. Anal. 42, 2320–2341 (2005)
Arioli, M., Georgoulis, E.H., Loghin, D.: Stopping criteria for adaptive finite element solvers. SIAM J. Sci. Comput. 35, A1537–A1559 (2013)
Arioli, M., Liesen, J., Miȩdlar, A., Strakoš, Z.: Interplay between discretization and algebraic computation in adaptive numerical solution of elliptic PDE problems. GAMM Mitt. 36, 102–129 (2013)
Babuška, I., Strouboulis, T.: The Finite Element Method and Its Reliability, Numerical Mathematics and Scientific Computation. The Clarendon Press, New York (2001)
Becker, R., Johnson, C., Rannacher, R.: Adaptive error control for multigrid finite element methods. Computing 55, 271–288 (1995)
Becker, R., Mao, S.: Convergence and quasi-optimal complexity of a simple adaptive finite element method, M2AN Math. Model. Numer. Anal. 43, 1203–1219 (2009)
Becker, R., Mao, S., Shi, Z.: A convergent nonconforming adaptive finite element method with quasi-optimal complexity. SIAM J. Numer. Anal. 47, 4639–4659 (2010)
Berndt, M., Manteuffel, T.A., McCormick, S.F.: Local error estimates and adaptive refinement for first-order system least squares (FOSLS). Electron. Trans. Numer. Anal. 6, 35–43 (1997). Special issue on multilevel methods (Copper Mountain, CO, 1997)
Braess, D., Pillwein, V., Schöberl, J.: Equilibrated residual error estimates are \(p\)-robust. Comput. Methods Appl. Mech. Engrg. 198, 1189–1197 (2009)
Braess, D., Schöberl, J.: Equilibrated residual error estimator for edge elements. Math. Comp. 77, 651–672 (2008)
Burstedde, C., Kunoth, A.: A wavelet-based nested iteration-inexact conjugate gradient algorithm for adaptively solving elliptic PDEs. Numer. Algorithms 48, 161–188 (2008)
Calvetti, D., Morigi, S., Reichel, L., Sgallari, F.: Computable error bounds and estimates for the conjugate gradient method. Numer. Algorithms 25, 75–88 (2000)
Cancès, C., Pop, I.S., Vohralík, M.: An a posteriori error estimate for vertex-centered finite volume discretizations of immiscible incompressible two-phase flow. Math. Comp. 83, 153–188 (2014)
Carstensen, C., Funken, S.A.: Fully reliable localized error control in the FEM. SIAM J. Sci. Comput. 21, 1465–1484 (2000)
Ciarlet, P.G.: The finite element method for elliptic problems. In: Classics in Applied Mathematics, vol. 40 . Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA (2002). Reprint of the 1978 original (North-Holland, Amsterdam)
Ciarlet, P.G.: Linear and Nonlinear Functional Analysis with Applications. Society for Industrial and Applied Mathematics, Philadelphia, PA (2013)
Dahlquist, G., Golub, G.H., Nash, S.G.: Bounds for the Error in Linear Systems. In: Semi-infinite Programming (Proceedings of Workshop, Bad Honnef 1978) Lecture Notes in Control and Information Science, vol. 15, pp. 154–172 . Springer, Berlin (1979)
Destuynder, P., Métivet, B.: Explicit error bounds in a conforming finite element method. Math. Comp. 68, 1379–1396 (1999)
Deuflhard, P.: Cascadic conjugate gradient methods for elliptic partial differential equations: algorithm and numerical results. In: Domain decomposition methods in scientific and engineering computing (University Park, PA, 1993) vol. 180 of Contemp. Math, pp. 29–42. American Mathematical Society, Providence, RI (1994)
Dolean, V., Jolivet, P., Nataf, F.: An Introduction to Domain Decomposition Methods: Algorithms, Theory, and Parallel Implementation. Other Titles in Applied Mathematics. SIAM, Philadelphia (2015)
Dolejší, V., Šebestová, I., Vohralík, M.: Algebraic and discretization error estimation by equilibrated fluxes for discontinuous Galerkin methods on nonmatching grids. J. Sci. Comput. 64, 1–34 (2015)
