Abstract
We prove exponential rates of convergence of hp-version finite element methods on geometric meshes consisting of hexahedral elements for linear, second-order elliptic boundary value problems in axiparallel polyhedral domains. We extend and generalize our earlier work for homogeneous Dirichlet boundary conditions and uniform isotropic polynomial degrees to mixed Dirichlet–Neumann boundary conditions and to anisotropic, which increase linearly over mesh layers away from edges and vertices. In particular, we construct \(H^1\)-conforming quasi-interpolation operators with N degrees of freedom and prove exponential consistency bounds \(\exp (-b\root 5 \of {N})\) for piecewise analytic functions with singularities at edges, vertices and interfaces of boundary conditions, based on countably normed classes of weighted Sobolev spaces with non-homogeneous weights in the vicinity of Neumann edges.
Similar content being viewed by others
References
I. Babuška and B. Q. Guo. The \(h\)-\(p\) version of the finite element method for domains with curved boundaries. SIAM J. Numer. Anal., 25(4):837–861, 1988.
I. Babuška and B. Q. Guo. Regularity of the solution of elliptic problems with piecewise analytic data. Part I. Boundary value problems for linear elliptic equation of second order. SIAM J. Math. Anal., 19(1):172–203, 1988.
I. Babuška and B. Q. Guo. Approximation properties of the \(h\)-\(p\) version of the finite element method. Comput. Methods Appl. Mech. Engrg., 133(3-4):319–346, 1996.
P. Binev, A. Cohen, W. Dahmen, R. DeVore, G. Petrova, and P. Wojtaszczyk. Convergence rates for greedy algorithms in reduced basis methods. SIAM J. Math. Anal., 43(3):1457–1472, 2011.
M. Costabel, M. Dauge, and S. Nicaise. Analytic regularity for linear elliptic systems in polygons and polyhedra.Math. Models Methods Appl. Sci., 22(8), 2012. doi:10.1142/S0218202512500157.
W. Dahmen and K. Scherer. Best approximation by piecewise polynomials with variable knots and degrees. J. Approx. Theory, 26(1):1–13, 1979.
L. Demkowicz, J. Kurtz, D. Pardo, M. Paszynski, W. Rachowicz, and A. Zdunek. Computing with \(hp\) -Adaptive Finite Elements. Volume 2, Frontiers: Three Dimensional Elliptic and Maxwell Problems with Applications. Chapman & Hall/CRC Applied Mathematics and Nonlinear Science Series. Chapman & Hall/CRC, November 2007.
P. Dutt, Akhlaq Husain, A. S. Vasudeva Murthy, and C. S. Upadhyay. \(h\)-\(p\) spectral element methods for three dimensional elliptic problems on non-smooth domains, Part-I: Regularity estimates and stability theorem. Proc. Indian Acad. Sci. Math. Sci., 125(2):239–270, 2015.
P. Dutt, Akhlaq Husain, A. S. Vasudeva Murthy, and C. S. Upadhyay. \(h\)-\(p\) spectral element methods for three dimensional elliptic problems on non-smooth domains, Part-II: Proof of stability theorem. Proc. Indian Acad. Sci. Math. Sci., 125(3):413–447, 2015.
A. Ern and J.-L. Guermond. Theory and Practice of Finite Elements, volume 159 of Applied Mathematical Sciences. Springer, 2004.
W. Gui and I. Babuška. The \(h,\;p\) and \(h\)-\(p\) versions of the finite element method in one dimension. II. The error analysis of the \(h\)- and \(h\)-\(p\) versions. Numer. Math., 49(6):613–657, 1986.
B. Q. Guo. The \(h\)-\(p\) version of the finite element method for solving boundary value problems in polyhedral domains. In Boundary Value Problems and Integral Equations in Nonsmooth Domains, volume 167 of Lecture Notes in Pure and Appl. Math., pages 101–120. Dekker, New York, 1995.
B. Q. Guo and I. Babuška. The \(hp\)-version of the finite element method. Part I: The basic approximation results. Comp. Mech., 1:21–41, 1986.
B. Q. Guo and I. Babuška. The \(hp\)-version of the finite element method. Part II: General results and applications. Comp. Mech., 1:203–220, 1986.
B. Q. Guo and I. Babuška. Regularity of the solutions for elliptic problems on nonsmooth domains in \(\mathbb{R}^3\). I. Countably normed spaces on polyhedral domains. Proc. Roy. Soc. Edinburgh Sect. A, 127(1):77–126, 1997.
B. Q. Guo and I. Babuška. Regularity of the solutions for elliptic problems on nonsmooth domains in \(\mathbb{R}^3\). II. Regularity in neighbourhoods of edges. Proc. Roy. Soc. Edinburgh Sect. A, 127(3):517–545, 1997.
Y. Maday, O. Mula, A.T. Patera, and M. Yano. The generalized empirical interpolation method: Stability theory on Hilbert spaces with an application to the Stokes equation. Comput. Methods Appl. Mech. Engrg., 287:310–334, 2015.
K. Scherer. On optimal global error bounds obtained by scaled local error estimates. Numer. Math., 36:257–277, 1981.
D. Schötzau and Ch. Schwab. Exponential convergence in a Galerkin least squares \(hp\)-FEM for Stokes flow. IMA J. Numer. Anal., 21(1):53–80, 2001.