Dolejší, V., Ern, A., Vohralík, M.: \(hp\)-adaptation driven by polynomial-degree-robust a posteriori error estimates for elliptic problems. SIAM J. Sci. Comput. 38, A3220–A3246 (2016)
Dörfler, W.: A convergent adaptive algorithm for Poisson’s equation. SIAM J. Numer. Anal. 33, 1106–1124 (1996)
Ern, A., Vohralík, M.: Adaptive inexact Newton methods with a posteriori stopping criteria for nonlinear diffusion PDEs. SIAM J. Sci. Comput. 35, A1761–A1791 (2013)
Ern, A., Vohralík, M.: Polynomial-degree-robust a posteriori estimates in a unified setting for conforming, nonconforming, discontinuous Galerkin, and mixed discretizations. SIAM J. Numer. Anal. 53, 1058–1081 (2015)
Ern, A., Vohralík, M.: Stable broken \(H^1\) and \({\varvec {H}}({\rm div})\) polynomial extensions for polynomial-degree-robust potential and flux reconstruction in three space dimensions. HAL Preprint 01422204, submitted for publication (2016)
Gergelits, T., Strakoš, Z.: Composite convergence bounds based on Chebyshev polynomials and finite precision conjugate gradient computations. Numer. Algorithms 65, 759–782 (2014)
Golub, G.H., Meurant, G.: Matrices, moments and quadrature. In: Numerical Analysis 1993 (Dundee, 1993), vol. 303 of Pitman Research Notes in Mathematics Series, pp. 105–156. Longman Sci. Tech., Harlow (1994)
Golub, G.H., Meurant, G.: Matrices, moments and quadrature. II. How to compute the norm of the error in iterative methods. BIT 37, 687–705 (1997)
Golub, G.H., Strakoš, Z.: Estimates in quadratic formulas. Numer. Algorithms 8, 241–268 (1994)
Harbrecht, H., Schneider, R.: On error estimation in finite element methods without having Galerkin orthogonality, Berichtsreihe des SFB 611 457, Universität Bonn (2009)
Hestenes, M.R., Stiefel, E.: Methods of conjugate gradients for solving linear systems. J. Res. Nat. Bur. Stand. 49, 409–436 (1952)
Hiptmair, R.: Operator preconditioning. Comput. Math. Appl. 52, 699–706 (2006)
Jiránek, P., Strakoš, Z., Vohralík, M.: A posteriori error estimates including algebraic error and stopping criteria for iterative solvers. SIAM J. Sci. Comput. 32, 1567–1590 (2010)
Kellogg, R.B.: On the Poisson equation with intersecting interfaces. Appl. Anal. 4, 101–129 (1974/75) (collection of articles dedicated to Nikolai Ivanovich Muskhelishvili)
Liesen, J., Strakoš, Z.: Krylov Subspace Methods: Principles and Analysis, Numerical Mathematics and Scientific Computation. Oxford University Press, Oxford (2013)
Luce, R., Wohlmuth, B.I.: A local a posteriori error estimator based on equilibrated fluxes. SIAM J. Numer. Anal. 42, 1394–1414 (2004)
Málek, J., Strakoš, Z.: Preconditioning and the conjugate gradient method in the context of solving PDEs, vol. 1 of SIAM Spotlights. Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA (2015)
Meurant, G.: The computation of bounds for the norm of the error in the conjugate gradient algorithm. Numer. Algorithms 16, 77–87 (1998)
Meurant, G.: Numerical experiments in computing bounds for the norm of the error in the preconditioned conjugate gradient algorithm. Numer. Algorithms 22, 353–365 (1999)
Meurant, G., Strakoš, Z.: The Lanczos and conjugate gradient algorithms in finite precision arithmetic. Acta Numer. 15, 471–542 (2006)
Meurant, G., Tichý, P.: On computing quadrature-based bounds for the \(A\)-norm of the error in conjugate gradients. Numer. Algorithms 62, 163–191 (2013)
Morin, P., Nochetto, R.H., Siebert, K.G.: Convergence of adaptive finite element methods. SIAM Rev. 44, 631–658 (2003). Revised reprint of “Data oscillation and convergence of adaptive FEM” [SIAM J. Numer. Anal. 38 (2000), no. 2, 466–488]
Papež, J.: Algebraic Error in Matrix Computations in the Context of Numerical Solution of Partial Differential Equations. Ph.D. thesis, Charles University, Prague, Nov (2016)
Papež, J., Strakoš, Z.: On a residual-based a posteriori error estimator for the total error. IMA J. Numer. Anal. (2017) (accepted for publication)
Papež, J., Liesen, J., Strakoš, Z.: Distribution of the discretization and algebraic error in numerical solution of partial differential equations. Linear Algebra Appl. 449, 89–114 (2014)
Papež, J., Rüde, U., Vohralík, M., Wohlmuth, B.: Sharp algebraic and total a posteriori error bounds for hp nite elements via a multilevel approach. (2017) (in preparation)
Patera, A.T., Rønquist, E.M.: A general output bound result: application to discretization and iteration error estimation and control. Math. Models Methods Appl. Sci. 11, 685–712 (2001)
Payne, L.E., Weinberger, H.F.: An optimal Poincaré inequality for convex domains. Arch. Ration. Mech. Anal. 5, 286–292 (1960)
Rannacher, R.: Error control in finite element computations. An introduction to error estimation and mesh-size adaptation. In: Error control and adaptivity in scientic computing (Antalya, 1998), vol. 536 of NATO Science Series C: Mathematical and Physical Sciences, pp. 247–278. Kluwer Acad. Publ., Dordrecht (1999)
Rektorys, K.: Variational Methods in Mathematics, Science and Engineering, 2nd edn. D. Reidel Publishing Co., Dordrecht-Boston, Mass (1980). (translated from the Czech by Michael Basch)
Repin, S.: A Posteriori Estimates for Partial Differential Equations. Radon Series on Computational and Applied Mathematics, vol. 4. Walter de Gruyter GmbH & Co. KG, Berlin (2008)
Shaidurov, V.V.: Some estimates of the rate of convergence for the cascadic conjugate-gradient method. Comput. Math. Appl. 31, 161–171 (1996)
Silvester, D.J., Simoncini, V.: An optimal iterative solver for symmetric indefinite systems stemming from mixed approximation. ACM Trans. Math. Softw. 37, Art. 42, 22 (2011)
Stevenson, R.: Optimality of a standard adaptive finite element method. Found. Comput. Math. 7, 245–269 (2007)
Strakoš, Z., Tichý, P.: On error estimation in the conjugate gradient method and why it works in finite precision computations. Electron. Trans. Numer. Anal. 13, 56–80 (2002)
Strakoš, Z., Tichý, P.: Error estimation in preconditioned conjugate gradients. BIT 45, 789–817 (2005)
Veeser, A., Verfürth, R.: Poincaré constants for finite element stars. IMA J. Numer. Anal. 32, 30–47 (2012)
Verfürth, R.: A posteriori error estimation techniques for finite element methods. Numerical Mathematics and Scientific Computation. Oxford University Press, Oxford (2013)
Vohralík, M., Wheeler, M.F.: A posteriori error estimates, stopping criteria, and adaptivity for two-phase flows. Comput. Geosci. 17, 789–812 (2013)
Wohlmuth, B.I., Hoppe, R.H.W.: A comparison of a posteriori error estimators for mixed finite element discretizations by Raviart-Thomas elements. Math. Comp. 68, 1347–1378 (1999)
Acknowledgements
The authors wish to thank Ivana Pultarová, in particular for pointing out to us the inequality (5.9) including its proof. The authors are also grateful to anonymous referees for their numerous helpful comments.
Author information
Authors and Affiliations
Corresponding author
Additional information
This work was supported by the ERC-CZ Project LL1202 financed by the MŠMT of the Czech Republic, and by the Project 13-06684S of the Grant Agency of the Czech Republic. It has also received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (Grant agreement No. 647134 GATIPOR).