D. Schötzau and Ch. Schwab. Exponential convergence for \(hp\)-version and spectral finite element methods for elliptic problems in polyhedra. Math. Models Methods Appl. Sci., 25(9):1617–1661, 2015.
D. Schötzau, Ch. Schwab, and A. Toselli. Mixed \(hp\)-DGFEM for incompressible flows. SIAM J. Numer. Anal., 40(6):2171–2194, 2003.
D. Schötzau, Ch. Schwab, and A. Toselli. Mixed \(hp\)-DGFEM for incompressible flows. II. Geometric edge meshes. IMA J. Numer. Anal., 24(2):273–308, 2004.
D. Schötzau, Ch. Schwab, and T. P. Wihler. \(hp\)-DGFEM for second-order elliptic problems in polyhedra. I: Stability on geometric meshes. SIAM J. Numer. Anal., 51(3):1610–1633, 2013.
D. Schötzau, Ch. Schwab, and T. P. Wihler. \(hp\)-DGFEM for second-order elliptic problems in polyhedra. II: Exponential convergence. SIAM J. Numer. Anal., 51(4):2005–2035, 2013.
D. Schötzau, Ch. Schwab, and T. P. Wihler. \(hp\)-DGFEM for second-order mixed elliptic problems in polyhedra. Math. Comp., 85(299):1051–1083, 2016.
Ch. Schwab. \(p\) - and \(hp\) -FEM – Theory and Application to Solid and Fluid Mechanics. Oxford University Press, Oxford, 1998.
T. P. Wihler and M. Wirz. Mixed \(hp\)-discontinuous Galerkin FEM for linear elasticity and Stokes flow in three dimensions. Math. Models Methods Appl. Sci., 22(8), 2012. doi:10.1142/S0218202512500169.
L. Zhu, S. Giani, P. Houston, and D. Schötzau. Energy norm a posteriori error estimation for \(hp\)-adaptive discontinuous Galerkin methods for elliptic problems in three dimensions. Math. Models Methods Appl. Sci., 21(2):267–306, 2011.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Endre Süli.
Dedicated to Monique Dauge on the occasion of her 60th birthday.
This work was supported in part by the Natural Sciences and Engineering Research Council of Canada (NSERC) and the Swiss National Science Foundation (SNF).
Appendix: Proof of Theorem 4.3
Appendix: Proof of Theorem 4.3
We outline the major steps of the proof of Theorem 4.3.
1.1 Approximation Results
We first establish auxiliary approximation results.
1.1.1 Univariate Approximation Properties
The following consistency bound holds for the \(H^1\)-projector \(\pi _{p,1}\) in (4.1) on \(\widehat{I}=(-1,1)\); see [26, Corollary 3.15].
Lemma 8.1
Let \(p\ge 1\), \(\widehat{u}\in H^{s+1}(\widehat{I})\) and \(0\le s\le p\). Then there holds
Here,
where \(\Gamma \) is the Gamma function satisfying \(\Gamma (m+1)=m!\) for any \(m\in \mathbb {N}_0\).
The subsequent \(H^1\)-norm error bound holds for the \(L^2\)-projection \(\widehat{\pi }_{p,0}\) (see also [26, Theorem 3.11] for p-optimal bounds).
Lemma 8.2
Let \(p\ge 1\), \(\widehat{u}\in H^{s+1}(\widehat{I})\) and \(0\le s\le p\). Then there holds
Proof
We recall from [25, Lemma 5.1] that
With (8.4) and for \(p\ge 1\), we find that
Referring to (8.1) yields (8.3). \(\square \)
1.1.2 Approximation Properties of \(\widehat{\pi }_{\varvec{p},r}\)
We next derive approximation results for the tensor projectors in (4.3). On \(\widehat{K}=\widehat{I}^3\), we introduce the tensor-product space
endowed with the standard (tensor-product) norm \(\Vert \cdot \Vert _{H^{1}_{\mathrm {mix}}(\widehat{K})}\). Let \(K=K^\perp \otimes K^\perp \) be an axiparallel element, \(\varvec{p}_K=(p_K^\perp ,p_K^{\Vert })\) an elemental degree vector and \(r_K\in \{0,1\}\) an elemental conformity index in edge-parallel direction. For \(u:K \rightarrow \mathbb {R}\), we denote by \(\widehat{u}:=u\circ \Phi _K\) the pullback to the reference element \(\widehat{K}\). In this setting, the tensor projection \(\widehat{\pi }_{\varvec{p}_K,r_K}\widehat{u}=\widehat{\pi }^\perp _{p_K^\perp ,0}\otimes \widehat{\pi }^{\Vert }_{p_K^{\Vert },r_K}\widehat{u}\) defined in (4.3) satisfies the subsequent bounds.
Proposition 8.3
The error \(\widehat{\eta }^\perp _0=\widehat{u}- \widehat{\pi }^\perp _{p_K^\perp ,0}\widehat{u}\) in edge-perpendicular direction satisfies
for any \(0 \le s_K^\perp \le p_K^\perp \), with
The error \(\widehat{\eta }^{\Vert }=\widehat{u} -\widehat{\pi }_{p_K^{\Vert },r_K} \widehat{u}\) in edge-parallel direction satisfies
for any \(r_K=0,1\), \(|\varvec{\alpha }^\perp |\ge 0\), \(\alpha ^{\Vert }=0,1\), and \(0\le s_K^{\Vert }\le p_K^{\Vert }\).
Proof
We have
Hence, by the triangle inequality and the stability property (8.4) of the univariate \(L^2\)-projector \(\widehat{\pi }^{(1)}_{p_K^\perp ,0}\), we find
The univariate approximation properties (8.3) in Lemma 8.2 now imply
for any \(0\le s_K^\perp \le p_K^\perp \) and where we write \(\mathsf {D}^{\varvec{\alpha }}=\mathsf {D}^{(\alpha _1,\alpha _2,\alpha _3)}\) for a multi-index \(\varvec{\alpha }=(\alpha _1,\alpha _2,\alpha _3)\). This bound and a scaling argument as in [24, Section 5.1.4] yield the desired bound (8.6) for \(\widehat{\eta }^\perp _0\).
The bound for \(\widehat{\eta }^{\Vert }\) is an immediate consequence of the consistency bounds (8.1) (\(r_K=1\)) and (8.3) (\(r_K=0\)) applied in edge-parallel direction, combined again with a scaling argument as in [24, Section 5.1.4]. \(\square \)
1.1.3 Weighted Norm Estimates in Plane Domains
In plane domains perpendicular to edges, we shall use estimates in weighted spaces analogous to those in [13, Section 3]. To state them, let \(\mathfrak {K}\) be an axiparallel and shape-regular rectangle of diameter \(h_\mathfrak {K}\) which is affinely equivalent to the reference square \(\widehat{\mathfrak {K}}=\widehat{I}^2\). Let \({\varvec{c}}\) be a corner of \(\mathfrak {K}\) and set \(r(\varvec{x})=|\varvec{x}-{\varvec{c}}|\). For a weight exponent \(\beta \in [0,1)\), we denote by \(L^2_\beta (\mathfrak {K})\) the weighted \(L^2\)-space endowed with the weighted norm \(\Vert u\Vert _{L^2_\beta (\mathfrak {K})}:=\Vert r^{\beta }u\Vert _{L^2(\mathfrak {K})}\). For \(m= 1,2\), the weighted Sobolev space \(H^{m,m}_\beta (\mathfrak {K})\) is defined as the completion of all \(C^\infty (\overline{\mathfrak {K}})\)-functions with respect to the norm \(\Vert u\Vert ^2_{H^{m,m}_\beta (\mathfrak {K})}:=\Vert u\Vert ^2_{H^{m-1}(\mathfrak {K})}+|u|^2_{H^{m,m}_\beta (\mathfrak {K})}\), where \(|u|^2_{H^{m,m}_\beta (\mathfrak {K})}:=\sum _{|\varvec{\alpha }|=m}\Vert r^{\beta }\mathsf {D}^{\varvec{\alpha }} u\Vert ^2_{L^2(\mathfrak {K})}\). We denote by \(\pi ^2_{p,0}\) the \(L^2\)-projection onto the tensor-product polynomial space \(\mathbb {Q}_p(\mathfrak {K})\) obtained by mapping \(\widehat{\pi }^2_{p,0}\) on \(\widehat{\mathfrak {K}}\).
Lemma 8.4
Let \(\beta \in [0,1)\) be a weight exponent. For \(u\in H^{1,1}_\beta (\mathfrak {K})\) and \(p\ge 0\), there holds
Similarly, for \(u\in H^{2,2}_\beta (\mathfrak {K})\) and \(p\ge 1\), there holds
The implied constants depend on the aspect ratio of \(\mathfrak {K}\).
Proof
To prove (8.9), we apply the triangle inequality and the stability of the \(L^2\)-projection \(\pi ^2_{p,0}\) to obtain
The proof of (8.9) for \(p=0\) can then be found in [19, Proposition 27].
To show (8.10), upon scaling it is sufficient to consider the reference square \(\widehat{\mathfrak {K}}=(-1,1)^2\). We denote by \(\widehat{{\mathsf {p}}}^2_{1,0}\) the \(L^2\)-projection onto the polynomial space \(\mathbb {P}_1(\widehat{\mathfrak {K}})\). With the stability bound (8.4), it follows that
Hence, it suffices to prove (8.10) for \(\widehat{{\mathsf {p}}}^2_{1,0}\): We claim that there is a constant \(\widehat{C}>0\) independent of \(\widehat{u}\) such that
The bound (8.11) follows with standard arguments from the Peetre–Tartar lemma (see [10, Lemma A.38]) and the fact that the embedding \(H^{2,2}_\beta (\widehat{\mathfrak {K}}) \hookrightarrow H^1(\widehat{\mathfrak {K}})\) is compact (see [13, Lemma 3.4]). Invoking (8.11) for \(\widehat{u}-\widehat{{\mathsf {p}}}^2_{1,0}\widehat{u}\) and noting that \(|\widehat{{\mathsf {p}}}^2_{1,0}\widehat{u}|_{H^{2,2}_\beta (\mathfrak {K})}=0\), \(\widehat{{\mathsf {p}}}^2_{1,0}(\widehat{u}-\widehat{{\mathsf {p}}}^2_{1,0}\widehat{u})=0\), results in \(\Vert \widehat{u}-\widehat{{\mathsf {p}}}^2_{1,0}\widehat{u}\Vert _{H^{2,2}_\beta (\widehat{\mathfrak {K}})} \le \widehat{C} |\widehat{u}|_{H^{2,2}_\beta (\widehat{\mathfrak {K}})}\), which finishes the proof. \(\square \)
1.1.4 Edge-Parallel Interpolation
We construct univariate hp-projectors and establish exponential convergence bounds for univariate geometric refinements on the interval \(\omega =(0,1)\) toward \(x=0\). These results will be used for the hp-approximations along edges \({\varvec{e}}\in \mathcal E_{{\varvec{c}}}\) toward corners \({\varvec{c}}\in \mathcal C\).