Appendices
A Details of the flux reconstruction
In this appendix we present the construction of the flux \(\mathbf d ^{i}_h\). It follows [24, Section 6.2.4] (see also [10, 18]) with the difference in the construction of the algebraic residual representation \(r^{i}_h\) satisfying (2.7), which allows to bound the algebraic error in Theorem 3.
For \(K\in \mathcal {T}_h\), let \(\mathbf {RTN}^\mathrm {}_{p'}(K) \equiv [\mathbb {P}_{p'}(K)]^d + \mathbf {x} \mathbb {P}_{p'}(K)\) be the Raviart–Thomas–Nédélec finite element space of order \(p' \ge 0\). We set
and \(\mathbf {RTN}^\mathrm {}_{p'}(\mathcal {T}_h) \equiv \mathbf {RTN}^\mathrm {-1}_{p'}(\mathcal {T}_h) \cap \mathbf {H}(\mathrm {div}, \varOmega )\). We use a similar notation for these spaces on various patches. Let \(\mathbf {RTN}^\mathrm {N,0}_{p'}(\mathcal {T}_{\texttt {a}})\) be the subspace of \(\mathbf {RTN}^\mathrm {}_{p'}(\mathcal {T}_{\texttt {a}})\) with zero normal flux through the boundary \(\partial {\omega _\texttt {a}}\) for \(\texttt {a}\in {\mathcal {V}^\mathrm {int}_h}\) and through \(\partial {\omega _\texttt {a}}\backslash \partial \varOmega \) for \(\texttt {a}\in {\mathcal {V}^\mathrm {ext}_h}\) (corresponding to a homogeneous Neumann condition). Let \(\mathbb {P}_{p'}^*(\mathcal {T}_{\texttt {a}})\) be spanned by piecewise \({p'}\)th order polynomials on \(\mathcal {T}_{\texttt {a}}\), with zero mean on \(\mathcal {T}_{\texttt {a}}\) when \(\texttt {a}\in {\mathcal {V}^\mathrm {int}_h}\).
For all vertices \(\texttt {a}\in {\mathcal {V}^\mathrm {}_h}\), we first solve the following mixed finite element problems on the patches \(\mathcal {T}_{\texttt {a}}\): find \(\mathbf d ^{i}_{h,\texttt {a}}\in \mathbf {RTN}^\mathrm {N,0}_{p'}(\mathcal {T}_{\texttt {a}})\) and \(q_{h,\texttt {a}} \in \mathbb {P}_{p'}^*(\mathcal {T}_{\texttt {a}})\), \({p'} = p\) or \({p'} = p+1\), such that
for all \((\mathbf {v}_h, \chi _h) \in \mathbf {RTN}^\mathrm {N,0}_{p'}(\mathcal {T}_{\texttt {a}}) \times \mathbb {P}_{p'}^*(\mathcal {T}_{\texttt {a}})\). Then we set
We typically choose \(f_h\) to be the \(L^2(\varOmega )\)-orthogonal projection of f onto the space of the piecewise polynomials of degree \(p'\), and \(r^{i}_h\in \mathbb {P}_{p}(\mathcal {T}_h)\); see Sect. 5.1. Since \(\psi _{\texttt {a}}\in V_h\), (2.8) gives the Neumann compatibility condition of the problem (A.1a)–(A.1b),
Consequently, we can in (A.1b) take all test functions \(\chi _h \in \mathbb {P}_{p'}(\mathcal {T}_{\texttt {a}})\), which allows to show that \(\mathbf d ^{i}_h\) given by (A.1) satisfies (4.2), i.e., that \(\displaystyle {\nabla \cdot \,}\mathbf d ^{i}_h= f_h - r^{i}_h\) holds. Indeed, let \(K \in \mathcal {T}_h\) and let \(v_h \in \mathbb {P}_{p'}(K)\) be fixed. Since \(\sum _{\texttt {a}\in {\mathcal {V}^\mathrm {}_h}}{\psi _{\texttt {a}}|_K} = 1\) and \(\sum _{\texttt {a}\in {\mathcal {V}^\mathrm {}_h}}{ \nabla \psi _{\texttt {a}}|_K} = 0\) (\(\psi _{\texttt {a}}\) form a partition of unity on K), we infer
and (4.2) is proved as \(f_h - r^{i}_h\in \mathbb {P}_{p'}(\mathcal {T}_h)\).