In \(\omega \) and for \(\sigma \in (0,1)\), we introduce geometric meshes \(\mathcal T^{\ell }_\sigma =\{I_j\}_{j=1}^{\ell +1}\) with elements given by \(I_{1}=(0,\sigma ^\ell )\) and \(I_j=(\sigma ^{\ell +2-j},\sigma ^{\ell +1-j})\) for \(2\le j \le \ell +1\), respectively. We introduce the local mesh sizes \(h_1:=\sigma ^\ell \) and
Then, there is a constant \(\kappa >0\) solely depending on \(\sigma \in (0,1)\) with
On the geometric mesh \(\mathcal T^{\ell }_\sigma \), let \({\varvec{p}}^{\Vert }=(p_1^{\Vert },\ldots ,p^{\Vert }_{\ell +1})\in \mathbb {N}^{\ell +1}\) be an (edge-parallel) polynomial degree vector with \(p^{\Vert }_j = \max \{1, \lfloor \mathfrak {s}j \rfloor \}\), for \(\mathfrak {s}>0\) as in Sect. 3.2.1. We set \(|{\varvec{p}}^{\Vert } |=\max _{j=1}^{\ell +1} p^{\Vert }_j\) and introduce the space
For conformity indices \(r_j\in \{0,1\}\), we denote by \(\pi \) the projection onto \(V^{0}(\mathcal T^{\ell }_\sigma ,\varvec{p}^{\Vert })\), given on interval \(I_j\) as the (scaled) univariate projector \(\pi _{p^{\Vert }_j,r_j}:H^{r_j}(I_j) \rightarrow {\mathbb P}_{p^{\Vert }_j}(I_j)\). For \(u\in H^1(\omega )\), we define the approximation errors \(\eta := u - \pi u\), and set
Lemma 8.5
For a weight exponent \(\beta >0\), let \(u\in H^1(\omega )\) be such that
Then, for any conformity indices \(r_j\in \{0,1\}\), there exist \(b,C>0\) independent of \(\ell \ge 1\) such that \( \sum _{j=2}^{\ell +1} T_j [\eta ]^2 \le C \exp (-2 b\ell ). \)
Proof
Fix \(I_j\in {\mathcal T}_\sigma ^{\ell }\) for \(2\le j\le \ell +1\). A straightforward scaling argument yields , where as usual we denote by \(\widehat{v}\) the pullback operator from \(v|_{I_j}\) to the reference interval \(\widehat{I}=(-1,1)\). The bounds in (8.1) and (8.3) yield
for any \(1 \le s^{\Vert }_j \le p^{\Vert }_j\), where we exclude \(s^{\Vert }_j=0\) in (8.1), (8.3) to ensure that \(s \ge 2\) in (8.16). Scaling the right-hand side above back to element \(I_j\) results in
Moreover, by the equivalence (8.13),
By combining (8.17), (8.18) with the regularity assumption (8.16), we find that
for \(1 \le s^{\Vert }_j\le p^{\Vert }_j\). An interpolation argument as in [24, Lemma 5.8] shows that the bound (8.19) holds for any real \(s^{\Vert }_j\in [1,p^{\Vert }_j]\).
Next, we sum the bound (8.19) over all intervals \(2\le j\le \ell +1\). In view of (8.12), we obtain
By [24, Lemma 5.12], the bracket on the right-hand side above is exponentially small. Adjusting the constants to absorb \(|\varvec{p}^{\Vert }|^4\) finishes the proof. \(\square \)
Similarly, we obtain the following result.
Lemma 8.6
For a weight exponent \(\beta >0\), let \(u\in L^2(\omega )\) be such that
For any conformity indices \(r_j\in \{0,1\}\), there exist \(b,C > 0\) independent of \(\ell \ge 1\) such that \( \sum _{j=2}^{\ell +1} \Vert \eta \Vert ^2_{L^2(I_j)} \le C \exp (-2 b\ell ). \)
Proof
This follows as in Lemma 8.5 or [25, Proposition 5.5]. \(\square \)
1.2 Reference Corner–Edge Mesh
We consider the reference corner–edge mesh patch \(\widetilde{{\mathcal M}}^{\ell ,{\varvec{c}}{\varvec{e}}}_\sigma \) on \(\widetilde{Q}\) for \({\varvec{c}}\in \mathcal C\) and \({\varvec{e}}\in \mathcal E_{{\varvec{c}}}\); cf. Fig. 1 (right). As in [25, Section 7], it is sufficient to focus on the elements in \(\widetilde{{\mathcal M}}^{\ell ,{\varvec{c}}{\varvec{e}}}_\sigma \) near the corner–edge pair. To this end, we introduce the submesh \(\widetilde{\mathcal K}^{\ell ,{\varvec{c}}{\varvec{e}}}_\sigma \subset \widetilde{{\mathcal M}}^{\ell ,{\varvec{c}}{\varvec{e}}}_\sigma \) given by
where the sets \(\widetilde{\mathfrak L}^{ij}_{{\varvec{c}}{\varvec{e}}}\) stand for layers of elements with identical scaling properties with respect to \({\varvec{c}}\) and \({\varvec{e}}\); cf. [24, Section 5.2.4]. The index j indicates the number of the geometric mesh layers in edge-parallel direction along the edge \({\varvec{e}}\), whereas the index i indicates the number of mesh layers in direction perpendicular to \({\varvec{e}}\). In agreement with [25, Section 7.1], we split \(\widetilde{\mathcal K}^{\ell ,{\varvec{c}}{\varvec{e}}}_\sigma \) into interior elements away from \({\varvec{c}}\) and \({\varvec{e}}\), boundary layer elements along \({\varvec{e}}\) (but away from \({\varvec{c}}\)), and corner elements abutting at \({\varvec{c}}\). That is, we have \(\widetilde{\mathcal K}^{\ell ,{\varvec{c}}{\varvec{e}}}_\sigma = \widetilde{\mathfrak O}^\ell _{{\varvec{c}}{\varvec{e}}}\mathop {\cup }\limits ^{.}\widetilde{\mathfrak T}^\ell _{{\varvec{e}}}\mathop {\cup }\limits ^{.}\widetilde{\mathfrak T}^{\ell }_{{\varvec{c}}}\), with
Here, for \(2\le i,j\le \ell +1\), interior elements \(K\in \widetilde{\mathfrak L}^{ij}_{{\varvec{c}}{\varvec{e}}}\) satisfy
Similarly, boundary layer elements \(K\in \widetilde{\mathfrak L}_{{\varvec{c}}{\varvec{e}}}^{1j}\) satisfy
Finally, a corner element in the layer \(\widetilde{\mathfrak T}_{\varvec{c}}^\ell =\widetilde{\mathfrak L}^{11}_{{\varvec{c}}{\varvec{e}}}\) is isotropic with \(r_{\varvec{e}}|_K\lesssim h_K \simeq \sigma ^\ell \), and \(r_{\varvec{c}}|_K\lesssim h_K \simeq \sigma ^\ell \). The sets \(\widetilde{\mathfrak L}^{1j}_{{\varvec{c}}{\varvec{e}}}\) and \(\widetilde{\mathfrak L}^{11}_{{\varvec{c}}{\varvec{e}}}\) are in fact singletons, and without loss of generality \({K_j}\in \widetilde{\mathfrak L}^{1j}_{{\varvec{c}}{\varvec{e}}}\) can be written in the form
cf. (3.3), where \(K^\perp =(0,\sigma ^\ell )^2\), and the sequence \(\{ K_j^{\Vert }\}_{j=2}^{\ell +1}\) forms a one-dimensional geometric mesh \({\mathcal T}_\sigma ^{\ell }\) along the edge \({\varvec{e}}\) as in Sect. 8.1.4. The \(\mathfrak {s}\)-linearly increasing polynomial degree distributions on \(\widetilde{\mathcal K}_\sigma ^{\ell ,{\varvec{c}}{\varvec{e}}}\) in (8.21) are given by
In the sequel, we introduce the domain \(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell :=\big (\cup _{K\in \widetilde{\mathcal K}_\sigma ^{\ell ,{\varvec{c}}{\varvec{e}}}}\overline{K}\big )^\circ \). Analogously to (2.6) and for exponents \(\varvec{\beta }=\{\beta _{\varvec{c}},\beta _{\varvec{e}}\}\), we introduce the non-homogeneous reference corner–edge semi-norm on \(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell \):
for any \(k\ge 0\) and where \(r_{\varvec{c}}\) and \(r_{\varvec{e}}\) are the distances to \({\varvec{c}}\) and \({\varvec{e}}\), respectively, and \(\rho _{{\varvec{c}}{\varvec{e}}}=r_{\varvec{e}}/r_{\varvec{c}}\). For \(m>k_{\varvec{\beta }}\) as in (2.7), the weighted Sobolev spaces \(\widetilde{N}^m_{\varvec{\beta }}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\) are defined as in Sect. 2.2 with respect to the norms \(\Vert \cdot \Vert ^2_{\widetilde{N}^{m}_{\varvec{\beta }}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )}=\sum _{k=0}^m |\cdot |^2_{\widetilde{N}^k_{\varvec{\beta }}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )}\). The corresponding analytic reference class \(B_{\varvec{\beta }}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\) consists of all functions \(u:\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell \rightarrow \mathbb {R}\) such that \(u\in \widetilde{N}^k_{\varvec{\beta }}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\) for \(k>k_{\varvec{\beta }}\) and such that there is a constant \(d_u>0\) with
In the following, we restrict ourselves to the classes \(B_{-1-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\) and \(B_{-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\) for exponents \(\varvec{b}=\{b_{\varvec{c}},b_{\varvec{e}}\}\) in (0, 1) as in Remark 2.3. In the first case, we have \(k_{\varvec{\beta }}\in (1,2)\) and the norms on the right-hand in (8.27) are given by
Similarly, for the second analytic class \(B_{-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\), we have \(k_{\varvec{\beta }}\in (0,1)\) and the norms on the right-hand side of (8.27) take the form
In the axiparallel setting considered in the present paper, when functions \(u\in B_{-1-\varvec{b}}(\Omega )\) and \(u\in B_{-\varvec{b}}(\Omega )\) as in Theorem 4.3 are localized and scaled to \(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell \), they belong to the reference classes \(B_{-1 -\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\) and \(B_{-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\), respectively; cf. [20, Section 3.4].