We now briefly comment on the algorithmic construction of \(\mathbf d ^{i}_h\) in (A.1). Denote by \(\varvec{\varPhi }_{\texttt {a}}\) the basis of \(\mathbf {RTN}^\mathrm {N,0}_{p'}(\mathcal {T}_{\texttt {a}})\), and by \(\widetilde{\mathcal {X}}_{\texttt {a}}\) the basis of \(\mathbb {P}_{p'}^*(\mathcal {T}_{\texttt {a}})\), Then we construct \(\mathbf d ^{i}_h\) as
where \(\overline{{\mathsf {D}}}^i_{\texttt {a}}\) forms the part of the vector \({{\mathsf {D}}}^i_\texttt {a}\) solving the algebraic form of (A.1a)–(A.1b)
Here \(\left( \overline{{{\mathbf {\mathsf{{K}}}}}}_{\texttt {a}}\right) _{kj} = (\varvec{\phi }_{j},\, \varvec{\phi }_{k})_{\omega _\texttt {a}}\) and \(\big (\widetilde{{{\mathbf {\mathsf{{K}}}}}}_{\texttt {a}}\big )_{k\ell } = (\widetilde{\chi }_{\ell },\, \displaystyle {\nabla \cdot \,}\varvec{\phi }_{k})_{\omega _\texttt {a}}\) with \(\varvec{\phi }_{j}, \varvec{\phi }_{k} \in \varvec{\varPhi }_{\texttt {a}}\), \(\widetilde{\chi }_{\ell } \in \widetilde{\mathcal {X}}_{\texttt {a}}\). The right-hand side vector is given as
where
Since \(u^{i}_h= \varPsi {\mathsf {U}}^{i}\), where, recall, \(\varPsi \) is the basis of \(V_h\), we have \(u^{i}_h|_{{\omega _\texttt {a}}} = \varPsi _\texttt {a}{\mathsf {U}}^{i}_\texttt {a}\) for \(\varPsi _\texttt {a}\subset \varPsi \) a subset of basis functions that are nonvanishing on \({\omega _\texttt {a}}\) and \({\mathsf {U}}^{i}_\texttt {a}\) the associated entries of \({\mathsf {U}}^{i}\). Then
where \({\psi }_j \in \varPsi _{\texttt {a}},\, \varvec{\phi }_{k} \in \varvec{\varPhi }_{\texttt {a}},\, \widetilde{\chi }_{\ell } \in \widetilde{\mathcal {X}}_{\texttt {a}}\). Similarly, denoting by \({\mathcal {X}}_{\texttt {a}}\) the basis of \(\mathbb {P}_{p}(\mathcal {T}_{\texttt {a}})\), we have for the coefficient vector \(\widehat{{\mathsf {R}}}^i_\texttt {a}\) such that \(r^{i}_h|_{{\omega _\texttt {a}}} = {\mathcal {X}}_{\texttt {a}} \widehat{{\mathsf {R}}}^i_\texttt {a}\),
where \(\chi _j \in {\mathcal {X}}_{\texttt {a}}, \,\widetilde{\chi }_{\ell } \in \widetilde{\mathcal {X}}_{\texttt {a}}\). Consequently, the vector \({\mathsf {D}}^i_\texttt {a}\) can be assembled as
This means that we can solve the system with \({{\mathbf {\mathsf{{K}}}}}_\texttt {a}\) only once with multiple right-hand sides \([{\mathsf {E}}_{\texttt {a},f}, {{\mathbf {\mathsf{{E}}}}}_{\texttt {a},\varPsi _\texttt {a}}, {{\mathbf {\mathsf{{E}}}}}_{\texttt {a},\mathcal {X}_\texttt {a}}]\) prior the start of the iterative solution of (2.5) and, at any iteration i, get the local coefficients \(\overline{{\mathsf {D}}}^i_{\texttt {a}}\) of the flux reconstruction \(\mathbf d ^{i}_h\) simply by matrix-vector multiplication and summing the vectors. This is particularly appealing when the error estimator is evaluated many times (e.g. when many iterations of the algebraic solver are performed). Note that assembling \({{\mathbf {\mathsf{{K}}}}}_\texttt {a}\), \({\mathsf {E}}_{\texttt {a},f}\), \({{\mathbf {\mathsf{{E}}}}}_{\texttt {a},\varPsi _\texttt {a}}\), \({{\mathbf {\mathsf{{E}}}}}_{\texttt {a},\mathcal {X}_\texttt {a}}\), \(\texttt {a}\in {\mathcal {V}^\mathrm {}_h}\), and solving the systems corresponding to (A.3) can be done in parallel (indeed, the individual patch problems (A.2) are mutually independent). Also, this can be done independently of assembling the system (2.5).