To prove Theorem 4.3 in the reference corner–edge framework, it is now enough to bound the error contributions as in (3.34), (3.33) over \(\widetilde{\mathcal K}_\sigma ^{\ell ,{\varvec{c}}{\varvec{e}}}=\widetilde{\mathfrak O}^\ell _{{\varvec{c}}{\varvec{e}}}\mathop {\cup }\limits ^{.}\widetilde{\mathfrak T}^\ell _{\varvec{e}}\mathop {\cup }\limits ^{.}\widetilde{\mathfrak T}_{\varvec{c}}^\ell \).
Proposition 8.7
For \(b_{\varvec{c}},b_{\varvec{e}}\in (0,1)\), let \(u\in B_{-1-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\cap H^{1+\theta }(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\) for some \(\theta \in (0,1)\), and let \(\pi u=\pi _0^\perp \otimes \pi ^{\Vert } u\) be the base interpolant (4.4) over \(\widetilde{\mathcal K}_\sigma ^{\ell ,{\varvec{c}}{\varvec{e}}}\) for any conformity indices \(r_K\in \{0,1\}\). For the errors \(\eta \), \(\eta _0^\perp \), \(\eta ^{\Vert }\) in (4.7), we have
with \(b,C>0\) independent of \(\ell \).
Moreover, for \(b_{\varvec{c}},b_{\varvec{e}}\in (0,1)\), let \(u\in B_{-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\cap H^{\theta }(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\) for some \(\theta \in (0,1)\), and let \(\pi _0 u=\pi _0^\perp \otimes \pi ^{\Vert }_0 u\) be the \(L^2\)-projection over \(\widetilde{\mathcal K}_\sigma ^{\ell ,{\varvec{c}}{\varvec{e}}}\). For the errors \(\eta _0\), \(\eta ^\perp _0\), \(\eta _0^{\Vert }\) in (4.7), we have
with \(b,C>0\) independent of \(\ell \).
The remainder of this section is devoted to the proof of Proposition 8.7.
1.3 Proof of Proposition 8.7
We bound the errors in Proposition 8.7 separately for the set \(\widetilde{\mathfrak O}_{{\varvec{c}}{\varvec{e}}}^\ell \) (Propositions 8.9, 8.10), for \(\widetilde{\mathfrak T}_{\varvec{e}}^\ell \) (Propositions 8.11, 8.12), and for \(\widetilde{\mathfrak T}_{\varvec{c}}^\ell \) (Proposition 8.13).
1.3.1 Convergence on \(\widetilde{\mathfrak O}_{{\varvec{c}}{\varvec{e}}}^\ell \)
We begin our analysis by recalling essential scaling properties; see [24, Section 5.1.4].
Lemma 8.8
Let \(K=(0,h_K^\perp )^2\times (0,h_K^{\Vert })\) be of the form (3.3). Let \(v:K\rightarrow \mathbb {R}\), and \(\widehat{v}=v\circ \Phi _{K}^{-1}\). Then:
-
(i)
\(\Vert v\Vert ^2_{L^2(K)} \lesssim (h_K^\perp )^2 h_K^{\Vert }\Vert \widehat{v}\Vert ^2_{L^2(\widehat{K})}\).
-
(ii)
\((h_K^{\Vert })^{-2}\Vert v\Vert ^2_{L^2(K)}+\Vert \mathsf {D}_{\Vert } v\Vert ^2_{L^2(K)}\lesssim (h_K^\perp )^2 (h_K^{\Vert })^{-1}\big (\Vert \widehat{v}\Vert ^2_{L^2(\widehat{K})}+\Vert \widehat{\mathsf {D}}_{\Vert } \widehat{v}\Vert ^2_{L^2(\widehat{K})}\big )\).
-
(iii)
\((h_K^\perp )^{-2}\Vert v\Vert ^2_{L^2(K)}+\Vert \mathsf {D}_\perp v\Vert ^2_{L^2(K)}\lesssim h_K^{\Vert } \big ( \Vert \widehat{v}\Vert ^2_{L^2(\widehat{K})}+\Vert \widehat{\mathsf {D}}_\perp \widehat{v}\Vert ^2_{L^2(\widehat{K})}\big )\).
We bound \(\eta _0^\perp \) over \(\widetilde{\mathfrak O}_{{\varvec{c}}{\varvec{e}}}^\ell \) as follows.
Proposition 8.9
Let \(u\in B_{-1-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\), respectively \(u\in B_{-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\). Then there are constants \(b,C>0\) independent of \(\ell \) such that \(\Upsilon ^\perp _{\widetilde{\mathfrak O}_{{\varvec{c}}{\varvec{e}}}^\ell }[\eta _0^{\perp }]^2 \le C \exp (-2 b\ell )\), respectively \(\Vert \eta _0^\perp \Vert ^2_{L^2(\widetilde{\mathfrak O}_{{\varvec{c}}{\varvec{e}}}^\ell )}\le C \exp (-2 b\ell )\).
Proof
Let \(u \in B_{-1-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\). We consider an element \(K\in \widetilde{\mathfrak L}^{ij}_{{\varvec{c}}{\varvec{e}}}\) with \(2\le j\le \ell +1\) and \(2\le i \le j\); see (8.22). With Lemma 8.8 (observing that \(h_K^\perp \lesssim h_K^{\Vert }\)), the approximation results for \(\widehat{\eta }^\perp _0\) in Proposition 8.3 and (8.26), we conclude that
for \(1\le s_i^\perp \le p_i^\perp \), where \(E^\perp _{s_i^\perp }(K;u)\) is the expression in (8.7). Notice that here we exclude the choice \(s^\perp _i=0\) to ensure that \(|\varvec{\alpha }|\ge |\varvec{\alpha }^\perp |\ge 2\) in \(E^\perp _{s_i^\perp }(K;u)\). Thanks to the equivalences (8.24), we insert the appropriate weights as in (8.29) and obtain
for \(2\le s_i^\perp +1\le |\varvec{\alpha }^\perp |\le s_i^\perp +2\) and \(\alpha ^{\Vert }=0,1\). Hence, it follows that
The analytic regularity (8.28) then implies the existence of \(C>0\) such that
for all \(1\le s_i^\perp \le p_i^\perp \). Summing (8.31) over all layers in \(\widetilde{\mathfrak O}^\ell _{{\varvec{c}}{\varvec{e}}}\) in (8.22) in combination with (8.23) results in
By interpolating to real parameters \(s_i^\perp \in [1,p_i^\perp ]\) as in [24, Lemma 5.8], this sum is of the same form as \(S^\perp \) in the proof of [24, Proposition 5.17], and the assertion now follows from the arguments there and after adjusting the constants to absorb the algebraic loss in \(|\varvec{p}|\).
For \(u\in B_{-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\), we proceed similarly and note that
for \(1\le s_i^\perp \le p_i^\perp \). Hence, we obtain
The second bound follows as before. \(\square \)
Next, we establish the analog of Proposition 8.9 in edge-parallel direction.
Proposition 8.10
Let \(u\in B_{-1-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\), respectively \(u\in B_{-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\). Then there are constants \(b,C>0\) independent of \(\ell \) such that \( \Upsilon ^{\Vert }_{\widetilde{\mathfrak O}_{{\varvec{c}}{\varvec{e}}}^\ell }[\eta ^{\Vert }]^2 \le C \exp (-2 b\ell )\), respectively \(\Vert \eta _0^{\Vert }\Vert ^2_{L^2(\widetilde{\mathfrak O}_{{\varvec{c}}{\varvec{e}}}^\ell )} \le C \exp (-2 b\ell )\).
Proof
For \(u\in B_{-1-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\), we claim that
for any \(K\in \widetilde{\mathfrak O}^\ell _{{\varvec{c}}{\varvec{e}}}\) and \(1\le s_K^{\Vert }\le p_K^{\Vert }\). To prove (8.32), we start by employing Lemma 8.8 and the approximation property for \(\widehat{\eta }^{\Vert }\) in Proposition 8.3:
for any \(1 < s_K^{\Vert }\le p_K^{\Vert }\), where we again exclude the choice \(s_K^{\Vert }=0\) so that \(|\varvec{\alpha }|\ge 2\). We then insert suitable weights with the aid of (8.23), (8.29) to obtain
Hence,
By proceeding similarly, we find that, for \(|\varvec{\alpha }^\perp |=1\),
This establishes the bound (8.32).
For \(u\in B_{-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\), we use analogous arguments based on Lemma 8.8, Proposition 8.3 and (8.30). This results in
Next, we sum the bounds in (8.32), (8.33) over all layers of \(\widetilde{\mathfrak O}^{\ell }_{{\varvec{c}}{\varvec{e}}}\). By noticing (8.23), (8.26) and the analytic regularity (8.28), we conclude that
The terms in the sums above are independent of the inner index i. Interpolation to non-integer differentiation orders \(s_j^{\Vert }\in [1 , p_j^{\Vert }]\) as in [24, Lemma 5.8], applying [24, Lemma 5.12] and absorbing the algebraic loss in \(|\varvec{p}|\) complete the proof. \(\square \)
1.3.2 Convergence on \(\widetilde{\mathfrak T}^{\ell }_{{\varvec{e}}}\)
We first consider edge-perpendicular elements.
Proposition 8.11
Let \(u\in B_{-1-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\), respectively \(u\in B_{-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\). Then there are constants \(b,C>0\) independent of \(\ell \) such that \(\Upsilon ^\perp _{\widetilde{\mathfrak T}^\ell _{{\varvec{e}}}}[\eta _0^{\perp }]^2\le C\exp (-2b\ell )\), respectively \(\Vert \eta _0^\perp \Vert ^2_{L^2(\widetilde{\mathfrak T}^\ell _{{\varvec{e}}})}\le C\exp (-2b\ell )\).