B Efficiency of the total error bound
We prove in this appendix the global and local efficiency of the upper bound of Theorem 1, which follows and extends the results in [24, 25, 47]. To simplify the presentation, we require that the source term f is piecewise polynomial, \(f\in \mathbb {P}_{p'-1}(\mathcal {T}_h)\). Consequently, we choose \(f_h = f\), and the oscillation term vanishes, \(\eta _{{\text {osc}}}= 0\).
The following lemma extends [14, Theorem 3.1] and [9, p. 1191] (see also [25, Lemma 3.12]) to the inexact algebraic solver case considered in this paper. Recall the space \(H^1_\mathrm {*}({{\omega _\texttt {a}}})\) introduced in (4.11).
Lemma 1
Let \(\texttt {a}\in {\mathcal {V}^\mathrm {}_h}\) and let \(m_{\texttt {a}}\in H^1_\mathrm {*}({{\omega _\texttt {a}}})\) be the solution of
Then there holds
Proof
From (B.1) and since, for \(v\in H^1_\mathrm {*}({{\omega _\texttt {a}}})\), \(\psi _{\texttt {a}}v\in H^1_0({\omega _\texttt {a}})\), we have, employing (2.2),
The Cauchy–Schwarz inequality and the bound (4.13) give
Using (4.10), the Cauchy–Schwarz inequality, and (4.12),
Finally, using
and combining the above results yields the desired bound. \(\square \)
The following crucial result has been shown in [9, Theorem 7] (see also [25, Corollary 3.16]) in the two-dimensional case. The three-dimensional proof is in [26, Corollary 3.3].
Lemma 2
Let \(\mathbf{d}^{i}_{h,a}\) be given by (A.1) with \({p'=p+1}\) and let \(m_{\texttt {a}}\) be given by (B.1). Let \(f\in \mathbb {P}_{p}(\mathcal {T}_h)\). Then there exists a constant \(C_{\mathrm {st}, {\omega _\texttt {a}}}>0\) depending only on the shape of elements of the patch \(\mathcal {T}_{\texttt {a}}\) but not on their diameters such that
The constant \(C_{\mathrm {st}, {\omega _\texttt {a}}}\) is not computable. It can, however, be bounded from above considering a finite-dimensional subspace of \(H^1_\mathrm {*}({{\omega _\texttt {a}}})\) and solving the discrete version of the problem (B.1); see [25, Lemma 3.23]. Hereafter we denote
We now state the main result on the global efficiency of the estimators of Theorem 1, both for the global stopping criteria in the sense of [24, 34] and for the secure stopping criterion in the sense of (6.3), relying on the estimator \(\mu ^{i}_{{\text {total}}}\) of Theorem 2:
Theorem 7
(Global efficiency) Let the estimators of Theorem 1 satisfy the global stopping criteria
with positive parameters \(\gamma _{{\text {rem}}}\), \(\gamma _{{\text {alg}}}\) such that
Alternatively, instead of (B.3)–(B.4), let
without any requirement on \(\gamma _{{\text {rem}}}\), \(\gamma _{{\text {alg}}}\), supposing only
that is typically satisfied, apart possibly the coarsest meshes. Let the assumptions of Lemma 2 hold. Then the upper bound of Theorem 1 is globally efficient,
with the global efficiency constant
Recall that \(\mathcal {V}_K\) stands for the vertices of the element K and consider the functions \(m_{h,\texttt {a}}\) specified in Theorem 2. Then the local version of Theorem 7 proving the local efficiency under the local stopping criteria is as follows:
Theorem 8
(Local efficiency) Let, for an element \(K \in \mathcal {T}_h\), the estimators of Theorem 1 satisfy the local stopping criteria
with positive parameters \(\gamma _{{\text {rem}},K}\), \(\gamma _{{\text {alg}},K}\) such that
Alternatively, instead of (B.6)–(B.7), let, for all \(\texttt {a}\in {\mathcal {V}_K}\),
without any requirement on \(\gamma _{{\text {rem}},K}\), \(\gamma _{{\text {alg}},K}\), supposing only
that is typically satisfied, apart possibly the coarsest meshes. Let the assumptions of Lemma 2 hold. Then we have the local efficiency of the upper bound,
with the local efficiency constant
Proof of Theorem 7
From the flux construction (A.1) of \(\mathbf d ^{i}_h\), using (B.2),
as any element \(K\in \mathcal {T}_h\) has \(d+1\) vertices. From Lemma 1, we have
Therefore, using \(\left[ \sum _{\texttt {a}\in {\mathcal {V}^\mathrm {}_h}} \Vert z\Vert ^2_{{\omega _\texttt {a}}} \right] ^{1/2} = (d+1)^{1/2} \Vert z\Vert \),
From the stopping criteria (B.3),
and from (B.4),
Finally, we get the assertion for the stopping criteria (B.3),
The efficiency under the stopping criteria (B.5) actually does not request any restrictive assumptions of the form (B.4). Using (B.5b) and the bound of Theorem 2,
Now a combination with (B.9) and (B.5a) gives
so that the assertion for the stopping criteria (B.5) follows with the constant
\(\square \)
Proof of Theorem 8
For the proof of the local efficiency, we first note that
From Lemma 1,
Thus, under the stopping criteria (B.6),
From (B.7), we further obtain
so that finally
Let \(\widetilde{m}_{\texttt {a}}\in H^1_\mathrm {*}({{\omega _\texttt {a}}})\) be the solution of
in the continuous counterpart to \(m_{h,\texttt {a}}\) of Theorem 2 and similarly to (B.1). The fact that \(m_{h,\texttt {a}}\) is a projection of \(\widetilde{m}_{\texttt {a}}\) from \(H^1_\mathrm {*}({{\omega _\texttt {a}}})\) onto \(W^{\texttt {a}}_h\) gives \(\Vert \nabla m_{h,\texttt {a}}\Vert _{\omega _\texttt {a}}\le \Vert \nabla \widetilde{m}_{\texttt {a}}\Vert _{\omega _\texttt {a}}\). Proceeding as in the proof of Lemma 1 with \(r^{i}_h= 0\), we get the inequality \(\Vert \nabla \widetilde{m}_{\texttt {a}}\Vert _{{\omega _\texttt {a}}} \le C_{\mathrm {cont,PF}, {\omega _\texttt {a}}}\Vert \nabla (u-u^{i}_h)\Vert _{{\omega _\texttt {a}}}\), so that
Thus, under the secure local stopping criterion (B.8b), we obtain
and, employing (B.10) and (B.8a),
The claim in this case thus follows from
\(\square \)
Rights and permissions
About this article
Cite this article
Papež, J., Strakoš, Z. & Vohralík, M. Estimating and localizing the algebraic and total numerical errors using flux reconstructions. Numer. Math. 138, 681–721 (2018). https://doi.org/10.1007/s00211-017-0915-5
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00211-017-0915-5
Keywords
- Numerical solution of partial differential equations
- Finite element method
- A posteriori error estimation
- Algebraic error
- Discretization error
- Stopping criteria
- Spatial distribution of the error