Proof
Let \(K=K_j=K^\perp \times K_j^{\Vert }\) be an element of \(\widetilde{\mathfrak T}_{\varvec{e}}^\ell \) as in (8.25). We claim that
To show (8.34), let \(s=|\varvec{\alpha }^\perp |=0,1\). Applying the bound (8.10) with \(\beta =1-b_{\varvec{e}}\) (noting that \(p_K^\perp =\max \{1,\mathfrak {s}\}\) by (8.26)) and from (8.24), (8.29), we see that
where \(|\mathsf {D}^2_\perp v|^2=\sum _{|\varvec{\alpha }^\perp |=2}|\mathsf {D}^{\varvec{\alpha }^\perp }_\perp v|^2\). Thus, combining these estimates and expressing the mesh sizes in terms of \(\sigma \), see (8.24), we have
which yields (8.34). To prove (8.35), we similarly conclude that
For the bound (8.36), we employ an analogous argument based on (8.9) (with \(\beta =1-b_{\varvec{e}}\)). Indeed, with (8.24) and (8.30), we conclude that
which is (8.36).
The assertions now follow by summing the estimates (8.34), (8.35) and (8.36) over all elements \(K\in \widetilde{\mathfrak T}_{\varvec{e}}^\ell \) (i.e., over \(2 \le j\le \ell +1\)) and by suitably adjusting the constants. \(\square \)
A similar estimate holds for the approximation errors in direction parallel to \({\varvec{e}}\).
Proposition 8.12
Let \(u\in B_{-1-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\), respectively \(u\in B_{-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\). Then there are constants \(b,C>0\) independent of \(\ell \) such that \(\Upsilon ^{\Vert }_{\widetilde{\mathfrak T}^\ell _{{\varvec{e}}}}[\eta ^{\Vert }]^2\le C \exp (-2b\ell )\), respectively \(\Vert \eta _0^{\Vert }\Vert ^2_{L^2(\widetilde{\mathfrak T}^\ell _{{\varvec{e}}})}\le C \exp (-2b\ell )\).
Proof
For \(u\in B_{-1-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\) and \(|\varvec{\alpha }^\perp |=1\), we have \(u, \mathsf {D}_{\Vert } u, \mathsf {D}^{\varvec{\alpha }^\perp }_\perp u\in L^2(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^{\ell })\) due to (8.28), and there holds
Similarly, for \(u\in B_{-\varvec{b}}(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\) it follows with (8.30) that \(u\in L^2(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\) and
In view of (8.24), (8.25), these properties correspond to the one-dimensional analytic regularity assumptions considered in (8.16) and (8.20), respectively. Moreover, due to (8.26), the polynomial degrees \(p_K^{\Vert }\) along the edge \({\varvec{e}}\) are \(\mathfrak {s}\)-linearly increasing away from the corner \({\varvec{c}}\). Hence, Lemma 8.5, respectively Lemma 8.6 along with the tensor-product structure of the elements yield the assertions. \(\square \)
1.3.3 Convergence on \(\widetilde{\mathfrak T}_{\varvec{c}}^\ell \)
It remains to show exponential convergence on \(\widetilde{\mathfrak T}_{\varvec{c}}^\ell \).
Proposition 8.13
Let \(u\in H^{1+\theta }(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\), respectively \(u\in H^{\theta }(\widetilde{\Omega }_{{\varvec{c}}{\varvec{e}}}^\ell )\) for some \(\theta \in (0,1)\). Then there exist constants \(b,C>0\) independent of \(\ell \) such that \(\Upsilon ^{\Vert }_{\widetilde{\mathfrak T}_{\varvec{c}}^\ell }[\eta ]^2\le C\exp (-2b\ell )\), respectively \(\Vert \eta _0\Vert ^2_{L^2(\widetilde{\mathfrak T}_{\varvec{c}}^\ell )}\le C\exp (-2b\ell )\).
Proof
The element \(K \in \widetilde{\mathfrak T}^\ell _{\varvec{c}}\) is isotropic with \(h_K\simeq \sigma ^\ell \); cf. (8.22). Standard h-version approximation properties then show that \(N^{\Vert }_{K}[\eta ]^2 \lesssim h_K^{2\theta } \Vert u\Vert ^2_{H^{1+\theta }(K)}\), respectively \(\Vert \eta _0\Vert ^2_{L^2(K)} \lesssim h_K^{2\theta } \Vert u\Vert ^2_{H^{\theta }(K)}\). \(\square \)
Rights and permissions
About this article
Cite this article
Schötzau, D., Schwab, C. Exponential Convergence of hp-FEM for Elliptic Problems in Polyhedra: Mixed Boundary Conditions and Anisotropic Polynomial Degrees. Found Comput Math 18, 595–660 (2018). https://doi.org/10.1007/s10208-017-9349-9
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10208-017-9349-9
Keywords
- hp-FEM
- Second-order elliptic problems in polyhedra
- Mixed boundary conditions
- Anisotropic polynomial degrees
- Exponential convergence